1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
36 #include <scsi/scsi_dh.h>
38 #include <trace/events/scsi.h>
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
45 * Size of integrity metadata is usually small, 1 inline sg should
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define SCSI_INLINE_PROT_SG_CNT 0
50 #define SCSI_INLINE_SG_CNT 0
52 #define SCSI_INLINE_PROT_SG_CNT 1
53 #define SCSI_INLINE_SG_CNT 2
56 static struct kmem_cache
*scsi_sense_cache
;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex
);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
);
61 int scsi_init_sense_cache(struct Scsi_Host
*shost
)
65 mutex_lock(&scsi_sense_cache_mutex
);
66 if (!scsi_sense_cache
) {
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE
, 0, SLAB_HWCACHE_ALIGN
,
70 0, SCSI_SENSE_BUFFERSIZE
, NULL
);
71 if (!scsi_sense_cache
)
74 mutex_unlock(&scsi_sense_cache_mutex
);
79 scsi_set_blocked(struct scsi_cmnd
*cmd
, int reason
)
81 struct Scsi_Host
*host
= cmd
->device
->host
;
82 struct scsi_device
*device
= cmd
->device
;
83 struct scsi_target
*starget
= scsi_target(device
);
86 * Set the appropriate busy bit for the device/host.
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
99 case SCSI_MLQUEUE_HOST_BUSY
:
100 atomic_set(&host
->host_blocked
, host
->max_host_blocked
);
102 case SCSI_MLQUEUE_DEVICE_BUSY
:
103 case SCSI_MLQUEUE_EH_RETRY
:
104 atomic_set(&device
->device_blocked
,
105 device
->max_device_blocked
);
107 case SCSI_MLQUEUE_TARGET_BUSY
:
108 atomic_set(&starget
->target_blocked
,
109 starget
->max_target_blocked
);
114 static void scsi_mq_requeue_cmd(struct scsi_cmnd
*cmd
, unsigned long msecs
)
116 struct request
*rq
= scsi_cmd_to_rq(cmd
);
118 if (rq
->rq_flags
& RQF_DONTPREP
) {
119 rq
->rq_flags
&= ~RQF_DONTPREP
;
120 scsi_mq_uninit_cmd(cmd
);
125 blk_mq_requeue_request(rq
, false);
126 if (!scsi_host_in_recovery(cmd
->device
->host
))
127 blk_mq_delay_kick_requeue_list(rq
->q
, msecs
);
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
142 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, bool unbusy
)
144 struct scsi_device
*device
= cmd
->device
;
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
147 "Inserting command %p into mlqueue\n", cmd
));
149 scsi_set_blocked(cmd
, reason
);
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
156 scsi_device_unbusy(device
, cmd
);
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_mq_destroy_queue() finishes.
166 blk_mq_requeue_request(scsi_cmd_to_rq(cmd
),
167 !scsi_host_in_recovery(cmd
->device
->host
));
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
179 * Context: This could be called either from an interrupt context or a normal
182 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
184 __scsi_queue_insert(cmd
, reason
, true);
187 void scsi_failures_reset_retries(struct scsi_failures
*failures
)
189 struct scsi_failure
*failure
;
191 failures
->total_retries
= 0;
193 for (failure
= failures
->failure_definitions
; failure
->result
;
195 failure
->retries
= 0;
197 EXPORT_SYMBOL_GPL(scsi_failures_reset_retries
);
200 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
201 * @scmd: scsi_cmnd to check.
202 * @failures: scsi_failures struct that lists failures to check for.
204 * Returns -EAGAIN if the caller should retry else 0.
206 static int scsi_check_passthrough(struct scsi_cmnd
*scmd
,
207 struct scsi_failures
*failures
)
209 struct scsi_failure
*failure
;
210 struct scsi_sense_hdr sshdr
;
211 enum sam_status status
;
216 for (failure
= failures
->failure_definitions
; failure
->result
;
218 if (failure
->result
== SCMD_FAILURE_RESULT_ANY
)
221 if (host_byte(scmd
->result
) &&
222 host_byte(scmd
->result
) == host_byte(failure
->result
))
225 status
= status_byte(scmd
->result
);
229 if (failure
->result
== SCMD_FAILURE_STAT_ANY
&&
230 !scsi_status_is_good(scmd
->result
))
233 if (status
!= status_byte(failure
->result
))
236 if (status_byte(failure
->result
) != SAM_STAT_CHECK_CONDITION
||
237 failure
->sense
== SCMD_FAILURE_SENSE_ANY
)
240 if (!scsi_command_normalize_sense(scmd
, &sshdr
))
243 if (failure
->sense
!= sshdr
.sense_key
)
246 if (failure
->asc
== SCMD_FAILURE_ASC_ANY
)
249 if (failure
->asc
!= sshdr
.asc
)
252 if (failure
->ascq
== SCMD_FAILURE_ASCQ_ANY
||
253 failure
->ascq
== sshdr
.ascq
)
260 if (failure
->allowed
) {
261 if (failure
->allowed
== SCMD_FAILURE_NO_LIMIT
||
262 ++failure
->retries
<= failure
->allowed
)
265 if (failures
->total_allowed
== SCMD_FAILURE_NO_LIMIT
||
266 ++failures
->total_retries
<= failures
->total_allowed
)
274 * scsi_execute_cmd - insert request and wait for the result
277 * @opf: block layer request cmd_flags
278 * @buffer: data buffer
279 * @bufflen: len of buffer
280 * @timeout: request timeout in HZ
281 * @ml_retries: number of times SCSI midlayer will retry request
282 * @args: Optional args. See struct definition for field descriptions
284 * Returns the scsi_cmnd result field if a command was executed, or a negative
285 * Linux error code if we didn't get that far.
287 int scsi_execute_cmd(struct scsi_device
*sdev
, const unsigned char *cmd
,
288 blk_opf_t opf
, void *buffer
, unsigned int bufflen
,
289 int timeout
, int ml_retries
,
290 const struct scsi_exec_args
*args
)
292 static const struct scsi_exec_args default_args
;
294 struct scsi_cmnd
*scmd
;
298 args
= &default_args
;
299 else if (WARN_ON_ONCE(args
->sense
&&
300 args
->sense_len
!= SCSI_SENSE_BUFFERSIZE
))
304 req
= scsi_alloc_request(sdev
->request_queue
, opf
, args
->req_flags
);
309 ret
= blk_rq_map_kern(sdev
->request_queue
, req
,
310 buffer
, bufflen
, GFP_NOIO
);
314 scmd
= blk_mq_rq_to_pdu(req
);
315 scmd
->cmd_len
= COMMAND_SIZE(cmd
[0]);
316 memcpy(scmd
->cmnd
, cmd
, scmd
->cmd_len
);
317 scmd
->allowed
= ml_retries
;
318 scmd
->flags
|= args
->scmd_flags
;
319 req
->timeout
= timeout
;
320 req
->rq_flags
|= RQF_QUIET
;
323 * head injection *required* here otherwise quiesce won't work
325 blk_execute_rq(req
, true);
327 if (scsi_check_passthrough(scmd
, args
->failures
) == -EAGAIN
) {
328 blk_mq_free_request(req
);
333 * Some devices (USB mass-storage in particular) may transfer
334 * garbage data together with a residue indicating that the data
335 * is invalid. Prevent the garbage from being misinterpreted
336 * and prevent security leaks by zeroing out the excess data.
338 if (unlikely(scmd
->resid_len
> 0 && scmd
->resid_len
<= bufflen
))
339 memset(buffer
+ bufflen
- scmd
->resid_len
, 0, scmd
->resid_len
);
342 *args
->resid
= scmd
->resid_len
;
344 memcpy(args
->sense
, scmd
->sense_buffer
, SCSI_SENSE_BUFFERSIZE
);
346 scsi_normalize_sense(scmd
->sense_buffer
, scmd
->sense_len
,
351 blk_mq_free_request(req
);
355 EXPORT_SYMBOL(scsi_execute_cmd
);
358 * Wake up the error handler if necessary. Avoid as follows that the error
359 * handler is not woken up if host in-flight requests number ==
360 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
361 * with an RCU read lock in this function to ensure that this function in
362 * its entirety either finishes before scsi_eh_scmd_add() increases the
363 * host_failed counter or that it notices the shost state change made by
364 * scsi_eh_scmd_add().
366 static void scsi_dec_host_busy(struct Scsi_Host
*shost
, struct scsi_cmnd
*cmd
)
371 __clear_bit(SCMD_STATE_INFLIGHT
, &cmd
->state
);
372 if (unlikely(scsi_host_in_recovery(shost
))) {
373 unsigned int busy
= scsi_host_busy(shost
);
375 spin_lock_irqsave(shost
->host_lock
, flags
);
376 if (shost
->host_failed
|| shost
->host_eh_scheduled
)
377 scsi_eh_wakeup(shost
, busy
);
378 spin_unlock_irqrestore(shost
->host_lock
, flags
);
383 void scsi_device_unbusy(struct scsi_device
*sdev
, struct scsi_cmnd
*cmd
)
385 struct Scsi_Host
*shost
= sdev
->host
;
386 struct scsi_target
*starget
= scsi_target(sdev
);
388 scsi_dec_host_busy(shost
, cmd
);
390 if (starget
->can_queue
> 0)
391 atomic_dec(&starget
->target_busy
);
393 sbitmap_put(&sdev
->budget_map
, cmd
->budget_token
);
394 cmd
->budget_token
= -1;
398 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
399 * interrupts disabled.
401 static void scsi_kick_sdev_queue(struct scsi_device
*sdev
, void *data
)
403 struct scsi_device
*current_sdev
= data
;
405 if (sdev
!= current_sdev
)
406 blk_mq_run_hw_queues(sdev
->request_queue
, true);
410 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
411 * and call blk_run_queue for all the scsi_devices on the target -
412 * including current_sdev first.
414 * Called with *no* scsi locks held.
416 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
418 struct Scsi_Host
*shost
= current_sdev
->host
;
419 struct scsi_target
*starget
= scsi_target(current_sdev
);
422 spin_lock_irqsave(shost
->host_lock
, flags
);
423 starget
->starget_sdev_user
= NULL
;
424 spin_unlock_irqrestore(shost
->host_lock
, flags
);
427 * Call blk_run_queue for all LUNs on the target, starting with
428 * current_sdev. We race with others (to set starget_sdev_user),
429 * but in most cases, we will be first. Ideally, each LU on the
430 * target would get some limited time or requests on the target.
432 blk_mq_run_hw_queues(current_sdev
->request_queue
,
433 shost
->queuecommand_may_block
);
435 spin_lock_irqsave(shost
->host_lock
, flags
);
436 if (!starget
->starget_sdev_user
)
437 __starget_for_each_device(starget
, current_sdev
,
438 scsi_kick_sdev_queue
);
439 spin_unlock_irqrestore(shost
->host_lock
, flags
);
442 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
444 if (scsi_device_busy(sdev
) >= sdev
->queue_depth
)
446 if (atomic_read(&sdev
->device_blocked
) > 0)
451 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
453 if (starget
->can_queue
> 0) {
454 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
456 if (atomic_read(&starget
->target_blocked
) > 0)
462 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
464 if (atomic_read(&shost
->host_blocked
) > 0)
466 if (shost
->host_self_blocked
)
471 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
473 LIST_HEAD(starved_list
);
474 struct scsi_device
*sdev
;
477 spin_lock_irqsave(shost
->host_lock
, flags
);
478 list_splice_init(&shost
->starved_list
, &starved_list
);
480 while (!list_empty(&starved_list
)) {
481 struct request_queue
*slq
;
484 * As long as shost is accepting commands and we have
485 * starved queues, call blk_run_queue. scsi_request_fn
486 * drops the queue_lock and can add us back to the
489 * host_lock protects the starved_list and starved_entry.
490 * scsi_request_fn must get the host_lock before checking
491 * or modifying starved_list or starved_entry.
493 if (scsi_host_is_busy(shost
))
496 sdev
= list_entry(starved_list
.next
,
497 struct scsi_device
, starved_entry
);
498 list_del_init(&sdev
->starved_entry
);
499 if (scsi_target_is_busy(scsi_target(sdev
))) {
500 list_move_tail(&sdev
->starved_entry
,
501 &shost
->starved_list
);
506 * Once we drop the host lock, a racing scsi_remove_device()
507 * call may remove the sdev from the starved list and destroy
508 * it and the queue. Mitigate by taking a reference to the
509 * queue and never touching the sdev again after we drop the
510 * host lock. Note: if __scsi_remove_device() invokes
511 * blk_mq_destroy_queue() before the queue is run from this
512 * function then blk_run_queue() will return immediately since
513 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
515 slq
= sdev
->request_queue
;
516 if (!blk_get_queue(slq
))
518 spin_unlock_irqrestore(shost
->host_lock
, flags
);
520 blk_mq_run_hw_queues(slq
, false);
523 spin_lock_irqsave(shost
->host_lock
, flags
);
525 /* put any unprocessed entries back */
526 list_splice(&starved_list
, &shost
->starved_list
);
527 spin_unlock_irqrestore(shost
->host_lock
, flags
);
531 * scsi_run_queue - Select a proper request queue to serve next.
532 * @q: last request's queue
534 * The previous command was completely finished, start a new one if possible.
536 static void scsi_run_queue(struct request_queue
*q
)
538 struct scsi_device
*sdev
= q
->queuedata
;
540 if (scsi_target(sdev
)->single_lun
)
541 scsi_single_lun_run(sdev
);
542 if (!list_empty(&sdev
->host
->starved_list
))
543 scsi_starved_list_run(sdev
->host
);
545 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
546 blk_mq_kick_requeue_list(q
);
549 void scsi_requeue_run_queue(struct work_struct
*work
)
551 struct scsi_device
*sdev
;
552 struct request_queue
*q
;
554 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
555 q
= sdev
->request_queue
;
559 void scsi_run_host_queues(struct Scsi_Host
*shost
)
561 struct scsi_device
*sdev
;
563 shost_for_each_device(sdev
, shost
)
564 scsi_run_queue(sdev
->request_queue
);
567 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
569 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd
))) {
570 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
572 if (drv
->uninit_command
)
573 drv
->uninit_command(cmd
);
577 void scsi_free_sgtables(struct scsi_cmnd
*cmd
)
579 if (cmd
->sdb
.table
.nents
)
580 sg_free_table_chained(&cmd
->sdb
.table
,
582 if (scsi_prot_sg_count(cmd
))
583 sg_free_table_chained(&cmd
->prot_sdb
->table
,
584 SCSI_INLINE_PROT_SG_CNT
);
586 EXPORT_SYMBOL_GPL(scsi_free_sgtables
);
588 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
590 scsi_free_sgtables(cmd
);
591 scsi_uninit_cmd(cmd
);
594 static void scsi_run_queue_async(struct scsi_device
*sdev
)
596 if (scsi_host_in_recovery(sdev
->host
))
599 if (scsi_target(sdev
)->single_lun
||
600 !list_empty(&sdev
->host
->starved_list
)) {
601 kblockd_schedule_work(&sdev
->requeue_work
);
604 * smp_mb() present in sbitmap_queue_clear() or implied in
605 * .end_io is for ordering writing .device_busy in
606 * scsi_device_unbusy() and reading sdev->restarts.
608 int old
= atomic_read(&sdev
->restarts
);
611 * ->restarts has to be kept as non-zero if new budget
614 * No need to run queue when either another re-run
615 * queue wins in updating ->restarts or a new budget
618 if (old
&& atomic_cmpxchg(&sdev
->restarts
, old
, 0) == old
)
619 blk_mq_run_hw_queues(sdev
->request_queue
, true);
623 /* Returns false when no more bytes to process, true if there are more */
624 static bool scsi_end_request(struct request
*req
, blk_status_t error
,
627 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
628 struct scsi_device
*sdev
= cmd
->device
;
629 struct request_queue
*q
= sdev
->request_queue
;
631 if (blk_update_request(req
, error
, bytes
))
635 if (blk_queue_add_random(q
))
636 add_disk_randomness(req
->q
->disk
);
638 if (!blk_rq_is_passthrough(req
)) {
639 WARN_ON_ONCE(!(cmd
->flags
& SCMD_INITIALIZED
));
640 cmd
->flags
&= ~SCMD_INITIALIZED
;
644 * Calling rcu_barrier() is not necessary here because the
645 * SCSI error handler guarantees that the function called by
646 * call_rcu() has been called before scsi_end_request() is
649 destroy_rcu_head(&cmd
->rcu
);
652 * In the MQ case the command gets freed by __blk_mq_end_request,
653 * so we have to do all cleanup that depends on it earlier.
655 * We also can't kick the queues from irq context, so we
656 * will have to defer it to a workqueue.
658 scsi_mq_uninit_cmd(cmd
);
661 * queue is still alive, so grab the ref for preventing it
662 * from being cleaned up during running queue.
664 percpu_ref_get(&q
->q_usage_counter
);
666 __blk_mq_end_request(req
, error
);
668 scsi_run_queue_async(sdev
);
670 percpu_ref_put(&q
->q_usage_counter
);
675 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
676 * @result: scsi error code
678 * Translate a SCSI result code into a blk_status_t value.
680 static blk_status_t
scsi_result_to_blk_status(int result
)
683 * Check the scsi-ml byte first in case we converted a host or status
686 switch (scsi_ml_byte(result
)) {
689 case SCSIML_STAT_RESV_CONFLICT
:
690 return BLK_STS_RESV_CONFLICT
;
691 case SCSIML_STAT_NOSPC
:
692 return BLK_STS_NOSPC
;
693 case SCSIML_STAT_MED_ERROR
:
694 return BLK_STS_MEDIUM
;
695 case SCSIML_STAT_TGT_FAILURE
:
696 return BLK_STS_TARGET
;
697 case SCSIML_STAT_DL_TIMEOUT
:
698 return BLK_STS_DURATION_LIMIT
;
701 switch (host_byte(result
)) {
703 if (scsi_status_is_good(result
))
705 return BLK_STS_IOERR
;
706 case DID_TRANSPORT_FAILFAST
:
707 case DID_TRANSPORT_MARGINAL
:
708 return BLK_STS_TRANSPORT
;
710 return BLK_STS_IOERR
;
715 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
716 * @rq: request to examine
719 * A request could be merge of IOs which require different failure
720 * handling. This function determines the number of bytes which
721 * can be failed from the beginning of the request without
722 * crossing into area which need to be retried further.
725 * The number of bytes to fail.
727 static unsigned int scsi_rq_err_bytes(const struct request
*rq
)
729 blk_opf_t ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
730 unsigned int bytes
= 0;
733 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
734 return blk_rq_bytes(rq
);
737 * Currently the only 'mixing' which can happen is between
738 * different fastfail types. We can safely fail portions
739 * which have all the failfast bits that the first one has -
740 * the ones which are at least as eager to fail as the first
743 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
744 if ((bio
->bi_opf
& ff
) != ff
)
746 bytes
+= bio
->bi_iter
.bi_size
;
749 /* this could lead to infinite loop */
750 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
754 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd
*cmd
)
756 struct request
*req
= scsi_cmd_to_rq(cmd
);
757 unsigned long wait_for
;
759 if (cmd
->allowed
== SCSI_CMD_RETRIES_NO_LIMIT
)
762 wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
763 if (time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
764 scmd_printk(KERN_ERR
, cmd
, "timing out command, waited %lus\n",
772 * When ALUA transition state is returned, reprep the cmd to
773 * use the ALUA handler's transition timeout. Delay the reprep
774 * 1 sec to avoid aggressive retries of the target in that
777 #define ALUA_TRANSITION_REPREP_DELAY 1000
779 /* Helper for scsi_io_completion() when special action required. */
780 static void scsi_io_completion_action(struct scsi_cmnd
*cmd
, int result
)
782 struct request
*req
= scsi_cmd_to_rq(cmd
);
784 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_DELAYED_REPREP
,
785 ACTION_RETRY
, ACTION_DELAYED_RETRY
} action
;
786 struct scsi_sense_hdr sshdr
;
788 bool sense_current
= true; /* false implies "deferred sense" */
789 blk_status_t blk_stat
;
791 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
793 sense_current
= !scsi_sense_is_deferred(&sshdr
);
795 blk_stat
= scsi_result_to_blk_status(result
);
797 if (host_byte(result
) == DID_RESET
) {
798 /* Third party bus reset or reset for error recovery
799 * reasons. Just retry the command and see what
802 action
= ACTION_RETRY
;
803 } else if (sense_valid
&& sense_current
) {
804 switch (sshdr
.sense_key
) {
806 if (cmd
->device
->removable
) {
807 /* Detected disc change. Set a bit
808 * and quietly refuse further access.
810 cmd
->device
->changed
= 1;
811 action
= ACTION_FAIL
;
813 /* Must have been a power glitch, or a
814 * bus reset. Could not have been a
815 * media change, so we just retry the
816 * command and see what happens.
818 action
= ACTION_RETRY
;
821 case ILLEGAL_REQUEST
:
822 /* If we had an ILLEGAL REQUEST returned, then
823 * we may have performed an unsupported
824 * command. The only thing this should be
825 * would be a ten byte read where only a six
826 * byte read was supported. Also, on a system
827 * where READ CAPACITY failed, we may have
828 * read past the end of the disk.
830 if ((cmd
->device
->use_10_for_rw
&&
831 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
832 (cmd
->cmnd
[0] == READ_10
||
833 cmd
->cmnd
[0] == WRITE_10
)) {
834 /* This will issue a new 6-byte command. */
835 cmd
->device
->use_10_for_rw
= 0;
836 action
= ACTION_REPREP
;
837 } else if (sshdr
.asc
== 0x10) /* DIX */ {
838 action
= ACTION_FAIL
;
839 blk_stat
= BLK_STS_PROTECTION
;
840 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
841 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
842 action
= ACTION_FAIL
;
843 blk_stat
= BLK_STS_TARGET
;
845 action
= ACTION_FAIL
;
847 case ABORTED_COMMAND
:
848 action
= ACTION_FAIL
;
849 if (sshdr
.asc
== 0x10) /* DIF */
850 blk_stat
= BLK_STS_PROTECTION
;
853 /* If the device is in the process of becoming
854 * ready, or has a temporary blockage, retry.
856 if (sshdr
.asc
== 0x04) {
857 switch (sshdr
.ascq
) {
858 case 0x01: /* becoming ready */
859 case 0x04: /* format in progress */
860 case 0x05: /* rebuild in progress */
861 case 0x06: /* recalculation in progress */
862 case 0x07: /* operation in progress */
863 case 0x08: /* Long write in progress */
864 case 0x09: /* self test in progress */
865 case 0x11: /* notify (enable spinup) required */
866 case 0x14: /* space allocation in progress */
867 case 0x1a: /* start stop unit in progress */
868 case 0x1b: /* sanitize in progress */
869 case 0x1d: /* configuration in progress */
870 case 0x24: /* depopulation in progress */
871 case 0x25: /* depopulation restore in progress */
872 action
= ACTION_DELAYED_RETRY
;
874 case 0x0a: /* ALUA state transition */
875 action
= ACTION_DELAYED_REPREP
;
878 action
= ACTION_FAIL
;
882 action
= ACTION_FAIL
;
884 case VOLUME_OVERFLOW
:
885 /* See SSC3rXX or current. */
886 action
= ACTION_FAIL
;
889 action
= ACTION_FAIL
;
890 if ((sshdr
.asc
== 0x0C && sshdr
.ascq
== 0x12) ||
891 (sshdr
.asc
== 0x55 &&
892 (sshdr
.ascq
== 0x0E || sshdr
.ascq
== 0x0F))) {
893 /* Insufficient zone resources */
894 blk_stat
= BLK_STS_ZONE_OPEN_RESOURCE
;
900 action
= ACTION_FAIL
;
904 action
= ACTION_FAIL
;
906 if (action
!= ACTION_FAIL
&& scsi_cmd_runtime_exceeced(cmd
))
907 action
= ACTION_FAIL
;
911 /* Give up and fail the remainder of the request */
912 if (!(req
->rq_flags
& RQF_QUIET
)) {
913 static DEFINE_RATELIMIT_STATE(_rs
,
914 DEFAULT_RATELIMIT_INTERVAL
,
915 DEFAULT_RATELIMIT_BURST
);
917 if (unlikely(scsi_logging_level
))
919 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT
,
920 SCSI_LOG_MLCOMPLETE_BITS
);
923 * if logging is enabled the failure will be printed
924 * in scsi_log_completion(), so avoid duplicate messages
926 if (!level
&& __ratelimit(&_rs
)) {
927 scsi_print_result(cmd
, NULL
, FAILED
);
929 scsi_print_sense(cmd
);
930 scsi_print_command(cmd
);
933 if (!scsi_end_request(req
, blk_stat
, scsi_rq_err_bytes(req
)))
937 scsi_mq_requeue_cmd(cmd
, 0);
939 case ACTION_DELAYED_REPREP
:
940 scsi_mq_requeue_cmd(cmd
, ALUA_TRANSITION_REPREP_DELAY
);
943 /* Retry the same command immediately */
944 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, false);
946 case ACTION_DELAYED_RETRY
:
947 /* Retry the same command after a delay */
948 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, false);
954 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
955 * new result that may suppress further error checking. Also modifies
956 * *blk_statp in some cases.
958 static int scsi_io_completion_nz_result(struct scsi_cmnd
*cmd
, int result
,
959 blk_status_t
*blk_statp
)
962 bool sense_current
= true; /* false implies "deferred sense" */
963 struct request
*req
= scsi_cmd_to_rq(cmd
);
964 struct scsi_sense_hdr sshdr
;
966 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
968 sense_current
= !scsi_sense_is_deferred(&sshdr
);
970 if (blk_rq_is_passthrough(req
)) {
973 * SG_IO wants current and deferred errors
975 cmd
->sense_len
= min(8 + cmd
->sense_buffer
[7],
976 SCSI_SENSE_BUFFERSIZE
);
979 *blk_statp
= scsi_result_to_blk_status(result
);
980 } else if (blk_rq_bytes(req
) == 0 && sense_current
) {
982 * Flush commands do not transfers any data, and thus cannot use
983 * good_bytes != blk_rq_bytes(req) as the signal for an error.
984 * This sets *blk_statp explicitly for the problem case.
986 *blk_statp
= scsi_result_to_blk_status(result
);
989 * Recovered errors need reporting, but they're always treated as
990 * success, so fiddle the result code here. For passthrough requests
991 * we already took a copy of the original into sreq->result which
992 * is what gets returned to the user
994 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
995 bool do_print
= true;
997 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
998 * skip print since caller wants ATA registers. Only occurs
999 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
1001 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
1003 else if (req
->rq_flags
& RQF_QUIET
)
1006 scsi_print_sense(cmd
);
1008 /* for passthrough, *blk_statp may be set */
1009 *blk_statp
= BLK_STS_OK
;
1012 * Another corner case: the SCSI status byte is non-zero but 'good'.
1013 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1014 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1015 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1016 * intermediate statuses (both obsolete in SAM-4) as good.
1018 if ((result
& 0xff) && scsi_status_is_good(result
)) {
1020 *blk_statp
= BLK_STS_OK
;
1026 * scsi_io_completion - Completion processing for SCSI commands.
1027 * @cmd: command that is finished.
1028 * @good_bytes: number of processed bytes.
1030 * We will finish off the specified number of sectors. If we are done, the
1031 * command block will be released and the queue function will be goosed. If we
1032 * are not done then we have to figure out what to do next:
1034 * a) We can call scsi_mq_requeue_cmd(). The request will be
1035 * unprepared and put back on the queue. Then a new command will
1036 * be created for it. This should be used if we made forward
1037 * progress, or if we want to switch from READ(10) to READ(6) for
1040 * b) We can call scsi_io_completion_action(). The request will be
1041 * put back on the queue and retried using the same command as
1042 * before, possibly after a delay.
1044 * c) We can call scsi_end_request() with blk_stat other than
1045 * BLK_STS_OK, to fail the remainder of the request.
1047 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
1049 int result
= cmd
->result
;
1050 struct request
*req
= scsi_cmd_to_rq(cmd
);
1051 blk_status_t blk_stat
= BLK_STS_OK
;
1053 if (unlikely(result
)) /* a nz result may or may not be an error */
1054 result
= scsi_io_completion_nz_result(cmd
, result
, &blk_stat
);
1057 * Next deal with any sectors which we were able to correctly
1060 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
1061 "%u sectors total, %d bytes done.\n",
1062 blk_rq_sectors(req
), good_bytes
));
1065 * Failed, zero length commands always need to drop down
1066 * to retry code. Fast path should return in this block.
1068 if (likely(blk_rq_bytes(req
) > 0 || blk_stat
== BLK_STS_OK
)) {
1069 if (likely(!scsi_end_request(req
, blk_stat
, good_bytes
)))
1070 return; /* no bytes remaining */
1073 /* Kill remainder if no retries. */
1074 if (unlikely(blk_stat
&& scsi_noretry_cmd(cmd
))) {
1075 if (scsi_end_request(req
, blk_stat
, blk_rq_bytes(req
)))
1077 "Bytes remaining after failed, no-retry command");
1082 * If there had been no error, but we have leftover bytes in the
1083 * request just queue the command up again.
1085 if (likely(result
== 0))
1086 scsi_mq_requeue_cmd(cmd
, 0);
1088 scsi_io_completion_action(cmd
, result
);
1091 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device
*sdev
,
1094 return sdev
->dma_drain_len
&& blk_rq_is_passthrough(rq
) &&
1095 !op_is_write(req_op(rq
)) &&
1096 sdev
->host
->hostt
->dma_need_drain(rq
);
1100 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1101 * @cmd: SCSI command data structure to initialize.
1103 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1107 * * BLK_STS_OK - on success
1108 * * BLK_STS_RESOURCE - if the failure is retryable
1109 * * BLK_STS_IOERR - if the failure is fatal
1111 blk_status_t
scsi_alloc_sgtables(struct scsi_cmnd
*cmd
)
1113 struct scsi_device
*sdev
= cmd
->device
;
1114 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1115 unsigned short nr_segs
= blk_rq_nr_phys_segments(rq
);
1116 struct scatterlist
*last_sg
= NULL
;
1118 bool need_drain
= scsi_cmd_needs_dma_drain(sdev
, rq
);
1121 if (WARN_ON_ONCE(!nr_segs
))
1122 return BLK_STS_IOERR
;
1125 * Make sure there is space for the drain. The driver must adjust
1126 * max_hw_segments to be prepared for this.
1132 * If sg table allocation fails, requeue request later.
1134 if (unlikely(sg_alloc_table_chained(&cmd
->sdb
.table
, nr_segs
,
1135 cmd
->sdb
.table
.sgl
, SCSI_INLINE_SG_CNT
)))
1136 return BLK_STS_RESOURCE
;
1139 * Next, walk the list, and fill in the addresses and sizes of
1142 count
= __blk_rq_map_sg(rq
->q
, rq
, cmd
->sdb
.table
.sgl
, &last_sg
);
1144 if (blk_rq_bytes(rq
) & rq
->q
->dma_pad_mask
) {
1145 unsigned int pad_len
=
1146 (rq
->q
->dma_pad_mask
& ~blk_rq_bytes(rq
)) + 1;
1148 last_sg
->length
+= pad_len
;
1149 cmd
->extra_len
+= pad_len
;
1153 sg_unmark_end(last_sg
);
1154 last_sg
= sg_next(last_sg
);
1155 sg_set_buf(last_sg
, sdev
->dma_drain_buf
, sdev
->dma_drain_len
);
1156 sg_mark_end(last_sg
);
1158 cmd
->extra_len
+= sdev
->dma_drain_len
;
1162 BUG_ON(count
> cmd
->sdb
.table
.nents
);
1163 cmd
->sdb
.table
.nents
= count
;
1164 cmd
->sdb
.length
= blk_rq_payload_bytes(rq
);
1166 if (blk_integrity_rq(rq
)) {
1167 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1170 if (WARN_ON_ONCE(!prot_sdb
)) {
1172 * This can happen if someone (e.g. multipath)
1173 * queues a command to a device on an adapter
1174 * that does not support DIX.
1176 ret
= BLK_STS_IOERR
;
1177 goto out_free_sgtables
;
1180 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1182 if (sg_alloc_table_chained(&prot_sdb
->table
, ivecs
,
1183 prot_sdb
->table
.sgl
,
1184 SCSI_INLINE_PROT_SG_CNT
)) {
1185 ret
= BLK_STS_RESOURCE
;
1186 goto out_free_sgtables
;
1189 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1190 prot_sdb
->table
.sgl
);
1191 BUG_ON(count
> ivecs
);
1192 BUG_ON(count
> queue_max_integrity_segments(rq
->q
));
1194 cmd
->prot_sdb
= prot_sdb
;
1195 cmd
->prot_sdb
->table
.nents
= count
;
1200 scsi_free_sgtables(cmd
);
1203 EXPORT_SYMBOL(scsi_alloc_sgtables
);
1206 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1207 * @rq: Request associated with the SCSI command to be initialized.
1209 * This function initializes the members of struct scsi_cmnd that must be
1210 * initialized before request processing starts and that won't be
1211 * reinitialized if a SCSI command is requeued.
1213 static void scsi_initialize_rq(struct request
*rq
)
1215 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1217 memset(cmd
->cmnd
, 0, sizeof(cmd
->cmnd
));
1218 cmd
->cmd_len
= MAX_COMMAND_SIZE
;
1220 init_rcu_head(&cmd
->rcu
);
1221 cmd
->jiffies_at_alloc
= jiffies
;
1225 struct request
*scsi_alloc_request(struct request_queue
*q
, blk_opf_t opf
,
1226 blk_mq_req_flags_t flags
)
1230 rq
= blk_mq_alloc_request(q
, opf
, flags
);
1232 scsi_initialize_rq(rq
);
1235 EXPORT_SYMBOL_GPL(scsi_alloc_request
);
1238 * Only called when the request isn't completed by SCSI, and not freed by
1241 static void scsi_cleanup_rq(struct request
*rq
)
1243 if (rq
->rq_flags
& RQF_DONTPREP
) {
1244 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq
));
1245 rq
->rq_flags
&= ~RQF_DONTPREP
;
1249 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1250 void scsi_init_command(struct scsi_device
*dev
, struct scsi_cmnd
*cmd
)
1252 struct request
*rq
= scsi_cmd_to_rq(cmd
);
1254 if (!blk_rq_is_passthrough(rq
) && !(cmd
->flags
& SCMD_INITIALIZED
)) {
1255 cmd
->flags
|= SCMD_INITIALIZED
;
1256 scsi_initialize_rq(rq
);
1260 INIT_LIST_HEAD(&cmd
->eh_entry
);
1261 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1264 static blk_status_t
scsi_setup_scsi_cmnd(struct scsi_device
*sdev
,
1265 struct request
*req
)
1267 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1270 * Passthrough requests may transfer data, in which case they must
1271 * a bio attached to them. Or they might contain a SCSI command
1272 * that does not transfer data, in which case they may optionally
1273 * submit a request without an attached bio.
1276 blk_status_t ret
= scsi_alloc_sgtables(cmd
);
1277 if (unlikely(ret
!= BLK_STS_OK
))
1280 BUG_ON(blk_rq_bytes(req
));
1282 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1285 cmd
->transfersize
= blk_rq_bytes(req
);
1290 scsi_device_state_check(struct scsi_device
*sdev
, struct request
*req
)
1292 switch (sdev
->sdev_state
) {
1296 case SDEV_TRANSPORT_OFFLINE
:
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1302 if (!sdev
->offline_already
) {
1303 sdev
->offline_already
= true;
1304 sdev_printk(KERN_ERR
, sdev
,
1305 "rejecting I/O to offline device\n");
1307 return BLK_STS_IOERR
;
1310 * If the device is fully deleted, we refuse to
1311 * process any commands as well.
1313 sdev_printk(KERN_ERR
, sdev
,
1314 "rejecting I/O to dead device\n");
1315 return BLK_STS_IOERR
;
1317 case SDEV_CREATED_BLOCK
:
1318 return BLK_STS_RESOURCE
;
1321 * If the device is blocked we only accept power management
1324 if (req
&& WARN_ON_ONCE(!(req
->rq_flags
& RQF_PM
)))
1325 return BLK_STS_RESOURCE
;
1329 * For any other not fully online state we only allow
1330 * power management commands.
1332 if (req
&& !(req
->rq_flags
& RQF_PM
))
1333 return BLK_STS_OFFLINE
;
1339 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1340 * and return the token else return -1.
1342 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1343 struct scsi_device
*sdev
)
1347 token
= sbitmap_get(&sdev
->budget_map
);
1351 if (!atomic_read(&sdev
->device_blocked
))
1355 * Only unblock if no other commands are pending and
1356 * if device_blocked has decreased to zero
1358 if (scsi_device_busy(sdev
) > 1 ||
1359 atomic_dec_return(&sdev
->device_blocked
) > 0) {
1360 sbitmap_put(&sdev
->budget_map
, token
);
1364 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1365 "unblocking device at zero depth\n"));
1371 * scsi_target_queue_ready: checks if there we can send commands to target
1372 * @sdev: scsi device on starget to check.
1374 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1375 struct scsi_device
*sdev
)
1377 struct scsi_target
*starget
= scsi_target(sdev
);
1380 if (starget
->single_lun
) {
1381 spin_lock_irq(shost
->host_lock
);
1382 if (starget
->starget_sdev_user
&&
1383 starget
->starget_sdev_user
!= sdev
) {
1384 spin_unlock_irq(shost
->host_lock
);
1387 starget
->starget_sdev_user
= sdev
;
1388 spin_unlock_irq(shost
->host_lock
);
1391 if (starget
->can_queue
<= 0)
1394 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1395 if (atomic_read(&starget
->target_blocked
) > 0) {
1400 * unblock after target_blocked iterates to zero
1402 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1405 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1406 "unblocking target at zero depth\n"));
1409 if (busy
>= starget
->can_queue
)
1415 spin_lock_irq(shost
->host_lock
);
1416 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1417 spin_unlock_irq(shost
->host_lock
);
1419 if (starget
->can_queue
> 0)
1420 atomic_dec(&starget
->target_busy
);
1425 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1426 * return 0. We must end up running the queue again whenever 0 is
1427 * returned, else IO can hang.
1429 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1430 struct Scsi_Host
*shost
,
1431 struct scsi_device
*sdev
,
1432 struct scsi_cmnd
*cmd
)
1434 if (atomic_read(&shost
->host_blocked
) > 0) {
1435 if (scsi_host_busy(shost
) > 0)
1439 * unblock after host_blocked iterates to zero
1441 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1445 shost_printk(KERN_INFO
, shost
,
1446 "unblocking host at zero depth\n"));
1449 if (shost
->host_self_blocked
)
1452 /* We're OK to process the command, so we can't be starved */
1453 if (!list_empty(&sdev
->starved_entry
)) {
1454 spin_lock_irq(shost
->host_lock
);
1455 if (!list_empty(&sdev
->starved_entry
))
1456 list_del_init(&sdev
->starved_entry
);
1457 spin_unlock_irq(shost
->host_lock
);
1460 __set_bit(SCMD_STATE_INFLIGHT
, &cmd
->state
);
1465 spin_lock_irq(shost
->host_lock
);
1466 if (list_empty(&sdev
->starved_entry
))
1467 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1468 spin_unlock_irq(shost
->host_lock
);
1470 scsi_dec_host_busy(shost
, cmd
);
1475 * Busy state exporting function for request stacking drivers.
1477 * For efficiency, no lock is taken to check the busy state of
1478 * shost/starget/sdev, since the returned value is not guaranteed and
1479 * may be changed after request stacking drivers call the function,
1480 * regardless of taking lock or not.
1482 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1483 * needs to return 'not busy'. Otherwise, request stacking drivers
1484 * may hold requests forever.
1486 static bool scsi_mq_lld_busy(struct request_queue
*q
)
1488 struct scsi_device
*sdev
= q
->queuedata
;
1489 struct Scsi_Host
*shost
;
1491 if (blk_queue_dying(q
))
1497 * Ignore host/starget busy state.
1498 * Since block layer does not have a concept of fairness across
1499 * multiple queues, congestion of host/starget needs to be handled
1502 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1509 * Block layer request completion callback. May be called from interrupt
1512 static void scsi_complete(struct request
*rq
)
1514 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1515 enum scsi_disposition disposition
;
1517 INIT_LIST_HEAD(&cmd
->eh_entry
);
1519 atomic_inc(&cmd
->device
->iodone_cnt
);
1521 atomic_inc(&cmd
->device
->ioerr_cnt
);
1523 disposition
= scsi_decide_disposition(cmd
);
1524 if (disposition
!= SUCCESS
&& scsi_cmd_runtime_exceeced(cmd
))
1525 disposition
= SUCCESS
;
1527 scsi_log_completion(cmd
, disposition
);
1529 switch (disposition
) {
1531 scsi_finish_command(cmd
);
1534 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1536 case ADD_TO_MLQUEUE
:
1537 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1540 scsi_eh_scmd_add(cmd
);
1546 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1547 * @cmd: command block we are dispatching.
1549 * Return: nonzero return request was rejected and device's queue needs to be
1552 static int scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
1554 struct Scsi_Host
*host
= cmd
->device
->host
;
1557 atomic_inc(&cmd
->device
->iorequest_cnt
);
1559 /* check if the device is still usable */
1560 if (unlikely(cmd
->device
->sdev_state
== SDEV_DEL
)) {
1561 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1562 * returns an immediate error upwards, and signals
1563 * that the device is no longer present */
1564 cmd
->result
= DID_NO_CONNECT
<< 16;
1568 /* Check to see if the scsi lld made this device blocked. */
1569 if (unlikely(scsi_device_blocked(cmd
->device
))) {
1571 * in blocked state, the command is just put back on
1572 * the device queue. The suspend state has already
1573 * blocked the queue so future requests should not
1574 * occur until the device transitions out of the
1577 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1578 "queuecommand : device blocked\n"));
1579 atomic_dec(&cmd
->device
->iorequest_cnt
);
1580 return SCSI_MLQUEUE_DEVICE_BUSY
;
1583 /* Store the LUN value in cmnd, if needed. */
1584 if (cmd
->device
->lun_in_cdb
)
1585 cmd
->cmnd
[1] = (cmd
->cmnd
[1] & 0x1f) |
1586 (cmd
->device
->lun
<< 5 & 0xe0);
1591 * Before we queue this command, check if the command
1592 * length exceeds what the host adapter can handle.
1594 if (cmd
->cmd_len
> cmd
->device
->host
->max_cmd_len
) {
1595 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1596 "queuecommand : command too long. "
1597 "cdb_size=%d host->max_cmd_len=%d\n",
1598 cmd
->cmd_len
, cmd
->device
->host
->max_cmd_len
));
1599 cmd
->result
= (DID_ABORT
<< 16);
1603 if (unlikely(host
->shost_state
== SHOST_DEL
)) {
1604 cmd
->result
= (DID_NO_CONNECT
<< 16);
1609 trace_scsi_dispatch_cmd_start(cmd
);
1610 rtn
= host
->hostt
->queuecommand(host
, cmd
);
1612 atomic_dec(&cmd
->device
->iorequest_cnt
);
1613 trace_scsi_dispatch_cmd_error(cmd
, rtn
);
1614 if (rtn
!= SCSI_MLQUEUE_DEVICE_BUSY
&&
1615 rtn
!= SCSI_MLQUEUE_TARGET_BUSY
)
1616 rtn
= SCSI_MLQUEUE_HOST_BUSY
;
1618 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1619 "queuecommand : request rejected\n"));
1628 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1629 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host
*shost
)
1631 return min_t(unsigned int, shost
->sg_tablesize
, SCSI_INLINE_SG_CNT
) *
1632 sizeof(struct scatterlist
);
1635 static blk_status_t
scsi_prepare_cmd(struct request
*req
)
1637 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1638 struct scsi_device
*sdev
= req
->q
->queuedata
;
1639 struct Scsi_Host
*shost
= sdev
->host
;
1640 bool in_flight
= test_bit(SCMD_STATE_INFLIGHT
, &cmd
->state
);
1641 struct scatterlist
*sg
;
1643 scsi_init_command(sdev
, cmd
);
1647 cmd
->prot_flags
= 0;
1649 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1651 cmd
->transfersize
= 0;
1652 cmd
->host_scribble
= NULL
;
1657 __set_bit(SCMD_STATE_INFLIGHT
, &cmd
->state
);
1660 * Only clear the driver-private command data if the LLD does not supply
1661 * a function to initialize that data.
1663 if (!shost
->hostt
->init_cmd_priv
)
1664 memset(cmd
+ 1, 0, shost
->hostt
->cmd_size
);
1666 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1667 if (blk_rq_bytes(req
))
1668 cmd
->sc_data_direction
= rq_dma_dir(req
);
1670 cmd
->sc_data_direction
= DMA_NONE
;
1672 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1673 cmd
->sdb
.table
.sgl
= sg
;
1675 if (scsi_host_get_prot(shost
)) {
1676 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1678 cmd
->prot_sdb
->table
.sgl
=
1679 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1683 * Special handling for passthrough commands, which don't go to the ULP
1686 if (blk_rq_is_passthrough(req
))
1687 return scsi_setup_scsi_cmnd(sdev
, req
);
1689 if (sdev
->handler
&& sdev
->handler
->prep_fn
) {
1690 blk_status_t ret
= sdev
->handler
->prep_fn(sdev
, req
);
1692 if (ret
!= BLK_STS_OK
)
1696 /* Usually overridden by the ULP */
1698 memset(cmd
->cmnd
, 0, sizeof(cmd
->cmnd
));
1699 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1702 static void scsi_done_internal(struct scsi_cmnd
*cmd
, bool complete_directly
)
1704 struct request
*req
= scsi_cmd_to_rq(cmd
);
1706 switch (cmd
->submitter
) {
1707 case SUBMITTED_BY_BLOCK_LAYER
:
1709 case SUBMITTED_BY_SCSI_ERROR_HANDLER
:
1710 return scsi_eh_done(cmd
);
1711 case SUBMITTED_BY_SCSI_RESET_IOCTL
:
1715 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd
)->q
)))
1717 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE
, &cmd
->state
)))
1719 trace_scsi_dispatch_cmd_done(cmd
);
1721 if (complete_directly
)
1722 blk_mq_complete_request_direct(req
, scsi_complete
);
1724 blk_mq_complete_request(req
);
1727 void scsi_done(struct scsi_cmnd
*cmd
)
1729 scsi_done_internal(cmd
, false);
1731 EXPORT_SYMBOL(scsi_done
);
1733 void scsi_done_direct(struct scsi_cmnd
*cmd
)
1735 scsi_done_internal(cmd
, true);
1737 EXPORT_SYMBOL(scsi_done_direct
);
1739 static void scsi_mq_put_budget(struct request_queue
*q
, int budget_token
)
1741 struct scsi_device
*sdev
= q
->queuedata
;
1743 sbitmap_put(&sdev
->budget_map
, budget_token
);
1747 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1748 * not change behaviour from the previous unplug mechanism, experimentation
1749 * may prove this needs changing.
1751 #define SCSI_QUEUE_DELAY 3
1753 static int scsi_mq_get_budget(struct request_queue
*q
)
1755 struct scsi_device
*sdev
= q
->queuedata
;
1756 int token
= scsi_dev_queue_ready(q
, sdev
);
1761 atomic_inc(&sdev
->restarts
);
1764 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1765 * .restarts must be incremented before .device_busy is read because the
1766 * code in scsi_run_queue_async() depends on the order of these operations.
1768 smp_mb__after_atomic();
1771 * If all in-flight requests originated from this LUN are completed
1772 * before reading .device_busy, sdev->device_busy will be observed as
1773 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1774 * soon. Otherwise, completion of one of these requests will observe
1775 * the .restarts flag, and the request queue will be run for handling
1776 * this request, see scsi_end_request().
1778 if (unlikely(scsi_device_busy(sdev
) == 0 &&
1779 !scsi_device_blocked(sdev
)))
1780 blk_mq_delay_run_hw_queues(sdev
->request_queue
, SCSI_QUEUE_DELAY
);
1784 static void scsi_mq_set_rq_budget_token(struct request
*req
, int token
)
1786 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1788 cmd
->budget_token
= token
;
1791 static int scsi_mq_get_rq_budget_token(struct request
*req
)
1793 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1795 return cmd
->budget_token
;
1798 static blk_status_t
scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1799 const struct blk_mq_queue_data
*bd
)
1801 struct request
*req
= bd
->rq
;
1802 struct request_queue
*q
= req
->q
;
1803 struct scsi_device
*sdev
= q
->queuedata
;
1804 struct Scsi_Host
*shost
= sdev
->host
;
1805 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1809 WARN_ON_ONCE(cmd
->budget_token
< 0);
1812 * If the device is not in running state we will reject some or all
1815 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1816 ret
= scsi_device_state_check(sdev
, req
);
1817 if (ret
!= BLK_STS_OK
)
1818 goto out_put_budget
;
1821 ret
= BLK_STS_RESOURCE
;
1822 if (!scsi_target_queue_ready(shost
, sdev
))
1823 goto out_put_budget
;
1824 if (unlikely(scsi_host_in_recovery(shost
))) {
1825 if (cmd
->flags
& SCMD_FAIL_IF_RECOVERING
)
1826 ret
= BLK_STS_OFFLINE
;
1827 goto out_dec_target_busy
;
1829 if (!scsi_host_queue_ready(q
, shost
, sdev
, cmd
))
1830 goto out_dec_target_busy
;
1832 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
1833 ret
= scsi_prepare_cmd(req
);
1834 if (ret
!= BLK_STS_OK
)
1835 goto out_dec_host_busy
;
1836 req
->rq_flags
|= RQF_DONTPREP
;
1838 clear_bit(SCMD_STATE_COMPLETE
, &cmd
->state
);
1841 cmd
->flags
&= SCMD_PRESERVED_FLAGS
;
1842 if (sdev
->simple_tags
)
1843 cmd
->flags
|= SCMD_TAGGED
;
1845 cmd
->flags
|= SCMD_LAST
;
1847 scsi_set_resid(cmd
, 0);
1848 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
1849 cmd
->submitter
= SUBMITTED_BY_BLOCK_LAYER
;
1851 blk_mq_start_request(req
);
1852 reason
= scsi_dispatch_cmd(cmd
);
1854 scsi_set_blocked(cmd
, reason
);
1855 ret
= BLK_STS_RESOURCE
;
1856 goto out_dec_host_busy
;
1862 scsi_dec_host_busy(shost
, cmd
);
1863 out_dec_target_busy
:
1864 if (scsi_target(sdev
)->can_queue
> 0)
1865 atomic_dec(&scsi_target(sdev
)->target_busy
);
1867 scsi_mq_put_budget(q
, cmd
->budget_token
);
1868 cmd
->budget_token
= -1;
1872 case BLK_STS_RESOURCE
:
1873 case BLK_STS_ZONE_RESOURCE
:
1874 if (scsi_device_blocked(sdev
))
1875 ret
= BLK_STS_DEV_RESOURCE
;
1878 cmd
->result
= DID_BUS_BUSY
<< 16;
1879 if (req
->rq_flags
& RQF_DONTPREP
)
1880 scsi_mq_uninit_cmd(cmd
);
1883 if (unlikely(!scsi_device_online(sdev
)))
1884 cmd
->result
= DID_NO_CONNECT
<< 16;
1886 cmd
->result
= DID_ERROR
<< 16;
1888 * Make sure to release all allocated resources when
1889 * we hit an error, as we will never see this command
1892 if (req
->rq_flags
& RQF_DONTPREP
)
1893 scsi_mq_uninit_cmd(cmd
);
1894 scsi_run_queue_async(sdev
);
1900 static int scsi_mq_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1901 unsigned int hctx_idx
, unsigned int numa_node
)
1903 struct Scsi_Host
*shost
= set
->driver_data
;
1904 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1905 struct scatterlist
*sg
;
1909 kmem_cache_alloc_node(scsi_sense_cache
, GFP_KERNEL
, numa_node
);
1910 if (!cmd
->sense_buffer
)
1913 if (scsi_host_get_prot(shost
)) {
1914 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) +
1915 shost
->hostt
->cmd_size
;
1916 cmd
->prot_sdb
= (void *)sg
+ scsi_mq_inline_sgl_size(shost
);
1919 if (shost
->hostt
->init_cmd_priv
) {
1920 ret
= shost
->hostt
->init_cmd_priv(shost
, cmd
);
1922 kmem_cache_free(scsi_sense_cache
, cmd
->sense_buffer
);
1928 static void scsi_mq_exit_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1929 unsigned int hctx_idx
)
1931 struct Scsi_Host
*shost
= set
->driver_data
;
1932 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1934 if (shost
->hostt
->exit_cmd_priv
)
1935 shost
->hostt
->exit_cmd_priv(shost
, cmd
);
1936 kmem_cache_free(scsi_sense_cache
, cmd
->sense_buffer
);
1940 static int scsi_mq_poll(struct blk_mq_hw_ctx
*hctx
, struct io_comp_batch
*iob
)
1942 struct Scsi_Host
*shost
= hctx
->driver_data
;
1944 if (shost
->hostt
->mq_poll
)
1945 return shost
->hostt
->mq_poll(shost
, hctx
->queue_num
);
1950 static int scsi_init_hctx(struct blk_mq_hw_ctx
*hctx
, void *data
,
1951 unsigned int hctx_idx
)
1953 struct Scsi_Host
*shost
= data
;
1955 hctx
->driver_data
= shost
;
1959 static void scsi_map_queues(struct blk_mq_tag_set
*set
)
1961 struct Scsi_Host
*shost
= container_of(set
, struct Scsi_Host
, tag_set
);
1963 if (shost
->hostt
->map_queues
)
1964 return shost
->hostt
->map_queues(shost
);
1965 blk_mq_map_queues(&set
->map
[HCTX_TYPE_DEFAULT
]);
1968 void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
1970 struct device
*dev
= shost
->dma_dev
;
1973 * this limit is imposed by hardware restrictions
1975 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
1978 if (scsi_host_prot_dma(shost
)) {
1979 shost
->sg_prot_tablesize
=
1980 min_not_zero(shost
->sg_prot_tablesize
,
1981 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
1982 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
1983 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
1986 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
1987 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1988 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1990 blk_queue_max_segment_size(q
, shost
->max_segment_size
);
1991 blk_queue_virt_boundary(q
, shost
->virt_boundary_mask
);
1992 dma_set_max_seg_size(dev
, queue_max_segment_size(q
));
1995 * Set a reasonable default alignment: The larger of 32-byte (dword),
1996 * which is a common minimum for HBAs, and the minimum DMA alignment,
1997 * which is set by the platform.
1999 * Devices that require a bigger alignment can increase it later.
2001 blk_queue_dma_alignment(q
, max(4, dma_get_cache_alignment()) - 1);
2003 EXPORT_SYMBOL_GPL(__scsi_init_queue
);
2005 static const struct blk_mq_ops scsi_mq_ops_no_commit
= {
2006 .get_budget
= scsi_mq_get_budget
,
2007 .put_budget
= scsi_mq_put_budget
,
2008 .queue_rq
= scsi_queue_rq
,
2009 .complete
= scsi_complete
,
2010 .timeout
= scsi_timeout
,
2011 #ifdef CONFIG_BLK_DEBUG_FS
2012 .show_rq
= scsi_show_rq
,
2014 .init_request
= scsi_mq_init_request
,
2015 .exit_request
= scsi_mq_exit_request
,
2016 .cleanup_rq
= scsi_cleanup_rq
,
2017 .busy
= scsi_mq_lld_busy
,
2018 .map_queues
= scsi_map_queues
,
2019 .init_hctx
= scsi_init_hctx
,
2020 .poll
= scsi_mq_poll
,
2021 .set_rq_budget_token
= scsi_mq_set_rq_budget_token
,
2022 .get_rq_budget_token
= scsi_mq_get_rq_budget_token
,
2026 static void scsi_commit_rqs(struct blk_mq_hw_ctx
*hctx
)
2028 struct Scsi_Host
*shost
= hctx
->driver_data
;
2030 shost
->hostt
->commit_rqs(shost
, hctx
->queue_num
);
2033 static const struct blk_mq_ops scsi_mq_ops
= {
2034 .get_budget
= scsi_mq_get_budget
,
2035 .put_budget
= scsi_mq_put_budget
,
2036 .queue_rq
= scsi_queue_rq
,
2037 .commit_rqs
= scsi_commit_rqs
,
2038 .complete
= scsi_complete
,
2039 .timeout
= scsi_timeout
,
2040 #ifdef CONFIG_BLK_DEBUG_FS
2041 .show_rq
= scsi_show_rq
,
2043 .init_request
= scsi_mq_init_request
,
2044 .exit_request
= scsi_mq_exit_request
,
2045 .cleanup_rq
= scsi_cleanup_rq
,
2046 .busy
= scsi_mq_lld_busy
,
2047 .map_queues
= scsi_map_queues
,
2048 .init_hctx
= scsi_init_hctx
,
2049 .poll
= scsi_mq_poll
,
2050 .set_rq_budget_token
= scsi_mq_set_rq_budget_token
,
2051 .get_rq_budget_token
= scsi_mq_get_rq_budget_token
,
2054 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2056 unsigned int cmd_size
, sgl_size
;
2057 struct blk_mq_tag_set
*tag_set
= &shost
->tag_set
;
2059 sgl_size
= max_t(unsigned int, sizeof(struct scatterlist
),
2060 scsi_mq_inline_sgl_size(shost
));
2061 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2062 if (scsi_host_get_prot(shost
))
2063 cmd_size
+= sizeof(struct scsi_data_buffer
) +
2064 sizeof(struct scatterlist
) * SCSI_INLINE_PROT_SG_CNT
;
2066 memset(tag_set
, 0, sizeof(*tag_set
));
2067 if (shost
->hostt
->commit_rqs
)
2068 tag_set
->ops
= &scsi_mq_ops
;
2070 tag_set
->ops
= &scsi_mq_ops_no_commit
;
2071 tag_set
->nr_hw_queues
= shost
->nr_hw_queues
? : 1;
2072 tag_set
->nr_maps
= shost
->nr_maps
? : 1;
2073 tag_set
->queue_depth
= shost
->can_queue
;
2074 tag_set
->cmd_size
= cmd_size
;
2075 tag_set
->numa_node
= dev_to_node(shost
->dma_dev
);
2076 tag_set
->flags
= BLK_MQ_F_SHOULD_MERGE
;
2078 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost
->hostt
->tag_alloc_policy
);
2079 if (shost
->queuecommand_may_block
)
2080 tag_set
->flags
|= BLK_MQ_F_BLOCKING
;
2081 tag_set
->driver_data
= shost
;
2082 if (shost
->host_tagset
)
2083 tag_set
->flags
|= BLK_MQ_F_TAG_HCTX_SHARED
;
2085 return blk_mq_alloc_tag_set(tag_set
);
2088 void scsi_mq_free_tags(struct kref
*kref
)
2090 struct Scsi_Host
*shost
= container_of(kref
, typeof(*shost
),
2093 blk_mq_free_tag_set(&shost
->tag_set
);
2094 complete(&shost
->tagset_freed
);
2098 * scsi_device_from_queue - return sdev associated with a request_queue
2099 * @q: The request queue to return the sdev from
2101 * Return the sdev associated with a request queue or NULL if the
2102 * request_queue does not reference a SCSI device.
2104 struct scsi_device
*scsi_device_from_queue(struct request_queue
*q
)
2106 struct scsi_device
*sdev
= NULL
;
2108 if (q
->mq_ops
== &scsi_mq_ops_no_commit
||
2109 q
->mq_ops
== &scsi_mq_ops
)
2110 sdev
= q
->queuedata
;
2111 if (!sdev
|| !get_device(&sdev
->sdev_gendev
))
2117 * pktcdvd should have been integrated into the SCSI layers, but for historical
2118 * reasons like the old IDE driver it isn't. This export allows it to safely
2119 * probe if a given device is a SCSI one and only attach to that.
2121 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2122 EXPORT_SYMBOL_GPL(scsi_device_from_queue
);
2126 * scsi_block_requests - Utility function used by low-level drivers to prevent
2127 * further commands from being queued to the device.
2128 * @shost: host in question
2130 * There is no timer nor any other means by which the requests get unblocked
2131 * other than the low-level driver calling scsi_unblock_requests().
2133 void scsi_block_requests(struct Scsi_Host
*shost
)
2135 shost
->host_self_blocked
= 1;
2137 EXPORT_SYMBOL(scsi_block_requests
);
2140 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2141 * further commands to be queued to the device.
2142 * @shost: host in question
2144 * There is no timer nor any other means by which the requests get unblocked
2145 * other than the low-level driver calling scsi_unblock_requests(). This is done
2146 * as an API function so that changes to the internals of the scsi mid-layer
2147 * won't require wholesale changes to drivers that use this feature.
2149 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2151 shost
->host_self_blocked
= 0;
2152 scsi_run_host_queues(shost
);
2154 EXPORT_SYMBOL(scsi_unblock_requests
);
2156 void scsi_exit_queue(void)
2158 kmem_cache_destroy(scsi_sense_cache
);
2162 * scsi_mode_select - issue a mode select
2163 * @sdev: SCSI device to be queried
2164 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2165 * @sp: Save page bit (0 == don't save, 1 == save)
2166 * @buffer: request buffer (may not be smaller than eight bytes)
2167 * @len: length of request buffer.
2168 * @timeout: command timeout
2169 * @retries: number of retries before failing
2170 * @data: returns a structure abstracting the mode header data
2171 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2172 * must be SCSI_SENSE_BUFFERSIZE big.
2174 * Returns zero if successful; negative error number or scsi
2178 int scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
,
2179 unsigned char *buffer
, int len
, int timeout
, int retries
,
2180 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2182 unsigned char cmd
[10];
2183 unsigned char *real_buffer
;
2184 const struct scsi_exec_args exec_args
= {
2189 memset(cmd
, 0, sizeof(cmd
));
2190 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2193 * Use MODE SELECT(10) if the device asked for it or if the mode page
2194 * and the mode select header cannot fit within the maximumm 255 bytes
2195 * of the MODE SELECT(6) command.
2197 if (sdev
->use_10_for_ms
||
2199 data
->block_descriptor_length
> 255) {
2200 if (len
> 65535 - 8)
2202 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2205 memcpy(real_buffer
+ 8, buffer
, len
);
2209 real_buffer
[2] = data
->medium_type
;
2210 real_buffer
[3] = data
->device_specific
;
2211 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2213 put_unaligned_be16(data
->block_descriptor_length
,
2216 cmd
[0] = MODE_SELECT_10
;
2217 put_unaligned_be16(len
, &cmd
[7]);
2222 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2225 memcpy(real_buffer
+ 4, buffer
, len
);
2228 real_buffer
[1] = data
->medium_type
;
2229 real_buffer
[2] = data
->device_specific
;
2230 real_buffer
[3] = data
->block_descriptor_length
;
2232 cmd
[0] = MODE_SELECT
;
2236 ret
= scsi_execute_cmd(sdev
, cmd
, REQ_OP_DRV_OUT
, real_buffer
, len
,
2237 timeout
, retries
, &exec_args
);
2241 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2244 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2245 * @sdev: SCSI device to be queried
2246 * @dbd: set to prevent mode sense from returning block descriptors
2247 * @modepage: mode page being requested
2248 * @subpage: sub-page of the mode page being requested
2249 * @buffer: request buffer (may not be smaller than eight bytes)
2250 * @len: length of request buffer.
2251 * @timeout: command timeout
2252 * @retries: number of retries before failing
2253 * @data: returns a structure abstracting the mode header data
2254 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2255 * must be SCSI_SENSE_BUFFERSIZE big.
2257 * Returns zero if successful, or a negative error number on failure
2260 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
, int subpage
,
2261 unsigned char *buffer
, int len
, int timeout
, int retries
,
2262 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2264 unsigned char cmd
[12];
2268 struct scsi_sense_hdr my_sshdr
;
2269 struct scsi_failure failure_defs
[] = {
2271 .sense
= UNIT_ATTENTION
,
2272 .asc
= SCMD_FAILURE_ASC_ANY
,
2273 .ascq
= SCMD_FAILURE_ASCQ_ANY
,
2275 .result
= SAM_STAT_CHECK_CONDITION
,
2279 struct scsi_failures failures
= {
2280 .failure_definitions
= failure_defs
,
2282 const struct scsi_exec_args exec_args
= {
2283 /* caller might not be interested in sense, but we need it */
2284 .sshdr
= sshdr
? : &my_sshdr
,
2285 .failures
= &failures
,
2288 memset(data
, 0, sizeof(*data
));
2289 memset(&cmd
[0], 0, 12);
2291 dbd
= sdev
->set_dbd_for_ms
? 8 : dbd
;
2292 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2296 sshdr
= exec_args
.sshdr
;
2299 use_10_for_ms
= sdev
->use_10_for_ms
|| len
> 255;
2301 if (use_10_for_ms
) {
2302 if (len
< 8 || len
> 65535)
2305 cmd
[0] = MODE_SENSE_10
;
2306 put_unaligned_be16(len
, &cmd
[7]);
2312 cmd
[0] = MODE_SENSE
;
2317 memset(buffer
, 0, len
);
2319 result
= scsi_execute_cmd(sdev
, cmd
, REQ_OP_DRV_IN
, buffer
, len
,
2320 timeout
, retries
, &exec_args
);
2324 /* This code looks awful: what it's doing is making sure an
2325 * ILLEGAL REQUEST sense return identifies the actual command
2326 * byte as the problem. MODE_SENSE commands can return
2327 * ILLEGAL REQUEST if the code page isn't supported */
2329 if (!scsi_status_is_good(result
)) {
2330 if (scsi_sense_valid(sshdr
)) {
2331 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2332 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2334 * Invalid command operation code: retry using
2335 * MODE SENSE(6) if this was a MODE SENSE(10)
2336 * request, except if the request mode page is
2337 * too large for MODE SENSE single byte
2338 * allocation length field.
2340 if (use_10_for_ms
) {
2343 sdev
->use_10_for_ms
= 0;
2350 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2351 (modepage
== 6 || modepage
== 8))) {
2352 /* Initio breakage? */
2355 data
->medium_type
= 0;
2356 data
->device_specific
= 0;
2358 data
->block_descriptor_length
= 0;
2359 } else if (use_10_for_ms
) {
2360 data
->length
= get_unaligned_be16(&buffer
[0]) + 2;
2361 data
->medium_type
= buffer
[2];
2362 data
->device_specific
= buffer
[3];
2363 data
->longlba
= buffer
[4] & 0x01;
2364 data
->block_descriptor_length
= get_unaligned_be16(&buffer
[6]);
2366 data
->length
= buffer
[0] + 1;
2367 data
->medium_type
= buffer
[1];
2368 data
->device_specific
= buffer
[2];
2369 data
->block_descriptor_length
= buffer
[3];
2371 data
->header_length
= header_length
;
2375 EXPORT_SYMBOL(scsi_mode_sense
);
2378 * scsi_test_unit_ready - test if unit is ready
2379 * @sdev: scsi device to change the state of.
2380 * @timeout: command timeout
2381 * @retries: number of retries before failing
2382 * @sshdr: outpout pointer for decoded sense information.
2384 * Returns zero if unsuccessful or an error if TUR failed. For
2385 * removable media, UNIT_ATTENTION sets ->changed flag.
2388 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2389 struct scsi_sense_hdr
*sshdr
)
2392 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2394 const struct scsi_exec_args exec_args
= {
2399 /* try to eat the UNIT_ATTENTION if there are enough retries */
2401 result
= scsi_execute_cmd(sdev
, cmd
, REQ_OP_DRV_IN
, NULL
, 0,
2402 timeout
, 1, &exec_args
);
2403 if (sdev
->removable
&& result
> 0 && scsi_sense_valid(sshdr
) &&
2404 sshdr
->sense_key
== UNIT_ATTENTION
)
2406 } while (result
> 0 && scsi_sense_valid(sshdr
) &&
2407 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2411 EXPORT_SYMBOL(scsi_test_unit_ready
);
2414 * scsi_device_set_state - Take the given device through the device state model.
2415 * @sdev: scsi device to change the state of.
2416 * @state: state to change to.
2418 * Returns zero if successful or an error if the requested
2419 * transition is illegal.
2422 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2424 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2426 if (state
== oldstate
)
2432 case SDEV_CREATED_BLOCK
:
2443 case SDEV_TRANSPORT_OFFLINE
:
2456 case SDEV_TRANSPORT_OFFLINE
:
2464 case SDEV_TRANSPORT_OFFLINE
:
2479 case SDEV_CREATED_BLOCK
:
2488 case SDEV_CREATED_BLOCK
:
2503 case SDEV_TRANSPORT_OFFLINE
:
2515 case SDEV_TRANSPORT_OFFLINE
:
2518 case SDEV_CREATED_BLOCK
:
2526 sdev
->offline_already
= false;
2527 sdev
->sdev_state
= state
;
2531 SCSI_LOG_ERROR_RECOVERY(1,
2532 sdev_printk(KERN_ERR
, sdev
,
2533 "Illegal state transition %s->%s",
2534 scsi_device_state_name(oldstate
),
2535 scsi_device_state_name(state
))
2539 EXPORT_SYMBOL(scsi_device_set_state
);
2542 * scsi_evt_emit - emit a single SCSI device uevent
2543 * @sdev: associated SCSI device
2544 * @evt: event to emit
2546 * Send a single uevent (scsi_event) to the associated scsi_device.
2548 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2553 switch (evt
->evt_type
) {
2554 case SDEV_EVT_MEDIA_CHANGE
:
2555 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2557 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2558 scsi_rescan_device(sdev
);
2559 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2561 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2562 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2564 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2565 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2567 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2568 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2570 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2571 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2573 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2574 envp
[idx
++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2576 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2577 envp
[idx
++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2586 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2590 * scsi_evt_thread - send a uevent for each scsi event
2591 * @work: work struct for scsi_device
2593 * Dispatch queued events to their associated scsi_device kobjects
2596 void scsi_evt_thread(struct work_struct
*work
)
2598 struct scsi_device
*sdev
;
2599 enum scsi_device_event evt_type
;
2600 LIST_HEAD(event_list
);
2602 sdev
= container_of(work
, struct scsi_device
, event_work
);
2604 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2605 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2606 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2609 struct scsi_event
*evt
;
2610 struct list_head
*this, *tmp
;
2611 unsigned long flags
;
2613 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2614 list_splice_init(&sdev
->event_list
, &event_list
);
2615 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2617 if (list_empty(&event_list
))
2620 list_for_each_safe(this, tmp
, &event_list
) {
2621 evt
= list_entry(this, struct scsi_event
, node
);
2622 list_del(&evt
->node
);
2623 scsi_evt_emit(sdev
, evt
);
2630 * sdev_evt_send - send asserted event to uevent thread
2631 * @sdev: scsi_device event occurred on
2632 * @evt: event to send
2634 * Assert scsi device event asynchronously.
2636 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2638 unsigned long flags
;
2641 /* FIXME: currently this check eliminates all media change events
2642 * for polled devices. Need to update to discriminate between AN
2643 * and polled events */
2644 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2650 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2651 list_add_tail(&evt
->node
, &sdev
->event_list
);
2652 schedule_work(&sdev
->event_work
);
2653 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2655 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2658 * sdev_evt_alloc - allocate a new scsi event
2659 * @evt_type: type of event to allocate
2660 * @gfpflags: GFP flags for allocation
2662 * Allocates and returns a new scsi_event.
2664 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2667 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2671 evt
->evt_type
= evt_type
;
2672 INIT_LIST_HEAD(&evt
->node
);
2674 /* evt_type-specific initialization, if any */
2676 case SDEV_EVT_MEDIA_CHANGE
:
2677 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2678 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2679 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2680 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2681 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2682 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2683 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2691 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2694 * sdev_evt_send_simple - send asserted event to uevent thread
2695 * @sdev: scsi_device event occurred on
2696 * @evt_type: type of event to send
2697 * @gfpflags: GFP flags for allocation
2699 * Assert scsi device event asynchronously, given an event type.
2701 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2702 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2704 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2706 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2711 sdev_evt_send(sdev
, evt
);
2713 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2716 * scsi_device_quiesce - Block all commands except power management.
2717 * @sdev: scsi device to quiesce.
2719 * This works by trying to transition to the SDEV_QUIESCE state
2720 * (which must be a legal transition). When the device is in this
2721 * state, only power management requests will be accepted, all others will
2724 * Must be called with user context, may sleep.
2726 * Returns zero if unsuccessful or an error if not.
2729 scsi_device_quiesce(struct scsi_device
*sdev
)
2731 struct request_queue
*q
= sdev
->request_queue
;
2735 * It is allowed to call scsi_device_quiesce() multiple times from
2736 * the same context but concurrent scsi_device_quiesce() calls are
2739 WARN_ON_ONCE(sdev
->quiesced_by
&& sdev
->quiesced_by
!= current
);
2741 if (sdev
->quiesced_by
== current
)
2746 blk_mq_freeze_queue(q
);
2748 * Ensure that the effect of blk_set_pm_only() will be visible
2749 * for percpu_ref_tryget() callers that occur after the queue
2750 * unfreeze even if the queue was already frozen before this function
2751 * was called. See also https://lwn.net/Articles/573497/.
2754 blk_mq_unfreeze_queue(q
);
2756 mutex_lock(&sdev
->state_mutex
);
2757 err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2759 sdev
->quiesced_by
= current
;
2761 blk_clear_pm_only(q
);
2762 mutex_unlock(&sdev
->state_mutex
);
2766 EXPORT_SYMBOL(scsi_device_quiesce
);
2769 * scsi_device_resume - Restart user issued commands to a quiesced device.
2770 * @sdev: scsi device to resume.
2772 * Moves the device from quiesced back to running and restarts the
2775 * Must be called with user context, may sleep.
2777 void scsi_device_resume(struct scsi_device
*sdev
)
2779 /* check if the device state was mutated prior to resume, and if
2780 * so assume the state is being managed elsewhere (for example
2781 * device deleted during suspend)
2783 mutex_lock(&sdev
->state_mutex
);
2784 if (sdev
->sdev_state
== SDEV_QUIESCE
)
2785 scsi_device_set_state(sdev
, SDEV_RUNNING
);
2786 if (sdev
->quiesced_by
) {
2787 sdev
->quiesced_by
= NULL
;
2788 blk_clear_pm_only(sdev
->request_queue
);
2790 mutex_unlock(&sdev
->state_mutex
);
2792 EXPORT_SYMBOL(scsi_device_resume
);
2795 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2797 scsi_device_quiesce(sdev
);
2801 scsi_target_quiesce(struct scsi_target
*starget
)
2803 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2805 EXPORT_SYMBOL(scsi_target_quiesce
);
2808 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2810 scsi_device_resume(sdev
);
2814 scsi_target_resume(struct scsi_target
*starget
)
2816 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2818 EXPORT_SYMBOL(scsi_target_resume
);
2820 static int __scsi_internal_device_block_nowait(struct scsi_device
*sdev
)
2822 if (scsi_device_set_state(sdev
, SDEV_BLOCK
))
2823 return scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2828 void scsi_start_queue(struct scsi_device
*sdev
)
2830 if (cmpxchg(&sdev
->queue_stopped
, 1, 0))
2831 blk_mq_unquiesce_queue(sdev
->request_queue
);
2834 static void scsi_stop_queue(struct scsi_device
*sdev
)
2837 * The atomic variable of ->queue_stopped covers that
2838 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2840 * The caller needs to wait until quiesce is done.
2842 if (!cmpxchg(&sdev
->queue_stopped
, 0, 1))
2843 blk_mq_quiesce_queue_nowait(sdev
->request_queue
);
2847 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2848 * @sdev: device to block
2850 * Pause SCSI command processing on the specified device. Does not sleep.
2852 * Returns zero if successful or a negative error code upon failure.
2855 * This routine transitions the device to the SDEV_BLOCK state (which must be
2856 * a legal transition). When the device is in this state, command processing
2857 * is paused until the device leaves the SDEV_BLOCK state. See also
2858 * scsi_internal_device_unblock_nowait().
2860 int scsi_internal_device_block_nowait(struct scsi_device
*sdev
)
2862 int ret
= __scsi_internal_device_block_nowait(sdev
);
2865 * The device has transitioned to SDEV_BLOCK. Stop the
2866 * block layer from calling the midlayer with this device's
2870 scsi_stop_queue(sdev
);
2873 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait
);
2876 * scsi_device_block - try to transition to the SDEV_BLOCK state
2877 * @sdev: device to block
2878 * @data: dummy argument, ignored
2880 * Pause SCSI command processing on the specified device. Callers must wait
2881 * until all ongoing scsi_queue_rq() calls have finished after this function
2885 * This routine transitions the device to the SDEV_BLOCK state (which must be
2886 * a legal transition). When the device is in this state, command processing
2887 * is paused until the device leaves the SDEV_BLOCK state. See also
2888 * scsi_internal_device_unblock().
2890 static void scsi_device_block(struct scsi_device
*sdev
, void *data
)
2893 enum scsi_device_state state
;
2895 mutex_lock(&sdev
->state_mutex
);
2896 err
= __scsi_internal_device_block_nowait(sdev
);
2897 state
= sdev
->sdev_state
;
2900 * scsi_stop_queue() must be called with the state_mutex
2901 * held. Otherwise a simultaneous scsi_start_queue() call
2902 * might unquiesce the queue before we quiesce it.
2904 scsi_stop_queue(sdev
);
2906 mutex_unlock(&sdev
->state_mutex
);
2908 WARN_ONCE(err
, "%s: failed to block %s in state %d\n",
2909 __func__
, dev_name(&sdev
->sdev_gendev
), state
);
2913 * scsi_internal_device_unblock_nowait - resume a device after a block request
2914 * @sdev: device to resume
2915 * @new_state: state to set the device to after unblocking
2917 * Restart the device queue for a previously suspended SCSI device. Does not
2920 * Returns zero if successful or a negative error code upon failure.
2923 * This routine transitions the device to the SDEV_RUNNING state or to one of
2924 * the offline states (which must be a legal transition) allowing the midlayer
2925 * to goose the queue for this device.
2927 int scsi_internal_device_unblock_nowait(struct scsi_device
*sdev
,
2928 enum scsi_device_state new_state
)
2930 switch (new_state
) {
2932 case SDEV_TRANSPORT_OFFLINE
:
2939 * Try to transition the scsi device to SDEV_RUNNING or one of the
2940 * offlined states and goose the device queue if successful.
2942 switch (sdev
->sdev_state
) {
2944 case SDEV_TRANSPORT_OFFLINE
:
2945 sdev
->sdev_state
= new_state
;
2947 case SDEV_CREATED_BLOCK
:
2948 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2949 new_state
== SDEV_OFFLINE
)
2950 sdev
->sdev_state
= new_state
;
2952 sdev
->sdev_state
= SDEV_CREATED
;
2960 scsi_start_queue(sdev
);
2964 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait
);
2967 * scsi_internal_device_unblock - resume a device after a block request
2968 * @sdev: device to resume
2969 * @new_state: state to set the device to after unblocking
2971 * Restart the device queue for a previously suspended SCSI device. May sleep.
2973 * Returns zero if successful or a negative error code upon failure.
2976 * This routine transitions the device to the SDEV_RUNNING state or to one of
2977 * the offline states (which must be a legal transition) allowing the midlayer
2978 * to goose the queue for this device.
2980 static int scsi_internal_device_unblock(struct scsi_device
*sdev
,
2981 enum scsi_device_state new_state
)
2985 mutex_lock(&sdev
->state_mutex
);
2986 ret
= scsi_internal_device_unblock_nowait(sdev
, new_state
);
2987 mutex_unlock(&sdev
->state_mutex
);
2993 target_block(struct device
*dev
, void *data
)
2995 if (scsi_is_target_device(dev
))
2996 starget_for_each_device(to_scsi_target(dev
), NULL
,
3002 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3003 * @dev: a parent device of one or more scsi_target devices
3004 * @shost: the Scsi_Host to which this device belongs
3006 * Iterate over all children of @dev, which should be scsi_target devices,
3007 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3008 * ongoing scsi_queue_rq() calls to finish. May sleep.
3011 * @dev must not itself be a scsi_target device.
3014 scsi_block_targets(struct Scsi_Host
*shost
, struct device
*dev
)
3016 WARN_ON_ONCE(scsi_is_target_device(dev
));
3017 device_for_each_child(dev
, NULL
, target_block
);
3018 blk_mq_wait_quiesce_done(&shost
->tag_set
);
3020 EXPORT_SYMBOL_GPL(scsi_block_targets
);
3023 device_unblock(struct scsi_device
*sdev
, void *data
)
3025 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
3029 target_unblock(struct device
*dev
, void *data
)
3031 if (scsi_is_target_device(dev
))
3032 starget_for_each_device(to_scsi_target(dev
), data
,
3038 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
3040 if (scsi_is_target_device(dev
))
3041 starget_for_each_device(to_scsi_target(dev
), &new_state
,
3044 device_for_each_child(dev
, &new_state
, target_unblock
);
3046 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
3049 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3050 * @shost: device to block
3052 * Pause SCSI command processing for all logical units associated with the SCSI
3053 * host and wait until pending scsi_queue_rq() calls have finished.
3055 * Returns zero if successful or a negative error code upon failure.
3058 scsi_host_block(struct Scsi_Host
*shost
)
3060 struct scsi_device
*sdev
;
3064 * Call scsi_internal_device_block_nowait so we can avoid
3065 * calling synchronize_rcu() for each LUN.
3067 shost_for_each_device(sdev
, shost
) {
3068 mutex_lock(&sdev
->state_mutex
);
3069 ret
= scsi_internal_device_block_nowait(sdev
);
3070 mutex_unlock(&sdev
->state_mutex
);
3072 scsi_device_put(sdev
);
3077 /* Wait for ongoing scsi_queue_rq() calls to finish. */
3078 blk_mq_wait_quiesce_done(&shost
->tag_set
);
3082 EXPORT_SYMBOL_GPL(scsi_host_block
);
3085 scsi_host_unblock(struct Scsi_Host
*shost
, int new_state
)
3087 struct scsi_device
*sdev
;
3090 shost_for_each_device(sdev
, shost
) {
3091 ret
= scsi_internal_device_unblock(sdev
, new_state
);
3093 scsi_device_put(sdev
);
3099 EXPORT_SYMBOL_GPL(scsi_host_unblock
);
3102 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3103 * @sgl: scatter-gather list
3104 * @sg_count: number of segments in sg
3105 * @offset: offset in bytes into sg, on return offset into the mapped area
3106 * @len: bytes to map, on return number of bytes mapped
3108 * Returns virtual address of the start of the mapped page
3110 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
3111 size_t *offset
, size_t *len
)
3114 size_t sg_len
= 0, len_complete
= 0;
3115 struct scatterlist
*sg
;
3118 WARN_ON(!irqs_disabled());
3120 for_each_sg(sgl
, sg
, sg_count
, i
) {
3121 len_complete
= sg_len
; /* Complete sg-entries */
3122 sg_len
+= sg
->length
;
3123 if (sg_len
> *offset
)
3127 if (unlikely(i
== sg_count
)) {
3128 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
3130 __func__
, sg_len
, *offset
, sg_count
);
3135 /* Offset starting from the beginning of first page in this sg-entry */
3136 *offset
= *offset
- len_complete
+ sg
->offset
;
3138 /* Assumption: contiguous pages can be accessed as "page + i" */
3139 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
3140 *offset
&= ~PAGE_MASK
;
3142 /* Bytes in this sg-entry from *offset to the end of the page */
3143 sg_len
= PAGE_SIZE
- *offset
;
3147 return kmap_atomic(page
);
3149 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3152 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3153 * @virt: virtual address to be unmapped
3155 void scsi_kunmap_atomic_sg(void *virt
)
3157 kunmap_atomic(virt
);
3159 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3161 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3163 atomic_inc(&sdev
->disk_events_disable_depth
);
3165 EXPORT_SYMBOL(sdev_disable_disk_events
);
3167 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3169 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3171 atomic_dec(&sdev
->disk_events_disable_depth
);
3173 EXPORT_SYMBOL(sdev_enable_disk_events
);
3175 static unsigned char designator_prio(const unsigned char *d
)
3178 /* not associated with LUN */
3182 /* invalid length */
3186 * Order of preference for lun descriptor:
3187 * - SCSI name string
3188 * - NAA IEEE Registered Extended
3189 * - EUI-64 based 16-byte
3190 * - EUI-64 based 12-byte
3191 * - NAA IEEE Registered
3192 * - NAA IEEE Extended
3193 * - EUI-64 based 8-byte
3194 * - SCSI name string (truncated)
3196 * as longer descriptors reduce the likelyhood
3197 * of identification clashes.
3200 switch (d
[1] & 0xf) {
3202 /* SCSI name string, variable-length UTF-8 */
3205 switch (d
[4] >> 4) {
3207 /* NAA registered extended */
3210 /* NAA registered */
3216 /* NAA locally assigned */
3225 /* EUI64-based, 16 byte */
3228 /* EUI64-based, 12 byte */
3231 /* EUI64-based, 8 byte */
3248 * scsi_vpd_lun_id - return a unique device identification
3249 * @sdev: SCSI device
3250 * @id: buffer for the identification
3251 * @id_len: length of the buffer
3253 * Copies a unique device identification into @id based
3254 * on the information in the VPD page 0x83 of the device.
3255 * The string will be formatted as a SCSI name string.
3257 * Returns the length of the identification or error on failure.
3258 * If the identifier is longer than the supplied buffer the actual
3259 * identifier length is returned and the buffer is not zero-padded.
3261 int scsi_vpd_lun_id(struct scsi_device
*sdev
, char *id
, size_t id_len
)
3265 const unsigned char *d
, *cur_id_str
;
3266 const struct scsi_vpd
*vpd_pg83
;
3267 int id_size
= -EINVAL
;
3270 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3276 /* The id string must be at least 20 bytes + terminating NULL byte */
3282 memset(id
, 0, id_len
);
3283 for (d
= vpd_pg83
->data
+ 4;
3284 d
< vpd_pg83
->data
+ vpd_pg83
->len
;
3286 u8 prio
= designator_prio(d
);
3288 if (prio
== 0 || cur_id_prio
> prio
)
3291 switch (d
[1] & 0xf) {
3294 if (cur_id_size
> d
[3])
3298 if (cur_id_size
+ 4 > id_len
)
3299 cur_id_size
= id_len
- 4;
3301 id_size
= snprintf(id
, id_len
, "t10.%*pE",
3302 cur_id_size
, cur_id_str
);
3309 switch (cur_id_size
) {
3311 id_size
= snprintf(id
, id_len
,
3316 id_size
= snprintf(id
, id_len
,
3321 id_size
= snprintf(id
, id_len
,
3334 switch (cur_id_size
) {
3336 id_size
= snprintf(id
, id_len
,
3341 id_size
= snprintf(id
, id_len
,
3350 /* SCSI name string */
3351 if (cur_id_size
> d
[3])
3353 /* Prefer others for truncated descriptor */
3354 if (d
[3] > id_len
) {
3356 if (cur_id_prio
> prio
)
3360 cur_id_size
= id_size
= d
[3];
3362 if (cur_id_size
>= id_len
)
3363 cur_id_size
= id_len
- 1;
3364 memcpy(id
, cur_id_str
, cur_id_size
);
3374 EXPORT_SYMBOL(scsi_vpd_lun_id
);
3377 * scsi_vpd_tpg_id - return a target port group identifier
3378 * @sdev: SCSI device
3380 * Returns the Target Port Group identifier from the information
3381 * froom VPD page 0x83 of the device.
3383 * Returns the identifier or error on failure.
3385 int scsi_vpd_tpg_id(struct scsi_device
*sdev
, int *rel_id
)
3387 const unsigned char *d
;
3388 const struct scsi_vpd
*vpd_pg83
;
3389 int group_id
= -EAGAIN
, rel_port
= -1;
3392 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3398 d
= vpd_pg83
->data
+ 4;
3399 while (d
< vpd_pg83
->data
+ vpd_pg83
->len
) {
3400 switch (d
[1] & 0xf) {
3402 /* Relative target port */
3403 rel_port
= get_unaligned_be16(&d
[6]);
3406 /* Target port group */
3407 group_id
= get_unaligned_be16(&d
[6]);
3416 if (group_id
>= 0 && rel_id
&& rel_port
!= -1)
3421 EXPORT_SYMBOL(scsi_vpd_tpg_id
);
3424 * scsi_build_sense - build sense data for a command
3425 * @scmd: scsi command for which the sense should be formatted
3426 * @desc: Sense format (non-zero == descriptor format,
3427 * 0 == fixed format)
3429 * @asc: Additional sense code
3430 * @ascq: Additional sense code qualifier
3433 void scsi_build_sense(struct scsi_cmnd
*scmd
, int desc
, u8 key
, u8 asc
, u8 ascq
)
3435 scsi_build_sense_buffer(desc
, scmd
->sense_buffer
, key
, asc
, ascq
);
3436 scmd
->result
= SAM_STAT_CHECK_CONDITION
;
3438 EXPORT_SYMBOL_GPL(scsi_build_sense
);
3440 #ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3441 #include "scsi_lib_test.c"