d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
f2b2f31707
freebsd-dev/sys/dev/hifn/hifn7751.c

2744 lines
72 KiB
C
Raw Normal View History

Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/* $OpenBSD: hifn7751.c,v 1.120 2002/05/17 00:33:34 deraadt Exp $ */
Start each of the license/copyright comments with /*-, minor shuffle of lines
2005-01-06 01:43:34 +00:00
/*-
sys/dev: further adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 2-Clause license, however the tool I was using misidentified many licenses so this was mostly a manual - error prone - task. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts.
2017-11-27 14:52:40 +00:00
* SPDX-License-Identifier: BSD-3-Clause
*
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
* Invertex AEON / Hifn 7751 driver
* Copyright (c) 1999 Invertex Inc. All rights reserved.
* Copyright (c) 1999 Theo de Raadt
* Copyright (c) 2000-2001 Network Security Technologies, Inc.
* http://www.netsec.net
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
* Copyright (c) 2003 Hifn Inc.
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
*
* This driver is based on a previous driver by Invertex, for which they
* requested: Please send any comments, feedback, bug-fixes, or feature
* requests to software@invertex.com.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Effort sponsored in part by the Defense Advanced Research Projects
* Agency (DARPA) and Air Force Research Laboratory, Air Force
* Materiel Command, USAF, under agreement number F30602-01-2-0537.
*/
Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
* Driver for various Hifn encryption processors.
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
*/
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
#include "opt_hifn.h"
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/errno.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
Add missing <sys/module.h> includes
2004-05-30 20:08:47 +00:00
#include <sys/module.h>
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#include <sys/mbuf.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/sysctl.h>
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
#include <sys/uio.h>
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#include <vm/vm.h>
#include <vm/pmap.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <sys/bus.h>
#include <sys/rman.h>
#include <opencrypto/cryptodev.h>
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
#include <opencrypto/xform_auth.h>
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#include <sys/random.h>
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
#include <sys/kobj.h>
#include "cryptodev_if.h"
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Prefer new location of pci include files (which have only been in the tree for two or more years now), except in a few places where there's code to be compatible with older versions of FreeBSD.
2003-08-22 06:00:27 +00:00
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
#ifdef HIFN_RNDTEST
#include <dev/rndtest/rndtest.h>
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#include <dev/hifn/hifn7751reg.h>
#include <dev/hifn/hifn7751var.h>
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
#ifdef HIFN_VULCANDEV
#include <sys/conf.h>
#include <sys/uio.h>
static struct cdevsw vulcanpk_cdevsw; /* forward declaration */
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* Prototypes and count for the pci_device structure
*/
static int hifn_probe(device_t);
static int hifn_attach(device_t);
static int hifn_detach(device_t);
static int hifn_suspend(device_t);
static int hifn_resume(device_t);
shutdown returns an int
2009-02-05 19:37:49 +00:00
static int hifn_shutdown(device_t);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
static int hifn_probesession(device_t, const struct crypto_session_params *);
static int hifn_newsession(device_t, crypto_session_t,
const struct crypto_session_params *);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
static int hifn_process(device_t, struct cryptop *, int);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
static device_method_t hifn_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, hifn_probe),
DEVMETHOD(device_attach, hifn_attach),
DEVMETHOD(device_detach, hifn_detach),
DEVMETHOD(device_suspend, hifn_suspend),
DEVMETHOD(device_resume, hifn_resume),
DEVMETHOD(device_shutdown, hifn_shutdown),
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
/* crypto device methods */
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
DEVMETHOD(cryptodev_probesession, hifn_probesession),
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
DEVMETHOD(cryptodev_newsession, hifn_newsession),
DEVMETHOD(cryptodev_process, hifn_process),
- There's no need to overwrite the default device method with the default one. Interestingly, these are actually the default for quite some time (bus_generic_driver_added(9) since r52045 and bus_generic_print_child(9) since r52045) but even recently added device drivers do this unnecessarily. Discussed with: jhb, marcel - While at it, use DEVMETHOD_END. Discussed with: jhb - Also while at it, use __FBSDID.
2011-11-22 21:28:20 +00:00
DEVMETHOD_END
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
};
static driver_t hifn_driver = {
"hifn",
hifn_methods,
sizeof (struct hifn_softc)
};
static devclass_t hifn_devclass;
DRIVER_MODULE(hifn, pci, hifn_driver, hifn_devclass, 0, 0);
Module-ize the 'core' crypto stuff. This may still need to be compiled into the kernel by default (if required), but other modules can now depend() on this. Fix inter-module dependancy. Earlier version OK'ed by: sam
2002-10-16 14:31:34 +00:00
MODULE_DEPEND(hifn, crypto, 1, 1, 1);
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
#ifdef HIFN_RNDTEST
MODULE_DEPEND(hifn, rndtest, 1, 1, 1);
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
static void hifn_reset_board(struct hifn_softc *, int);
static void hifn_reset_puc(struct hifn_softc *);
static void hifn_puc_wait(struct hifn_softc *);
static int hifn_enable_crypto(struct hifn_softc *);
static void hifn_set_retry(struct hifn_softc *sc);
static void hifn_init_dma(struct hifn_softc *);
static void hifn_init_pci_registers(struct hifn_softc *);
static int hifn_sramsize(struct hifn_softc *);
static int hifn_dramsize(struct hifn_softc *);
static int hifn_ramtype(struct hifn_softc *);
static void hifn_sessions(struct hifn_softc *);
static void hifn_intr(void *);
static u_int hifn_write_command(struct hifn_command *, u_int8_t *);
static u_int32_t hifn_next_signature(u_int32_t a, u_int cnt);
static void hifn_callback(struct hifn_softc *, struct hifn_command *, u_int8_t *);
static int hifn_crypto(struct hifn_softc *, struct hifn_command *, struct cryptop *, int);
static int hifn_readramaddr(struct hifn_softc *, int, u_int8_t *);
static int hifn_writeramaddr(struct hifn_softc *, int, u_int8_t *);
static int hifn_dmamap_load_src(struct hifn_softc *, struct hifn_command *);
static int hifn_dmamap_load_dst(struct hifn_softc *, struct hifn_command *);
static int hifn_init_pubrng(struct hifn_softc *);
static void hifn_rng(void *);
static void hifn_tick(void *);
static void hifn_abort(struct hifn_softc *);
static void hifn_alloc_slot(struct hifn_softc *, int *, int *, int *, int *);
static void hifn_write_reg_0(struct hifn_softc *, bus_size_t, u_int32_t);
static void hifn_write_reg_1(struct hifn_softc *, bus_size_t, u_int32_t);
Consistently use __inline instead of __inline__ as the former is an empty macro in <sys/cdefs.h> for compilers without support for inline.
2004-07-04 16:11:03 +00:00
static __inline u_int32_t
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
READ_REG_0(struct hifn_softc *sc, bus_size_t reg)
{
u_int32_t v = bus_space_read_4(sc->sc_st0, sc->sc_sh0, reg);
sc->sc_bar0_lastreg = (bus_size_t) -1;
return (v);
}
#define WRITE_REG_0(sc, reg, val) hifn_write_reg_0(sc, reg, val)
Consistently use __inline instead of __inline__ as the former is an empty macro in <sys/cdefs.h> for compilers without support for inline.
2004-07-04 16:11:03 +00:00
static __inline u_int32_t
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
READ_REG_1(struct hifn_softc *sc, bus_size_t reg)
{
u_int32_t v = bus_space_read_4(sc->sc_st1, sc->sc_sh1, reg);
sc->sc_bar1_lastreg = (bus_size_t) -1;
return (v);
}
#define WRITE_REG_1(sc, reg, val) hifn_write_reg_1(sc, reg, val)
Mark more nodes as CTLFLAG_MPSAFE or CTLFLAG_NEEDGIANT (17 of many) r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are still not MPSAFE (or already are but aren’t properly marked). Use it in preparation for a general review of all nodes. This is non-functional change that adds annotations to SYSCTL_NODE and SYSCTL_PROC nodes using one of the soon-to-be-required flags. Mark all obvious cases as MPSAFE. All entries that haven't been marked as MPSAFE before are by default marked as NEEDGIANT Approved by: kib (mentor, blanket) Commented by: kib, gallatin, melifaro Differential Revision: https://reviews.freebsd.org/D23718
2020-02-26 14:26:36 +00:00
static SYSCTL_NODE(_hw, OID_AUTO, hifn, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"Hifn driver parameters");
move sysctl's under hw.hifn; change max batching default to 1 (no batching) based on what I learned from the Broadcom h/w
2003-01-20 21:11:56 +00:00
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#ifdef HIFN_DEBUG
static int hifn_debug = 0;
move sysctl's under hw.hifn; change max batching default to 1 (no batching) based on what I learned from the Broadcom h/w
2003-01-20 21:11:56 +00:00
SYSCTL_INT(_hw_hifn, OID_AUTO, debug, CTLFLAG_RW, &hifn_debug,
0, "control debugging msgs");
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#endif
static struct hifn_stats hifnstats;
move sysctl's under hw.hifn; change max batching default to 1 (no batching) based on what I learned from the Broadcom h/w
2003-01-20 21:11:56 +00:00
SYSCTL_STRUCT(_hw_hifn, OID_AUTO, stats, CTLFLAG_RD, &hifnstats,
hifn_stats, "driver statistics");
gack; revert previous commit; not only did I read things backward but I was in the wrong tree
2003-03-11 22:23:19 +00:00
static int hifn_maxbatch = 1;
move sysctl's under hw.hifn; change max batching default to 1 (no batching) based on what I learned from the Broadcom h/w
2003-01-20 21:11:56 +00:00
SYSCTL_INT(_hw_hifn, OID_AUTO, maxbatch, CTLFLAG_RW, &hifn_maxbatch,
0, "max ops to batch w/o interrupt");
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* Probe for a supported device. The PCI vendor and device
* IDs are used to detect devices we know how to handle.
*/
static int
hifn_probe(device_t dev)
{
if (pci_get_vendor(dev) == PCI_VENDOR_INVERTEX &&
pci_get_device(dev) == PCI_PRODUCT_INVERTEX_AEON)
Use BUS_PROBE_DEFAULT for pci probe return value
2005-03-05 18:30:12 +00:00
return (BUS_PROBE_DEFAULT);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (pci_get_vendor(dev) == PCI_VENDOR_HIFN &&
(pci_get_device(dev) == PCI_PRODUCT_HIFN_7751 ||
pci_get_device(dev) == PCI_PRODUCT_HIFN_7951 ||
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
pci_get_device(dev) == PCI_PRODUCT_HIFN_7955 ||
pci_get_device(dev) == PCI_PRODUCT_HIFN_7956 ||
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
pci_get_device(dev) == PCI_PRODUCT_HIFN_7811))
Use BUS_PROBE_DEFAULT for pci probe return value
2005-03-05 18:30:12 +00:00
return (BUS_PROBE_DEFAULT);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (pci_get_vendor(dev) == PCI_VENDOR_NETSEC &&
pci_get_device(dev) == PCI_PRODUCT_NETSEC_7751)
Use BUS_PROBE_DEFAULT for pci probe return value
2005-03-05 18:30:12 +00:00
return (BUS_PROBE_DEFAULT);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (ENXIO);
}
static void
hifn_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
bus_addr_t *paddr = (bus_addr_t*) arg;
*paddr = segs->ds_addr;
}
static const char*
hifn_partname(struct hifn_softc *sc)
{
/* XXX sprintf numbers when not decoded */
switch (pci_get_vendor(sc->sc_dev)) {
case PCI_VENDOR_HIFN:
switch (pci_get_device(sc->sc_dev)) {
case PCI_PRODUCT_HIFN_6500: return "Hifn 6500";
case PCI_PRODUCT_HIFN_7751: return "Hifn 7751";
case PCI_PRODUCT_HIFN_7811: return "Hifn 7811";
case PCI_PRODUCT_HIFN_7951: return "Hifn 7951";
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
case PCI_PRODUCT_HIFN_7955: return "Hifn 7955";
case PCI_PRODUCT_HIFN_7956: return "Hifn 7956";
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
return "Hifn unknown-part";
case PCI_VENDOR_INVERTEX:
switch (pci_get_device(sc->sc_dev)) {
case PCI_PRODUCT_INVERTEX_AEON: return "Invertex AEON";
}
return "Invertex unknown-part";
case PCI_VENDOR_NETSEC:
switch (pci_get_device(sc->sc_dev)) {
case PCI_PRODUCT_NETSEC_7751: return "NetSec 7751";
}
return "NetSec unknown-part";
}
return "Unknown-vendor unknown-part";
}
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
static void
default_harvest(struct rndtest_state *rsp, void *buf, u_int count)
{
Huge cleanup of random(4) code. * GENERAL - Update copyright. - Make kernel options for RANDOM_YARROW and RANDOM_DUMMY. Set neither to ON, which means we want Fortuna - If there is no 'device random' in the kernel, there will be NO random(4) device in the kernel, and the KERN_ARND sysctl will return nothing. With RANDOM_DUMMY there will be a random(4) that always blocks. - Repair kern.arandom (KERN_ARND sysctl). The old version went through arc4random(9) and was a bit weird. - Adjust arc4random stirring a bit - the existing code looks a little suspect. - Fix the nasty pre- and post-read overloading by providing explictit functions to do these tasks. - Redo read_random(9) so as to duplicate random(4)'s read internals. This makes it a first-class citizen rather than a hack. - Move stuff out of locked regions when it does not need to be there. - Trim RANDOM_DEBUG printfs. Some are excess to requirement, some behind boot verbose. - Use SYSINIT to sequence the startup. - Fix init/deinit sysctl stuff. - Make relevant sysctls also tunables. - Add different harvesting "styles" to allow for different requirements (direct, queue, fast). - Add harvesting of FFS atime events. This needs to be checked for weighing down the FS code. - Add harvesting of slab allocator events. This needs to be checked for weighing down the allocator code. - Fix the random(9) manpage. - Loadable modules are not present for now. These will be re-engineered when the dust settles. - Use macros for locks. - Fix comments. * src/share/man/... - Update the man pages. * src/etc/... - The startup/shutdown work is done in D2924. * src/UPDATING - Add UPDATING announcement. * src/sys/dev/random/build.sh - Add copyright. - Add libz for unit tests. * src/sys/dev/random/dummy.c - Remove; no longer needed. Functionality incorporated into randomdev.*. * live_entropy_sources.c live_entropy_sources.h - Remove; content moved. - move content to randomdev.[ch] and optimise. * src/sys/dev/random/random_adaptors.c src/sys/dev/random/random_adaptors.h - Remove; plugability is no longer used. Compile-time algorithm selection is the way to go. * src/sys/dev/random/random_harvestq.c src/sys/dev/random/random_harvestq.h - Add early (re)boot-time randomness caching. * src/sys/dev/random/randomdev_soft.c src/sys/dev/random/randomdev_soft.h - Remove; no longer needed. * src/sys/dev/random/uint128.h - Provide a fake uint128_t; if a real one ever arrived, we can use that instead. All that is needed here is N=0, N++, N==0, and some localised trickery is used to manufacture a 128-bit 0ULLL. * src/sys/dev/random/unit_test.c src/sys/dev/random/unit_test.h - Improve unit tests; previously the testing human needed clairvoyance; now the test will do a basic check of compressibility. Clairvoyant talent is still a good idea. - This is still a long way off a proper unit test. * src/sys/dev/random/fortuna.c src/sys/dev/random/fortuna.h - Improve messy union to just uint128_t. - Remove unneeded 'static struct fortuna_start_cache'. - Tighten up up arithmetic. - Provide a method to allow eternal junk to be introduced; harden it against blatant by compress/hashing. - Assert that locks are held correctly. - Fix the nasty pre- and post-read overloading by providing explictit functions to do these tasks. - Turn into self-sufficient module (no longer requires randomdev_soft.[ch]) * src/sys/dev/random/yarrow.c src/sys/dev/random/yarrow.h - Improve messy union to just uint128_t. - Remove unneeded 'staic struct start_cache'. - Tighten up up arithmetic. - Provide a method to allow eternal junk to be introduced; harden it against blatant by compress/hashing. - Assert that locks are held correctly. - Fix the nasty pre- and post-read overloading by providing explictit functions to do these tasks. - Turn into self-sufficient module (no longer requires randomdev_soft.[ch]) - Fix some magic numbers elsewhere used as FAST and SLOW. Differential Revision: https://reviews.freebsd.org/D2025 Reviewed by: vsevolod,delphij,rwatson,trasz,jmg Approved by: so (delphij)
2015-06-30 17:00:45 +00:00
/* MarkM: FIX!! Check that this does not swamp the harvester! */
Remove the Yarrow PRNG algorithm option in accordance with due notice given in random(4). This includes updating of the relevant man pages, and no-longer-used harvesting parameters. Ensure that the pseudo-unit-test still does something useful, now also with the "other" algorithm instead of Yarrow. PR: 230870 Reviewed by: cem Approved by: so(delphij,gtetlow) Approved by: re(marius) Differential Revision: https://reviews.freebsd.org/D16898
2018-08-26 12:51:46 +00:00
random_harvest_queue(buf, count, RANDOM_PURE_HIFN);
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
}
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
static u_int
checkmaxmin(device_t dev, const char *what, u_int v, u_int min, u_int max)
{
if (v > max) {
device_printf(dev, "Warning, %s %u out of range, "
"using max %u\n", what, v, max);
v = max;
} else if (v < min) {
device_printf(dev, "Warning, %s %u out of range, "
"using min %u\n", what, v, min);
v = min;
}
return v;
}
/*
* Select PLL configuration for 795x parts. This is complicated in
* that we cannot determine the optimal parameters without user input.
* The reference clock is derived from an external clock through a
* multiplier. The external clock is either the host bus (i.e. PCI)
* or an external clock generator. When using the PCI bus we assume
* the clock is either 33 or 66 MHz; for an external source we cannot
* tell the speed.
*
* PLL configuration is done with a string: "pci" for PCI bus, or "ext"
* for an external source, followed by the frequency. We calculate
* the appropriate multiplier and PLL register contents accordingly.
* When no configuration is given we default to "pci66" since that
* always will allow the card to work. If a card is using the PCI
* bus clock and in a 33MHz slot then it will be operating at half
* speed until the correct information is provided.
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
*
* We use a default setting of "ext66" because according to Mike Ham
* of HiFn, almost every board in existence has an external crystal
* populated at 66Mhz. Using PCI can be a problem on modern motherboards,
* because PCI33 can have clocks from 0 to 33Mhz, and some have
* non-PCI-compliant spread-spectrum clocks, which can confuse the pll.
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
*/
static void
hifn_getpllconfig(device_t dev, u_int *pll)
{
const char *pllspec;
u_int freq, mul, fl, fh;
u_int32_t pllconfig;
char *nxt;
if (resource_string_value("hifn", device_get_unit(dev),
"pllconfig", &pllspec))
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
pllspec = "ext66";
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
fl = 33, fh = 66;
pllconfig = 0;
if (strncmp(pllspec, "ext", 3) == 0) {
pllspec += 3;
pllconfig |= HIFN_PLL_REF_SEL;
switch (pci_get_device(dev)) {
case PCI_PRODUCT_HIFN_7955:
case PCI_PRODUCT_HIFN_7956:
fl = 20, fh = 100;
break;
#ifdef notyet
case PCI_PRODUCT_HIFN_7954:
fl = 20, fh = 66;
break;
#endif
}
} else if (strncmp(pllspec, "pci", 3) == 0)
pllspec += 3;
freq = strtoul(pllspec, &nxt, 10);
if (nxt == pllspec)
freq = 66;
else
freq = checkmaxmin(dev, "frequency", freq, fl, fh);
/*
* Calculate multiplier. We target a Fck of 266 MHz,
* allowing only even values, possibly rounded down.
* Multipliers > 8 must set the charge pump current.
*/
mul = checkmaxmin(dev, "PLL divisor", (266 / freq) &~ 1, 2, 12);
pllconfig |= (mul / 2 - 1) << HIFN_PLL_ND_SHIFT;
if (mul > 8)
pllconfig |= HIFN_PLL_IS;
*pll = pllconfig;
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* Attach an interface that successfully probed.
*/
static int
hifn_attach(device_t dev)
{
struct hifn_softc *sc = device_get_softc(dev);
caddr_t kva;
int rseg, rid;
char rbase;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
uint16_t rev;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_dev = dev;
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
mtx_init(&sc->sc_mtx, device_get_nameunit(dev), "hifn driver", MTX_DEF);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/* XXX handle power management */
/*
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
* The 7951 and 795x have a random number generator and
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
* public key support; note this.
*/
if (pci_get_vendor(dev) == PCI_VENDOR_HIFN &&
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
(pci_get_device(dev) == PCI_PRODUCT_HIFN_7951 ||
pci_get_device(dev) == PCI_PRODUCT_HIFN_7955 ||
pci_get_device(dev) == PCI_PRODUCT_HIFN_7956))
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_flags = HIFN_HAS_RNG | HIFN_HAS_PUBLIC;
/*
* The 7811 has a random number generator and
* we also note it's identity 'cuz of some quirks.
*/
if (pci_get_vendor(dev) == PCI_VENDOR_HIFN &&
pci_get_device(dev) == PCI_PRODUCT_HIFN_7811)
sc->sc_flags |= HIFN_IS_7811 | HIFN_HAS_RNG;
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
/*
* The 795x parts support AES.
*/
if (pci_get_vendor(dev) == PCI_VENDOR_HIFN &&
(pci_get_device(dev) == PCI_PRODUCT_HIFN_7955 ||
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
pci_get_device(dev) == PCI_PRODUCT_HIFN_7956)) {
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
sc->sc_flags |= HIFN_IS_7956 | HIFN_HAS_AES;
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
/*
* Select PLL configuration. This depends on the
* bus and board design and must be manually configured
* if the default setting is unacceptable.
*/
hifn_getpllconfig(dev, &sc->sc_pllconfig);
}
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* Setup PCI resources. Note that we record the bus
* tag and handle for each register mapping, this is
* used by the READ_REG_0, WRITE_REG_0, READ_REG_1,
* and WRITE_REG_1 macros throughout the driver.
*/
No need to zero the softc. It's allocated with M_ZERO. Use pci_enable_busmaster instead of setting PCIM_CMD_BUSMASTEREN directly. There's no need to set PCIM_CMD_MEMEN. The bit is set when a SYS_RES_MEMORY resource is activated. Remove redundant pci_* function calls from suspend/resume methods. The bus driver already saves and restores the PCI configuration. Write 1 byte instead of 4 when setting the HIFN_TRDY_TIMEOUT register. It is only 1 byte according to the specification. Reviewed by: jhb Approved by: kib (mentor)
2010-12-18 14:24:24 +00:00
pci_enable_busmaster(dev);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
rid = HIFN_BAR0;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
sc->sc_bar0res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
RF_ACTIVE);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (sc->sc_bar0res == NULL) {
device_printf(dev, "cannot map bar%d register space\n", 0);
goto fail_pci;
}
sc->sc_st0 = rman_get_bustag(sc->sc_bar0res);
sc->sc_sh0 = rman_get_bushandle(sc->sc_bar0res);
sc->sc_bar0_lastreg = (bus_size_t) -1;
rid = HIFN_BAR1;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
sc->sc_bar1res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
RF_ACTIVE);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (sc->sc_bar1res == NULL) {
device_printf(dev, "cannot map bar%d register space\n", 1);
goto fail_io0;
}
sc->sc_st1 = rman_get_bustag(sc->sc_bar1res);
sc->sc_sh1 = rman_get_bushandle(sc->sc_bar1res);
sc->sc_bar1_lastreg = (bus_size_t) -1;
hifn_set_retry(sc);
/*
* Setup the area where the Hifn DMA's descriptors
* and associated data structures.
*/
Convert a number of drivers to obtaining their parent DMA tag from their PCI device attachment.
2012-03-12 08:03:51 +00:00
if (bus_dma_tag_create(bus_get_dma_tag(dev), /* PCI parent */
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
1, 0, /* alignment,boundary */
BUS_SPACE_MAXADDR_32BIT, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
HIFN_MAX_DMALEN, /* maxsize */
MAX_SCATTER, /* nsegments */
HIFN_MAX_SEGLEN, /* maxsegsize */
BUS_DMA_ALLOCNOW, /* flags */
Mega busdma API commit. Add two new arguments to bus_dma_tag_create(): lockfunc and lockfuncarg. Lockfunc allows a driver to provide a function for managing its locking semantics while using busdma. At the moment, this is used for the asynchronous busdma_swi and callback mechanism. Two lockfunc implementations are provided: busdma_lock_mutex() performs standard mutex operations on the mutex that is specified from lockfuncarg. dftl_lock() is a panic implementation and is defaulted to when NULL, NULL are passed to bus_dma_tag_create(). The only time that NULL, NULL should ever be used is when the driver ensures that bus_dmamap_load() will not be deferred. Drivers that do not provide their own locking can pass busdma_lock_mutex,&Giant args in order to preserve the former behaviour. sparc64 and powerpc do not provide real busdma_swi functions, so this is largely a noop on those platforms. The busdma_swi on is64 is not properly locked yet, so warnings will be emitted on this platform when busdma callback deferrals happen. If anyone gets panics or warnings from dflt_lock() being called, please let me know right away. Reviewed by: tmm, gibbs
2003-07-01 15:52:06 +00:00
NULL, /* lockfunc */
NULL, /* lockarg */
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
&sc->sc_dmat)) {
device_printf(dev, "cannot allocate DMA tag\n");
goto fail_io1;
}
if (bus_dmamap_create(sc->sc_dmat, BUS_DMA_NOWAIT, &sc->sc_dmamap)) {
device_printf(dev, "cannot create dma map\n");
bus_dma_tag_destroy(sc->sc_dmat);
goto fail_io1;
}
if (bus_dmamem_alloc(sc->sc_dmat, (void**) &kva, BUS_DMA_NOWAIT, &sc->sc_dmamap)) {
device_printf(dev, "cannot alloc dma buffer\n");
bus_dmamap_destroy(sc->sc_dmat, sc->sc_dmamap);
bus_dma_tag_destroy(sc->sc_dmat);
goto fail_io1;
}
if (bus_dmamap_load(sc->sc_dmat, sc->sc_dmamap, kva,
sizeof (*sc->sc_dma),
hifn_dmamap_cb, &sc->sc_dma_physaddr,
BUS_DMA_NOWAIT)) {
device_printf(dev, "cannot load dma map\n");
bus_dmamem_free(sc->sc_dmat, kva, sc->sc_dmamap);
bus_dma_tag_destroy(sc->sc_dmat);
goto fail_io1;
}
sc->sc_dma = (struct hifn_dma *)kva;
bzero(sc->sc_dma, sizeof(*sc->sc_dma));
NULL fallout
2003-12-25 01:23:38 +00:00
KASSERT(sc->sc_st0 != 0, ("hifn_attach: null bar0 tag!"));
KASSERT(sc->sc_sh0 != 0, ("hifn_attach: null bar0 handle!"));
KASSERT(sc->sc_st1 != 0, ("hifn_attach: null bar1 tag!"));
KASSERT(sc->sc_sh1 != 0, ("hifn_attach: null bar1 handle!"));
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* Reset the board and do the ``secret handshake''
* to enable the crypto support. Then complete the
* initialization procedure by setting up the interrupt
* and hooking in to the system crypto support so we'll
* get used for system services like the crypto device,
* IPsec, RNG device, etc.
*/
hifn_reset_board(sc, 0);
if (hifn_enable_crypto(sc) != 0) {
device_printf(dev, "crypto enabling failed\n");
goto fail_mem;
}
hifn_reset_puc(sc);
hifn_init_dma(sc);
hifn_init_pci_registers(sc);
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
/* XXX can't dynamically determine ram type for 795x; force dram */
if (sc->sc_flags & HIFN_IS_7956)
sc->sc_drammodel = 1;
else if (hifn_ramtype(sc))
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
goto fail_mem;
if (sc->sc_drammodel == 0)
hifn_sramsize(sc);
else
hifn_dramsize(sc);
/*
* Workaround for NetSec 7751 rev A: half ram size because two
* of the address lines were left floating
*/
if (pci_get_vendor(dev) == PCI_VENDOR_NETSEC &&
pci_get_device(dev) == PCI_PRODUCT_NETSEC_7751 &&
pci_get_revid(dev) == 0x61) /*XXX???*/
sc->sc_ramsize >>= 1;
/*
* Arrange the interrupt line.
*/
rid = 0;
Convert callers to the new bus_alloc_resource_any(9) API. Submitted by: Mark Santcroos <marks@ripe.net> Reviewed by: imp, dfr, bde
2004-03-17 17:50:55 +00:00
sc->sc_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
RF_SHAREABLE|RF_ACTIVE);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (sc->sc_irq == NULL) {
device_printf(dev, "could not map interrupt\n");
goto fail_mem;
}
/*
* NB: Network code assumes we are blocked with splimp()
* so make sure the IRQ is marked appropriately.
*/
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
if (bus_setup_intr(dev, sc->sc_irq, INTR_TYPE_NET | INTR_MPSAFE,
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +00:00
NULL, hifn_intr, sc, &sc->sc_intrhand)) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
device_printf(dev, "could not setup interrupt\n");
goto fail_intr2;
}
hifn_sessions(sc);
/*
* NB: Keep only the low 16 bits; this masks the chip id
* from the 7951.
*/
rev = READ_REG_1(sc, HIFN_1_REVID) & 0xffff;
rseg = sc->sc_ramsize / 1024;
rbase = 'K';
if (sc->sc_ramsize >= (1024 * 1024)) {
rbase = 'M';
rseg /= 1024;
}
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
device_printf(sc->sc_dev, "%s, rev %u, %d%cB %cram",
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifn_partname(sc), rev,
Move session state from on-card memory to host memory so we no longer are constrained to a small number of sessions by the small on-card memories found in newer devices. This is really a stopgap solution as having session state in main memory incurs a (small but noticeable) performance penalty. The better solution is to manage session state so that it's cached on chip. Obtained from: openbsd
2004-10-15 00:36:07 +00:00
rseg, rbase, sc->sc_drammodel ? 'd' : 's');
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
if (sc->sc_flags & HIFN_IS_7956)
printf(", pll=0x%x<%s clk, %ux mult>",
sc->sc_pllconfig,
sc->sc_pllconfig & HIFN_PLL_REF_SEL ? "ext" : "pci",
2 + 2*((sc->sc_pllconfig & HIFN_PLL_ND) >> 11));
printf("\n");
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
WRITE_REG_0(sc, HIFN_0_PUCNFG,
READ_REG_0(sc, HIFN_0_PUCNFG) | HIFN_PUCNFG_CHIPID);
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
sc->sc_ena = READ_REG_0(sc, HIFN_0_PUSTAT) & HIFN_PUSTAT_CHIPENA;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (sc->sc_ena) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
case HIFN_PUSTAT_ENA_2:
case HIFN_PUSTAT_ENA_1:
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
sc->sc_cid = crypto_get_driverid(dev,
sizeof(struct hifn_session), CRYPTOCAP_F_HARDWARE);
if (sc->sc_cid < 0) {
device_printf(dev, "could not get crypto driver id\n");
goto fail_intr;
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
break;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
if (sc->sc_flags & (HIFN_HAS_PUBLIC | HIFN_HAS_RNG))
hifn_init_pubrng(sc);
CALLOUT_MPSAFE has lost its meaning since r141428, i.e., for more than ten years for head. However, it is continuously misused as the mpsafe argument for callout_init(9). Deprecate the flag and clean up callout_init() calls to make them more consistent. Differential Revision: https://reviews.freebsd.org/D2613 Reviewed by: jhb MFC after: 2 weeks
2015-05-22 17:05:21 +00:00
callout_init(&sc->sc_tickto, 1);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
callout_reset(&sc->sc_tickto, hz, hifn_tick, sc);
return (0);
fail_intr:
bus_teardown_intr(dev, sc->sc_irq, sc->sc_intrhand);
fail_intr2:
/* XXX don't store rid */
bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
fail_mem:
bus_dmamap_unload(sc->sc_dmat, sc->sc_dmamap);
bus_dmamem_free(sc->sc_dmat, sc->sc_dma, sc->sc_dmamap);
bus_dma_tag_destroy(sc->sc_dmat);
/* Turn off DMA polling */
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MSTRESET |
HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE);
fail_io1:
bus_release_resource(dev, SYS_RES_MEMORY, HIFN_BAR1, sc->sc_bar1res);
fail_io0:
bus_release_resource(dev, SYS_RES_MEMORY, HIFN_BAR0, sc->sc_bar0res);
fail_pci:
mtx_destroy(&sc->sc_mtx);
return (ENXIO);
}
/*
* Detach an interface that successfully probed.
*/
static int
hifn_detach(device_t dev)
{
struct hifn_softc *sc = device_get_softc(dev);
KASSERT(sc != NULL, ("hifn_detach: null software carrier!"));
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
/* disable interrupts */
WRITE_REG_1(sc, HIFN_1_DMA_IER, 0);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*XXX other resources */
callout_stop(&sc->sc_tickto);
callout_stop(&sc->sc_rngto);
detach RNG test facility on device detach
2003-06-04 17:58:20 +00:00
#ifdef HIFN_RNDTEST
if (sc->sc_rndtest)
fix typo (how did this get committed?)
2003-06-04 23:12:32 +00:00
rndtest_detach(sc->sc_rndtest);
detach RNG test facility on device detach
2003-06-04 17:58:20 +00:00
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/* Turn off DMA polling */
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MSTRESET |
HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE);
crypto_unregister_all(sc->sc_cid);
bus_generic_detach(dev); /*XXX should be no children, right? */
bus_teardown_intr(dev, sc->sc_irq, sc->sc_intrhand);
/* XXX don't store rid */
bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
bus_dmamap_unload(sc->sc_dmat, sc->sc_dmamap);
bus_dmamem_free(sc->sc_dmat, sc->sc_dma, sc->sc_dmamap);
bus_dma_tag_destroy(sc->sc_dmat);
bus_release_resource(dev, SYS_RES_MEMORY, HIFN_BAR1, sc->sc_bar1res);
bus_release_resource(dev, SYS_RES_MEMORY, HIFN_BAR0, sc->sc_bar0res);
mtx_destroy(&sc->sc_mtx);
return (0);
}
/*
* Stop all chip I/O so that the kernel's probe routines don't
* get confused by errant DMAs when rebooting.
*/
shutdown returns an int
2009-02-05 19:37:49 +00:00
static int
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifn_shutdown(device_t dev)
{
#ifdef notyet
hifn_stop(device_get_softc(dev));
#endif
shutdown returns an int
2009-02-05 19:37:49 +00:00
return (0);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
/*
* Device suspend routine. Stop the interface and save some PCI
* settings in case the BIOS doesn't restore them properly on
* resume.
*/
static int
hifn_suspend(device_t dev)
{
struct hifn_softc *sc = device_get_softc(dev);
#ifdef notyet
hifn_stop(sc);
#endif
sc->sc_suspended = 1;
return (0);
}
/*
* Device resume routine. Restore some PCI settings in case the BIOS
* doesn't, re-enable busmastering, and restart the interface if
* appropriate.
*/
static int
hifn_resume(device_t dev)
{
struct hifn_softc *sc = device_get_softc(dev);
#ifdef notyet
/* reinitialize interface if necessary */
if (ifp->if_flags & IFF_UP)
rl_init(sc);
#endif
sc->sc_suspended = 0;
return (0);
}
static int
hifn_init_pubrng(struct hifn_softc *sc)
{
u_int32_t r;
int i;
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
#ifdef HIFN_RNDTEST
sc->sc_rndtest = rndtest_attach(sc->sc_dev);
if (sc->sc_rndtest)
sc->sc_harvest = rndtest_harvest;
else
sc->sc_harvest = default_harvest;
#else
sc->sc_harvest = default_harvest;
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if ((sc->sc_flags & HIFN_IS_7811) == 0) {
/* Reset 7951 public key/rng engine */
WRITE_REG_1(sc, HIFN_1_PUB_RESET,
READ_REG_1(sc, HIFN_1_PUB_RESET) | HIFN_PUBRST_RESET);
for (i = 0; i < 100; i++) {
DELAY(1000);
if ((READ_REG_1(sc, HIFN_1_PUB_RESET) &
HIFN_PUBRST_RESET) == 0)
break;
}
if (i == 100) {
device_printf(sc->sc_dev, "public key init failed\n");
return (1);
}
}
/* Enable the rng, if available */
if (sc->sc_flags & HIFN_HAS_RNG) {
if (sc->sc_flags & HIFN_IS_7811) {
r = READ_REG_1(sc, HIFN_1_7811_RNGENA);
if (r & HIFN_7811_RNGENA_ENA) {
r &= ~HIFN_7811_RNGENA_ENA;
WRITE_REG_1(sc, HIFN_1_7811_RNGENA, r);
}
WRITE_REG_1(sc, HIFN_1_7811_RNGCFG,
HIFN_7811_RNGCFG_DEFL);
r |= HIFN_7811_RNGENA_ENA;
WRITE_REG_1(sc, HIFN_1_7811_RNGENA, r);
} else
WRITE_REG_1(sc, HIFN_1_RNG_CONFIG,
READ_REG_1(sc, HIFN_1_RNG_CONFIG) |
HIFN_RNGCFG_ENA);
sc->sc_rngfirst = 1;
if (hz >= 100)
sc->sc_rnghz = hz / 100;
else
sc->sc_rnghz = 1;
CALLOUT_MPSAFE has lost its meaning since r141428, i.e., for more than ten years for head. However, it is continuously misused as the mpsafe argument for callout_init(9). Deprecate the flag and clean up callout_init() calls to make them more consistent. Differential Revision: https://reviews.freebsd.org/D2613 Reviewed by: jhb MFC after: 2 weeks
2015-05-22 17:05:21 +00:00
callout_init(&sc->sc_rngto, 1);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
callout_reset(&sc->sc_rngto, sc->sc_rnghz, hifn_rng, sc);
}
/* Enable public key engine, if available */
if (sc->sc_flags & HIFN_HAS_PUBLIC) {
WRITE_REG_1(sc, HIFN_1_PUB_IEN, HIFN_PUBIEN_DONE);
sc->sc_dmaier |= HIFN_DMAIER_PUBDONE;
WRITE_REG_1(sc, HIFN_1_DMA_IER, sc->sc_dmaier);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
#ifdef HIFN_VULCANDEV
sc->sc_pkdev = make_dev(&vulcanpk_cdevsw, 0,
UID_ROOT, GID_WHEEL, 0666,
"vulcanpk");
sc->sc_pkdev->si_drv1 = sc;
#endif
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
return (0);
}
static void
hifn_rng(void *vsc)
{
#define RANDOM_BITS(n) (n)*sizeof (u_int32_t), (n)*sizeof (u_int32_t)*NBBY, 0
struct hifn_softc *sc = vsc;
u_int32_t sts, num[2];
int i;
if (sc->sc_flags & HIFN_IS_7811) {
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
/* ONLY VALID ON 7811!!!! */
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
for (i = 0; i < 5; i++) {
sts = READ_REG_1(sc, HIFN_1_7811_RNGSTS);
if (sts & HIFN_7811_RNGSTS_UFL) {
device_printf(sc->sc_dev,
"RNG underflow: disabling\n");
return;
}
if ((sts & HIFN_7811_RNGSTS_RDY) == 0)
break;
/*
* There are at least two words in the RNG FIFO
* at this point.
*/
num[0] = READ_REG_1(sc, HIFN_1_7811_RNGDAT);
num[1] = READ_REG_1(sc, HIFN_1_7811_RNGDAT);
/* NB: discard first data read */
if (sc->sc_rngfirst)
sc->sc_rngfirst = 0;
else
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
(*sc->sc_harvest)(sc->sc_rndtest,
num, sizeof (num));
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
} else {
num[0] = READ_REG_1(sc, HIFN_1_RNG_DATA);
/* NB: discard first data read */
if (sc->sc_rngfirst)
sc->sc_rngfirst = 0;
else
o add crypto driver glue for using the new rndtest driver/module; this is conditional in each driver on foo_RNDTEST being defined_ o bring HIFN_DEBUG and UBSEC_DEBUG out to be visible options; they control the debugging printfs that are set with hw.foo.debug (e.g. hw.hifn.debug)
2003-03-11 22:47:06 +00:00
(*sc->sc_harvest)(sc->sc_rndtest,
num, sizeof (num[0]));
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
callout_reset(&sc->sc_rngto, sc->sc_rnghz, hifn_rng, sc);
#undef RANDOM_BITS
}
static void
hifn_puc_wait(struct hifn_softc *sc)
{
int i;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
int reg = HIFN_0_PUCTRL;
if (sc->sc_flags & HIFN_IS_7956) {
reg = HIFN_0_PUCTRL2;
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
for (i = 5000; i > 0; i--) {
DELAY(1);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
if (!(READ_REG_0(sc, reg) & HIFN_PUCTRL_RESET))
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
break;
}
if (!i)
device_printf(sc->sc_dev, "proc unit did not reset\n");
}
/*
* Reset the processing unit.
*/
static void
hifn_reset_puc(struct hifn_softc *sc)
{
/* Reset processing unit */
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
int reg = HIFN_0_PUCTRL;
if (sc->sc_flags & HIFN_IS_7956) {
reg = HIFN_0_PUCTRL2;
}
WRITE_REG_0(sc, reg, HIFN_PUCTRL_DMAENA);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifn_puc_wait(sc);
}
/*
* Set the Retry and TRDY registers; note that we set them to
* zero because the 7811 locks up when forced to retry (section
* 3.6 of "Specification Update SU-0014-04". Not clear if we
* should do this for all Hifn parts, but it doesn't seem to hurt.
*/
static void
hifn_set_retry(struct hifn_softc *sc)
{
/* NB: RETRY only responds to 8-bit reads/writes */
pci_write_config(sc->sc_dev, HIFN_RETRY_TIMEOUT, 0, 1);
No need to zero the softc. It's allocated with M_ZERO. Use pci_enable_busmaster instead of setting PCIM_CMD_BUSMASTEREN directly. There's no need to set PCIM_CMD_MEMEN. The bit is set when a SYS_RES_MEMORY resource is activated. Remove redundant pci_* function calls from suspend/resume methods. The bus driver already saves and restores the PCI configuration. Write 1 byte instead of 4 when setting the HIFN_TRDY_TIMEOUT register. It is only 1 byte according to the specification. Reviewed by: jhb Approved by: kib (mentor)
2010-12-18 14:24:24 +00:00
pci_write_config(sc->sc_dev, HIFN_TRDY_TIMEOUT, 0, 1);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
/*
* Resets the board. Values in the regesters are left as is
* from the reset (i.e. initial values are assigned elsewhere).
*/
static void
hifn_reset_board(struct hifn_softc *sc, int full)
{
u_int32_t reg;
/*
* Set polling in the DMA configuration register to zero. 0x7 avoids
* resetting the board and zeros out the other fields.
*/
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MSTRESET |
HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE);
/*
* Now that polling has been disabled, we have to wait 1 ms
* before resetting the board.
*/
DELAY(1000);
/* Reset the DMA unit */
if (full) {
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MODE);
DELAY(1000);
} else {
WRITE_REG_1(sc, HIFN_1_DMA_CNFG,
HIFN_DMACNFG_MODE | HIFN_DMACNFG_MSTRESET);
hifn_reset_puc(sc);
}
KASSERT(sc->sc_dma != NULL, ("hifn_reset_board: null DMA tag!"));
bzero(sc->sc_dma, sizeof(*sc->sc_dma));
/* Bring dma unit out of reset */
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MSTRESET |
HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE);
hifn_puc_wait(sc);
hifn_set_retry(sc);
if (sc->sc_flags & HIFN_IS_7811) {
for (reg = 0; reg < 1000; reg++) {
if (READ_REG_1(sc, HIFN_1_7811_MIPSRST) &
HIFN_MIPSRST_CRAMINIT)
break;
DELAY(1000);
}
if (reg == 1000)
printf(": cram init timeout\n");
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
} else {
/* set up DMA configuration register #2 */
/* turn off all PK and BAR0 swaps */
WRITE_REG_1(sc, HIFN_1_DMA_CNFG2,
(3 << HIFN_DMACNFG2_INIT_WRITE_BURST_SHIFT)|
(3 << HIFN_DMACNFG2_INIT_READ_BURST_SHIFT)|
(2 << HIFN_DMACNFG2_TGT_WRITE_BURST_SHIFT)|
(2 << HIFN_DMACNFG2_TGT_READ_BURST_SHIFT));
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
static u_int32_t
hifn_next_signature(u_int32_t a, u_int cnt)
{
int i;
u_int32_t v;
for (i = 0; i < cnt; i++) {
/* get the parity */
v = a & 0x80080125;
v ^= v >> 16;
v ^= v >> 8;
v ^= v >> 4;
v ^= v >> 2;
v ^= v >> 1;
a = (v & 1) ^ (a << 1);
}
return a;
}
struct pci2id {
u_short pci_vendor;
u_short pci_prod;
char card_id[13];
};
static struct pci2id pci2id[] = {
{
PCI_VENDOR_HIFN,
PCI_PRODUCT_HIFN_7951,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
}, {
PCI_VENDOR_HIFN,
PCI_PRODUCT_HIFN_7955,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
}, {
PCI_VENDOR_HIFN,
PCI_PRODUCT_HIFN_7956,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}, {
PCI_VENDOR_NETSEC,
PCI_PRODUCT_NETSEC_7751,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
}, {
PCI_VENDOR_INVERTEX,
PCI_PRODUCT_INVERTEX_AEON,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
}, {
PCI_VENDOR_HIFN,
PCI_PRODUCT_HIFN_7811,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
}, {
/*
* Other vendors share this PCI ID as well, such as
* http://www.powercrypt.com, and obviously they also
* use the same key.
*/
PCI_VENDOR_HIFN,
PCI_PRODUCT_HIFN_7751,
{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00 }
},
};
/*
* Checks to see if crypto is already enabled. If crypto isn't enable,
* "hifn_enable_crypto" is called to enable it. The check is important,
* as enabling crypto twice will lock the board.
*/
static int
hifn_enable_crypto(struct hifn_softc *sc)
{
u_int32_t dmacfg, ramcfg, encl, addr, i;
char *offtbl = NULL;
sys/dev: use our nitems() macro when it is avaliable through param.h. No functional change, only trivial cases are done in this sweep, Drivers that can get further enhancements will be done independently. Discussed in: freebsd-current
2016-04-19 23:37:24 +00:00
for (i = 0; i < nitems(pci2id); i++) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (pci2id[i].pci_vendor == pci_get_vendor(sc->sc_dev) &&
pci2id[i].pci_prod == pci_get_device(sc->sc_dev)) {
offtbl = pci2id[i].card_id;
break;
}
}
if (offtbl == NULL) {
device_printf(sc->sc_dev, "Unknown card!\n");
return (1);
}
ramcfg = READ_REG_0(sc, HIFN_0_PUCNFG);
dmacfg = READ_REG_1(sc, HIFN_1_DMA_CNFG);
/*
* The RAM config register's encrypt level bit needs to be set before
* every read performed on the encryption level register.
*/
WRITE_REG_0(sc, HIFN_0_PUCNFG, ramcfg | HIFN_PUCNFG_CHIPID);
encl = READ_REG_0(sc, HIFN_0_PUSTAT) & HIFN_PUSTAT_CHIPENA;
/*
* Make sure we don't re-unlock. Two unlocks kills chip until the
* next reboot.
*/
if (encl == HIFN_PUSTAT_ENA_1 || encl == HIFN_PUSTAT_ENA_2) {
#ifdef HIFN_DEBUG
if (hifn_debug)
device_printf(sc->sc_dev,
"Strong crypto already enabled!\n");
#endif
goto report;
}
if (encl != 0 && encl != HIFN_PUSTAT_ENA_0) {
#ifdef HIFN_DEBUG
if (hifn_debug)
device_printf(sc->sc_dev,
"Unknown encryption level 0x%x\n", encl);
#endif
return 1;
}
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_UNLOCK |
HIFN_DMACNFG_MSTRESET | HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE);
DELAY(1000);
addr = READ_REG_1(sc, HIFN_UNLOCK_SECRET1);
DELAY(1000);
WRITE_REG_1(sc, HIFN_UNLOCK_SECRET2, 0);
DELAY(1000);
for (i = 0; i <= 12; i++) {
addr = hifn_next_signature(addr, offtbl[i] + 0x101);
WRITE_REG_1(sc, HIFN_UNLOCK_SECRET2, addr);
DELAY(1000);
}
WRITE_REG_0(sc, HIFN_0_PUCNFG, ramcfg | HIFN_PUCNFG_CHIPID);
encl = READ_REG_0(sc, HIFN_0_PUSTAT) & HIFN_PUSTAT_CHIPENA;
#ifdef HIFN_DEBUG
if (hifn_debug) {
if (encl != HIFN_PUSTAT_ENA_1 && encl != HIFN_PUSTAT_ENA_2)
device_printf(sc->sc_dev, "Engine is permanently "
"locked until next system reset!\n");
else
device_printf(sc->sc_dev, "Engine enabled "
"successfully!\n");
}
#endif
report:
WRITE_REG_0(sc, HIFN_0_PUCNFG, ramcfg);
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, dmacfg);
switch (encl) {
case HIFN_PUSTAT_ENA_1:
case HIFN_PUSTAT_ENA_2:
break;
case HIFN_PUSTAT_ENA_0:
default:
device_printf(sc->sc_dev, "disabled");
break;
}
return 0;
}
/*
* Give initial values to the registers listed in the "Register Space"
* section of the HIFN Software Development reference manual.
*/
static void
hifn_init_pci_registers(struct hifn_softc *sc)
{
/* write fixed values needed by the Initialization registers */
WRITE_REG_0(sc, HIFN_0_PUCTRL, HIFN_PUCTRL_DMAENA);
WRITE_REG_0(sc, HIFN_0_FIFOCNFG, HIFN_FIFOCNFG_THRESHOLD);
WRITE_REG_0(sc, HIFN_0_PUIER, HIFN_PUIER_DSTOVER);
/* write all 4 ring address registers */
WRITE_REG_1(sc, HIFN_1_DMA_CRAR, sc->sc_dma_physaddr +
offsetof(struct hifn_dma, cmdr[0]));
WRITE_REG_1(sc, HIFN_1_DMA_SRAR, sc->sc_dma_physaddr +
offsetof(struct hifn_dma, srcr[0]));
WRITE_REG_1(sc, HIFN_1_DMA_DRAR, sc->sc_dma_physaddr +
offsetof(struct hifn_dma, dstr[0]));
WRITE_REG_1(sc, HIFN_1_DMA_RRAR, sc->sc_dma_physaddr +
offsetof(struct hifn_dma, resr[0]));
DELAY(2000);
/* write status register */
WRITE_REG_1(sc, HIFN_1_DMA_CSR,
HIFN_DMACSR_D_CTRL_DIS | HIFN_DMACSR_R_CTRL_DIS |
HIFN_DMACSR_S_CTRL_DIS | HIFN_DMACSR_C_CTRL_DIS |
HIFN_DMACSR_D_ABORT | HIFN_DMACSR_D_DONE | HIFN_DMACSR_D_LAST |
HIFN_DMACSR_D_WAIT | HIFN_DMACSR_D_OVER |
HIFN_DMACSR_R_ABORT | HIFN_DMACSR_R_DONE | HIFN_DMACSR_R_LAST |
HIFN_DMACSR_R_WAIT | HIFN_DMACSR_R_OVER |
HIFN_DMACSR_S_ABORT | HIFN_DMACSR_S_DONE | HIFN_DMACSR_S_LAST |
HIFN_DMACSR_S_WAIT |
HIFN_DMACSR_C_ABORT | HIFN_DMACSR_C_DONE | HIFN_DMACSR_C_LAST |
HIFN_DMACSR_C_WAIT |
HIFN_DMACSR_ENGINE |
((sc->sc_flags & HIFN_HAS_PUBLIC) ?
HIFN_DMACSR_PUBDONE : 0) |
((sc->sc_flags & HIFN_IS_7811) ?
HIFN_DMACSR_ILLW | HIFN_DMACSR_ILLR : 0));
sc->sc_d_busy = sc->sc_r_busy = sc->sc_s_busy = sc->sc_c_busy = 0;
sc->sc_dmaier |= HIFN_DMAIER_R_DONE | HIFN_DMAIER_C_ABORT |
HIFN_DMAIER_D_OVER | HIFN_DMAIER_R_OVER |
HIFN_DMAIER_S_ABORT | HIFN_DMAIER_D_ABORT | HIFN_DMAIER_R_ABORT |
((sc->sc_flags & HIFN_IS_7811) ?
HIFN_DMAIER_ILLW | HIFN_DMAIER_ILLR : 0);
sc->sc_dmaier &= ~HIFN_DMAIER_C_WAIT;
WRITE_REG_1(sc, HIFN_1_DMA_IER, sc->sc_dmaier);
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
if (sc->sc_flags & HIFN_IS_7956) {
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
u_int32_t pll;
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
WRITE_REG_0(sc, HIFN_0_PUCNFG, HIFN_PUCNFG_COMPSING |
HIFN_PUCNFG_TCALLPHASES |
HIFN_PUCNFG_TCDRVTOTEM | HIFN_PUCNFG_BUS32);
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
/* turn off the clocks and insure bypass is set */
pll = READ_REG_1(sc, HIFN_1_PLL);
pll = (pll &~ (HIFN_PLL_PK_CLK_SEL | HIFN_PLL_PE_CLK_SEL))
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
| HIFN_PLL_BP | HIFN_PLL_MBSET;
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
WRITE_REG_1(sc, HIFN_1_PLL, pll);
DELAY(10*1000); /* 10ms */
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
/* change configuration */
pll = (pll &~ HIFN_PLL_CONFIG) | sc->sc_pllconfig;
WRITE_REG_1(sc, HIFN_1_PLL, pll);
DELAY(10*1000); /* 10ms */
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
Update support for 795x parts: o rework pll setup code to follow h/w specification o add hint.hifn.X.pllconfig to specify reference clock setup requirements; default is pci66 which means the clock is derived from the PCI bus clock and the card resides in a 66MHz slot Tested on 7955 and 7956 cards; support for 7954 cards not enabled since we have no cards to test against. In collaboration with Poul-Henning Kamp. Reviewed by: phk MFC after: 1 week
2005-01-19 17:03:35 +00:00
/* disable bypass */
pll &= ~HIFN_PLL_BP;
WRITE_REG_1(sc, HIFN_1_PLL, pll);
/* enable clocks with new configuration */
pll |= HIFN_PLL_PK_CLK_SEL | HIFN_PLL_PE_CLK_SEL;
WRITE_REG_1(sc, HIFN_1_PLL, pll);
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
} else {
WRITE_REG_0(sc, HIFN_0_PUCNFG, HIFN_PUCNFG_COMPSING |
HIFN_PUCNFG_DRFR_128 | HIFN_PUCNFG_TCALLPHASES |
HIFN_PUCNFG_TCDRVTOTEM | HIFN_PUCNFG_BUS32 |
(sc->sc_drammodel ? HIFN_PUCNFG_DRAM : HIFN_PUCNFG_SRAM));
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
WRITE_REG_0(sc, HIFN_0_PUISR, HIFN_PUISR_DSTOVER);
WRITE_REG_1(sc, HIFN_1_DMA_CNFG, HIFN_DMACNFG_MSTRESET |
HIFN_DMACNFG_DMARESET | HIFN_DMACNFG_MODE | HIFN_DMACNFG_LAST |
((HIFN_POLL_FREQUENCY << 16 ) & HIFN_DMACNFG_POLLFREQ) |
((HIFN_POLL_SCALAR << 8) & HIFN_DMACNFG_POLLINVAL));
}
/*
* The maximum number of sessions supported by the card
* is dependent on the amount of context ram, which
* encryption algorithms are enabled, and how compression
* is configured. This should be configured before this
* routine is called.
*/
static void
hifn_sessions(struct hifn_softc *sc)
{
u_int32_t pucnfg;
int ctxsize;
pucnfg = READ_REG_0(sc, HIFN_0_PUCNFG);
if (pucnfg & HIFN_PUCNFG_COMPSING) {
if (pucnfg & HIFN_PUCNFG_ENCCNFG)
ctxsize = 128;
else
ctxsize = 512;
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
/*
* 7955/7956 has internal context memory of 32K
*/
if (sc->sc_flags & HIFN_IS_7956)
sc->sc_maxses = 32768 / ctxsize;
else
sc->sc_maxses = 1 +
((sc->sc_ramsize - 32768) / ctxsize);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
} else
sc->sc_maxses = sc->sc_ramsize / 16384;
if (sc->sc_maxses > 2048)
sc->sc_maxses = 2048;
}
/*
* Determine ram type (sram or dram). Board should be just out of a reset
* state when this is called.
*/
static int
hifn_ramtype(struct hifn_softc *sc)
{
u_int8_t data[8], dataexpect[8];
int i;
for (i = 0; i < sizeof(data); i++)
data[i] = dataexpect[i] = 0x55;
if (hifn_writeramaddr(sc, 0, data))
return (-1);
if (hifn_readramaddr(sc, 0, data))
return (-1);
if (bcmp(data, dataexpect, sizeof(data)) != 0) {
sc->sc_drammodel = 1;
return (0);
}
for (i = 0; i < sizeof(data); i++)
data[i] = dataexpect[i] = 0xaa;
if (hifn_writeramaddr(sc, 0, data))
return (-1);
if (hifn_readramaddr(sc, 0, data))
return (-1);
if (bcmp(data, dataexpect, sizeof(data)) != 0) {
sc->sc_drammodel = 1;
return (0);
}
return (0);
}
#define HIFN_SRAM_MAX (32 << 20)
#define HIFN_SRAM_STEP_SIZE 16384
#define HIFN_SRAM_GRANULARITY (HIFN_SRAM_MAX / HIFN_SRAM_STEP_SIZE)
static int
hifn_sramsize(struct hifn_softc *sc)
{
u_int32_t a;
u_int8_t data[8];
u_int8_t dataexpect[sizeof(data)];
int32_t i;
for (i = 0; i < sizeof(data); i++)
data[i] = dataexpect[i] = i ^ 0x5a;
for (i = HIFN_SRAM_GRANULARITY - 1; i >= 0; i--) {
a = i * HIFN_SRAM_STEP_SIZE;
bcopy(&i, data, sizeof(i));
hifn_writeramaddr(sc, a, data);
}
for (i = 0; i < HIFN_SRAM_GRANULARITY; i++) {
a = i * HIFN_SRAM_STEP_SIZE;
bcopy(&i, dataexpect, sizeof(i));
if (hifn_readramaddr(sc, a, data) < 0)
return (0);
if (bcmp(data, dataexpect, sizeof(data)) != 0)
return (0);
sc->sc_ramsize = a + HIFN_SRAM_STEP_SIZE;
}
return (0);
}
/*
* XXX For dram boards, one should really try all of the
* HIFN_PUCNFG_DSZ_*'s. This just assumes that PUCNFG
* is already set up correctly.
*/
static int
hifn_dramsize(struct hifn_softc *sc)
{
u_int32_t cnfg;
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
if (sc->sc_flags & HIFN_IS_7956) {
/*
* 7955/7956 have a fixed internal ram of only 32K.
*/
sc->sc_ramsize = 32768;
} else {
cnfg = READ_REG_0(sc, HIFN_0_PUCNFG) &
HIFN_PUCNFG_DRAMMASK;
sc->sc_ramsize = 1 << ((cnfg >> 13) + 18);
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (0);
}
static void
hifn_alloc_slot(struct hifn_softc *sc, int *cmdp, int *srcp, int *dstp, int *resp)
{
struct hifn_dma *dma = sc->sc_dma;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_cmdi == HIFN_D_CMD_RSIZE) {
sc->sc_cmdi = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->cmdr[HIFN_D_CMD_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_CMDR_SYNC(sc, HIFN_D_CMD_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
*cmdp = sc->sc_cmdi++;
sc->sc_cmdk = sc->sc_cmdi;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_srci == HIFN_D_SRC_RSIZE) {
sc->sc_srci = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->srcr[HIFN_D_SRC_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_SRCR_SYNC(sc, HIFN_D_SRC_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
*srcp = sc->sc_srci++;
sc->sc_srck = sc->sc_srci;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_dsti == HIFN_D_DST_RSIZE) {
sc->sc_dsti = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->dstr[HIFN_D_DST_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_DSTR_SYNC(sc, HIFN_D_DST_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
*dstp = sc->sc_dsti++;
sc->sc_dstk = sc->sc_dsti;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_resi == HIFN_D_RES_RSIZE) {
sc->sc_resi = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->resr[HIFN_D_RES_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_RESR_SYNC(sc, HIFN_D_RES_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
*resp = sc->sc_resi++;
sc->sc_resk = sc->sc_resi;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
static int
hifn_writeramaddr(struct hifn_softc *sc, int addr, u_int8_t *data)
{
struct hifn_dma *dma = sc->sc_dma;
hifn_base_command_t wc;
const u_int32_t masks = HIFN_D_VALID | HIFN_D_LAST | HIFN_D_MASKDONEIRQ;
int r, cmdi, resi, srci, dsti;
wc.masks = htole16(3 << 13);
wc.session_num = htole16(addr >> 14);
wc.total_source_count = htole16(8);
wc.total_dest_count = htole16(addr & 0x3fff);
hifn_alloc_slot(sc, &cmdi, &srci, &dsti, &resi);
WRITE_REG_1(sc, HIFN_1_DMA_CSR,
HIFN_DMACSR_C_CTRL_ENA | HIFN_DMACSR_S_CTRL_ENA |
HIFN_DMACSR_D_CTRL_ENA | HIFN_DMACSR_R_CTRL_ENA);
/* build write command */
bzero(dma->command_bufs[cmdi], HIFN_MAX_COMMAND);
*(hifn_base_command_t *)dma->command_bufs[cmdi] = wc;
bcopy(data, &dma->test_src, sizeof(dma->test_src));
dma->srcr[srci].p = htole32(sc->sc_dma_physaddr
+ offsetof(struct hifn_dma, test_src));
dma->dstr[dsti].p = htole32(sc->sc_dma_physaddr
+ offsetof(struct hifn_dma, test_dst));
dma->cmdr[cmdi].l = htole32(16 | masks);
dma->srcr[srci].l = htole32(8 | masks);
dma->dstr[dsti].l = htole32(4 | masks);
dma->resr[resi].l = htole32(4 | masks);
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
for (r = 10000; r >= 0; r--) {
DELAY(10);
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if ((dma->resr[resi].l & htole32(HIFN_D_VALID)) == 0)
break;
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
if (r == 0) {
device_printf(sc->sc_dev, "writeramaddr -- "
"result[%d](addr %d) still valid\n", resi, addr);
r = -1;
return (-1);
} else
r = 0;
WRITE_REG_1(sc, HIFN_1_DMA_CSR,
HIFN_DMACSR_C_CTRL_DIS | HIFN_DMACSR_S_CTRL_DIS |
HIFN_DMACSR_D_CTRL_DIS | HIFN_DMACSR_R_CTRL_DIS);
return (r);
}
static int
hifn_readramaddr(struct hifn_softc *sc, int addr, u_int8_t *data)
{
struct hifn_dma *dma = sc->sc_dma;
hifn_base_command_t rc;
const u_int32_t masks = HIFN_D_VALID | HIFN_D_LAST | HIFN_D_MASKDONEIRQ;
int r, cmdi, srci, dsti, resi;
rc.masks = htole16(2 << 13);
rc.session_num = htole16(addr >> 14);
rc.total_source_count = htole16(addr & 0x3fff);
rc.total_dest_count = htole16(8);
hifn_alloc_slot(sc, &cmdi, &srci, &dsti, &resi);
WRITE_REG_1(sc, HIFN_1_DMA_CSR,
HIFN_DMACSR_C_CTRL_ENA | HIFN_DMACSR_S_CTRL_ENA |
HIFN_DMACSR_D_CTRL_ENA | HIFN_DMACSR_R_CTRL_ENA);
bzero(dma->command_bufs[cmdi], HIFN_MAX_COMMAND);
*(hifn_base_command_t *)dma->command_bufs[cmdi] = rc;
dma->srcr[srci].p = htole32(sc->sc_dma_physaddr +
offsetof(struct hifn_dma, test_src));
dma->test_src = 0;
dma->dstr[dsti].p = htole32(sc->sc_dma_physaddr +
offsetof(struct hifn_dma, test_dst));
dma->test_dst = 0;
dma->cmdr[cmdi].l = htole32(8 | masks);
dma->srcr[srci].l = htole32(8 | masks);
dma->dstr[dsti].l = htole32(8 | masks);
dma->resr[resi].l = htole32(HIFN_MAX_RESULT | masks);
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
for (r = 10000; r >= 0; r--) {
DELAY(10);
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if ((dma->resr[resi].l & htole32(HIFN_D_VALID)) == 0)
break;
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
if (r == 0) {
device_printf(sc->sc_dev, "readramaddr -- "
"result[%d](addr %d) still valid\n", resi, addr);
r = -1;
} else {
r = 0;
bcopy(&dma->test_dst, data, sizeof(dma->test_dst));
}
WRITE_REG_1(sc, HIFN_1_DMA_CSR,
HIFN_DMACSR_C_CTRL_DIS | HIFN_DMACSR_S_CTRL_DIS |
HIFN_DMACSR_D_CTRL_DIS | HIFN_DMACSR_R_CTRL_DIS);
return (r);
}
/*
* Initialize the descriptor rings.
*/
static void
hifn_init_dma(struct hifn_softc *sc)
{
struct hifn_dma *dma = sc->sc_dma;
int i;
hifn_set_retry(sc);
/* initialize static pointer values */
for (i = 0; i < HIFN_D_CMD_RSIZE; i++)
dma->cmdr[i].p = htole32(sc->sc_dma_physaddr +
offsetof(struct hifn_dma, command_bufs[i][0]));
for (i = 0; i < HIFN_D_RES_RSIZE; i++)
dma->resr[i].p = htole32(sc->sc_dma_physaddr +
offsetof(struct hifn_dma, result_bufs[i][0]));
dma->cmdr[HIFN_D_CMD_RSIZE].p =
htole32(sc->sc_dma_physaddr + offsetof(struct hifn_dma, cmdr[0]));
dma->srcr[HIFN_D_SRC_RSIZE].p =
htole32(sc->sc_dma_physaddr + offsetof(struct hifn_dma, srcr[0]));
dma->dstr[HIFN_D_DST_RSIZE].p =
htole32(sc->sc_dma_physaddr + offsetof(struct hifn_dma, dstr[0]));
dma->resr[HIFN_D_RES_RSIZE].p =
htole32(sc->sc_dma_physaddr + offsetof(struct hifn_dma, resr[0]));
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdu = sc->sc_srcu = sc->sc_dstu = sc->sc_resu = 0;
sc->sc_cmdi = sc->sc_srci = sc->sc_dsti = sc->sc_resi = 0;
sc->sc_cmdk = sc->sc_srck = sc->sc_dstk = sc->sc_resk = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
/*
* Writes out the raw command buffer space. Returns the
* command buffer size.
*/
static u_int
hifn_write_command(struct hifn_command *cmd, u_int8_t *buf)
{
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
struct cryptop *crp;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
u_int8_t *buf_pos;
hifn_base_command_t *base_cmd;
hifn_mac_command_t *mac_cmd;
hifn_crypt_command_t *cry_cmd;
Remove support for the ARC4 algorithm from OCF. There are no longer any in-kernel consumers. The software implementation was also a non-functional stub. Reviewed by: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D24771
2020-05-11 21:17:08 +00:00
int using_mac, using_crypt, ivlen;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
u_int32_t dlen, slen;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
crp = cmd->crp;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
buf_pos = buf;
using_mac = cmd->base_masks & HIFN_BASE_CMD_MAC;
using_crypt = cmd->base_masks & HIFN_BASE_CMD_CRYPT;
base_cmd = (hifn_base_command_t *)buf_pos;
base_cmd->masks = htole16(cmd->base_masks);
slen = cmd->src_mapsize;
if (cmd->sloplen)
dlen = cmd->dst_mapsize - cmd->sloplen + sizeof(u_int32_t);
else
dlen = cmd->dst_mapsize;
base_cmd->total_source_count = htole16(slen & HIFN_BASE_CMD_LENMASK_LO);
base_cmd->total_dest_count = htole16(dlen & HIFN_BASE_CMD_LENMASK_LO);
dlen >>= 16;
slen >>= 16;
Move session state from on-card memory to host memory so we no longer are constrained to a small number of sessions by the small on-card memories found in newer devices. This is really a stopgap solution as having session state in main memory incurs a (small but noticeable) performance penalty. The better solution is to manage session state so that it's cached on chip. Obtained from: openbsd
2004-10-15 00:36:07 +00:00
base_cmd->session_num = htole16(
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
((slen << HIFN_BASE_CMD_SRCLEN_S) & HIFN_BASE_CMD_SRCLEN_M) |
((dlen << HIFN_BASE_CMD_DSTLEN_S) & HIFN_BASE_CMD_DSTLEN_M));
buf_pos += sizeof(hifn_base_command_t);
if (using_mac) {
mac_cmd = (hifn_mac_command_t *)buf_pos;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
dlen = crp->crp_aad_length + crp->crp_payload_length;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
mac_cmd->source_count = htole16(dlen & 0xffff);
dlen >>= 16;
mac_cmd->masks = htole16(cmd->mac_masks |
((dlen << HIFN_MAC_CMD_SRCLEN_S) & HIFN_MAC_CMD_SRCLEN_M));
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (crp->crp_aad_length != 0)
mac_cmd->header_skip = htole16(crp->crp_aad_start);
else
mac_cmd->header_skip = htole16(crp->crp_payload_start);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
mac_cmd->reserved = 0;
buf_pos += sizeof(hifn_mac_command_t);
}
if (using_crypt) {
cry_cmd = (hifn_crypt_command_t *)buf_pos;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
dlen = crp->crp_payload_length;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cry_cmd->source_count = htole16(dlen & 0xffff);
dlen >>= 16;
cry_cmd->masks = htole16(cmd->cry_masks |
((dlen << HIFN_CRYPT_CMD_SRCLEN_S) & HIFN_CRYPT_CMD_SRCLEN_M));
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
cry_cmd->header_skip = htole16(crp->crp_payload_length);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cry_cmd->reserved = 0;
buf_pos += sizeof(hifn_crypt_command_t);
}
if (using_mac && cmd->mac_masks & HIFN_MAC_CMD_NEW_KEY) {
bcopy(cmd->mac, buf_pos, HIFN_MAC_KEY_LENGTH);
buf_pos += HIFN_MAC_KEY_LENGTH;
}
if (using_crypt && cmd->cry_masks & HIFN_CRYPT_CMD_NEW_KEY) {
switch (cmd->cry_masks & HIFN_CRYPT_CMD_ALG_MASK) {
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
case HIFN_CRYPT_CMD_ALG_AES:
/*
* AES keys are variable 128, 192 and
* 256 bits (16, 24 and 32 bytes).
*/
bcopy(cmd->ck, buf_pos, cmd->cklen);
buf_pos += cmd->cklen;
break;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
}
if (using_crypt && cmd->cry_masks & HIFN_CRYPT_CMD_NEW_IV) {
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
switch (cmd->cry_masks & HIFN_CRYPT_CMD_ALG_MASK) {
case HIFN_CRYPT_CMD_ALG_AES:
ivlen = HIFN_AES_IV_LENGTH;
break;
default:
ivlen = HIFN_IV_LENGTH;
break;
}
bcopy(cmd->iv, buf_pos, ivlen);
buf_pos += ivlen;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
if ((cmd->base_masks & (HIFN_BASE_CMD_MAC|HIFN_BASE_CMD_CRYPT)) == 0) {
bzero(buf_pos, 8);
buf_pos += 8;
}
return (buf_pos - buf);
}
static int
hifn_dmamap_aligned(struct hifn_operand *op)
{
int i;
for (i = 0; i < op->nsegs; i++) {
if (op->segs[i].ds_addr & 3)
return (0);
if ((i != (op->nsegs - 1)) && (op->segs[i].ds_len & 3))
return (0);
}
return (1);
}
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
static __inline int
hifn_dmamap_dstwrap(struct hifn_softc *sc, int idx)
{
struct hifn_dma *dma = sc->sc_dma;
if (++idx == HIFN_D_DST_RSIZE) {
dma->dstr[idx].l = htole32(HIFN_D_VALID | HIFN_D_JUMP |
HIFN_D_MASKDONEIRQ);
HIFN_DSTR_SYNC(sc, idx,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
idx = 0;
}
return (idx);
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
static int
hifn_dmamap_load_dst(struct hifn_softc *sc, struct hifn_command *cmd)
{
struct hifn_dma *dma = sc->sc_dma;
struct hifn_operand *dst = &cmd->dst;
u_int32_t p, l;
int idx, used = 0, i;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
idx = sc->sc_dsti;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
for (i = 0; i < dst->nsegs - 1; i++) {
dma->dstr[idx].p = htole32(dst->segs[i].ds_addr);
dma->dstr[idx].l = htole32(HIFN_D_VALID |
HIFN_D_MASKDONEIRQ | dst->segs[i].ds_len);
HIFN_DSTR_SYNC(sc, idx,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
used++;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
idx = hifn_dmamap_dstwrap(sc, idx);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
if (cmd->sloplen == 0) {
p = dst->segs[i].ds_addr;
l = HIFN_D_VALID | HIFN_D_MASKDONEIRQ | HIFN_D_LAST |
dst->segs[i].ds_len;
} else {
p = sc->sc_dma_physaddr +
offsetof(struct hifn_dma, slop[cmd->slopidx]);
l = HIFN_D_VALID | HIFN_D_MASKDONEIRQ | HIFN_D_LAST |
sizeof(u_int32_t);
if ((dst->segs[i].ds_len - cmd->sloplen) != 0) {
dma->dstr[idx].p = htole32(dst->segs[i].ds_addr);
dma->dstr[idx].l = htole32(HIFN_D_VALID |
HIFN_D_MASKDONEIRQ |
(dst->segs[i].ds_len - cmd->sloplen));
HIFN_DSTR_SYNC(sc, idx,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
used++;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
idx = hifn_dmamap_dstwrap(sc, idx);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
}
dma->dstr[idx].p = htole32(p);
dma->dstr[idx].l = htole32(l);
HIFN_DSTR_SYNC(sc, idx, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
used++;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
idx = hifn_dmamap_dstwrap(sc, idx);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_dsti = idx;
sc->sc_dstu += used;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (idx);
}
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
static __inline int
hifn_dmamap_srcwrap(struct hifn_softc *sc, int idx)
{
struct hifn_dma *dma = sc->sc_dma;
if (++idx == HIFN_D_SRC_RSIZE) {
dma->srcr[idx].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_SRCR_SYNC(sc, HIFN_D_SRC_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
idx = 0;
}
return (idx);
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
static int
hifn_dmamap_load_src(struct hifn_softc *sc, struct hifn_command *cmd)
{
struct hifn_dma *dma = sc->sc_dma;
struct hifn_operand *src = &cmd->src;
int idx, i;
u_int32_t last = 0;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
idx = sc->sc_srci;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
for (i = 0; i < src->nsegs; i++) {
if (i == src->nsegs - 1)
last = HIFN_D_LAST;
dma->srcr[idx].p = htole32(src->segs[i].ds_addr);
dma->srcr[idx].l = htole32(src->segs[i].ds_len |
HIFN_D_VALID | HIFN_D_MASKDONEIRQ | last);
HIFN_SRCR_SYNC(sc, idx,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
idx = hifn_dmamap_srcwrap(sc, idx);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_srci = idx;
sc->sc_srcu += src->nsegs;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (idx);
}
static void
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
hifn_op_cb(void* arg, bus_dma_segment_t *seg, int nsegs, int error)
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
{
struct hifn_operand *op = arg;
KASSERT(nsegs <= MAX_SCATTER,
("hifn_op_cb: too many DMA segments (%u > %u) "
"returned when mapping operand", nsegs, MAX_SCATTER));
op->nsegs = nsegs;
bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
}
static int
hifn_crypto(
struct hifn_softc *sc,
struct hifn_command *cmd,
struct cryptop *crp,
int hint)
{
struct hifn_dma *dma = sc->sc_dma;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
u_int32_t cmdlen, csr;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
int cmdi, resi, err = 0;
/*
* need 1 cmd, and 1 res
*
* NB: check this first since it's easy.
*/
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_LOCK(sc);
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if ((sc->sc_cmdu + 1) > HIFN_D_CMD_RSIZE ||
(sc->sc_resu + 1) > HIFN_D_RES_RSIZE) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#ifdef HIFN_DEBUG
if (hifn_debug) {
device_printf(sc->sc_dev,
"cmd/result exhaustion, cmdu %u resu %u\n",
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdu, sc->sc_resu);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
#endif
hifnstats.hst_nomem_cr++;
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_UNLOCK(sc);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (ERESTART);
}
if (bus_dmamap_create(sc->sc_dmat, BUS_DMA_NOWAIT, &cmd->src_map)) {
hifnstats.hst_nomem_map++;
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_UNLOCK(sc);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (ENOMEM);
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (bus_dmamap_load_crp(sc->sc_dmat, cmd->src_map, crp, hifn_op_cb,
&cmd->src, BUS_DMA_NOWAIT)) {
hifnstats.hst_nomem_load++;
err = ENOMEM;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
goto err_srcmap1;
}
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
cmd->src_mapsize = crypto_buffer_len(&crp->crp_buf);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (hifn_dmamap_aligned(&cmd->src)) {
cmd->sloplen = cmd->src_mapsize & 3;
cmd->dst = cmd->src;
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
} else if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
int totlen, len;
struct mbuf *m, *m0, *mlast;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
KASSERT(cmd->dst_m == NULL,
("hifn_crypto: dst_m initialized improperly"));
hifnstats.hst_unaligned++;
/*
* Source is not aligned on a longword boundary.
* Copy the data to insure alignment. If we fail
* to allocate mbufs or clusters while doing this
* we return ERESTART so the operation is requeued
* at the crypto later, but only if there are
* ops already posted to the hardware; otherwise we
* have no guarantee that we'll be re-entered.
*/
totlen = cmd->src_mapsize;
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
if (crp->crp_buf.cb_mbuf->m_flags & M_PKTHDR) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
len = MHLEN;
MGETHDR(m0, M_NOWAIT, MT_DATA);
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
if (m0 && !m_dup_pkthdr(m0, crp->crp_buf.cb_mbuf,
M_NOWAIT)) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
m_free(m0);
m0 = NULL;
}
} else {
len = MLEN;
MGET(m0, M_NOWAIT, MT_DATA);
}
if (m0 == NULL) {
hifnstats.hst_nomem_mbuf++;
err = sc->sc_cmdu ? ERESTART : ENOMEM;
goto err_srcmap;
}
if (totlen >= MINCLSIZE) {
if (!(MCLGET(m0, M_NOWAIT))) {
hifnstats.hst_nomem_mcl++;
err = sc->sc_cmdu ? ERESTART : ENOMEM;
m_freem(m0);
goto err_srcmap;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
len = MCLBYTES;
}
totlen -= len;
m0->m_pkthdr.len = m0->m_len = len;
mlast = m0;
while (totlen > 0) {
MGET(m, M_NOWAIT, MT_DATA);
if (m == NULL) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifnstats.hst_nomem_mbuf++;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
err = sc->sc_cmdu ? ERESTART : ENOMEM;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
m_freem(m0);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
goto err_srcmap;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
len = MLEN;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (totlen >= MINCLSIZE) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (!(MCLGET(m, M_NOWAIT))) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifnstats.hst_nomem_mcl++;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
err = sc->sc_cmdu ? ERESTART : ENOMEM;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
mlast->m_next = m;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
m_freem(m0);
goto err_srcmap;
}
len = MCLBYTES;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
m->m_len = len;
m0->m_pkthdr.len += len;
totlen -= len;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
mlast->m_next = m;
mlast = m;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
cmd->dst_m = m0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (bus_dmamap_create(sc->sc_dmat, BUS_DMA_NOWAIT,
&cmd->dst_map)) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifnstats.hst_nomem_map++;
err = ENOMEM;
goto err_srcmap;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (bus_dmamap_load_mbuf_sg(sc->sc_dmat, cmd->dst_map, m0,
cmd->dst_segs, &cmd->dst_nsegs, 0)) {
hifnstats.hst_nomem_map++;
err = ENOMEM;
goto err_dstmap1;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
cmd->dst_mapsize = m0->m_pkthdr.len;
} else {
err = EINVAL;
goto err_srcmap;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
#ifdef HIFN_DEBUG
if (hifn_debug) {
device_printf(sc->sc_dev,
"Entering cmd: stat %8x ien %8x u %d/%d/%d/%d n %d/%d\n",
READ_REG_1(sc, HIFN_1_DMA_CSR),
READ_REG_1(sc, HIFN_1_DMA_IER),
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdu, sc->sc_srcu, sc->sc_dstu, sc->sc_resu,
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd->src_nsegs, cmd->dst_nsegs);
}
#endif
if (cmd->src_map == cmd->dst_map) {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
} else {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_PREWRITE);
bus_dmamap_sync(sc->sc_dmat, cmd->dst_map,
BUS_DMASYNC_PREREAD);
}
/*
* need N src, and N dst
*/
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if ((sc->sc_srcu + cmd->src_nsegs) > HIFN_D_SRC_RSIZE ||
(sc->sc_dstu + cmd->dst_nsegs + 1) > HIFN_D_DST_RSIZE) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#ifdef HIFN_DEBUG
if (hifn_debug) {
device_printf(sc->sc_dev,
"src/dst exhaustion, srcu %u+%u dstu %u+%u\n",
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_srcu, cmd->src_nsegs,
sc->sc_dstu, cmd->dst_nsegs);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
#endif
hifnstats.hst_nomem_sd++;
err = ERESTART;
goto err_dstmap;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_cmdi == HIFN_D_CMD_RSIZE) {
sc->sc_cmdi = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->cmdr[HIFN_D_CMD_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_CMDR_SYNC(sc, HIFN_D_CMD_RSIZE,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
cmdi = sc->sc_cmdi++;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmdlen = hifn_write_command(cmd, dma->command_bufs[cmdi]);
HIFN_CMD_SYNC(sc, cmdi, BUS_DMASYNC_PREWRITE);
/* .p for command/result already set */
dma->cmdr[cmdi].l = htole32(cmdlen | HIFN_D_VALID | HIFN_D_LAST |
HIFN_D_MASKDONEIRQ);
HIFN_CMDR_SYNC(sc, cmdi,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdu++;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* We don't worry about missing an interrupt (which a "command wait"
* interrupt salvages us from), unless there is more than one command
* in the queue.
*/
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_cmdu > 1) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_dmaier |= HIFN_DMAIER_C_WAIT;
WRITE_REG_1(sc, HIFN_1_DMA_IER, sc->sc_dmaier);
}
hifnstats.hst_ipackets++;
hifnstats.hst_ibytes += cmd->src_mapsize;
hifn_dmamap_load_src(sc, cmd);
/*
* Unlike other descriptors, we don't mask done interrupt from
* result descriptor.
*/
#ifdef HIFN_DEBUG
if (hifn_debug)
printf("load res\n");
#endif
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_resi == HIFN_D_RES_RSIZE) {
sc->sc_resi = 0;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
dma->resr[HIFN_D_RES_RSIZE].l = htole32(HIFN_D_VALID |
HIFN_D_JUMP | HIFN_D_MASKDONEIRQ);
HIFN_RESR_SYNC(sc, HIFN_D_RES_RSIZE,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
resi = sc->sc_resi++;
KASSERT(sc->sc_hifn_commands[resi] == NULL,
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
("hifn_crypto: command slot %u busy", resi));
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_hifn_commands[resi] = cmd;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
HIFN_RES_SYNC(sc, resi, BUS_DMASYNC_PREREAD);
if ((hint & CRYPTO_HINT_MORE) && sc->sc_curbatch < hifn_maxbatch) {
dma->resr[resi].l = htole32(HIFN_MAX_RESULT |
HIFN_D_VALID | HIFN_D_LAST | HIFN_D_MASKDONEIRQ);
sc->sc_curbatch++;
if (sc->sc_curbatch > hifnstats.hst_maxbatch)
hifnstats.hst_maxbatch = sc->sc_curbatch;
hifnstats.hst_totbatch++;
} else {
dma->resr[resi].l = htole32(HIFN_MAX_RESULT |
HIFN_D_VALID | HIFN_D_LAST);
sc->sc_curbatch = 0;
}
HIFN_RESR_SYNC(sc, resi,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_resu++;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (cmd->sloplen)
cmd->slopidx = resi;
hifn_dmamap_load_dst(sc, cmd);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
csr = 0;
if (sc->sc_c_busy == 0) {
csr |= HIFN_DMACSR_C_CTRL_ENA;
sc->sc_c_busy = 1;
}
if (sc->sc_s_busy == 0) {
csr |= HIFN_DMACSR_S_CTRL_ENA;
sc->sc_s_busy = 1;
}
if (sc->sc_r_busy == 0) {
csr |= HIFN_DMACSR_R_CTRL_ENA;
sc->sc_r_busy = 1;
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (sc->sc_d_busy == 0) {
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
csr |= HIFN_DMACSR_D_CTRL_ENA;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_d_busy = 1;
}
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
if (csr)
WRITE_REG_1(sc, HIFN_1_DMA_CSR, csr);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#ifdef HIFN_DEBUG
if (hifn_debug) {
device_printf(sc->sc_dev, "command: stat %8x ier %8x\n",
READ_REG_1(sc, HIFN_1_DMA_CSR),
READ_REG_1(sc, HIFN_1_DMA_IER));
}
#endif
sc->sc_active = 5;
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_UNLOCK(sc);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
KASSERT(err == 0, ("hifn_crypto: success with error %u", err));
return (err); /* success */
err_dstmap:
if (cmd->src_map != cmd->dst_map)
bus_dmamap_unload(sc->sc_dmat, cmd->dst_map);
err_dstmap1:
if (cmd->src_map != cmd->dst_map)
bus_dmamap_destroy(sc->sc_dmat, cmd->dst_map);
err_srcmap:
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (cmd->dst_m != NULL)
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
m_freem(cmd->dst_m);
}
bus_dmamap_unload(sc->sc_dmat, cmd->src_map);
err_srcmap1:
bus_dmamap_destroy(sc->sc_dmat, cmd->src_map);
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_UNLOCK(sc);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (err);
}
static void
hifn_tick(void* vsc)
{
struct hifn_softc *sc = vsc;
HIFN_LOCK(sc);
if (sc->sc_active == 0) {
u_int32_t r = 0;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_cmdu == 0 && sc->sc_c_busy) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_c_busy = 0;
r |= HIFN_DMACSR_C_CTRL_DIS;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_srcu == 0 && sc->sc_s_busy) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_s_busy = 0;
r |= HIFN_DMACSR_S_CTRL_DIS;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_dstu == 0 && sc->sc_d_busy) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_d_busy = 0;
r |= HIFN_DMACSR_D_CTRL_DIS;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if (sc->sc_resu == 0 && sc->sc_r_busy) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
sc->sc_r_busy = 0;
r |= HIFN_DMACSR_R_CTRL_DIS;
}
if (r)
WRITE_REG_1(sc, HIFN_1_DMA_CSR, r);
} else
sc->sc_active--;
HIFN_UNLOCK(sc);
callout_reset(&sc->sc_tickto, hz, hifn_tick, sc);
}
static void
hifn_intr(void *arg)
{
struct hifn_softc *sc = arg;
struct hifn_dma *dma;
u_int32_t dmacsr, restart;
int i, u;
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
dmacsr = READ_REG_1(sc, HIFN_1_DMA_CSR);
/* Nothing in the DMA unit interrupted */
if ((dmacsr & sc->sc_dmaier) == 0)
return;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
HIFN_LOCK(sc);
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
dma = sc->sc_dma;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#ifdef HIFN_DEBUG
if (hifn_debug) {
device_printf(sc->sc_dev,
"irq: stat %08x ien %08x damier %08x i %d/%d/%d/%d k %d/%d/%d/%d u %d/%d/%d/%d\n",
dmacsr, READ_REG_1(sc, HIFN_1_DMA_IER), sc->sc_dmaier,
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdi, sc->sc_srci, sc->sc_dsti, sc->sc_resi,
sc->sc_cmdk, sc->sc_srck, sc->sc_dstk, sc->sc_resk,
sc->sc_cmdu, sc->sc_srcu, sc->sc_dstu, sc->sc_resu);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
#endif
WRITE_REG_1(sc, HIFN_1_DMA_CSR, dmacsr & sc->sc_dmaier);
if ((sc->sc_flags & HIFN_HAS_PUBLIC) &&
(dmacsr & HIFN_DMACSR_PUBDONE))
WRITE_REG_1(sc, HIFN_1_PUB_STATUS,
READ_REG_1(sc, HIFN_1_PUB_STATUS) | HIFN_PUBSTS_DONE);
restart = dmacsr & (HIFN_DMACSR_D_OVER | HIFN_DMACSR_R_OVER);
if (restart)
device_printf(sc->sc_dev, "overrun %x\n", dmacsr);
if (sc->sc_flags & HIFN_IS_7811) {
if (dmacsr & HIFN_DMACSR_ILLR)
device_printf(sc->sc_dev, "illegal read\n");
if (dmacsr & HIFN_DMACSR_ILLW)
device_printf(sc->sc_dev, "illegal write\n");
}
restart = dmacsr & (HIFN_DMACSR_C_ABORT | HIFN_DMACSR_S_ABORT |
HIFN_DMACSR_D_ABORT | HIFN_DMACSR_R_ABORT);
if (restart) {
device_printf(sc->sc_dev, "abort, resetting.\n");
hifnstats.hst_abort++;
hifn_abort(sc);
HIFN_UNLOCK(sc);
return;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
if ((dmacsr & HIFN_DMACSR_C_WAIT) && (sc->sc_cmdu == 0)) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
/*
* If no slots to process and we receive a "waiting on
* command" interrupt, we disable the "waiting on command"
* (by clearing it).
*/
sc->sc_dmaier &= ~HIFN_DMAIER_C_WAIT;
WRITE_REG_1(sc, HIFN_1_DMA_IER, sc->sc_dmaier);
}
/* clear the rings */
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
i = sc->sc_resk; u = sc->sc_resu;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
while (u != 0) {
HIFN_RESR_SYNC(sc, i,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if (dma->resr[i].l & htole32(HIFN_D_VALID)) {
HIFN_RESR_SYNC(sc, i,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
break;
}
if (i != HIFN_D_RES_RSIZE) {
struct hifn_command *cmd;
u_int8_t *macbuf = NULL;
HIFN_RES_SYNC(sc, i, BUS_DMASYNC_POSTREAD);
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
cmd = sc->sc_hifn_commands[i];
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
KASSERT(cmd != NULL,
("hifn_intr: null command slot %u", i));
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_hifn_commands[i] = NULL;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (cmd->base_masks & HIFN_BASE_CMD_MAC) {
macbuf = dma->result_bufs[i];
macbuf += 12;
}
hifn_callback(sc, cmd, macbuf);
hifnstats.hst_opackets++;
u--;
}
if (++i == (HIFN_D_RES_RSIZE + 1))
i = 0;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_resk = i; sc->sc_resu = u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
i = sc->sc_srck; u = sc->sc_srcu;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
while (u != 0) {
if (i == HIFN_D_SRC_RSIZE)
i = 0;
HIFN_SRCR_SYNC(sc, i,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if (dma->srcr[i].l & htole32(HIFN_D_VALID)) {
HIFN_SRCR_SYNC(sc, i,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
break;
}
i++, u--;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_srck = i; sc->sc_srcu = u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
i = sc->sc_cmdk; u = sc->sc_cmdu;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
while (u != 0) {
HIFN_CMDR_SYNC(sc, i,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if (dma->cmdr[i].l & htole32(HIFN_D_VALID)) {
HIFN_CMDR_SYNC(sc, i,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
break;
}
if (i != HIFN_D_CMD_RSIZE) {
u--;
HIFN_CMD_SYNC(sc, i, BUS_DMASYNC_POSTWRITE);
}
if (++i == (HIFN_D_CMD_RSIZE + 1))
i = 0;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdk = i; sc->sc_cmdu = u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Redo locking for better SMP suport: o adding locking to op submission o mark interrupt handler MPSAFE o don't use locking on detach; disabling interrupts should be sufficient o change mutex string names so witness printouts are more meaningful Note: locking is still pretty brute-force but it's probably not worth improving it given the relatively low performance of hifn-based cards.
2003-06-02 23:34:59 +00:00
HIFN_UNLOCK(sc);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (sc->sc_needwakeup) { /* XXX check high watermark */
int wakeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ);
#ifdef HIFN_DEBUG
if (hifn_debug)
device_printf(sc->sc_dev,
"wakeup crypto (%x) u %d/%d/%d/%d\n",
sc->sc_needwakeup,
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_cmdu, sc->sc_srcu, sc->sc_dstu, sc->sc_resu);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
#endif
sc->sc_needwakeup &= ~wakeup;
crypto_unblock(sc->sc_cid, wakeup);
}
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
static bool
hifn_auth_supported(struct hifn_softc *sc,
const struct crypto_session_params *csp)
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
{
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (sc->sc_ena) {
case HIFN_PUSTAT_ENA_2:
case HIFN_PUSTAT_ENA_1:
break;
default:
return (false);
}
switch (csp->csp_auth_alg) {
case CRYPTO_SHA1:
break;
case CRYPTO_SHA1_HMAC:
if (csp->csp_auth_klen > HIFN_MAC_KEY_LENGTH)
return (false);
break;
default:
return (false);
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
return (true);
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
static bool
hifn_cipher_supported(struct hifn_softc *sc,
const struct crypto_session_params *csp)
{
if (csp->csp_cipher_klen == 0)
return (false);
if (csp->csp_ivlen > HIFN_MAX_IV_LENGTH)
return (false);
switch (sc->sc_ena) {
case HIFN_PUSTAT_ENA_2:
switch (csp->csp_cipher_alg) {
case CRYPTO_AES_CBC:
if ((sc->sc_flags & HIFN_HAS_AES) == 0)
return (false);
switch (csp->csp_cipher_klen) {
case 128:
case 192:
case 256:
break;
default:
return (false);
}
return (true);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
return (false);
}
static int
hifn_probesession(device_t dev, const struct crypto_session_params *csp)
{
struct hifn_softc *sc;
sc = device_get_softc(dev);
if (csp->csp_flags != 0)
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (EINVAL);
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (csp->csp_mode) {
case CSP_MODE_DIGEST:
if (!hifn_auth_supported(sc, csp))
return (EINVAL);
break;
case CSP_MODE_CIPHER:
if (!hifn_cipher_supported(sc, csp))
return (EINVAL);
break;
case CSP_MODE_ETA:
if (!hifn_auth_supported(sc, csp) ||
!hifn_cipher_supported(sc, csp))
return (EINVAL);
break;
default:
return (EINVAL);
}
return (CRYPTODEV_PROBE_HARDWARE);
}
/*
* Allocate a new 'session'.
*/
static int
hifn_newsession(device_t dev, crypto_session_t cses,
const struct crypto_session_params *csp)
{
struct hifn_session *ses;
ses = crypto_get_driver_session(cses);
if (csp->csp_auth_alg != 0) {
if (csp->csp_auth_mlen == 0)
ses->hs_mlen = crypto_auth_hash(csp)->hashsize;
else
ses->hs_mlen = csp->csp_auth_mlen;
}
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
return (0);
}
/*
OpenCrypto: Convert sessions to opaque handles instead of integers Track session objects in the framework, and pass handles between the framework (OCF), consumers, and drivers. Avoid redundancy and complexity in individual drivers by allocating session memory in the framework and providing it to drivers in ::newsession(). Session handles are no longer integers with information encoded in various high bits. Use of the CRYPTO_SESID2FOO() macros should be replaced with the appropriate crypto_ses2foo() function on the opaque session handle. Convert OCF drivers (in particular, cryptosoft, as well as myriad others) to the opaque handle interface. Discard existing session tracking as much as possible (quick pass). There may be additional code ripe for deletion. Convert OCF consumers (ipsec, geom_eli, krb5, cryptodev) to handle-style interface. The conversion is largely mechnical. The change is documented in crypto.9. Inspired by https://lists.freebsd.org/pipermail/freebsd-arch/2018-January/018835.html . No objection from: ae (ipsec portion) Reported by: jhb
2018-07-18 00:56:25 +00:00
* XXX freesession routine should run a zero'd mac/encrypt key into context
* ram. to blow away any keys already stored there.
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
*/
static int
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
hifn_process(device_t dev, struct cryptop *crp, int hint)
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
{
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
const struct crypto_session_params *csp;
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
struct hifn_softc *sc = device_get_softc(dev);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
struct hifn_command *cmd = NULL;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
const void *mackey;
Retire the CRYPTO_F_IV_GENERATE flag. The sole in-tree user of this flag has been retired, so remove this complexity from all drivers. While here, add a helper routine drivers can use to read the current request's IV into a local buffer. Use this routine to replace duplicated code in nearly all drivers. Reviewed by: cem Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24450
2020-04-20 22:24:49 +00:00
int err, keylen;
OpenCrypto: Convert sessions to opaque handles instead of integers Track session objects in the framework, and pass handles between the framework (OCF), consumers, and drivers. Avoid redundancy and complexity in individual drivers by allocating session memory in the framework and providing it to drivers in ::newsession(). Session handles are no longer integers with information encoded in various high bits. Use of the CRYPTO_SESID2FOO() macros should be replaced with the appropriate crypto_ses2foo() function on the opaque session handle. Convert OCF drivers (in particular, cryptosoft, as well as myriad others) to the opaque handle interface. Discard existing session tracking as much as possible (quick pass). There may be additional code ripe for deletion. Convert OCF consumers (ipsec, geom_eli, krb5, cryptodev) to handle-style interface. The conversion is largely mechnical. The change is documented in crypto.9. Inspired by https://lists.freebsd.org/pipermail/freebsd-arch/2018-January/018835.html . No objection from: ae (ipsec portion) Reported by: jhb
2018-07-18 00:56:25 +00:00
struct hifn_session *ses;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
OpenCrypto: Convert sessions to opaque handles instead of integers Track session objects in the framework, and pass handles between the framework (OCF), consumers, and drivers. Avoid redundancy and complexity in individual drivers by allocating session memory in the framework and providing it to drivers in ::newsession(). Session handles are no longer integers with information encoded in various high bits. Use of the CRYPTO_SESID2FOO() macros should be replaced with the appropriate crypto_ses2foo() function on the opaque session handle. Convert OCF drivers (in particular, cryptosoft, as well as myriad others) to the opaque handle interface. Discard existing session tracking as much as possible (quick pass). There may be additional code ripe for deletion. Convert OCF consumers (ipsec, geom_eli, krb5, cryptodev) to handle-style interface. The conversion is largely mechnical. The change is documented in crypto.9. Inspired by https://lists.freebsd.org/pipermail/freebsd-arch/2018-January/018835.html . No objection from: ae (ipsec portion) Reported by: jhb
2018-07-18 00:56:25 +00:00
ses = crypto_get_driver_session(crp->crp_session);
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd = malloc(sizeof(struct hifn_command), M_DEVBUF, M_NOWAIT | M_ZERO);
if (cmd == NULL) {
hifnstats.hst_nomem++;
err = ENOMEM;
goto errout;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
csp = crypto_get_params(crp->crp_session);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
/*
* The driver only supports ETA requests where there is no
* gap between the AAD and payload.
*/
if (csp->csp_mode == CSP_MODE_ETA && crp->crp_aad_length != 0 &&
crp->crp_aad_start + crp->crp_aad_length !=
crp->crp_payload_start) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
err = EINVAL;
goto errout;
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (csp->csp_mode) {
case CSP_MODE_CIPHER:
case CSP_MODE_ETA:
if (!CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
cmd->base_masks |= HIFN_BASE_CMD_DECODE;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd->base_masks |= HIFN_BASE_CMD_CRYPT;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (csp->csp_cipher_alg) {
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
case CRYPTO_AES_CBC:
cmd->cry_masks |= HIFN_CRYPT_CMD_ALG_AES |
HIFN_CRYPT_CMD_MODE_CBC |
HIFN_CRYPT_CMD_NEW_IV;
break;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
default:
err = EINVAL;
goto errout;
}
Remove support for the ARC4 algorithm from OCF. There are no longer any in-kernel consumers. The software implementation was also a non-functional stub. Reviewed by: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D24771
2020-05-11 21:17:08 +00:00
crypto_read_iv(crp, cmd->iv);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (crp->crp_cipher_key != NULL)
cmd->ck = crp->crp_cipher_key;
else
cmd->ck = csp->csp_cipher_key;
cmd->cklen = csp->csp_cipher_klen;
Move session state from on-card memory to host memory so we no longer are constrained to a small number of sessions by the small on-card memories found in newer devices. This is really a stopgap solution as having session state in main memory incurs a (small but noticeable) performance penalty. The better solution is to manage session state so that it's cached on chip. Obtained from: openbsd
2004-10-15 00:36:07 +00:00
cmd->cry_masks |= HIFN_CRYPT_CMD_NEW_KEY;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
Add symmetric crypto support for the 7955 and 7956. Note performance is currently suboptimal. Submitted by: Rajesh Vaidyanath <RVaidyanath@hifn.com>
2003-10-08 20:25:47 +00:00
/*
* Need to specify the size for the AES key in the masks.
*/
if ((cmd->cry_masks & HIFN_CRYPT_CMD_ALG_MASK) ==
HIFN_CRYPT_CMD_ALG_AES) {
switch (cmd->cklen) {
case 16:
cmd->cry_masks |= HIFN_CRYPT_CMD_KSZ_128;
break;
case 24:
cmd->cry_masks |= HIFN_CRYPT_CMD_KSZ_192;
break;
case 32:
cmd->cry_masks |= HIFN_CRYPT_CMD_KSZ_256;
break;
default:
err = EINVAL;
goto errout;
}
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
break;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (csp->csp_mode) {
case CSP_MODE_DIGEST:
case CSP_MODE_ETA:
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd->base_masks |= HIFN_BASE_CMD_MAC;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
switch (csp->csp_auth_alg) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
case CRYPTO_SHA1:
cmd->mac_masks |= HIFN_MAC_CMD_ALG_SHA1 |
HIFN_MAC_CMD_RESULT | HIFN_MAC_CMD_MODE_HASH |
HIFN_MAC_CMD_POS_IPSEC;
break;
case CRYPTO_SHA1_HMAC:
cmd->mac_masks |= HIFN_MAC_CMD_ALG_SHA1 |
HIFN_MAC_CMD_RESULT | HIFN_MAC_CMD_MODE_HMAC |
HIFN_MAC_CMD_POS_IPSEC | HIFN_MAC_CMD_TRUNC;
break;
}
Remove MD5 HMAC from OCF. There are no in-kernel consumers. Reviewed by: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D24775
2020-05-11 22:08:08 +00:00
if (csp->csp_auth_alg == CRYPTO_SHA1_HMAC) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd->mac_masks |= HIFN_MAC_CMD_NEW_KEY;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (crp->crp_auth_key != NULL)
mackey = crp->crp_auth_key;
else
mackey = csp->csp_auth_key;
keylen = csp->csp_auth_klen;
bcopy(mackey, cmd->mac, keylen);
bzero(cmd->mac + keylen, HIFN_MAC_KEY_LENGTH - keylen);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
}
cmd->crp = crp;
OpenCrypto: Convert sessions to opaque handles instead of integers Track session objects in the framework, and pass handles between the framework (OCF), consumers, and drivers. Avoid redundancy and complexity in individual drivers by allocating session memory in the framework and providing it to drivers in ::newsession(). Session handles are no longer integers with information encoded in various high bits. Use of the CRYPTO_SESID2FOO() macros should be replaced with the appropriate crypto_ses2foo() function on the opaque session handle. Convert OCF drivers (in particular, cryptosoft, as well as myriad others) to the opaque handle interface. Discard existing session tracking as much as possible (quick pass). There may be additional code ripe for deletion. Convert OCF consumers (ipsec, geom_eli, krb5, cryptodev) to handle-style interface. The conversion is largely mechnical. The change is documented in crypto.9. Inspired by https://lists.freebsd.org/pipermail/freebsd-arch/2018-January/018835.html . No objection from: ae (ipsec portion) Reported by: jhb
2018-07-18 00:56:25 +00:00
cmd->session = ses;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
cmd->softc = sc;
err = hifn_crypto(sc, cmd, crp, hint);
if (!err) {
return 0;
} else if (err == ERESTART) {
/*
* There weren't enough resources to dispatch the request
* to the part. Notify the caller so they'll requeue this
* request and resubmit it again soon.
*/
#ifdef HIFN_DEBUG
if (hifn_debug)
device_printf(sc->sc_dev, "requeue request\n");
#endif
free(cmd, M_DEVBUF);
sc->sc_needwakeup |= CRYPTO_SYMQ;
return (err);
}
errout:
if (cmd != NULL)
free(cmd, M_DEVBUF);
if (err == EINVAL)
hifnstats.hst_invalid++;
else
hifnstats.hst_nomem++;
crp->crp_etype = err;
crypto_done(crp);
return (err);
}
static void
hifn_abort(struct hifn_softc *sc)
{
struct hifn_dma *dma = sc->sc_dma;
struct hifn_command *cmd;
struct cryptop *crp;
int i, u;
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
i = sc->sc_resk; u = sc->sc_resu;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
while (u != 0) {
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
cmd = sc->sc_hifn_commands[i];
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
KASSERT(cmd != NULL, ("hifn_abort: null command slot %u", i));
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_hifn_commands[i] = NULL;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
crp = cmd->crp;
if ((dma->resr[i].l & htole32(HIFN_D_VALID)) == 0) {
/* Salvage what we can. */
u_int8_t *macbuf;
if (cmd->base_masks & HIFN_BASE_CMD_MAC) {
macbuf = dma->result_bufs[i];
macbuf += 12;
} else
macbuf = NULL;
hifnstats.hst_opackets++;
hifn_callback(sc, cmd, macbuf);
} else {
if (cmd->src_map == cmd->dst_map) {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
} else {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_sync(sc->sc_dmat, cmd->dst_map,
BUS_DMASYNC_POSTREAD);
}
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (cmd->dst_m != NULL) {
m_freem(cmd->dst_m);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
/* non-shared buffers cannot be restarted */
if (cmd->src_map != cmd->dst_map) {
/*
* XXX should be EAGAIN, delayed until
* after the reset.
*/
crp->crp_etype = ENOMEM;
bus_dmamap_unload(sc->sc_dmat, cmd->dst_map);
bus_dmamap_destroy(sc->sc_dmat, cmd->dst_map);
} else
crp->crp_etype = ENOMEM;
bus_dmamap_unload(sc->sc_dmat, cmd->src_map);
bus_dmamap_destroy(sc->sc_dmat, cmd->src_map);
free(cmd, M_DEVBUF);
if (crp->crp_etype != EAGAIN)
crypto_done(crp);
}
if (++i == HIFN_D_RES_RSIZE)
i = 0;
u--;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_resk = i; sc->sc_resu = u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifn_reset_board(sc, 1);
hifn_init_dma(sc);
hifn_init_pci_registers(sc);
}
static void
hifn_callback(struct hifn_softc *sc, struct hifn_command *cmd, u_int8_t *macbuf)
{
struct hifn_dma *dma = sc->sc_dma;
struct cryptop *crp = cmd->crp;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
uint8_t macbuf2[SHA1_HASH_LEN];
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
struct mbuf *m;
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
int totlen, i, u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
if (cmd->src_map == cmd->dst_map) {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_POSTWRITE | BUS_DMASYNC_POSTREAD);
} else {
bus_dmamap_sync(sc->sc_dmat, cmd->src_map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_sync(sc->sc_dmat, cmd->dst_map,
BUS_DMASYNC_POSTREAD);
}
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (cmd->dst_m != NULL) {
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
totlen = cmd->src_mapsize;
for (m = cmd->dst_m; m != NULL; m = m->m_next) {
if (totlen < m->m_len) {
m->m_len = totlen;
totlen = 0;
} else
totlen -= m->m_len;
}
Add support for optional separate output buffers to in-kernel crypto. Some crypto consumers such as GELI and KTLS for file-backed sendfile need to store their output in a separate buffer from the input. Currently these consumers copy the contents of the input buffer into the output buffer and queue an in-place crypto operation on the output buffer. Using a separate output buffer avoids this copy. - Create a new 'struct crypto_buffer' describing a crypto buffer containing a type and type-specific fields. crp_ilen is gone, instead buffers that use a flat kernel buffer have a cb_buf_len field for their length. The length of other buffer types is inferred from the backing store (e.g. uio_resid for a uio). Requests now have two such structures: crp_buf for the input buffer, and crp_obuf for the output buffer. - Consumers now use helper functions (crypto_use_*, e.g. crypto_use_mbuf()) to configure the input buffer. If an output buffer is not configured, the request still modifies the input buffer in-place. A consumer uses a second set of helper functions (crypto_use_output_*) to configure an output buffer. - Consumers must request support for separate output buffers when creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are only permitted to queue a request with a separate output buffer on sessions with this flag set. Existing drivers already reject sessions with unknown flags, so this permits drivers to be modified to support this extension without requiring all drivers to change. - Several data-related functions now have matching versions that operate on an explicit buffer (e.g. crypto_apply_buf, crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf). - Most of the existing data-related functions operate on the input buffer. However crypto_copyback always writes to the output buffer if a request uses a separate output buffer. - For the regions in input/output buffers, the following conventions are followed: - AAD and IV are always present in input only and their fields are offsets into the input buffer. - payload is always present in both buffers. If a request uses a separate output buffer, it must set a new crp_payload_start_output field to the offset of the payload in the output buffer. - digest is in the input buffer for verify operations, and in the output buffer for compute operations. crp_digest_start is relative to the appropriate buffer. - Add a crypto buffer cursor abstraction. This is a more general form of some bits in the cryptosoft driver that tried to always use uio's. However, compared to the original code, this avoids rewalking the uio iovec array for requests with multiple vectors. It also avoids allocate an iovec array for mbufs and populating it by instead walking the mbuf chain directly. - Update the cryptosoft(4) driver to support separate output buffers making use of the cursor abstraction. Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 22:12:04 +00:00
cmd->dst_m->m_pkthdr.len =
crp->crp_buf.cb_mbuf->m_pkthdr.len;
m_freem(crp->crp_buf.cb_mbuf);
crp->crp_buf.cb_mbuf = cmd->dst_m;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
}
if (cmd->sloplen != 0) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
crypto_copyback(crp, cmd->src_mapsize - cmd->sloplen,
cmd->sloplen, &dma->slop[cmd->slopidx]);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
i = sc->sc_dstk; u = sc->sc_dstu;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
while (u != 0) {
if (i == HIFN_D_DST_RSIZE)
i = 0;
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
if (dma->dstr[i].l & htole32(HIFN_D_VALID)) {
bus_dmamap_sync(sc->sc_dmat, sc->sc_dmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
break;
}
i++, u--;
}
- Do not place variables to memory allocated by busdma. pre/post ops might invalidate cache and hence effectively void validity of values
2010-09-24 04:46:46 +00:00
sc->sc_dstk = i; sc->sc_dstu = u;
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
hifnstats.hst_obytes += cmd->dst_mapsize;
if (macbuf != NULL) {
Refactor driver and consumer interfaces for OCF (in-kernel crypto). - The linked list of cryptoini structures used in session initialization is replaced with a new flat structure: struct crypto_session_params. This session includes a new mode to define how the other fields should be interpreted. Available modes include: - COMPRESS (for compression/decompression) - CIPHER (for simply encryption/decryption) - DIGEST (computing and verifying digests) - AEAD (combined auth and encryption such as AES-GCM and AES-CCM) - ETA (combined auth and encryption using encrypt-then-authenticate) Additional modes could be added in the future (e.g. if we wanted to support TLS MtE for AES-CBC in the kernel we could add a new mode for that. TLS modes might also affect how AAD is interpreted, etc.) The flat structure also includes the key lengths and algorithms as before. However, code doesn't have to walk the linked list and switch on the algorithm to determine which key is the auth key vs encryption key. The 'csp_auth_*' fields are always used for auth keys and settings and 'csp_cipher_*' for cipher. (Compression algorithms are stored in csp_cipher_alg.) - Drivers no longer register a list of supported algorithms. This doesn't quite work when you factor in modes (e.g. a driver might support both AES-CBC and SHA2-256-HMAC separately but not combined for ETA). Instead, a new 'crypto_probesession' method has been added to the kobj interface for symmteric crypto drivers. This method returns a negative value on success (similar to how device_probe works) and the crypto framework uses this value to pick the "best" driver. There are three constants for hardware (e.g. ccr), accelerated software (e.g. aesni), and plain software (cryptosoft) that give preference in that order. One effect of this is that if you request only hardware when creating a new session, you will no longer get a session using accelerated software. Another effect is that the default setting to disallow software crypto via /dev/crypto now disables accelerated software. Once a driver is chosen, 'crypto_newsession' is invoked as before. - Crypto operations are now solely described by the flat 'cryptop' structure. The linked list of descriptors has been removed. A separate enum has been added to describe the type of data buffer in use instead of using CRYPTO_F_* flags to make it easier to add more types in the future if needed (e.g. wired userspace buffers for zero-copy). It will also make it easier to re-introduce separate input and output buffers (in-kernel TLS would benefit from this). Try to make the flags related to IV handling less insane: - CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv' member of the operation structure. If this flag is not set, the IV is stored in the data buffer at the 'crp_iv_start' offset. - CRYPTO_F_IV_GENERATE means that a random IV should be generated and stored into the data buffer. This cannot be used with CRYPTO_F_IV_SEPARATE. If a consumer wants to deal with explicit vs implicit IVs, etc. it can always generate the IV however it needs and store partial IVs in the buffer and the full IV/nonce in crp_iv and set CRYPTO_F_IV_SEPARATE. The layout of the buffer is now described via fields in cryptop. crp_aad_start and crp_aad_length define the boundaries of any AAD. Previously with GCM and CCM you defined an auth crd with this range, but for ETA your auth crd had to span both the AAD and plaintext (and they had to be adjacent). crp_payload_start and crp_payload_length define the boundaries of the plaintext/ciphertext. Modes that only do a single operation (COMPRESS, CIPHER, DIGEST) should only use this region and leave the AAD region empty. If a digest is present (or should be generated), it's starting location is marked by crp_digest_start. Instead of using the CRD_F_ENCRYPT flag to determine the direction of the operation, cryptop now includes an 'op' field defining the operation to perform. For digests I've added a new VERIFY digest mode which assumes a digest is present in the input and fails the request with EBADMSG if it doesn't match the internally-computed digest. GCM and CCM already assumed this, and the new AEAD mode requires this for decryption. The new ETA mode now also requires this for decryption, so IPsec and GELI no longer do their own authentication verification. Simple DIGEST operations can also do this, though there are no in-tree consumers. To eventually support some refcounting to close races, the session cookie is now passed to crypto_getop() and clients should no longer set crp_sesssion directly. - Assymteric crypto operation structures should be allocated via crypto_getkreq() and freed via crypto_freekreq(). This permits the crypto layer to track open asym requests and close races with a driver trying to unregister while asym requests are in flight. - crypto_copyback, crypto_copydata, crypto_apply, and crypto_contiguous_subsegment now accept the 'crp' object as the first parameter instead of individual members. This makes it easier to deal with different buffer types in the future as well as separate input and output buffers. It's also simpler for driver writers to use. - bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer. This understands the various types of buffers so that drivers that use DMA do not have to be aware of different buffer types. - Helper routines now exist to build an auth context for HMAC IPAD and OPAD. This reduces some duplicated work among drivers. - Key buffers are now treated as const throughout the framework and in device drivers. However, session key buffers provided when a session is created are expected to remain alive for the duration of the session. - GCM and CCM sessions now only specify a cipher algorithm and a cipher key. The redundant auth information is not needed or used. - For cryptosoft, split up the code a bit such that the 'process' callback now invokes a function pointer in the session. This function pointer is set based on the mode (in effect) though it simplifies a few edge cases that would otherwise be in the switch in 'process'. It does split up GCM vs CCM which I think is more readable even if there is some duplication. - I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC as an auth algorithm and updated cryptocheck to work with it. - Combined cipher and auth sessions via /dev/crypto now always use ETA mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored. This was actually documented as being true in crypto(4) before, but the code had not implemented this before I added the CIPHER_FIRST flag. - I have not yet updated /dev/crypto to be aware of explicit modes for sessions. I will probably do that at some point in the future as well as teach it about IV/nonce and tag lengths for AEAD so we can support all of the NIST KAT tests for GCM and CCM. - I've split up the exising crypto.9 manpage into several pages of which many are written from scratch. - I have converted all drivers and consumers in the tree and verified that they compile, but I have not tested all of them. I have tested the following drivers: - cryptosoft - aesni (AES only) - blake2 - ccr and the following consumers: - cryptodev - IPsec - ktls_ocf - GELI (lightly) I have not tested the following: - ccp - aesni with sha - hifn - kgssapi_krb5 - ubsec - padlock - safe - armv8_crypto (aarch64) - glxsb (i386) - sec (ppc) - cesa (armv7) - cryptocteon (mips64) - nlmsec (mips64) Discussed with: cem Relnotes: yes Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 18:25:23 +00:00
if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
crypto_copydata(crp, crp->crp_digest_start,
cmd->session->hs_mlen, macbuf2);
if (timingsafe_bcmp(macbuf, macbuf2,
cmd->session->hs_mlen) != 0)
crp->crp_etype = EBADMSG;
} else
crypto_copyback(crp, crp->crp_digest_start,
cmd->session->hs_mlen, macbuf);
Crypto device driver for hifn-based cards. Tested with 7951-, and 7811-based cards. Claimed to support many others. Obtained from: openbsd
2002-10-04 20:32:37 +00:00
}
if (cmd->src_map != cmd->dst_map) {
bus_dmamap_unload(sc->sc_dmat, cmd->dst_map);
bus_dmamap_destroy(sc->sc_dmat, cmd->dst_map);
}
bus_dmamap_unload(sc->sc_dmat, cmd->src_map);
bus_dmamap_destroy(sc->sc_dmat, cmd->src_map);
free(cmd, M_DEVBUF);
crypto_done(crp);
}
/*
* 7811 PB3 rev/2 parts lock-up on burst writes to Group 0
* and Group 1 registers; avoid conditions that could create
* burst writes by doing a read in between the writes.
*
* NB: The read we interpose is always to the same register;
* we do this because reading from an arbitrary (e.g. last)
* register may not always work.
*/
static void
hifn_write_reg_0(struct hifn_softc *sc, bus_size_t reg, u_int32_t val)
{
if (sc->sc_flags & HIFN_IS_7811) {
if (sc->sc_bar0_lastreg == reg - 4)
bus_space_read_4(sc->sc_st0, sc->sc_sh0, HIFN_0_PUCNFG);
sc->sc_bar0_lastreg = reg;
}
bus_space_write_4(sc->sc_st0, sc->sc_sh0, reg, val);
}
static void
hifn_write_reg_1(struct hifn_softc *sc, bus_size_t reg, u_int32_t val)
{
if (sc->sc_flags & HIFN_IS_7811) {
if (sc->sc_bar1_lastreg == reg - 4)
bus_space_read_4(sc->sc_st1, sc->sc_sh1, HIFN_1_REVID);
sc->sc_bar1_lastreg = reg;
}
bus_space_write_4(sc->sc_st1, sc->sc_sh1, reg, val);
}
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
#ifdef HIFN_VULCANDEV
/*
* this code provides support for mapping the PK engine's register
* into a userspace program.
*
*/
static int
Update d_mmap() to accept vm_ooffset_t and vm_memattr_t. This replaces d_mmap() with the d_mmap2() implementation and also changes the type of offset to vm_ooffset_t. Purge d_mmap2(). All driver modules will need to be rebuilt since D_VERSION is also bumped. Reviewed by: jhb@ MFC after: Not in this lifetime...
2009-12-29 21:51:28 +00:00
vulcanpk_mmap(struct cdev *dev, vm_ooffset_t offset,
vm_paddr_t *paddr, int nprot, vm_memattr_t *memattr)
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
{
struct hifn_softc *sc;
vm_paddr_t pd;
void *b;
sc = dev->si_drv1;
pd = rman_get_start(sc->sc_bar1res);
b = rman_get_virtual(sc->sc_bar1res);
#if 0
Update d_mmap() to accept vm_ooffset_t and vm_memattr_t. This replaces d_mmap() with the d_mmap2() implementation and also changes the type of offset to vm_ooffset_t. Purge d_mmap2(). All driver modules will need to be rebuilt since D_VERSION is also bumped. Reviewed by: jhb@ MFC after: Not in this lifetime...
2009-12-29 21:51:28 +00:00
printf("vpk mmap: %p(%016llx) offset=%lld\n", b,
(unsigned long long)pd, offset);
Overhaul driver/subsystem api's: o make all crypto drivers have a device_t; pseudo drivers like the s/w crypto driver synthesize one o change the api between the crypto subsystem and drivers to use kobj; cryptodev_if.m defines this api o use the fact that all crypto drivers now have a device_t to add support for specifying which of several potential devices to use when doing crypto operations o add new ioctls that allow user apps to select a specific crypto device to use (previous ioctls maintained for compatibility) o overhaul crypto subsystem code to eliminate lots of cruft and hide implementation details from drivers o bring in numerous fixes from Michale Richardson/hifn; mostly for 795x parts o add an optional mechanism for mmap'ing the hifn 795x public key h/w to user space for use by openssl (not enabled by default) o update crypto test tools to use new ioctl's and add cmd line options to specify a device to use for tests These changes will also enable much future work on improving the core crypto subsystem; including proper load balancing and interposing code between the core and drivers to dispatch small operations to the s/w driver as appropriate. These changes were instigated by the work of Michael Richardson. Reviewed by: pjd Approved by: re
2007-03-21 03:42:51 +00:00
hexdump(b, HIFN_1_PUB_MEMEND, "vpk", 0);
#endif
if (offset == 0) {
*paddr = pd;
return (0);
}
return (-1);
}
static struct cdevsw vulcanpk_cdevsw = {
.d_version = D_VERSION,
.d_mmap = vulcanpk_mmap,
.d_name = "vulcanpk",
};
#endif /* HIFN_VULCANDEV */
Reference in New Issue Copy Permalink