freebsd-dev/sys/kgssapi/krb5/krb5_mech.c

2130 lines
48 KiB
C
Raw Normal View History

Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* Copyright (c) 2008 Isilon Inc http://www.isilon.com/
* Authors: Doug Rabson <dfr@rabson.org>
* Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 OR CONTRIBUTORS 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.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/kobj.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <kgssapi/gssapi.h>
#include <kgssapi/gssapi_impl.h>
#include "kgss_if.h"
#include "kcrypto.h"
#define GSS_TOKEN_SENT_BY_ACCEPTOR 1
#define GSS_TOKEN_SEALED 2
#define GSS_TOKEN_ACCEPTOR_SUBKEY 4
static gss_OID_desc krb5_mech_oid =
{9, (void *) "\x2a\x86\x48\x86\xf7\x12\x01\x02\x02" };
struct krb5_data {
size_t kd_length;
void *kd_data;
};
struct krb5_keyblock {
uint16_t kk_type; /* encryption type */
struct krb5_data kk_key; /* key data */
};
struct krb5_address {
uint16_t ka_type;
struct krb5_data ka_addr;
};
/*
* The km_elem array is ordered so that the highest received sequence
* number is listed first.
*/
struct krb5_msg_order {
uint32_t km_flags;
uint32_t km_start;
uint32_t km_length;
uint32_t km_jitter_window;
uint32_t km_first_seq;
uint32_t *km_elem;
};
struct krb5_context {
struct _gss_ctx_id_t kc_common;
struct mtx kc_lock;
uint32_t kc_ac_flags;
uint32_t kc_ctx_flags;
uint32_t kc_more_flags;
#define LOCAL 1
#define OPEN 2
#define COMPAT_OLD_DES3 4
#define COMPAT_OLD_DES3_SELECTED 8
#define ACCEPTOR_SUBKEY 16
struct krb5_address kc_local_address;
struct krb5_address kc_remote_address;
uint16_t kc_local_port;
uint16_t kc_remote_port;
struct krb5_keyblock kc_keyblock;
struct krb5_keyblock kc_local_subkey;
struct krb5_keyblock kc_remote_subkey;
volatile uint32_t kc_local_seqnumber;
uint32_t kc_remote_seqnumber;
uint32_t kc_keytype;
uint32_t kc_cksumtype;
struct krb5_data kc_source_name;
struct krb5_data kc_target_name;
uint32_t kc_lifetime;
struct krb5_msg_order kc_msg_order;
struct krb5_key_state *kc_tokenkey;
struct krb5_key_state *kc_encryptkey;
struct krb5_key_state *kc_checksumkey;
struct krb5_key_state *kc_send_seal_Ke;
struct krb5_key_state *kc_send_seal_Ki;
struct krb5_key_state *kc_send_seal_Kc;
struct krb5_key_state *kc_send_sign_Kc;
struct krb5_key_state *kc_recv_seal_Ke;
struct krb5_key_state *kc_recv_seal_Ki;
struct krb5_key_state *kc_recv_seal_Kc;
struct krb5_key_state *kc_recv_sign_Kc;
};
static uint16_t
get_uint16(const uint8_t **pp, size_t *lenp)
{
const uint8_t *p = *pp;
uint16_t v;
if (*lenp < 2)
return (0);
v = (p[0] << 8) | p[1];
*pp = p + 2;
*lenp = *lenp - 2;
return (v);
}
static uint32_t
get_uint32(const uint8_t **pp, size_t *lenp)
{
const uint8_t *p = *pp;
uint32_t v;
if (*lenp < 4)
return (0);
v = (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
*pp = p + 4;
*lenp = *lenp - 4;
return (v);
}
static void
get_data(const uint8_t **pp, size_t *lenp, struct krb5_data *dp)
{
size_t sz = get_uint32(pp, lenp);
dp->kd_length = sz;
dp->kd_data = malloc(sz, M_GSSAPI, M_WAITOK);
if (*lenp < sz)
sz = *lenp;
bcopy(*pp, dp->kd_data, sz);
(*pp) += sz;
(*lenp) -= sz;
}
static void
delete_data(struct krb5_data *dp)
{
if (dp->kd_data) {
free(dp->kd_data, M_GSSAPI);
dp->kd_length = 0;
dp->kd_data = NULL;
}
}
static void
get_address(const uint8_t **pp, size_t *lenp, struct krb5_address *ka)
{
ka->ka_type = get_uint16(pp, lenp);
get_data(pp, lenp, &ka->ka_addr);
}
static void
delete_address(struct krb5_address *ka)
{
delete_data(&ka->ka_addr);
}
static void
get_keyblock(const uint8_t **pp, size_t *lenp, struct krb5_keyblock *kk)
{
kk->kk_type = get_uint16(pp, lenp);
get_data(pp, lenp, &kk->kk_key);
}
static void
delete_keyblock(struct krb5_keyblock *kk)
{
if (kk->kk_key.kd_data)
bzero(kk->kk_key.kd_data, kk->kk_key.kd_length);
delete_data(&kk->kk_key);
}
static void
copy_key(struct krb5_keyblock *from, struct krb5_keyblock **to)
{
if (from->kk_key.kd_length)
*to = from;
else
*to = NULL;
}
/*
* Return non-zero if we are initiator.
*/
static __inline int
is_initiator(struct krb5_context *kc)
{
return (kc->kc_more_flags & LOCAL);
}
/*
* Return non-zero if we are acceptor.
*/
static __inline int
is_acceptor(struct krb5_context *kc)
{
return !(kc->kc_more_flags & LOCAL);
}
static void
get_initiator_subkey(struct krb5_context *kc, struct krb5_keyblock **kdp)
{
if (is_initiator(kc))
copy_key(&kc->kc_local_subkey, kdp);
else
copy_key(&kc->kc_remote_subkey, kdp);
if (!*kdp)
copy_key(&kc->kc_keyblock, kdp);
}
static void
get_acceptor_subkey(struct krb5_context *kc, struct krb5_keyblock **kdp)
{
if (is_initiator(kc))
copy_key(&kc->kc_remote_subkey, kdp);
else
copy_key(&kc->kc_local_subkey, kdp);
}
static OM_uint32
get_keys(struct krb5_context *kc)
{
struct krb5_keyblock *keydata;
struct krb5_encryption_class *ec;
struct krb5_key_state *key;
int etype;
keydata = NULL;
get_acceptor_subkey(kc, &keydata);
if (!keydata)
if ((kc->kc_more_flags & ACCEPTOR_SUBKEY) == 0)
get_initiator_subkey(kc, &keydata);
if (!keydata)
return (GSS_S_FAILURE);
/*
* GSS-API treats all DES etypes the same and all DES3 etypes
* the same.
*/
switch (keydata->kk_type) {
case ETYPE_DES_CBC_CRC:
case ETYPE_DES_CBC_MD4:
case ETYPE_DES_CBC_MD5:
etype = ETYPE_DES_CBC_CRC;
break;
case ETYPE_DES3_CBC_MD5:
case ETYPE_DES3_CBC_SHA1:
case ETYPE_OLD_DES3_CBC_SHA1:
etype = ETYPE_DES3_CBC_SHA1;
break;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
default:
etype = keydata->kk_type;
}
ec = krb5_find_encryption_class(etype);
if (!ec)
return (GSS_S_FAILURE);
key = krb5_create_key(ec);
krb5_set_key(key, keydata->kk_key.kd_data);
kc->kc_tokenkey = key;
switch (etype) {
case ETYPE_DES_CBC_CRC:
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56: {
/*
* Single DES and ARCFOUR uses a 'derived' key (XOR
* with 0xf0) for encrypting wrap tokens. The original
* key is used for checksums and sequence numbers.
*/
struct krb5_key_state *ekey;
uint8_t *ekp, *kp;
int i;
ekey = krb5_create_key(ec);
ekp = ekey->ks_key;
kp = key->ks_key;
for (i = 0; i < ec->ec_keylen; i++)
ekp[i] = kp[i] ^ 0xf0;
krb5_set_key(ekey, ekp);
kc->kc_encryptkey = ekey;
refcount_acquire(&key->ks_refs);
kc->kc_checksumkey = key;
break;
}
case ETYPE_DES3_CBC_SHA1:
/*
* Triple DES uses a RFC 3961 style derived key with
* usage number KG_USAGE_SIGN for checksums. The
* original key is used for encryption and sequence
* numbers.
*/
kc->kc_checksumkey = krb5_get_checksum_key(key, KG_USAGE_SIGN);
refcount_acquire(&key->ks_refs);
kc->kc_encryptkey = key;
break;
default:
/*
* We need eight derived keys four for sending and
* four for receiving.
*/
if (is_initiator(kc)) {
/*
* We are initiator.
*/
kc->kc_send_seal_Ke = krb5_get_encryption_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_send_seal_Ki = krb5_get_integrity_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_send_seal_Kc = krb5_get_checksum_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_send_sign_Kc = krb5_get_checksum_key(key,
KG_USAGE_INITIATOR_SIGN);
kc->kc_recv_seal_Ke = krb5_get_encryption_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_recv_seal_Ki = krb5_get_integrity_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_recv_seal_Kc = krb5_get_checksum_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_recv_sign_Kc = krb5_get_checksum_key(key,
KG_USAGE_ACCEPTOR_SIGN);
} else {
/*
* We are acceptor.
*/
kc->kc_send_seal_Ke = krb5_get_encryption_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_send_seal_Ki = krb5_get_integrity_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_send_seal_Kc = krb5_get_checksum_key(key,
KG_USAGE_ACCEPTOR_SEAL);
kc->kc_send_sign_Kc = krb5_get_checksum_key(key,
KG_USAGE_ACCEPTOR_SIGN);
kc->kc_recv_seal_Ke = krb5_get_encryption_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_recv_seal_Ki = krb5_get_integrity_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_recv_seal_Kc = krb5_get_checksum_key(key,
KG_USAGE_INITIATOR_SEAL);
kc->kc_recv_sign_Kc = krb5_get_checksum_key(key,
KG_USAGE_INITIATOR_SIGN);
}
break;
}
return (GSS_S_COMPLETE);
}
static void
krb5_init(gss_ctx_id_t ctx)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
mtx_init(&kc->kc_lock, "krb5 gss lock", NULL, MTX_DEF);
}
static OM_uint32
krb5_import(gss_ctx_id_t ctx,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
enum sec_context_format format,
const gss_buffer_t context_token)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
OM_uint32 res;
const uint8_t *p = (const uint8_t *) context_token->value;
size_t len = context_token->length;
uint32_t flags;
int i;
/*
* We support heimdal 0.6 and heimdal 1.1
*/
if (format != KGSS_HEIMDAL_0_6 && format != KGSS_HEIMDAL_1_1)
return (GSS_S_DEFECTIVE_TOKEN);
#define SC_LOCAL_ADDRESS 1
#define SC_REMOTE_ADDRESS 2
#define SC_KEYBLOCK 4
#define SC_LOCAL_SUBKEY 8
#define SC_REMOTE_SUBKEY 16
/*
* Ensure that the token starts with krb5 oid.
*/
if (p[0] != 0x00 || p[1] != krb5_mech_oid.length
|| len < krb5_mech_oid.length + 2
|| bcmp(krb5_mech_oid.elements, p + 2,
krb5_mech_oid.length))
return (GSS_S_DEFECTIVE_TOKEN);
p += krb5_mech_oid.length + 2;
len -= krb5_mech_oid.length + 2;
flags = get_uint32(&p, &len);
kc->kc_ac_flags = get_uint32(&p, &len);
if (flags & SC_LOCAL_ADDRESS)
get_address(&p, &len, &kc->kc_local_address);
if (flags & SC_REMOTE_ADDRESS)
get_address(&p, &len, &kc->kc_remote_address);
kc->kc_local_port = get_uint16(&p, &len);
kc->kc_remote_port = get_uint16(&p, &len);
if (flags & SC_KEYBLOCK)
get_keyblock(&p, &len, &kc->kc_keyblock);
if (flags & SC_LOCAL_SUBKEY)
get_keyblock(&p, &len, &kc->kc_local_subkey);
if (flags & SC_REMOTE_SUBKEY)
get_keyblock(&p, &len, &kc->kc_remote_subkey);
kc->kc_local_seqnumber = get_uint32(&p, &len);
kc->kc_remote_seqnumber = get_uint32(&p, &len);
kc->kc_keytype = get_uint32(&p, &len);
kc->kc_cksumtype = get_uint32(&p, &len);
get_data(&p, &len, &kc->kc_source_name);
get_data(&p, &len, &kc->kc_target_name);
kc->kc_ctx_flags = get_uint32(&p, &len);
kc->kc_more_flags = get_uint32(&p, &len);
kc->kc_lifetime = get_uint32(&p, &len);
/*
* Heimdal 1.1 adds the message order stuff.
*/
if (format == KGSS_HEIMDAL_1_1) {
kc->kc_msg_order.km_flags = get_uint32(&p, &len);
kc->kc_msg_order.km_start = get_uint32(&p, &len);
kc->kc_msg_order.km_length = get_uint32(&p, &len);
kc->kc_msg_order.km_jitter_window = get_uint32(&p, &len);
kc->kc_msg_order.km_first_seq = get_uint32(&p, &len);
kc->kc_msg_order.km_elem =
malloc(kc->kc_msg_order.km_jitter_window * sizeof(uint32_t),
M_GSSAPI, M_WAITOK);
for (i = 0; i < kc->kc_msg_order.km_jitter_window; i++)
kc->kc_msg_order.km_elem[i] = get_uint32(&p, &len);
} else {
kc->kc_msg_order.km_flags = 0;
}
res = get_keys(kc);
if (GSS_ERROR(res))
return (res);
/*
* We don't need these anymore.
*/
delete_keyblock(&kc->kc_keyblock);
delete_keyblock(&kc->kc_local_subkey);
delete_keyblock(&kc->kc_remote_subkey);
return (GSS_S_COMPLETE);
}
static void
krb5_delete(gss_ctx_id_t ctx, gss_buffer_t output_token)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
delete_address(&kc->kc_local_address);
delete_address(&kc->kc_remote_address);
delete_keyblock(&kc->kc_keyblock);
delete_keyblock(&kc->kc_local_subkey);
delete_keyblock(&kc->kc_remote_subkey);
delete_data(&kc->kc_source_name);
delete_data(&kc->kc_target_name);
if (kc->kc_msg_order.km_elem)
free(kc->kc_msg_order.km_elem, M_GSSAPI);
if (output_token) {
output_token->length = 0;
output_token->value = NULL;
}
if (kc->kc_tokenkey) {
krb5_free_key(kc->kc_tokenkey);
if (kc->kc_encryptkey) {
krb5_free_key(kc->kc_encryptkey);
krb5_free_key(kc->kc_checksumkey);
} else {
krb5_free_key(kc->kc_send_seal_Ke);
krb5_free_key(kc->kc_send_seal_Ki);
krb5_free_key(kc->kc_send_seal_Kc);
krb5_free_key(kc->kc_send_sign_Kc);
krb5_free_key(kc->kc_recv_seal_Ke);
krb5_free_key(kc->kc_recv_seal_Ki);
krb5_free_key(kc->kc_recv_seal_Kc);
krb5_free_key(kc->kc_recv_sign_Kc);
}
}
mtx_destroy(&kc->kc_lock);
}
static gss_OID
krb5_mech_type(gss_ctx_id_t ctx)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
return (&krb5_mech_oid);
}
/*
* Make a token with the given type and length (the length includes
* the TOK_ID), initialising the token header appropriately. Return a
* pointer to the TOK_ID of the token. A new mbuf is allocated with
* the framing header plus hlen bytes of space.
*
* Format is as follows:
*
* 0x60 [APPLICATION 0] SEQUENCE
* DER encoded length length of oid + type + inner token length
* 0x06 NN <oid data> OID of mechanism type
* TT TT TOK_ID
* <inner token> data for inner token
*
* 1: der encoded length
*/
static void *
krb5_make_token(char tok_id[2], size_t hlen, size_t len, struct mbuf **mp)
{
size_t inside_len, len_len, tlen;
gss_OID oid = &krb5_mech_oid;
struct mbuf *m;
uint8_t *p;
inside_len = 2 + oid->length + len;
if (inside_len < 128)
len_len = 1;
else if (inside_len < 0x100)
len_len = 2;
else if (inside_len < 0x10000)
len_len = 3;
else if (inside_len < 0x1000000)
len_len = 4;
else
len_len = 5;
tlen = 1 + len_len + 2 + oid->length + hlen;
KASSERT(tlen <= MLEN, ("token head too large"));
MGET(m, M_WAITOK, MT_DATA);
M_ALIGN(m, tlen);
m->m_len = tlen;
p = (uint8_t *) m->m_data;
*p++ = 0x60;
switch (len_len) {
case 1:
*p++ = inside_len;
break;
case 2:
*p++ = 0x81;
*p++ = inside_len;
break;
case 3:
*p++ = 0x82;
*p++ = inside_len >> 8;
*p++ = inside_len;
break;
case 4:
*p++ = 0x83;
*p++ = inside_len >> 16;
*p++ = inside_len >> 8;
*p++ = inside_len;
break;
case 5:
*p++ = 0x84;
*p++ = inside_len >> 24;
*p++ = inside_len >> 16;
*p++ = inside_len >> 8;
*p++ = inside_len;
break;
}
*p++ = 0x06;
*p++ = oid->length;
bcopy(oid->elements, p, oid->length);
p += oid->length;
p[0] = tok_id[0];
p[1] = tok_id[1];
*mp = m;
return (p);
}
/*
* Verify a token, checking the inner token length and mechanism oid.
* pointer to the first byte of the TOK_ID. The length of the
* encapsulated data is checked to be at least len bytes; the actual
* length of the encapsulated data (including TOK_ID) is returned in
* *encap_len.
*
* If can_pullup is TRUE and the token header is fragmented, we will
* rearrange it.
*
* Format is as follows:
*
* 0x60 [APPLICATION 0] SEQUENCE
* DER encoded length length of oid + type + inner token length
* 0x06 NN <oid data> OID of mechanism type
* TT TT TOK_ID
* <inner token> data for inner token
*
* 1: der encoded length
*/
static void *
krb5_verify_token(char tok_id[2], size_t len, struct mbuf **mp,
size_t *encap_len, bool_t can_pullup)
{
struct mbuf *m;
size_t tlen, hlen, len_len, inside_len;
gss_OID oid = &krb5_mech_oid;
uint8_t *p;
m = *mp;
tlen = m_length(m, NULL);
if (tlen < 2)
return (NULL);
/*
* Ensure that at least the framing part of the token is
* contigous.
*/
if (m->m_len < 2) {
if (can_pullup)
*mp = m = m_pullup(m, 2);
else
return (NULL);
}
p = m->m_data;
if (*p++ != 0x60)
return (NULL);
if (*p < 0x80) {
inside_len = *p++;
len_len = 1;
} else {
/*
* Ensure there is enough space for the DER encoded length.
*/
len_len = (*p & 0x7f) + 1;
if (tlen < len_len + 1)
return (NULL);
if (m->m_len < len_len + 1) {
if (can_pullup)
*mp = m = m_pullup(m, len_len + 1);
else
return (NULL);
p = m->m_data + 1;
}
switch (*p++) {
case 0x81:
inside_len = *p++;
break;
case 0x82:
inside_len = (p[0] << 8) | p[1];
p += 2;
break;
case 0x83:
inside_len = (p[0] << 16) | (p[1] << 8) | p[2];
p += 3;
break;
case 0x84:
inside_len = (p[0] << 24) | (p[1] << 16)
| (p[2] << 8) | p[3];
p += 4;
break;
default:
return (NULL);
}
}
if (tlen != inside_len + len_len + 1)
return (NULL);
if (inside_len < 2 + oid->length + len)
return (NULL);
/*
* Now that we know the value of len_len, we can pullup the
* whole header. The header is 1 + len_len + 2 + oid->length +
* len bytes.
*/
hlen = 1 + len_len + 2 + oid->length + len;
if (m->m_len < hlen) {
if (can_pullup)
*mp = m = m_pullup(m, hlen);
else
return (NULL);
p = m->m_data + 1 + len_len;
}
if (*p++ != 0x06)
return (NULL);
if (*p++ != oid->length)
return (NULL);
if (bcmp(oid->elements, p, oid->length))
return (NULL);
p += oid->length;
if (p[0] != tok_id[0])
return (NULL);
if (p[1] != tok_id[1])
return (NULL);
*encap_len = inside_len - 2 - oid->length;
return (p);
}
static void
krb5_insert_seq(struct krb5_msg_order *mo, uint32_t seq, int index)
{
int i;
if (mo->km_length < mo->km_jitter_window)
mo->km_length++;
for (i = mo->km_length - 1; i > index; i--)
mo->km_elem[i] = mo->km_elem[i - 1];
mo->km_elem[index] = seq;
}
/*
* Check sequence numbers according to RFC 2743 section 1.2.3.
*/
static OM_uint32
krb5_sequence_check(struct krb5_context *kc, uint32_t seq)
{
OM_uint32 res = GSS_S_FAILURE;
struct krb5_msg_order *mo = &kc->kc_msg_order;
int check_sequence = mo->km_flags & GSS_C_SEQUENCE_FLAG;
int check_replay = mo->km_flags & GSS_C_REPLAY_FLAG;
int i;
mtx_lock(&kc->kc_lock);
/*
* Message is in-sequence with no gap.
*/
if (mo->km_length == 0 || seq == mo->km_elem[0] + 1) {
/*
* This message is received in-sequence with no gaps.
*/
krb5_insert_seq(mo, seq, 0);
res = GSS_S_COMPLETE;
goto out;
}
if (seq > mo->km_elem[0]) {
/*
* This message is received in-sequence with a gap.
*/
krb5_insert_seq(mo, seq, 0);
if (check_sequence)
res = GSS_S_GAP_TOKEN;
else
res = GSS_S_COMPLETE;
goto out;
}
if (seq < mo->km_elem[mo->km_length - 1]) {
if (check_replay && !check_sequence)
res = GSS_S_OLD_TOKEN;
else
res = GSS_S_UNSEQ_TOKEN;
goto out;
}
for (i = 0; i < mo->km_length; i++) {
if (mo->km_elem[i] == seq) {
res = GSS_S_DUPLICATE_TOKEN;
goto out;
}
if (mo->km_elem[i] < seq) {
/*
* We need to insert this seq here,
*/
krb5_insert_seq(mo, seq, i);
if (check_replay && !check_sequence)
res = GSS_S_COMPLETE;
else
res = GSS_S_UNSEQ_TOKEN;
goto out;
}
}
out:
mtx_unlock(&kc->kc_lock);
return (res);
}
static uint8_t sgn_alg_des_md5[] = { 0x00, 0x00 };
static uint8_t seal_alg_des[] = { 0x00, 0x00 };
static uint8_t sgn_alg_des3_sha1[] = { 0x04, 0x00 };
static uint8_t seal_alg_des3[] = { 0x02, 0x00 };
static uint8_t seal_alg_rc4[] = { 0x10, 0x00 };
static uint8_t sgn_alg_hmac_md5[] = { 0x11, 0x00 };
/*
* Return the size of the inner token given the use of the key's
* encryption class. For wrap tokens, the length of the padded
* plaintext will be added to this.
*/
static size_t
token_length(struct krb5_key_state *key)
{
return (16 + key->ks_class->ec_checksumlen);
}
static OM_uint32
krb5_get_mic_old(struct krb5_context *kc, struct mbuf *m,
struct mbuf **micp, uint8_t sgn_alg[2])
{
struct mbuf *mlast, *mic, *tm;
uint8_t *p, dir;
size_t tlen, mlen, cklen;
uint32_t seq;
char buf[8];
mlen = m_length(m, &mlast);
tlen = token_length(kc->kc_tokenkey);
p = krb5_make_token("\x01\x01", tlen, tlen, &mic);
p += 2; /* TOK_ID */
*p++ = sgn_alg[0]; /* SGN_ALG */
*p++ = sgn_alg[1];
*p++ = 0xff; /* filler */
*p++ = 0xff;
*p++ = 0xff;
*p++ = 0xff;
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the token header plus the
* message.
*/
cklen = kc->kc_checksumkey->ks_class->ec_checksumlen;
mic->m_len = p - (uint8_t *) mic->m_data;
mic->m_next = m;
MGET(tm, M_WAITOK, MT_DATA);
tm->m_len = cklen;
mlast->m_next = tm;
krb5_checksum(kc->kc_checksumkey, 15, mic, mic->m_len - 8,
8 + mlen, cklen);
bcopy(tm->m_data, p + 8, cklen);
mic->m_next = NULL;
mlast->m_next = NULL;
m_free(tm);
/*
* SND_SEQ:
*
* Take the four bytes of the sequence number least
* significant first followed by four bytes of direction
* marker (zero for initiator and 0xff for acceptor). Encrypt
* that data using the SGN_CKSUM as IV. Note: ARC4 wants the
* sequence number big-endian.
*/
seq = atomic_fetchadd_32(&kc->kc_local_seqnumber, 1);
if (sgn_alg[0] == 0x11) {
p[0] = (seq >> 24);
p[1] = (seq >> 16);
p[2] = (seq >> 8);
p[3] = (seq >> 0);
} else {
p[0] = (seq >> 0);
p[1] = (seq >> 8);
p[2] = (seq >> 16);
p[3] = (seq >> 24);
}
if (is_initiator(kc)) {
dir = 0;
} else {
dir = 0xff;
}
p[4] = dir;
p[5] = dir;
p[6] = dir;
p[7] = dir;
bcopy(p + 8, buf, 8);
/*
* Set the mic buffer to its final size so that the encrypt
* can see the SND_SEQ part.
*/
mic->m_len += 8 + cklen;
krb5_encrypt(kc->kc_tokenkey, mic, mic->m_len - cklen - 8, 8, buf, 8);
*micp = mic;
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_get_mic_new(struct krb5_context *kc, struct mbuf *m,
struct mbuf **micp)
{
struct krb5_key_state *key = kc->kc_send_sign_Kc;
struct mbuf *mlast, *mic;
uint8_t *p;
int flags;
size_t mlen, cklen;
uint32_t seq;
mlen = m_length(m, &mlast);
cklen = key->ks_class->ec_checksumlen;
KASSERT(16 + cklen <= MLEN, ("checksum too large for an mbuf"));
MGET(mic, M_WAITOK, MT_DATA);
M_ALIGN(mic, 16 + cklen);
mic->m_len = 16 + cklen;
p = mic->m_data;
/* TOK_ID */
p[0] = 0x04;
p[1] = 0x04;
/* Flags */
flags = 0;
if (is_acceptor(kc))
flags |= GSS_TOKEN_SENT_BY_ACCEPTOR;
if (kc->kc_more_flags & ACCEPTOR_SUBKEY)
flags |= GSS_TOKEN_ACCEPTOR_SUBKEY;
p[2] = flags;
/* Filler */
p[3] = 0xff;
p[4] = 0xff;
p[5] = 0xff;
p[6] = 0xff;
p[7] = 0xff;
/* SND_SEQ */
p[8] = 0;
p[9] = 0;
p[10] = 0;
p[11] = 0;
seq = atomic_fetchadd_32(&kc->kc_local_seqnumber, 1);
p[12] = (seq >> 24);
p[13] = (seq >> 16);
p[14] = (seq >> 8);
p[15] = (seq >> 0);
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the message plus the first
* 16 bytes of the token header.
*/
mlast->m_next = mic;
krb5_checksum(key, 0, m, 0, mlen + 16, cklen);
mlast->m_next = NULL;
*micp = mic;
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_get_mic(gss_ctx_id_t ctx, OM_uint32 *minor_status,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
gss_qop_t qop_req, struct mbuf *m, struct mbuf **micp)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*minor_status = 0;
if (qop_req != GSS_C_QOP_DEFAULT)
return (GSS_S_BAD_QOP);
if (time_uptime > kc->kc_lifetime)
return (GSS_S_CONTEXT_EXPIRED);
switch (kc->kc_tokenkey->ks_class->ec_type) {
case ETYPE_DES_CBC_CRC:
return (krb5_get_mic_old(kc, m, micp, sgn_alg_des_md5));
case ETYPE_DES3_CBC_SHA1:
return (krb5_get_mic_old(kc, m, micp, sgn_alg_des3_sha1));
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56:
return (krb5_get_mic_old(kc, m, micp, sgn_alg_hmac_md5));
default:
return (krb5_get_mic_new(kc, m, micp));
}
return (GSS_S_FAILURE);
}
static OM_uint32
krb5_verify_mic_old(struct krb5_context *kc, struct mbuf *m, struct mbuf *mic,
uint8_t sgn_alg[2])
{
struct mbuf *mlast, *tm;
uint8_t *p, *tp, dir;
size_t mlen, tlen, elen, miclen;
size_t cklen;
uint32_t seq;
mlen = m_length(m, &mlast);
tlen = token_length(kc->kc_tokenkey);
p = krb5_verify_token("\x01\x01", tlen, &mic, &elen, FALSE);
if (!p)
return (GSS_S_DEFECTIVE_TOKEN);
#if 0
/*
* Disable this check - heimdal-1.1 generates DES3 MIC tokens
* that are 2 bytes too big.
*/
if (elen != tlen)
return (GSS_S_DEFECTIVE_TOKEN);
#endif
/* TOK_ID */
p += 2;
/* SGN_ALG */
if (p[0] != sgn_alg[0] || p[1] != sgn_alg[1])
return (GSS_S_DEFECTIVE_TOKEN);
p += 2;
if (p[0] != 0xff || p[1] != 0xff || p[2] != 0xff || p[3] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
p += 4;
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the token header plus the
* message.
*/
cklen = kc->kc_checksumkey->ks_class->ec_checksumlen;
miclen = mic->m_len;
mic->m_len = p - (uint8_t *) mic->m_data;
mic->m_next = m;
MGET(tm, M_WAITOK, MT_DATA);
tm->m_len = cklen;
mlast->m_next = tm;
krb5_checksum(kc->kc_checksumkey, 15, mic, mic->m_len - 8,
8 + mlen, cklen);
mic->m_next = NULL;
mlast->m_next = NULL;
if (bcmp(tm->m_data, p + 8, cklen)) {
m_free(tm);
return (GSS_S_BAD_SIG);
}
/*
* SND_SEQ:
*
* Take the four bytes of the sequence number least
* significant first followed by four bytes of direction
* marker (zero for initiator and 0xff for acceptor). Encrypt
* that data using the SGN_CKSUM as IV. Note: ARC4 wants the
* sequence number big-endian.
*/
bcopy(p, tm->m_data, 8);
tm->m_len = 8;
krb5_decrypt(kc->kc_tokenkey, tm, 0, 8, p + 8, 8);
tp = tm->m_data;
if (sgn_alg[0] == 0x11) {
seq = tp[3] | (tp[2] << 8) | (tp[1] << 16) | (tp[0] << 24);
} else {
seq = tp[0] | (tp[1] << 8) | (tp[2] << 16) | (tp[3] << 24);
}
if (is_initiator(kc)) {
dir = 0xff;
} else {
dir = 0;
}
if (tp[4] != dir || tp[5] != dir || tp[6] != dir || tp[7] != dir) {
m_free(tm);
return (GSS_S_DEFECTIVE_TOKEN);
}
m_free(tm);
if (kc->kc_msg_order.km_flags &
(GSS_C_REPLAY_FLAG | GSS_C_SEQUENCE_FLAG)) {
return (krb5_sequence_check(kc, seq));
}
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_verify_mic_new(struct krb5_context *kc, struct mbuf *m, struct mbuf *mic)
{
OM_uint32 res;
struct krb5_key_state *key = kc->kc_recv_sign_Kc;
struct mbuf *mlast;
uint8_t *p;
int flags;
size_t mlen, cklen;
char buf[32];
mlen = m_length(m, &mlast);
cklen = key->ks_class->ec_checksumlen;
KASSERT(mic->m_next == NULL, ("MIC should be contiguous"));
if (mic->m_len != 16 + cklen)
return (GSS_S_DEFECTIVE_TOKEN);
p = mic->m_data;
/* TOK_ID */
if (p[0] != 0x04)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[1] != 0x04)
return (GSS_S_DEFECTIVE_TOKEN);
/* Flags */
flags = 0;
if (is_initiator(kc))
flags |= GSS_TOKEN_SENT_BY_ACCEPTOR;
if (kc->kc_more_flags & ACCEPTOR_SUBKEY)
flags |= GSS_TOKEN_ACCEPTOR_SUBKEY;
if (p[2] != flags)
return (GSS_S_DEFECTIVE_TOKEN);
/* Filler */
if (p[3] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[4] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[5] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[6] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[7] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
/* SND_SEQ */
if (kc->kc_msg_order.km_flags &
(GSS_C_REPLAY_FLAG | GSS_C_SEQUENCE_FLAG)) {
uint32_t seq;
if (p[8] || p[9] || p[10] || p[11]) {
res = GSS_S_UNSEQ_TOKEN;
} else {
seq = (p[12] << 24) | (p[13] << 16)
| (p[14] << 8) | p[15];
res = krb5_sequence_check(kc, seq);
}
if (GSS_ERROR(res))
return (res);
} else {
res = GSS_S_COMPLETE;
}
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the message plus the first
* 16 bytes of the token header.
*/
m_copydata(mic, 16, cklen, buf);
mlast->m_next = mic;
krb5_checksum(key, 0, m, 0, mlen + 16, cklen);
mlast->m_next = NULL;
if (bcmp(buf, p + 16, cklen)) {
return (GSS_S_BAD_SIG);
}
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_verify_mic(gss_ctx_id_t ctx, OM_uint32 *minor_status,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct mbuf *m, struct mbuf *mic, gss_qop_t *qop_state)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*minor_status = 0;
if (qop_state)
*qop_state = GSS_C_QOP_DEFAULT;
if (time_uptime > kc->kc_lifetime)
return (GSS_S_CONTEXT_EXPIRED);
switch (kc->kc_tokenkey->ks_class->ec_type) {
case ETYPE_DES_CBC_CRC:
return (krb5_verify_mic_old(kc, m, mic, sgn_alg_des_md5));
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56:
return (krb5_verify_mic_old(kc, m, mic, sgn_alg_hmac_md5));
case ETYPE_DES3_CBC_SHA1:
return (krb5_verify_mic_old(kc, m, mic, sgn_alg_des3_sha1));
default:
return (krb5_verify_mic_new(kc, m, mic));
}
return (GSS_S_FAILURE);
}
static OM_uint32
krb5_wrap_old(struct krb5_context *kc, int conf_req_flag,
struct mbuf **mp, int *conf_state,
uint8_t sgn_alg[2], uint8_t seal_alg[2])
{
struct mbuf *m, *mlast, *tm, *cm, *pm;
size_t mlen, tlen, padlen, datalen;
uint8_t *p, dir;
size_t cklen;
uint8_t buf[8];
uint32_t seq;
/*
* How many trailing pad bytes do we need?
*/
m = *mp;
mlen = m_length(m, &mlast);
tlen = kc->kc_tokenkey->ks_class->ec_msgblocklen;
padlen = tlen - (mlen % tlen);
/*
* The data part of the token has eight bytes of random
* confounder prepended and followed by up to eight bytes of
* padding bytes each of which is set to the number of padding
* bytes.
*/
datalen = mlen + 8 + padlen;
tlen = token_length(kc->kc_tokenkey);
p = krb5_make_token("\x02\x01", tlen, datalen + tlen, &tm);
p += 2; /* TOK_ID */
*p++ = sgn_alg[0]; /* SGN_ALG */
*p++ = sgn_alg[1];
if (conf_req_flag) {
*p++ = seal_alg[0]; /* SEAL_ALG */
*p++ = seal_alg[1];
} else {
*p++ = 0xff; /* SEAL_ALG = none */
*p++ = 0xff;
}
*p++ = 0xff; /* filler */
*p++ = 0xff;
/*
* Copy the padded message data.
*/
if (M_LEADINGSPACE(m) >= 8) {
m->m_data -= 8;
m->m_len += 8;
} else {
MGET(cm, M_WAITOK, MT_DATA);
cm->m_len = 8;
cm->m_next = m;
m = cm;
}
arc4rand(m->m_data, 8, 0);
if (M_TRAILINGSPACE(mlast) >= padlen) {
memset(mlast->m_data + mlast->m_len, padlen, padlen);
mlast->m_len += padlen;
} else {
MGET(pm, M_WAITOK, MT_DATA);
memset(pm->m_data, padlen, padlen);
pm->m_len = padlen;
mlast->m_next = pm;
mlast = pm;
}
tm->m_next = m;
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the token header plus the
* padded message. Fiddle with tm->m_len so that we only
* checksum the 8 bytes of head that we care about.
*/
cklen = kc->kc_checksumkey->ks_class->ec_checksumlen;
tlen = tm->m_len;
tm->m_len = p - (uint8_t *) tm->m_data;
MGET(cm, M_WAITOK, MT_DATA);
cm->m_len = cklen;
mlast->m_next = cm;
krb5_checksum(kc->kc_checksumkey, 13, tm, tm->m_len - 8,
datalen + 8, cklen);
tm->m_len = tlen;
mlast->m_next = NULL;
bcopy(cm->m_data, p + 8, cklen);
m_free(cm);
/*
* SND_SEQ:
*
* Take the four bytes of the sequence number least
* significant first (most significant first for ARCFOUR)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* followed by four bytes of direction marker (zero for
* initiator and 0xff for acceptor). Encrypt that data using
* the SGN_CKSUM as IV.
*/
seq = atomic_fetchadd_32(&kc->kc_local_seqnumber, 1);
if (sgn_alg[0] == 0x11) {
p[0] = (seq >> 24);
p[1] = (seq >> 16);
p[2] = (seq >> 8);
p[3] = (seq >> 0);
} else {
p[0] = (seq >> 0);
p[1] = (seq >> 8);
p[2] = (seq >> 16);
p[3] = (seq >> 24);
}
if (is_initiator(kc)) {
dir = 0;
} else {
dir = 0xff;
}
p[4] = dir;
p[5] = dir;
p[6] = dir;
p[7] = dir;
krb5_encrypt(kc->kc_tokenkey, tm, p - (uint8_t *) tm->m_data,
8, p + 8, 8);
if (conf_req_flag) {
/*
* Encrypt the padded message with an IV of zero for
* DES and DES3, or an IV of the sequence number in
* big-endian format for ARCFOUR.
*/
if (seal_alg[0] == 0x10) {
buf[0] = (seq >> 24);
buf[1] = (seq >> 16);
buf[2] = (seq >> 8);
buf[3] = (seq >> 0);
krb5_encrypt(kc->kc_encryptkey, m, 0, datalen,
buf, 4);
} else {
krb5_encrypt(kc->kc_encryptkey, m, 0, datalen,
NULL, 0);
}
}
if (conf_state)
*conf_state = conf_req_flag;
*mp = tm;
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_wrap_new(struct krb5_context *kc, int conf_req_flag,
struct mbuf **mp, int *conf_state)
{
struct krb5_key_state *Ke = kc->kc_send_seal_Ke;
struct krb5_key_state *Ki = kc->kc_send_seal_Ki;
struct krb5_key_state *Kc = kc->kc_send_seal_Kc;
const struct krb5_encryption_class *ec = Ke->ks_class;
struct mbuf *m, *mlast, *tm;
uint8_t *p;
int flags, EC;
size_t mlen, blen, mblen, cklen, ctlen;
uint32_t seq;
static char zpad[32];
m = *mp;
mlen = m_length(m, &mlast);
blen = ec->ec_blocklen;
mblen = ec->ec_msgblocklen;
cklen = ec->ec_checksumlen;
if (conf_req_flag) {
/*
* For sealed messages, we need space for 16 bytes of
* header, blen confounder, plaintext, padding, copy
* of header and checksum.
*
* We pad to mblen (which may be different from
* blen). If the encryption class is using CTS, mblen
* will be one (i.e. no padding required).
*/
if (mblen > 1)
EC = mlen % mblen;
else
EC = 0;
ctlen = blen + mlen + EC + 16;
/*
* Put initial header and confounder before the
* message.
*/
M_PREPEND(m, 16 + blen, M_WAITOK);
/*
* Append padding + copy of header and checksum. Try
* to fit this into the end of the original message,
* otherwise allocate a trailer.
*/
if (M_TRAILINGSPACE(mlast) >= EC + 16 + cklen) {
tm = NULL;
mlast->m_len += EC + 16 + cklen;
} else {
MGET(tm, M_WAITOK, MT_DATA);
tm->m_len = EC + 16 + cklen;
mlast->m_next = tm;
}
} else {
/*
* For unsealed messages, we need 16 bytes of header
* plus space for the plaintext and a checksum. EC is
* set to the checksum size. We leave space in tm for
* a copy of the header - this will be trimmed later.
*/
M_PREPEND(m, 16, M_WAITOK);
MGET(tm, M_WAITOK, MT_DATA);
tm->m_len = cklen + 16;
mlast->m_next = tm;
ctlen = 0;
EC = cklen;
}
p = m->m_data;
/* TOK_ID */
p[0] = 0x05;
p[1] = 0x04;
/* Flags */
flags = 0;
if (conf_req_flag)
flags = GSS_TOKEN_SEALED;
if (is_acceptor(kc))
flags |= GSS_TOKEN_SENT_BY_ACCEPTOR;
if (kc->kc_more_flags & ACCEPTOR_SUBKEY)
flags |= GSS_TOKEN_ACCEPTOR_SUBKEY;
p[2] = flags;
/* Filler */
p[3] = 0xff;
/* EC + RRC - set to zero initially */
p[4] = 0;
p[5] = 0;
p[6] = 0;
p[7] = 0;
/* SND_SEQ */
p[8] = 0;
p[9] = 0;
p[10] = 0;
p[11] = 0;
seq = atomic_fetchadd_32(&kc->kc_local_seqnumber, 1);
p[12] = (seq >> 24);
p[13] = (seq >> 16);
p[14] = (seq >> 8);
p[15] = (seq >> 0);
if (conf_req_flag) {
/*
* Encrypt according to RFC 4121 section 4.2 and RFC
* 3961 section 5.3. Note: we don't generate tokens
* with RRC values other than zero. If we did, we
* should zero RRC in the copied header.
*/
arc4rand(p + 16, blen, 0);
if (EC) {
m_copyback(m, 16 + blen + mlen, EC, zpad);
}
m_copyback(m, 16 + blen + mlen + EC, 16, p);
krb5_checksum(Ki, 0, m, 16, ctlen, cklen);
krb5_encrypt(Ke, m, 16, ctlen, NULL, 0);
} else {
/*
* The plaintext message is followed by a checksum of
* the plaintext plus a version of the header where EC
* and RRC are set to zero. Also, the original EC must
* be our checksum size.
*/
bcopy(p, tm->m_data, 16);
krb5_checksum(Kc, 0, m, 16, mlen + 16, cklen);
tm->m_data += 16;
tm->m_len -= 16;
}
/*
* Finally set EC to its actual value
*/
p[4] = EC >> 8;
p[5] = EC;
*mp = m;
return (GSS_S_COMPLETE);
}
static OM_uint32
krb5_wrap(gss_ctx_id_t ctx, OM_uint32 *minor_status,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
int conf_req_flag, gss_qop_t qop_req,
struct mbuf **mp, int *conf_state)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*minor_status = 0;
if (conf_state)
*conf_state = 0;
if (qop_req != GSS_C_QOP_DEFAULT)
return (GSS_S_BAD_QOP);
if (time_uptime > kc->kc_lifetime)
return (GSS_S_CONTEXT_EXPIRED);
switch (kc->kc_tokenkey->ks_class->ec_type) {
case ETYPE_DES_CBC_CRC:
return (krb5_wrap_old(kc, conf_req_flag,
mp, conf_state, sgn_alg_des_md5, seal_alg_des));
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56:
return (krb5_wrap_old(kc, conf_req_flag,
mp, conf_state, sgn_alg_hmac_md5, seal_alg_rc4));
case ETYPE_DES3_CBC_SHA1:
return (krb5_wrap_old(kc, conf_req_flag,
mp, conf_state, sgn_alg_des3_sha1, seal_alg_des3));
default:
return (krb5_wrap_new(kc, conf_req_flag, mp, conf_state));
}
return (GSS_S_FAILURE);
}
static void
m_trim(struct mbuf *m, int len)
{
struct mbuf *n;
int off;
if (m == NULL)
return;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
n = m_getptr(m, len, &off);
if (n) {
n->m_len = off;
if (n->m_next) {
m_freem(n->m_next);
n->m_next = NULL;
}
}
}
static OM_uint32
krb5_unwrap_old(struct krb5_context *kc, struct mbuf **mp, int *conf_state,
uint8_t sgn_alg[2], uint8_t seal_alg[2])
{
OM_uint32 res;
struct mbuf *m, *mlast, *hm, *cm, *n;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
uint8_t *p, dir;
size_t mlen, tlen, elen, datalen, padlen;
size_t cklen;
uint8_t buf[32];
uint32_t seq;
int i, conf;
m = *mp;
mlen = m_length(m, &mlast);
tlen = token_length(kc->kc_tokenkey);
cklen = kc->kc_tokenkey->ks_class->ec_checksumlen;
p = krb5_verify_token("\x02\x01", tlen, &m, &elen, TRUE);
*mp = m;
if (!p)
return (GSS_S_DEFECTIVE_TOKEN);
datalen = elen - tlen;
/*
* Trim the framing header first to make life a little easier
* later.
*/
m_adj(m, p - (uint8_t *) m->m_data);
/* TOK_ID */
p += 2;
/* SGN_ALG */
if (p[0] != sgn_alg[0] || p[1] != sgn_alg[1])
return (GSS_S_DEFECTIVE_TOKEN);
p += 2;
/* SEAL_ALG */
if (p[0] == seal_alg[0] && p[1] == seal_alg[1])
conf = 1;
else if (p[0] == 0xff && p[1] == 0xff)
conf = 0;
else
return (GSS_S_DEFECTIVE_TOKEN);
p += 2;
if (p[0] != 0xff || p[1] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
p += 2;
/*
* SND_SEQ:
*
* Take the four bytes of the sequence number least
* significant first (most significant for ARCFOUR) followed
* by four bytes of direction marker (zero for initiator and
* 0xff for acceptor). Encrypt that data using the SGN_CKSUM
* as IV.
*/
krb5_decrypt(kc->kc_tokenkey, m, 8, 8, p + 8, 8);
if (sgn_alg[0] == 0x11) {
seq = p[3] | (p[2] << 8) | (p[1] << 16) | (p[0] << 24);
} else {
seq = p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
}
if (is_initiator(kc)) {
dir = 0xff;
} else {
dir = 0;
}
if (p[4] != dir || p[5] != dir || p[6] != dir || p[7] != dir)
return (GSS_S_DEFECTIVE_TOKEN);
if (kc->kc_msg_order.km_flags &
(GSS_C_REPLAY_FLAG | GSS_C_SEQUENCE_FLAG)) {
res = krb5_sequence_check(kc, seq);
if (GSS_ERROR(res))
return (res);
} else {
res = GSS_S_COMPLETE;
}
/*
* If the token was encrypted, decode it in-place.
*/
if (conf) {
/*
* Decrypt the padded message with an IV of zero for
* DES and DES3 or an IV of the big-endian encoded
* sequence number for ARCFOUR.
*/
if (seal_alg[0] == 0x10) {
krb5_decrypt(kc->kc_encryptkey, m, 16 + cklen,
datalen, p, 4);
} else {
krb5_decrypt(kc->kc_encryptkey, m, 16 + cklen,
datalen, NULL, 0);
}
}
if (conf_state)
*conf_state = conf;
/*
* Check the trailing pad bytes.
* RFC1964 specifies between 1<->8 bytes, each with a binary value
* equal to the number of bytes.
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*/
if (mlast->m_len > 0)
padlen = mlast->m_data[mlast->m_len - 1];
else {
n = m_getptr(m, tlen + datalen - 1, &i);
/*
* When the position is exactly equal to the # of data bytes
* in the mbuf list, m_getptr() will return the last mbuf in
* the list and an off == m_len for that mbuf, so that case
* needs to be checked as well as a NULL return.
*/
if (n == NULL || n->m_len == i)
return (GSS_S_DEFECTIVE_TOKEN);
padlen = n->m_data[i];
}
if (padlen < 1 || padlen > 8 || padlen > tlen + datalen)
return (GSS_S_DEFECTIVE_TOKEN);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
m_copydata(m, tlen + datalen - padlen, padlen, buf);
for (i = 0; i < padlen; i++) {
if (buf[i] != padlen) {
return (GSS_S_DEFECTIVE_TOKEN);
}
}
/*
* SGN_CKSUM:
*
* Calculate the keyed checksum of the token header plus the
* padded message. We do a little mbuf surgery to trim out the
* parts we don't want to checksum.
*/
hm = m;
*mp = m = m_split(m, 16 + cklen, M_WAITOK);
mlast = m_last(m);
hm->m_len = 8;
hm->m_next = m;
MGET(cm, M_WAITOK, MT_DATA);
cm->m_len = cklen;
mlast->m_next = cm;
krb5_checksum(kc->kc_checksumkey, 13, hm, 0, datalen + 8, cklen);
hm->m_next = NULL;
mlast->m_next = NULL;
if (bcmp(cm->m_data, hm->m_data + 16, cklen)) {
m_freem(hm);
m_free(cm);
return (GSS_S_BAD_SIG);
}
m_freem(hm);
m_free(cm);
/*
* Trim off the confounder and padding.
*/
m_adj(m, 8);
if (mlast->m_len >= padlen) {
mlast->m_len -= padlen;
} else {
m_trim(m, datalen - 8 - padlen);
}
*mp = m;
return (res);
}
static OM_uint32
krb5_unwrap_new(struct krb5_context *kc, struct mbuf **mp, int *conf_state)
{
OM_uint32 res;
struct krb5_key_state *Ke = kc->kc_recv_seal_Ke;
struct krb5_key_state *Ki = kc->kc_recv_seal_Ki;
struct krb5_key_state *Kc = kc->kc_recv_seal_Kc;
const struct krb5_encryption_class *ec = Ke->ks_class;
struct mbuf *m, *mlast, *hm, *cm;
uint8_t *p, *pp;
int sealed, flags, EC, RRC;
size_t blen, cklen, ctlen, mlen, plen, tlen;
char buf[32], buf2[32];
m = *mp;
mlen = m_length(m, &mlast);
if (mlen <= 16)
return (GSS_S_DEFECTIVE_TOKEN);
if (m->m_len < 16) {
m = m_pullup(m, 16);
*mp = m;
}
p = m->m_data;
/* TOK_ID */
if (p[0] != 0x05)
return (GSS_S_DEFECTIVE_TOKEN);
if (p[1] != 0x04)
return (GSS_S_DEFECTIVE_TOKEN);
/* Flags */
sealed = p[2] & GSS_TOKEN_SEALED;
flags = sealed;
if (is_initiator(kc))
flags |= GSS_TOKEN_SENT_BY_ACCEPTOR;
if (kc->kc_more_flags & ACCEPTOR_SUBKEY)
flags |= GSS_TOKEN_ACCEPTOR_SUBKEY;
if (p[2] != flags)
return (GSS_S_DEFECTIVE_TOKEN);
/* Filler */
if (p[3] != 0xff)
return (GSS_S_DEFECTIVE_TOKEN);
/* EC + RRC */
EC = (p[4] << 8) + p[5];
RRC = (p[6] << 8) + p[7];
/* SND_SEQ */
if (kc->kc_msg_order.km_flags &
(GSS_C_REPLAY_FLAG | GSS_C_SEQUENCE_FLAG)) {
uint32_t seq;
if (p[8] || p[9] || p[10] || p[11]) {
res = GSS_S_UNSEQ_TOKEN;
} else {
seq = (p[12] << 24) | (p[13] << 16)
| (p[14] << 8) | p[15];
res = krb5_sequence_check(kc, seq);
}
if (GSS_ERROR(res))
return (res);
} else {
res = GSS_S_COMPLETE;
}
/*
* Separate the header before dealing with RRC. We only need
* to keep the header if the message isn't encrypted.
*/
if (sealed) {
hm = NULL;
m_adj(m, 16);
} else {
hm = m;
*mp = m = m_split(m, 16, M_WAITOK);
mlast = m_last(m);
}
/*
* Undo the effects of RRC by rotating left.
*/
if (RRC > 0) {
struct mbuf *rm;
size_t rlen;
rlen = mlen - 16;
if (RRC <= sizeof(buf) && m->m_len >= rlen) {
/*
* Simple case, just rearrange the bytes in m.
*/
bcopy(m->m_data, buf, RRC);
bcopy(m->m_data + RRC, m->m_data, rlen - RRC);
bcopy(buf, m->m_data + rlen - RRC, RRC);
} else {
/*
* More complicated - rearrange the mbuf
* chain.
*/
rm = m;
*mp = m = m_split(m, RRC, M_WAITOK);
m_cat(m, rm);
mlast = rm;
}
}
blen = ec->ec_blocklen;
cklen = ec->ec_checksumlen;
if (sealed) {
/*
* Decrypt according to RFC 4121 section 4.2 and RFC
* 3961 section 5.3. The message must be large enough
* for a blocksize confounder, at least one block of
* cyphertext and a checksum.
*/
if (mlen < 16 + 2*blen + cklen)
return (GSS_S_DEFECTIVE_TOKEN);
ctlen = mlen - 16 - cklen;
krb5_decrypt(Ke, m, 0, ctlen, NULL, 0);
/*
* The size of the plaintext is ctlen minus blocklen
* (for the confounder), 16 (for the copy of the token
* header) and EC (for the filler). The actual
* plaintext starts after the confounder.
*/
plen = ctlen - blen - 16 - EC;
pp = p + 16 + blen;
/*
* Checksum the padded plaintext.
*/
m_copydata(m, ctlen, cklen, buf);
krb5_checksum(Ki, 0, m, 0, ctlen, cklen);
m_copydata(m, ctlen, cklen, buf2);
if (bcmp(buf, buf2, cklen))
return (GSS_S_BAD_SIG);
/*
* Trim the message back to just plaintext.
*/
m_adj(m, blen);
tlen = 16 + EC + cklen;
if (mlast->m_len >= tlen) {
mlast->m_len -= tlen;
} else {
m_trim(m, plen);
}
} else {
/*
* The plaintext message is followed by a checksum of
* the plaintext plus a version of the header where EC
* and RRC are set to zero. Also, the original EC must
* be our checksum size.
*/
if (mlen < 16 + cklen || EC != cklen)
return (GSS_S_DEFECTIVE_TOKEN);
/*
* The size of the plaintext is simply the message
* size less header and checksum. The plaintext starts
* right after the header (which we have saved in hm).
*/
plen = mlen - 16 - cklen;
/*
* Insert a copy of the header (with EC and RRC set to
* zero) between the plaintext message and the
* checksum.
*/
p = hm->m_data;
p[4] = p[5] = p[6] = p[7] = 0;
cm = m_split(m, plen, M_WAITOK);
mlast = m_last(m);
m->m_next = hm;
hm->m_next = cm;
bcopy(cm->m_data, buf, cklen);
krb5_checksum(Kc, 0, m, 0, plen + 16, cklen);
if (bcmp(cm->m_data, buf, cklen))
return (GSS_S_BAD_SIG);
/*
* The checksum matches, discard all buf the plaintext.
*/
mlast->m_next = NULL;
m_freem(hm);
}
if (conf_state)
*conf_state = (sealed != 0);
return (res);
}
static OM_uint32
krb5_unwrap(gss_ctx_id_t ctx, OM_uint32 *minor_status,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct mbuf **mp, int *conf_state, gss_qop_t *qop_state)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
OM_uint32 maj_stat;
*minor_status = 0;
if (qop_state)
*qop_state = GSS_C_QOP_DEFAULT;
if (conf_state)
*conf_state = 0;
if (time_uptime > kc->kc_lifetime)
return (GSS_S_CONTEXT_EXPIRED);
switch (kc->kc_tokenkey->ks_class->ec_type) {
case ETYPE_DES_CBC_CRC:
maj_stat = krb5_unwrap_old(kc, mp, conf_state,
sgn_alg_des_md5, seal_alg_des);
break;
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56:
maj_stat = krb5_unwrap_old(kc, mp, conf_state,
sgn_alg_hmac_md5, seal_alg_rc4);
break;
case ETYPE_DES3_CBC_SHA1:
maj_stat = krb5_unwrap_old(kc, mp, conf_state,
sgn_alg_des3_sha1, seal_alg_des3);
break;
default:
maj_stat = krb5_unwrap_new(kc, mp, conf_state);
break;
}
if (GSS_ERROR(maj_stat)) {
m_freem(*mp);
*mp = NULL;
}
return (maj_stat);
}
static OM_uint32
krb5_wrap_size_limit(gss_ctx_id_t ctx, OM_uint32 *minor_status,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
int conf_req_flag, gss_qop_t qop_req, OM_uint32 req_output_size,
OM_uint32 *max_input_size)
{
struct krb5_context *kc = (struct krb5_context *)ctx;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
const struct krb5_encryption_class *ec;
OM_uint32 overhead;
*minor_status = 0;
*max_input_size = 0;
if (qop_req != GSS_C_QOP_DEFAULT)
return (GSS_S_BAD_QOP);
ec = kc->kc_tokenkey->ks_class;
switch (ec->ec_type) {
case ETYPE_DES_CBC_CRC:
case ETYPE_DES3_CBC_SHA1:
case ETYPE_ARCFOUR_HMAC_MD5:
case ETYPE_ARCFOUR_HMAC_MD5_56:
/*
* up to 5 bytes for [APPLICATION 0] SEQUENCE
* 2 + krb5 oid length
* 8 bytes of header
* 8 bytes of confounder
* maximum of 8 bytes of padding
* checksum
*/
overhead = 5 + 2 + krb5_mech_oid.length;
overhead += 8 + 8 + ec->ec_msgblocklen;
overhead += ec->ec_checksumlen;
break;
default:
if (conf_req_flag) {
/*
* 16 byts of header
* blocklen bytes of confounder
* up to msgblocklen - 1 bytes of padding
* 16 bytes for copy of header
* checksum
*/
overhead = 16 + ec->ec_blocklen;
overhead += ec->ec_msgblocklen - 1;
overhead += 16;
overhead += ec->ec_checksumlen;
} else {
/*
* 16 bytes of header plus checksum.
*/
overhead = 16 + ec->ec_checksumlen;
}
}
*max_input_size = req_output_size - overhead;
return (GSS_S_COMPLETE);
}
static kobj_method_t krb5_methods[] = {
KOBJMETHOD(kgss_init, krb5_init),
KOBJMETHOD(kgss_import, krb5_import),
KOBJMETHOD(kgss_delete, krb5_delete),
KOBJMETHOD(kgss_mech_type, krb5_mech_type),
KOBJMETHOD(kgss_get_mic, krb5_get_mic),
KOBJMETHOD(kgss_verify_mic, krb5_verify_mic),
KOBJMETHOD(kgss_wrap, krb5_wrap),
KOBJMETHOD(kgss_unwrap, krb5_unwrap),
KOBJMETHOD(kgss_wrap_size_limit, krb5_wrap_size_limit),
{ 0, 0 }
};
static struct kobj_class krb5_class = {
"kerberosv5",
krb5_methods,
sizeof(struct krb5_context)
};
/*
* Kernel module glue
*/
static int
kgssapi_krb5_modevent(module_t mod, int type, void *data)
{
switch (type) {
case MOD_LOAD:
kgss_install_mech(&krb5_mech_oid, "kerberosv5", &krb5_class);
break;
case MOD_UNLOAD:
kgss_uninstall_mech(&krb5_mech_oid);
break;
}
return (0);
}
static moduledata_t kgssapi_krb5_mod = {
"kgssapi_krb5",
kgssapi_krb5_modevent,
NULL,
};
DECLARE_MODULE(kgssapi_krb5, kgssapi_krb5_mod, SI_SUB_VFS, SI_ORDER_ANY);
MODULE_DEPEND(kgssapi_krb5, kgssapi, 1, 1, 1);
MODULE_DEPEND(kgssapi_krb5, crypto, 1, 1, 1);
MODULE_DEPEND(kgssapi_krb5, rc4, 1, 1, 1);
MODULE_VERSION(kgssapi_krb5, 1);