freebsd-dev/sys/dev/iscsi/icl.h

174 lines
4.9 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2012 The FreeBSD Foundation
*
* This software was developed by Edward Tomasz Napierala under sponsorship
* from the FreeBSD Foundation.
*
* 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.
*
* $FreeBSD$
*/
#ifndef ICL_H
#define ICL_H
/*
* iSCSI Common Layer. It's used by both the initiator and target to send
* and receive iSCSI PDUs.
*/
#include <sys/types.h>
#include <sys/kobj.h>
#include <sys/condvar.h>
#include <sys/sysctl.h>
SYSCTL_DECL(_kern_icl);
extern int icl_debug;
#define ICL_DEBUG(X, ...) \
do { \
if (icl_debug > 1) \
printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
} while (0)
#define ICL_WARN(X, ...) \
do { \
if (icl_debug > 0) { \
printf("WARNING: %s: " X "\n", \
__func__, ## __VA_ARGS__); \
} \
} while (0)
struct icl_conn;
struct ccb_scsiio;
union ctl_io;
struct icl_pdu {
STAILQ_ENTRY(icl_pdu) ip_next;
struct icl_conn *ip_conn;
struct iscsi_bhs *ip_bhs;
struct mbuf *ip_bhs_mbuf;
size_t ip_ahs_len;
struct mbuf *ip_ahs_mbuf;
size_t ip_data_len;
struct mbuf *ip_data_mbuf;
/*
* User (initiator or provider) private fields.
*/
Implement zero-copy iSCSI target transmission/read. Add ICL_NOCOPY flag to icl_pdu_append_data(), specifying that the method can just reference the data buffer instead of immediately copying it. Extend the offload KPI with optional PDU queue method, allowing to specify completion callback, called when all the data referenced by above has been transferred and won't be accessed any more (the buffers can be freed). Implement the above functionality in software iSCSI driver using mbufs with external storage and reference counter. Note that some NICs (ixl(4)) may keep the mbuf in TX queue for a long time, so CTL has to be ready. Add optional method to struct ctl_scsiio for buffer reference counting. Implement it for CTL block backend, allowing to delay free of the struct ctl_be_block_io and memory it references as needed. In first reincarnation of the patch I tried to delay whole I/O as it is done for FibreChannel, that was cleaner, but due to the above callback delays I had to rewrite it this way to not leave LUN referenced potentially for hours or more. All together on sequential read from ZFS ARC this saves about 30% of CPU time and memory bandwidth by avoiding one of 3 memory copies (the other two are from ZFS ARC to DMU cache and then from DMU cache to CTL buffers). On tests with 2x Xeon Silver 4114 this allows to reach full line rate of 100GigE NIC. Tests with Gold CPUs and two 100GigE NICs are stil TBD, but expectations to saturate them are pretty high. ;) Discussed with: Chelsio Sponsored by: iXsystems, Inc.
2020-06-08 20:53:57 +00:00
void *ip_prv0;
void *ip_prv1;
};
#define ICL_CONN_STATE_INVALID 0
#define ICL_CONN_STATE_BHS 1
#define ICL_CONN_STATE_AHS 2
#define ICL_CONN_STATE_HEADER_DIGEST 3
#define ICL_CONN_STATE_DATA 4
#define ICL_CONN_STATE_DATA_DIGEST 5
Implement zero-copy iSCSI target transmission/read. Add ICL_NOCOPY flag to icl_pdu_append_data(), specifying that the method can just reference the data buffer instead of immediately copying it. Extend the offload KPI with optional PDU queue method, allowing to specify completion callback, called when all the data referenced by above has been transferred and won't be accessed any more (the buffers can be freed). Implement the above functionality in software iSCSI driver using mbufs with external storage and reference counter. Note that some NICs (ixl(4)) may keep the mbuf in TX queue for a long time, so CTL has to be ready. Add optional method to struct ctl_scsiio for buffer reference counting. Implement it for CTL block backend, allowing to delay free of the struct ctl_be_block_io and memory it references as needed. In first reincarnation of the patch I tried to delay whole I/O as it is done for FibreChannel, that was cleaner, but due to the above callback delays I had to rewrite it this way to not leave LUN referenced potentially for hours or more. All together on sequential read from ZFS ARC this saves about 30% of CPU time and memory bandwidth by avoiding one of 3 memory copies (the other two are from ZFS ARC to DMU cache and then from DMU cache to CTL buffers). On tests with 2x Xeon Silver 4114 this allows to reach full line rate of 100GigE NIC. Tests with Gold CPUs and two 100GigE NICs are stil TBD, but expectations to saturate them are pretty high. ;) Discussed with: Chelsio Sponsored by: iXsystems, Inc.
2020-06-08 20:53:57 +00:00
#define ICL_NOCOPY (1 << 30)
struct icl_conn {
KOBJ_FIELDS;
struct mtx *ic_lock;
struct socket *ic_socket;
#ifdef DIAGNOSTIC
volatile u_int ic_outstanding_pdus;
#endif
STAILQ_HEAD(, icl_pdu) ic_to_send;
bool ic_check_send_space;
size_t ic_receive_len;
int ic_receive_state;
struct icl_pdu *ic_receive_pdu;
struct cv ic_send_cv;
struct cv ic_receive_cv;
bool ic_header_crc32c;
bool ic_data_crc32c;
bool ic_send_running;
bool ic_receive_running;
size_t ic_max_data_segment_length;
size_t ic_maxtags;
bool ic_disconnecting;
bool ic_iser;
bool ic_unmapped;
const char *ic_name;
const char *ic_offload;
void (*ic_receive)(struct icl_pdu *);
void (*ic_error)(struct icl_conn *);
/*
* User (initiator or provider) private fields.
*/
void *ic_prv0;
};
struct icl_drv_limits {
int idl_max_recv_data_segment_length;
int idl_max_send_data_segment_length;
int idl_max_burst_length;
int idl_first_burst_length;
int spare[4];
};
Implement zero-copy iSCSI target transmission/read. Add ICL_NOCOPY flag to icl_pdu_append_data(), specifying that the method can just reference the data buffer instead of immediately copying it. Extend the offload KPI with optional PDU queue method, allowing to specify completion callback, called when all the data referenced by above has been transferred and won't be accessed any more (the buffers can be freed). Implement the above functionality in software iSCSI driver using mbufs with external storage and reference counter. Note that some NICs (ixl(4)) may keep the mbuf in TX queue for a long time, so CTL has to be ready. Add optional method to struct ctl_scsiio for buffer reference counting. Implement it for CTL block backend, allowing to delay free of the struct ctl_be_block_io and memory it references as needed. In first reincarnation of the patch I tried to delay whole I/O as it is done for FibreChannel, that was cleaner, but due to the above callback delays I had to rewrite it this way to not leave LUN referenced potentially for hours or more. All together on sequential read from ZFS ARC this saves about 30% of CPU time and memory bandwidth by avoiding one of 3 memory copies (the other two are from ZFS ARC to DMU cache and then from DMU cache to CTL buffers). On tests with 2x Xeon Silver 4114 this allows to reach full line rate of 100GigE NIC. Tests with Gold CPUs and two 100GigE NICs are stil TBD, but expectations to saturate them are pretty high. ;) Discussed with: Chelsio Sponsored by: iXsystems, Inc.
2020-06-08 20:53:57 +00:00
typedef void (*icl_pdu_cb)(struct icl_pdu *, int error);
struct icl_conn *icl_new_conn(const char *offload, bool iser, const char *name,
struct mtx *lock);
int icl_limits(const char *offload, bool iser,
struct icl_drv_limits *idl);
int icl_register(const char *offload, bool iser, int priority,
int (*limits)(struct icl_drv_limits *),
struct icl_conn *(*new_conn)(const char *, struct mtx *));
int icl_unregister(const char *offload, bool rdma);
#ifdef ICL_KERNEL_PROXY
struct sockaddr;
struct icl_listen;
/*
* Target part.
*/
struct icl_listen *icl_listen_new(void (*accept_cb)(struct socket *,
struct sockaddr *, int));
void icl_listen_free(struct icl_listen *il);
int icl_listen_add(struct icl_listen *il, bool rdma,
int domain, int socktype, int protocol,
struct sockaddr *sa, int portal_id);
int icl_listen_remove(struct icl_listen *il, struct sockaddr *sa);
/*
* Those two are not a public API; only to be used between icl_soft.c
* and icl_soft_proxy.c.
*/
int icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so);
int icl_soft_proxy_connect(struct icl_conn *ic, int domain,
int socktype, int protocol, struct sockaddr *from_sa,
struct sockaddr *to_sa);
#endif /* ICL_KERNEL_PROXY */
#endif /* !ICL_H */