2
0
mirror of https://github.com/edk2-porting/linux-next.git synced 2024-12-21 03:33:59 +08:00
linux-next/drivers/scsi/libiscsi_tcp.c

1215 lines
32 KiB
C
Raw Normal View History

/*
* iSCSI over TCP/IP Data-Path lib
*
* Copyright (C) 2004 Dmitry Yusupov
* Copyright (C) 2004 Alex Aizman
* Copyright (C) 2005 - 2006 Mike Christie
* Copyright (C) 2006 Red Hat, Inc. All rights reserved.
* maintained by open-iscsi@googlegroups.com
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published
* by the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* See the file COPYING included with this distribution for more details.
*
* Credits:
* Christoph Hellwig
* FUJITA Tomonori
* Arne Redlich
* Zhenyu Wang
*/
#include <linux/types.h>
#include <linux/list.h>
#include <linux/inet.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
#include <linux/file.h>
#include <linux/blkdev.h>
#include <linux/crypto.h>
#include <linux/delay.h>
#include <linux/kfifo.h>
#include <linux/scatterlist.h>
#include <linux/module.h>
#include <net/tcp.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi.h>
#include <scsi/scsi_transport_iscsi.h>
#include "iscsi_tcp.h"
MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
"Dmitry Yusupov <dmitry_yus@yahoo.com>, "
"Alex Aizman <itn780@yahoo.com>");
MODULE_DESCRIPTION("iSCSI/TCP data-path");
MODULE_LICENSE("GPL");
static int iscsi_dbg_libtcp;
module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
"module. Set to 1 to turn on, and zero to turn off. Default "
"is off.");
#define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...) \
do { \
if (iscsi_dbg_libtcp) \
iscsi_conn_printk(KERN_INFO, _conn, \
"%s " dbg_fmt, \
__func__, ##arg); \
} while (0);
static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment);
/*
* Scatterlist handling: inside the iscsi_segment, we
* remember an index into the scatterlist, and set data/size
* to the current scatterlist entry. For highmem pages, we
* kmap as needed.
*
* Note that the page is unmapped when we return from
* TCP's data_ready handler, so we may end up mapping and
* unmapping the same page repeatedly. The whole reason
* for this is that we shouldn't keep the page mapped
* outside the softirq.
*/
/**
* iscsi_tcp_segment_init_sg - init indicated scatterlist entry
* @segment: the buffer object
* @sg: scatterlist
* @offset: byte offset into that sg entry
*
* This function sets up the segment so that subsequent
* data is copied to the indicated sg entry, at the given
* offset.
*/
static inline void
iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
struct scatterlist *sg, unsigned int offset)
{
segment->sg = sg;
segment->sg_offset = offset;
segment->size = min(sg->length - offset,
segment->total_size - segment->total_copied);
segment->data = NULL;
}
/**
* iscsi_tcp_segment_map - map the current S/G page
* @segment: iscsi_segment
* @recv: 1 if called from recv path
*
* We only need to possibly kmap data if scatter lists are being used,
* because the iscsi passthrough and internal IO paths will never use high
* mem pages.
*/
static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
{
struct scatterlist *sg;
if (segment->data != NULL || !segment->sg)
return;
sg = segment->sg;
BUG_ON(segment->sg_mapped);
BUG_ON(sg->length == 0);
/*
* If the page count is greater than one it is ok to send
* to the network layer's zero copy send path. If not we
* have to go the slow sendmsg path. We always map for the
* recv path.
*/
if (page_count(sg_page(sg)) >= 1 && !recv)
return;
if (recv) {
segment->atomic_mapped = true;
segment->sg_mapped = kmap_atomic(sg_page(sg));
} else {
segment->atomic_mapped = false;
/* the xmit path can sleep with the page mapped so use kmap */
segment->sg_mapped = kmap(sg_page(sg));
}
segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
}
void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
{
if (segment->sg_mapped) {
if (segment->atomic_mapped)
kunmap_atomic(segment->sg_mapped);
else
kunmap(sg_page(segment->sg));
segment->sg_mapped = NULL;
segment->data = NULL;
}
}
EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
/*
* Splice the digest buffer into the buffer
*/
static inline void
iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
{
segment->data = digest;
segment->digest_len = ISCSI_DIGEST_SIZE;
segment->total_size += ISCSI_DIGEST_SIZE;
segment->size = ISCSI_DIGEST_SIZE;
segment->copied = 0;
segment->sg = NULL;
segment->hash = NULL;
}
/**
* iscsi_tcp_segment_done - check whether the segment is complete
* @tcp_conn: iscsi tcp connection
* @segment: iscsi segment to check
* @recv: set to one of this is called from the recv path
* @copied: number of bytes copied
*
* Check if we're done receiving this segment. If the receive
* buffer is full but we expect more data, move on to the
* next entry in the scatterlist.
*
* If the amount of data we received isn't a multiple of 4,
* we will transparently receive the pad bytes, too.
*
* This function must be re-entrant.
*/
int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment, int recv,
unsigned copied)
{
struct scatterlist sg;
unsigned int pad;
ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
segment->copied, copied, segment->size,
recv ? "recv" : "xmit");
if (segment->hash && copied) {
/*
* If a segment is kmapd we must unmap it before sending
* to the crypto layer since that will try to kmap it again.
*/
iscsi_tcp_segment_unmap(segment);
if (!segment->data) {
sg_init_table(&sg, 1);
sg_set_page(&sg, sg_page(segment->sg), copied,
segment->copied + segment->sg_offset +
segment->sg->offset);
} else
sg_init_one(&sg, segment->data + segment->copied,
copied);
crypto_hash_update(segment->hash, &sg, copied);
}
segment->copied += copied;
if (segment->copied < segment->size) {
iscsi_tcp_segment_map(segment, recv);
return 0;
}
segment->total_copied += segment->copied;
segment->copied = 0;
segment->size = 0;
/* Unmap the current scatterlist page, if there is one. */
iscsi_tcp_segment_unmap(segment);
/* Do we have more scatterlist entries? */
ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
segment->total_copied, segment->total_size);
if (segment->total_copied < segment->total_size) {
/* Proceed to the next entry in the scatterlist. */
iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
0);
iscsi_tcp_segment_map(segment, recv);
BUG_ON(segment->size == 0);
return 0;
}
/* Do we need to handle padding? */
if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
pad = iscsi_padding(segment->total_copied);
if (pad != 0) {
ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
"consume %d pad bytes\n", pad);
segment->total_size += pad;
segment->size = pad;
segment->data = segment->padbuf;
return 0;
}
}
/*
* Set us up for transferring the data digest. hdr digest
* is completely handled in hdr done function.
*/
if (segment->hash) {
crypto_hash_final(segment->hash, segment->digest);
iscsi_tcp_segment_splice_digest(segment,
recv ? segment->recv_digest : segment->digest);
return 0;
}
return 1;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
/**
* iscsi_tcp_segment_recv - copy data to segment
* @tcp_conn: the iSCSI TCP connection
* @segment: the buffer to copy to
* @ptr: data pointer
* @len: amount of data available
*
* This function copies up to @len bytes to the
* given buffer, and returns the number of bytes
* consumed, which can actually be less than @len.
*
* If hash digest is enabled, the function will update the
* hash while copying.
* Combining these two operations doesn't buy us a lot (yet),
* but in the future we could implement combined copy+crc,
* just way we do for network layer checksums.
*/
static int
iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment, const void *ptr,
unsigned int len)
{
unsigned int copy = 0, copied = 0;
while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
if (copied == len) {
ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
"copied %d bytes\n", len);
break;
}
copy = min(len - copied, segment->size - segment->copied);
ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
memcpy(segment->data + segment->copied, ptr + copied, copy);
copied += copy;
}
return copied;
}
inline void
iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
unsigned char digest[ISCSI_DIGEST_SIZE])
{
struct scatterlist sg;
sg_init_one(&sg, hdr, hdrlen);
crypto_hash_digest(hash, &sg, hdrlen, digest);
}
EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
static inline int
iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment)
{
if (!segment->digest_len)
return 1;
if (memcmp(segment->recv_digest, segment->digest,
segment->digest_len)) {
ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
return 0;
}
return 1;
}
/*
* Helper function to set up segment buffer
*/
static inline void
__iscsi_segment_init(struct iscsi_segment *segment, size_t size,
iscsi_segment_done_fn_t *done, struct hash_desc *hash)
{
memset(segment, 0, sizeof(*segment));
segment->total_size = size;
segment->done = done;
if (hash) {
segment->hash = hash;
crypto_hash_init(hash);
}
}
inline void
iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
size_t size, iscsi_segment_done_fn_t *done,
struct hash_desc *hash)
{
__iscsi_segment_init(segment, size, done, hash);
segment->data = data;
segment->size = size;
}
EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
inline int
iscsi_segment_seek_sg(struct iscsi_segment *segment,
struct scatterlist *sg_list, unsigned int sg_count,
unsigned int offset, size_t size,
iscsi_segment_done_fn_t *done, struct hash_desc *hash)
{
struct scatterlist *sg;
unsigned int i;
__iscsi_segment_init(segment, size, done, hash);
for_each_sg(sg_list, sg, sg_count, i) {
if (offset < sg->length) {
iscsi_tcp_segment_init_sg(segment, sg, offset);
return 0;
}
offset -= sg->length;
}
return ISCSI_ERR_DATA_OFFSET;
}
EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
/**
* iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
* @tcp_conn: iscsi connection to prep for
*
* This function always passes NULL for the hash argument, because when this
* function is called we do not yet know the final size of the header and want
* to delay the digest processing until we know that.
*/
void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
{
ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
"(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
"digest enabled" : "digest disabled");
iscsi_segment_init_linear(&tcp_conn->in.segment,
tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
iscsi_tcp_hdr_recv_done, NULL);
}
EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
/*
* Handle incoming reply to any other type of command
*/
static int
iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment)
{
struct iscsi_conn *conn = tcp_conn->iscsi_conn;
int rc = 0;
if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
return ISCSI_ERR_DATA_DGST;
rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
conn->data, tcp_conn->in.datalen);
if (rc)
return rc;
iscsi_tcp_hdr_recv_prep(tcp_conn);
return 0;
}
static void
iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
{
struct iscsi_conn *conn = tcp_conn->iscsi_conn;
struct hash_desc *rx_hash = NULL;
if (conn->datadgst_en &&
!(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
rx_hash = tcp_conn->rx_hash;
iscsi_segment_init_linear(&tcp_conn->in.segment,
conn->data, tcp_conn->in.datalen,
iscsi_tcp_data_recv_done, rx_hash);
}
/**
* iscsi_tcp_cleanup_task - free tcp_task resources
* @task: iscsi task
*
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
* must be called with session back_lock
*/
void iscsi_tcp_cleanup_task(struct iscsi_task *task)
{
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct iscsi_r2t_info *r2t;
/* nothing to do for mgmt */
if (!task->sc)
return;
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock_bh(&tcp_task->queue2pool);
/* flush task's r2t queues */
while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
sizeof(void*));
ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
}
r2t = tcp_task->r2t;
if (r2t != NULL) {
kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
sizeof(void*));
tcp_task->r2t = NULL;
}
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock_bh(&tcp_task->queue2pool);
}
EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
/**
* iscsi_tcp_data_in - SCSI Data-In Response processing
* @conn: iscsi connection
* @task: scsi command task
*/
static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
{
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
int datasn = be32_to_cpu(rhdr->datasn);
unsigned total_in_length = scsi_in(task->sc)->length;
/*
* lib iscsi will update this in the completion handling if there
* is status.
*/
if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
if (tcp_conn->in.datalen == 0)
return 0;
if (tcp_task->exp_datasn != datasn) {
ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
"\n", tcp_task->exp_datasn, datasn);
return ISCSI_ERR_DATASN;
}
tcp_task->exp_datasn++;
tcp_task->data_offset = be32_to_cpu(rhdr->offset);
if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
"total_length_in(%d)\n", tcp_task->data_offset,
tcp_conn->in.datalen, total_in_length);
return ISCSI_ERR_DATA_OFFSET;
}
conn->datain_pdus_cnt++;
return 0;
}
/**
* iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
* @conn: iscsi connection
* @task: scsi command task
*/
static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
{
struct iscsi_session *session = conn->session;
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
struct iscsi_r2t_info *r2t;
int r2tsn = be32_to_cpu(rhdr->r2tsn);
u32 data_length;
u32 data_offset;
int rc;
if (tcp_conn->in.datalen) {
iscsi_conn_printk(KERN_ERR, conn,
"invalid R2t with datalen %d\n",
tcp_conn->in.datalen);
return ISCSI_ERR_DATALEN;
}
if (tcp_task->exp_datasn != r2tsn){
ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
tcp_task->exp_datasn, r2tsn);
return ISCSI_ERR_R2TSN;
}
/* fill-in new R2T associated with the task */
iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
iscsi_conn_printk(KERN_INFO, conn,
"dropping R2T itt %d in recovery.\n",
task->itt);
return 0;
}
data_length = be32_to_cpu(rhdr->data_length);
if (data_length == 0) {
iscsi_conn_printk(KERN_ERR, conn,
"invalid R2T with zero data len\n");
return ISCSI_ERR_DATALEN;
}
if (data_length > session->max_burst)
ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
"burst %u. Attempting to execute request.\n",
data_length, session->max_burst);
data_offset = be32_to_cpu(rhdr->data_offset);
if (data_offset + data_length > scsi_out(task->sc)->length) {
iscsi_conn_printk(KERN_ERR, conn,
"invalid R2T with data len %u at offset %u "
"and total length %d\n", data_length,
data_offset, scsi_out(task->sc)->length);
return ISCSI_ERR_DATALEN;
}
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock(&tcp_task->pool2queue);
rc = kfifo_out(&tcp_task->r2tpool.queue, (void *)&r2t, sizeof(void *));
if (!rc) {
iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
"Target has sent more R2Ts than it "
"negotiated for or driver has leaked.\n");
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&tcp_task->pool2queue);
return ISCSI_ERR_PROTO;
}
r2t->exp_statsn = rhdr->statsn;
r2t->data_length = data_length;
r2t->data_offset = data_offset;
r2t->ttt = rhdr->ttt; /* no flip */
r2t->datasn = 0;
r2t->sent = 0;
tcp_task->exp_datasn = r2tsn + 1;
kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
conn->r2t_pdus_cnt++;
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&tcp_task->pool2queue);
iscsi_requeue_task(task);
return 0;
}
/*
* Handle incoming reply to DataIn command
*/
static int
iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment)
{
struct iscsi_conn *conn = tcp_conn->iscsi_conn;
struct iscsi_hdr *hdr = tcp_conn->in.hdr;
int rc;
if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
return ISCSI_ERR_DATA_DGST;
/* check for non-exceptional status */
if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
if (rc)
return rc;
}
iscsi_tcp_hdr_recv_prep(tcp_conn);
return 0;
}
/**
* iscsi_tcp_hdr_dissect - process PDU header
* @conn: iSCSI connection
* @hdr: PDU header
*
* This function analyzes the header of the PDU received,
* and performs several sanity checks. If the PDU is accompanied
* by data, the receive buffer is set up to copy the incoming data
* to the correct location.
*/
static int
iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
{
int rc = 0, opcode, ahslen;
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
struct iscsi_task *task;
/* verify PDU length */
tcp_conn->in.datalen = ntoh24(hdr->dlength);
if (tcp_conn->in.datalen > conn->max_recv_dlength) {
iscsi_conn_printk(KERN_ERR, conn,
"iscsi_tcp: datalen %d > %d\n",
tcp_conn->in.datalen, conn->max_recv_dlength);
return ISCSI_ERR_DATALEN;
}
/* Additional header segments. So far, we don't
* process additional headers.
*/
ahslen = hdr->hlength << 2;
opcode = hdr->opcode & ISCSI_OPCODE_MASK;
/* verify itt (itt encoding: age+cid+itt) */
rc = iscsi_verify_itt(conn, hdr->itt);
if (rc)
return rc;
ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
opcode, ahslen, tcp_conn->in.datalen);
switch(opcode) {
case ISCSI_OP_SCSI_DATA_IN:
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock(&conn->session->back_lock);
task = iscsi_itt_to_ctask(conn, hdr->itt);
if (!task)
rc = ISCSI_ERR_BAD_ITT;
else
rc = iscsi_tcp_data_in(conn, task);
if (rc) {
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&conn->session->back_lock);
break;
}
if (tcp_conn->in.datalen) {
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct hash_desc *rx_hash = NULL;
struct scsi_data_buffer *sdb = scsi_in(task->sc);
/*
* Setup copy of Data-In into the Scsi_Cmnd
* Scatterlist case:
* We set up the iscsi_segment to point to the next
* scatterlist entry to copy to. As we go along,
* we move on to the next scatterlist entry and
* update the digest per-entry.
*/
if (conn->datadgst_en &&
!(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
rx_hash = tcp_conn->rx_hash;
ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
"offset=%d, datalen=%d)\n",
tcp_task->data_offset,
tcp_conn->in.datalen);
task->last_xfer = jiffies;
rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
sdb->table.sgl,
sdb->table.nents,
tcp_task->data_offset,
tcp_conn->in.datalen,
iscsi_tcp_process_data_in,
rx_hash);
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&conn->session->back_lock);
return rc;
}
rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&conn->session->back_lock);
break;
case ISCSI_OP_SCSI_CMD_RSP:
if (tcp_conn->in.datalen) {
iscsi_tcp_data_recv_prep(tcp_conn);
return 0;
}
rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
break;
case ISCSI_OP_R2T:
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock(&conn->session->back_lock);
task = iscsi_itt_to_ctask(conn, hdr->itt);
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&conn->session->back_lock);
if (!task)
rc = ISCSI_ERR_BAD_ITT;
else if (ahslen)
rc = ISCSI_ERR_AHSLEN;
else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
task->last_xfer = jiffies;
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock(&conn->session->frwd_lock);
rc = iscsi_tcp_r2t_rsp(conn, task);
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock(&conn->session->frwd_lock);
} else
rc = ISCSI_ERR_PROTO;
break;
case ISCSI_OP_LOGIN_RSP:
case ISCSI_OP_TEXT_RSP:
case ISCSI_OP_REJECT:
case ISCSI_OP_ASYNC_EVENT:
/*
* It is possible that we could get a PDU with a buffer larger
* than 8K, but there are no targets that currently do this.
* For now we fail until we find a vendor that needs it
*/
if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
iscsi_conn_printk(KERN_ERR, conn,
"iscsi_tcp: received buffer of "
"len %u but conn buffer is only %u "
"(opcode %0x)\n",
tcp_conn->in.datalen,
ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
rc = ISCSI_ERR_PROTO;
break;
}
/* If there's data coming in with the response,
* receive it to the connection's buffer.
*/
if (tcp_conn->in.datalen) {
iscsi_tcp_data_recv_prep(tcp_conn);
return 0;
}
/* fall through */
case ISCSI_OP_LOGOUT_RSP:
case ISCSI_OP_NOOP_IN:
case ISCSI_OP_SCSI_TMFUNC_RSP:
rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
break;
default:
rc = ISCSI_ERR_BAD_OPCODE;
break;
}
if (rc == 0) {
/* Anything that comes with data should have
* been handled above. */
if (tcp_conn->in.datalen)
return ISCSI_ERR_PROTO;
iscsi_tcp_hdr_recv_prep(tcp_conn);
}
return rc;
}
/**
* iscsi_tcp_hdr_recv_done - process PDU header
*
* This is the callback invoked when the PDU header has
* been received. If the header is followed by additional
* header segments, we go back for more data.
*/
static int
iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
struct iscsi_segment *segment)
{
struct iscsi_conn *conn = tcp_conn->iscsi_conn;
struct iscsi_hdr *hdr;
/* Check if there are additional header segments
* *prior* to computing the digest, because we
* may need to go back to the caller for more.
*/
hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
/* Bump the header length - the caller will
* just loop around and get the AHS for us, and
* call again. */
unsigned int ahslen = hdr->hlength << 2;
/* Make sure we don't overflow */
if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
return ISCSI_ERR_AHSLEN;
segment->total_size += ahslen;
segment->size += ahslen;
return 0;
}
/* We're done processing the header. See if we're doing
* header digests; if so, set up the recv_digest buffer
* and go back for more. */
if (conn->hdrdgst_en &&
!(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
if (segment->digest_len == 0) {
/*
* Even if we offload the digest processing we
* splice it in so we can increment the skb/segment
* counters in preparation for the data segment.
*/
iscsi_tcp_segment_splice_digest(segment,
segment->recv_digest);
return 0;
}
iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
segment->total_copied - ISCSI_DIGEST_SIZE,
segment->digest);
if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
return ISCSI_ERR_HDR_DGST;
}
tcp_conn->in.hdr = hdr;
return iscsi_tcp_hdr_dissect(conn, hdr);
}
/**
* iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
* @tcp_conn: iscsi tcp conn
*
* returns non zero if we are currently processing or setup to process
* a header.
*/
inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
{
return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
/**
* iscsi_tcp_recv_skb - Process skb
* @conn: iscsi connection
* @skb: network buffer with header and/or data segment
* @offset: offset in skb
* @offload: bool indicating if transfer was offloaded
*
* Will return status of transfer in status. And will return
* number of bytes copied.
*/
int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
unsigned int offset, bool offloaded, int *status)
{
struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
struct iscsi_segment *segment = &tcp_conn->in.segment;
struct skb_seq_state seq;
unsigned int consumed = 0;
int rc = 0;
ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
/*
* Update for each skb instead of pdu, because over slow networks a
* data_in's data could take a while to read in. We also want to
* account for r2ts.
*/
conn->last_recv = jiffies;
if (unlikely(conn->suspend_rx)) {
ISCSI_DBG_TCP(conn, "Rx suspended!\n");
*status = ISCSI_TCP_SUSPENDED;
return 0;
}
if (offloaded) {
segment->total_copied = segment->total_size;
goto segment_done;
}
skb_prepare_seq_read(skb, offset, skb->len, &seq);
while (1) {
unsigned int avail;
const u8 *ptr;
avail = skb_seq_read(consumed, &ptr, &seq);
if (avail == 0) {
ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
consumed);
*status = ISCSI_TCP_SKB_DONE;
goto skb_done;
}
BUG_ON(segment->copied >= segment->size);
ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
avail);
rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
BUG_ON(rc == 0);
consumed += rc;
if (segment->total_copied >= segment->total_size) {
skb_abort_seq_read(&seq);
goto segment_done;
}
}
segment_done:
*status = ISCSI_TCP_SEGMENT_DONE;
ISCSI_DBG_TCP(conn, "segment done\n");
rc = segment->done(tcp_conn, segment);
if (rc != 0) {
*status = ISCSI_TCP_CONN_ERR;
ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
iscsi_conn_failure(conn, rc);
return 0;
}
/* The done() functions sets up the next segment. */
skb_done:
conn->rxdata_octets += consumed;
return consumed;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
/**
* iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
* @conn: iscsi connection
* @task: scsi command task
* @sc: scsi command
*/
int iscsi_tcp_task_init(struct iscsi_task *task)
{
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct iscsi_conn *conn = task->conn;
struct scsi_cmnd *sc = task->sc;
int err;
if (!sc) {
/*
* mgmt tasks do not have a scatterlist since they come
* in from the iscsi interface.
*/
ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
return conn->session->tt->init_pdu(task, 0, task->data_count);
}
BUG_ON(kfifo_len(&tcp_task->r2tqueue));
tcp_task->exp_datasn = 0;
/* Prepare PDU, optionally w/ immediate data */
ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
task->itt, task->imm_count, task->unsol_r2t.data_length);
err = conn->session->tt->init_pdu(task, 0, task->imm_count);
if (err)
return err;
task->imm_count = 0;
return 0;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
{
struct iscsi_tcp_task *tcp_task = task->dd_data;
struct iscsi_r2t_info *r2t = NULL;
if (iscsi_task_has_unsol_data(task))
r2t = &task->unsol_r2t;
else {
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock_bh(&tcp_task->queue2pool);
if (tcp_task->r2t) {
r2t = tcp_task->r2t;
/* Continue with this R2T? */
if (r2t->data_length <= r2t->sent) {
ISCSI_DBG_TCP(task->conn,
" done with r2t %p\n", r2t);
kfifo_in(&tcp_task->r2tpool.queue,
(void *)&tcp_task->r2t,
sizeof(void *));
tcp_task->r2t = r2t = NULL;
}
}
if (r2t == NULL) {
if (kfifo_out(&tcp_task->r2tqueue,
(void *)&tcp_task->r2t, sizeof(void *)) !=
sizeof(void *))
r2t = NULL;
else
r2t = tcp_task->r2t;
}
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_unlock_bh(&tcp_task->queue2pool);
}
return r2t;
}
/**
* iscsi_tcp_task_xmit - xmit normal PDU task
* @task: iscsi command task
*
* We're expected to return 0 when everything was transmitted successfully,
* -EAGAIN if there's still data in the queue, or != 0 for any other kind
* of error.
*/
int iscsi_tcp_task_xmit(struct iscsi_task *task)
{
struct iscsi_conn *conn = task->conn;
struct iscsi_session *session = conn->session;
struct iscsi_r2t_info *r2t;
int rc = 0;
flush:
/* Flush any pending data first. */
rc = session->tt->xmit_pdu(task);
if (rc < 0)
return rc;
/* mgmt command */
if (!task->sc) {
if (task->hdr->itt == RESERVED_ITT)
iscsi_put_task(task);
return 0;
}
/* Are we done already? */
if (task->sc->sc_data_direction != DMA_TO_DEVICE)
return 0;
r2t = iscsi_tcp_get_curr_r2t(task);
if (r2t == NULL) {
/* Waiting for more R2Ts to arrive. */
ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
return 0;
}
rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
if (rc)
return rc;
iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
r2t, r2t->datasn - 1, task->hdr->itt,
r2t->data_offset + r2t->sent, r2t->data_count);
rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
r2t->data_count);
if (rc) {
iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
return rc;
}
r2t->sent += r2t->data_count;
goto flush;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
struct iscsi_cls_conn *
iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
uint32_t conn_idx)
{
struct iscsi_conn *conn;
struct iscsi_cls_conn *cls_conn;
struct iscsi_tcp_conn *tcp_conn;
cls_conn = iscsi_conn_setup(cls_session,
sizeof(*tcp_conn) + dd_data_size, conn_idx);
if (!cls_conn)
return NULL;
conn = cls_conn->dd_data;
/*
* due to strange issues with iser these are not set
* in iscsi_conn_setup
*/
conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
tcp_conn = conn->dd_data;
tcp_conn->iscsi_conn = conn;
tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
return cls_conn;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
{
iscsi_conn_teardown(cls_conn);
}
EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
{
int i;
int cmd_i;
/*
* initialize per-task: R2T pool and xmit queue
*/
for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
struct iscsi_task *task = session->cmds[cmd_i];
struct iscsi_tcp_task *tcp_task = task->dd_data;
/*
* pre-allocated x2 as much r2ts to handle race when
* target acks DataOut faster than we data_xmit() queues
* could replenish r2tqueue.
*/
/* R2T pool */
if (iscsi_pool_init(&tcp_task->r2tpool,
session->max_r2t * 2, NULL,
sizeof(struct iscsi_r2t_info))) {
goto r2t_alloc_fail;
}
/* R2T xmit queue */
kfifo: move struct kfifo in place This is a new generic kernel FIFO implementation. The current kernel fifo API is not very widely used, because it has to many constrains. Only 17 files in the current 2.6.31-rc5 used it. FIFO's are like list's a very basic thing and a kfifo API which handles the most use case would save a lot of development time and memory resources. I think this are the reasons why kfifo is not in use: - The API is to simple, important functions are missing - A fifo can be only allocated dynamically - There is a requirement of a spinlock whether you need it or not - There is no support for data records inside a fifo So I decided to extend the kfifo in a more generic way without blowing up the API to much. The new API has the following benefits: - Generic usage: For kernel internal use and/or device driver. - Provide an API for the most use case. - Slim API: The whole API provides 25 functions. - Linux style habit. - DECLARE_KFIFO, DEFINE_KFIFO and INIT_KFIFO Macros - Direct copy_to_user from the fifo and copy_from_user into the fifo. - The kfifo itself is an in place member of the using data structure, this save an indirection access and does not waste the kernel allocator. - Lockless access: if only one reader and one writer is active on the fifo, which is the common use case, no additional locking is necessary. - Remove spinlock - give the user the freedom of choice what kind of locking to use if one is required. - Ability to handle records. Three type of records are supported: - Variable length records between 0-255 bytes, with a record size field of 1 bytes. - Variable length records between 0-65535 bytes, with a record size field of 2 bytes. - Fixed size records, which no record size field. - Preserve memory resource. - Performance! - Easy to use! This patch: Since most users want to have the kfifo as part of another object, reorganize the code to allow including struct kfifo in another data structure. This requires changing the kfifo_alloc and kfifo_init prototypes so that we pass an existing kfifo pointer into them. This patch changes the implementation and all existing users. [akpm@linux-foundation.org: fix warning] Signed-off-by: Stefani Seibold <stefani@seibold.net> Acked-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com> Acked-by: Andi Kleen <ak@linux.intel.com> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-22 06:37:26 +08:00
if (kfifo_alloc(&tcp_task->r2tqueue,
session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
iscsi_pool_free(&tcp_task->r2tpool);
goto r2t_alloc_fail;
}
[SCSI] libiscsi: Reduce locking contention in fast path Replace the session lock with two locks, a forward lock and a backwards lock named frwd_lock and back_lock respectively. The forward lock protects resources that change while sending a request to the target, such as cmdsn, queued_cmdsn, and allocating task from the commands' pool with kfifo_out. The backward lock protects resources that change while processing a response or in error path, such as cmdsn_exp, cmdsn_max, and returning tasks to the commands' pool with kfifo_in. Under a steady state fast-path situation, that is when one or more processes/threads submit IO to an iscsi device and a single kernel upcall (e.g softirq) is dealing with processing of responses without errors, this patch eliminates the contention between the queuecommand()/request response/scsi_done() flows associated with iscsi sessions. Between the forward and the backward locks exists a strict locking hierarchy. The mutual exclusion zone protected by the forward lock can enclose the mutual exclusion zone protected by the backward lock but not vice versa. For example, in iscsi_conn_teardown or in iscsi_xmit_data when there is a failure and __iscsi_put_task is called, the backward lock is taken while the forward lock is still taken. On the other hand, if in the RX path a nop is to be sent, for example in iscsi_handle_reject or __iscsi_complete_pdu than the forward lock is released and the backward lock is taken for the duration of iscsi_send_nopout, later the backward lock is released and the forward lock is retaken. libiscsi_tcp uses two kernel fifos the r2t pool and the r2t queue. The insertion and deletion from these queues didn't corespond to the assumption taken by the new forward/backwards session locking paradigm. That is, in iscsi_tcp_clenup_task which belongs to the RX (backwards) path, r2t is taken out from r2t queue and inserted to the r2t pool. In iscsi_tcp_get_curr_r2t which belong to the TX (forward) path, r2t is also inserted to the r2t pool and another r2t is pulled from r2t queue. Only in iscsi_tcp_r2t_rsp which is called in the RX path but can requeue to the TX path, r2t is taken from the r2t pool and inserted to the r2t queue. In order to cope with this situation, two spin locks were added, pool2queue and queue2pool. The former protects extracting from the r2t pool and inserting to the r2t queue, and the later protects the extracing from the r2t queue and inserting to the r2t pool. Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> [minor fix up to apply cleanly and compile fix] Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2014-02-07 14:41:38 +08:00
spin_lock_init(&tcp_task->pool2queue);
spin_lock_init(&tcp_task->queue2pool);
}
return 0;
r2t_alloc_fail:
for (i = 0; i < cmd_i; i++) {
struct iscsi_task *task = session->cmds[i];
struct iscsi_tcp_task *tcp_task = task->dd_data;
kfifo: move struct kfifo in place This is a new generic kernel FIFO implementation. The current kernel fifo API is not very widely used, because it has to many constrains. Only 17 files in the current 2.6.31-rc5 used it. FIFO's are like list's a very basic thing and a kfifo API which handles the most use case would save a lot of development time and memory resources. I think this are the reasons why kfifo is not in use: - The API is to simple, important functions are missing - A fifo can be only allocated dynamically - There is a requirement of a spinlock whether you need it or not - There is no support for data records inside a fifo So I decided to extend the kfifo in a more generic way without blowing up the API to much. The new API has the following benefits: - Generic usage: For kernel internal use and/or device driver. - Provide an API for the most use case. - Slim API: The whole API provides 25 functions. - Linux style habit. - DECLARE_KFIFO, DEFINE_KFIFO and INIT_KFIFO Macros - Direct copy_to_user from the fifo and copy_from_user into the fifo. - The kfifo itself is an in place member of the using data structure, this save an indirection access and does not waste the kernel allocator. - Lockless access: if only one reader and one writer is active on the fifo, which is the common use case, no additional locking is necessary. - Remove spinlock - give the user the freedom of choice what kind of locking to use if one is required. - Ability to handle records. Three type of records are supported: - Variable length records between 0-255 bytes, with a record size field of 1 bytes. - Variable length records between 0-65535 bytes, with a record size field of 2 bytes. - Fixed size records, which no record size field. - Preserve memory resource. - Performance! - Easy to use! This patch: Since most users want to have the kfifo as part of another object, reorganize the code to allow including struct kfifo in another data structure. This requires changing the kfifo_alloc and kfifo_init prototypes so that we pass an existing kfifo pointer into them. This patch changes the implementation and all existing users. [akpm@linux-foundation.org: fix warning] Signed-off-by: Stefani Seibold <stefani@seibold.net> Acked-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com> Acked-by: Andi Kleen <ak@linux.intel.com> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-22 06:37:26 +08:00
kfifo_free(&tcp_task->r2tqueue);
iscsi_pool_free(&tcp_task->r2tpool);
}
return -ENOMEM;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
{
int i;
for (i = 0; i < session->cmds_max; i++) {
struct iscsi_task *task = session->cmds[i];
struct iscsi_tcp_task *tcp_task = task->dd_data;
kfifo: move struct kfifo in place This is a new generic kernel FIFO implementation. The current kernel fifo API is not very widely used, because it has to many constrains. Only 17 files in the current 2.6.31-rc5 used it. FIFO's are like list's a very basic thing and a kfifo API which handles the most use case would save a lot of development time and memory resources. I think this are the reasons why kfifo is not in use: - The API is to simple, important functions are missing - A fifo can be only allocated dynamically - There is a requirement of a spinlock whether you need it or not - There is no support for data records inside a fifo So I decided to extend the kfifo in a more generic way without blowing up the API to much. The new API has the following benefits: - Generic usage: For kernel internal use and/or device driver. - Provide an API for the most use case. - Slim API: The whole API provides 25 functions. - Linux style habit. - DECLARE_KFIFO, DEFINE_KFIFO and INIT_KFIFO Macros - Direct copy_to_user from the fifo and copy_from_user into the fifo. - The kfifo itself is an in place member of the using data structure, this save an indirection access and does not waste the kernel allocator. - Lockless access: if only one reader and one writer is active on the fifo, which is the common use case, no additional locking is necessary. - Remove spinlock - give the user the freedom of choice what kind of locking to use if one is required. - Ability to handle records. Three type of records are supported: - Variable length records between 0-255 bytes, with a record size field of 1 bytes. - Variable length records between 0-65535 bytes, with a record size field of 2 bytes. - Fixed size records, which no record size field. - Preserve memory resource. - Performance! - Easy to use! This patch: Since most users want to have the kfifo as part of another object, reorganize the code to allow including struct kfifo in another data structure. This requires changing the kfifo_alloc and kfifo_init prototypes so that we pass an existing kfifo pointer into them. This patch changes the implementation and all existing users. [akpm@linux-foundation.org: fix warning] Signed-off-by: Stefani Seibold <stefani@seibold.net> Acked-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com> Acked-by: Andi Kleen <ak@linux.intel.com> Acked-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-22 06:37:26 +08:00
kfifo_free(&tcp_task->r2tqueue);
iscsi_pool_free(&tcp_task->r2tpool);
}
}
EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
int iscsi_tcp_set_max_r2t(struct iscsi_conn *conn, char *buf)
{
struct iscsi_session *session = conn->session;
unsigned short r2ts = 0;
sscanf(buf, "%hu", &r2ts);
if (session->max_r2t == r2ts)
return 0;
if (!r2ts || !is_power_of_2(r2ts))
return -EINVAL;
session->max_r2t = r2ts;
iscsi_tcp_r2tpool_free(session);
return iscsi_tcp_r2tpool_alloc(session);
}
EXPORT_SYMBOL_GPL(iscsi_tcp_set_max_r2t);
void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
struct iscsi_stats *stats)
{
struct iscsi_conn *conn = cls_conn->dd_data;
stats->txdata_octets = conn->txdata_octets;
stats->rxdata_octets = conn->rxdata_octets;
stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
stats->dataout_pdus = conn->dataout_pdus_cnt;
stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
stats->datain_pdus = conn->datain_pdus_cnt;
stats->r2t_pdus = conn->r2t_pdus_cnt;
stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
}
EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);