2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
42 static struct kmem_cache
*scsi_sdb_cache
;
43 static struct kmem_cache
*scsi_sense_cache
;
44 static struct kmem_cache
*scsi_sense_isadma_cache
;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex
);
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
);
49 static inline struct kmem_cache
*
50 scsi_select_sense_cache(bool unchecked_isa_dma
)
52 return unchecked_isa_dma
? scsi_sense_isadma_cache
: scsi_sense_cache
;
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma
,
56 unsigned char *sense_buffer
)
58 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma
),
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma
,
63 gfp_t gfp_mask
, int numa_node
)
65 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma
),
69 int scsi_init_sense_cache(struct Scsi_Host
*shost
)
71 struct kmem_cache
*cache
;
74 mutex_lock(&scsi_sense_cache_mutex
);
75 cache
= scsi_select_sense_cache(shost
->unchecked_isa_dma
);
79 if (shost
->unchecked_isa_dma
) {
80 scsi_sense_isadma_cache
=
81 kmem_cache_create("scsi_sense_cache(DMA)",
82 SCSI_SENSE_BUFFERSIZE
, 0,
83 SLAB_HWCACHE_ALIGN
| SLAB_CACHE_DMA
, NULL
);
84 if (!scsi_sense_isadma_cache
)
88 kmem_cache_create("scsi_sense_cache",
89 SCSI_SENSE_BUFFERSIZE
, 0, SLAB_HWCACHE_ALIGN
, NULL
);
90 if (!scsi_sense_cache
)
94 mutex_unlock(&scsi_sense_cache_mutex
);
99 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
100 * not change behaviour from the previous unplug mechanism, experimentation
101 * may prove this needs changing.
103 #define SCSI_QUEUE_DELAY 3
106 scsi_set_blocked(struct scsi_cmnd
*cmd
, int reason
)
108 struct Scsi_Host
*host
= cmd
->device
->host
;
109 struct scsi_device
*device
= cmd
->device
;
110 struct scsi_target
*starget
= scsi_target(device
);
113 * Set the appropriate busy bit for the device/host.
115 * If the host/device isn't busy, assume that something actually
116 * completed, and that we should be able to queue a command now.
118 * Note that the prior mid-layer assumption that any host could
119 * always queue at least one command is now broken. The mid-layer
120 * will implement a user specifiable stall (see
121 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
122 * if a command is requeued with no other commands outstanding
123 * either for the device or for the host.
126 case SCSI_MLQUEUE_HOST_BUSY
:
127 atomic_set(&host
->host_blocked
, host
->max_host_blocked
);
129 case SCSI_MLQUEUE_DEVICE_BUSY
:
130 case SCSI_MLQUEUE_EH_RETRY
:
131 atomic_set(&device
->device_blocked
,
132 device
->max_device_blocked
);
134 case SCSI_MLQUEUE_TARGET_BUSY
:
135 atomic_set(&starget
->target_blocked
,
136 starget
->max_target_blocked
);
141 static void scsi_mq_requeue_cmd(struct scsi_cmnd
*cmd
)
143 struct scsi_device
*sdev
= cmd
->device
;
145 if (cmd
->request
->rq_flags
& RQF_DONTPREP
) {
146 cmd
->request
->rq_flags
&= ~RQF_DONTPREP
;
147 scsi_mq_uninit_cmd(cmd
);
151 blk_mq_requeue_request(cmd
->request
, true);
152 put_device(&sdev
->sdev_gendev
);
156 * __scsi_queue_insert - private queue insertion
157 * @cmd: The SCSI command being requeued
158 * @reason: The reason for the requeue
159 * @unbusy: Whether the queue should be unbusied
161 * This is a private queue insertion. The public interface
162 * scsi_queue_insert() always assumes the queue should be unbusied
163 * because it's always called before the completion. This function is
164 * for a requeue after completion, which should only occur in this
167 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, int unbusy
)
169 struct scsi_device
*device
= cmd
->device
;
170 struct request_queue
*q
= device
->request_queue
;
173 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
174 "Inserting command %p into mlqueue\n", cmd
));
176 scsi_set_blocked(cmd
, reason
);
179 * Decrement the counters, since these commands are no longer
180 * active on the host/device.
183 scsi_device_unbusy(device
);
186 * Requeue this command. It will go before all other commands
187 * that are already in the queue. Schedule requeue work under
188 * lock such that the kblockd_schedule_work() call happens
189 * before blk_cleanup_queue() finishes.
193 scsi_mq_requeue_cmd(cmd
);
196 spin_lock_irqsave(q
->queue_lock
, flags
);
197 blk_requeue_request(q
, cmd
->request
);
198 kblockd_schedule_work(&device
->requeue_work
);
199 spin_unlock_irqrestore(q
->queue_lock
, flags
);
203 * Function: scsi_queue_insert()
205 * Purpose: Insert a command in the midlevel queue.
207 * Arguments: cmd - command that we are adding to queue.
208 * reason - why we are inserting command to queue.
210 * Lock status: Assumed that lock is not held upon entry.
214 * Notes: We do this for one of two cases. Either the host is busy
215 * and it cannot accept any more commands for the time being,
216 * or the device returned QUEUE_FULL and can accept no more
218 * Notes: This could be called either from an interrupt context or a
219 * normal process context.
221 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
223 __scsi_queue_insert(cmd
, reason
, 1);
228 * scsi_execute - insert request and wait for the result
231 * @data_direction: data direction
232 * @buffer: data buffer
233 * @bufflen: len of buffer
234 * @sense: optional sense buffer
235 * @sshdr: optional decoded sense header
236 * @timeout: request timeout in seconds
237 * @retries: number of times to retry request
238 * @flags: flags for ->cmd_flags
239 * @rq_flags: flags for ->rq_flags
240 * @resid: optional residual length
242 * Returns the scsi_cmnd result field if a command was executed, or a negative
243 * Linux error code if we didn't get that far.
245 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
246 int data_direction
, void *buffer
, unsigned bufflen
,
247 unsigned char *sense
, struct scsi_sense_hdr
*sshdr
,
248 int timeout
, int retries
, u64 flags
, req_flags_t rq_flags
,
252 struct scsi_request
*rq
;
253 int ret
= DRIVER_ERROR
<< 24;
255 req
= blk_get_request_flags(sdev
->request_queue
,
256 data_direction
== DMA_TO_DEVICE
?
257 REQ_OP_SCSI_OUT
: REQ_OP_SCSI_IN
, BLK_MQ_REQ_PREEMPT
);
262 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
263 buffer
, bufflen
, __GFP_RECLAIM
))
266 rq
->cmd_len
= COMMAND_SIZE(cmd
[0]);
267 memcpy(rq
->cmd
, cmd
, rq
->cmd_len
);
268 rq
->retries
= retries
;
269 req
->timeout
= timeout
;
270 req
->cmd_flags
|= flags
;
271 req
->rq_flags
|= rq_flags
| RQF_QUIET
;
274 * head injection *required* here otherwise quiesce won't work
276 blk_execute_rq(req
->q
, NULL
, req
, 1);
279 * Some devices (USB mass-storage in particular) may transfer
280 * garbage data together with a residue indicating that the data
281 * is invalid. Prevent the garbage from being misinterpreted
282 * and prevent security leaks by zeroing out the excess data.
284 if (unlikely(rq
->resid_len
> 0 && rq
->resid_len
<= bufflen
))
285 memset(buffer
+ (bufflen
- rq
->resid_len
), 0, rq
->resid_len
);
288 *resid
= rq
->resid_len
;
289 if (sense
&& rq
->sense_len
)
290 memcpy(sense
, rq
->sense
, SCSI_SENSE_BUFFERSIZE
);
292 scsi_normalize_sense(rq
->sense
, rq
->sense_len
, sshdr
);
295 blk_put_request(req
);
299 EXPORT_SYMBOL(scsi_execute
);
302 * Function: scsi_init_cmd_errh()
304 * Purpose: Initialize cmd fields related to error handling.
306 * Arguments: cmd - command that is ready to be queued.
308 * Notes: This function has the job of initializing a number of
309 * fields related to error handling. Typically this will
310 * be called once for each command, as required.
312 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
314 cmd
->serial_number
= 0;
315 scsi_set_resid(cmd
, 0);
316 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
317 if (cmd
->cmd_len
== 0)
318 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
322 * Decrement the host_busy counter and wake up the error handler if necessary.
323 * Avoid as follows that the error handler is not woken up if shost->host_busy
324 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
325 * with an RCU read lock in this function to ensure that this function in its
326 * entirety either finishes before scsi_eh_scmd_add() increases the
327 * host_failed counter or that it notices the shost state change made by
328 * scsi_eh_scmd_add().
330 static void scsi_dec_host_busy(struct Scsi_Host
*shost
)
335 atomic_dec(&shost
->host_busy
);
336 if (unlikely(scsi_host_in_recovery(shost
))) {
337 spin_lock_irqsave(shost
->host_lock
, flags
);
338 if (shost
->host_failed
|| shost
->host_eh_scheduled
)
339 scsi_eh_wakeup(shost
);
340 spin_unlock_irqrestore(shost
->host_lock
, flags
);
345 void scsi_device_unbusy(struct scsi_device
*sdev
)
347 struct Scsi_Host
*shost
= sdev
->host
;
348 struct scsi_target
*starget
= scsi_target(sdev
);
350 scsi_dec_host_busy(shost
);
352 if (starget
->can_queue
> 0)
353 atomic_dec(&starget
->target_busy
);
355 atomic_dec(&sdev
->device_busy
);
358 static void scsi_kick_queue(struct request_queue
*q
)
361 blk_mq_start_hw_queues(q
);
367 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
368 * and call blk_run_queue for all the scsi_devices on the target -
369 * including current_sdev first.
371 * Called with *no* scsi locks held.
373 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
375 struct Scsi_Host
*shost
= current_sdev
->host
;
376 struct scsi_device
*sdev
, *tmp
;
377 struct scsi_target
*starget
= scsi_target(current_sdev
);
380 spin_lock_irqsave(shost
->host_lock
, flags
);
381 starget
->starget_sdev_user
= NULL
;
382 spin_unlock_irqrestore(shost
->host_lock
, flags
);
385 * Call blk_run_queue for all LUNs on the target, starting with
386 * current_sdev. We race with others (to set starget_sdev_user),
387 * but in most cases, we will be first. Ideally, each LU on the
388 * target would get some limited time or requests on the target.
390 scsi_kick_queue(current_sdev
->request_queue
);
392 spin_lock_irqsave(shost
->host_lock
, flags
);
393 if (starget
->starget_sdev_user
)
395 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
396 same_target_siblings
) {
397 if (sdev
== current_sdev
)
399 if (scsi_device_get(sdev
))
402 spin_unlock_irqrestore(shost
->host_lock
, flags
);
403 scsi_kick_queue(sdev
->request_queue
);
404 spin_lock_irqsave(shost
->host_lock
, flags
);
406 scsi_device_put(sdev
);
409 spin_unlock_irqrestore(shost
->host_lock
, flags
);
412 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
414 if (atomic_read(&sdev
->device_busy
) >= sdev
->queue_depth
)
416 if (atomic_read(&sdev
->device_blocked
) > 0)
421 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
423 if (starget
->can_queue
> 0) {
424 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
426 if (atomic_read(&starget
->target_blocked
) > 0)
432 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
434 if (shost
->can_queue
> 0 &&
435 atomic_read(&shost
->host_busy
) >= shost
->can_queue
)
437 if (atomic_read(&shost
->host_blocked
) > 0)
439 if (shost
->host_self_blocked
)
444 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
446 LIST_HEAD(starved_list
);
447 struct scsi_device
*sdev
;
450 spin_lock_irqsave(shost
->host_lock
, flags
);
451 list_splice_init(&shost
->starved_list
, &starved_list
);
453 while (!list_empty(&starved_list
)) {
454 struct request_queue
*slq
;
457 * As long as shost is accepting commands and we have
458 * starved queues, call blk_run_queue. scsi_request_fn
459 * drops the queue_lock and can add us back to the
462 * host_lock protects the starved_list and starved_entry.
463 * scsi_request_fn must get the host_lock before checking
464 * or modifying starved_list or starved_entry.
466 if (scsi_host_is_busy(shost
))
469 sdev
= list_entry(starved_list
.next
,
470 struct scsi_device
, starved_entry
);
471 list_del_init(&sdev
->starved_entry
);
472 if (scsi_target_is_busy(scsi_target(sdev
))) {
473 list_move_tail(&sdev
->starved_entry
,
474 &shost
->starved_list
);
479 * Once we drop the host lock, a racing scsi_remove_device()
480 * call may remove the sdev from the starved list and destroy
481 * it and the queue. Mitigate by taking a reference to the
482 * queue and never touching the sdev again after we drop the
483 * host lock. Note: if __scsi_remove_device() invokes
484 * blk_cleanup_queue() before the queue is run from this
485 * function then blk_run_queue() will return immediately since
486 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
488 slq
= sdev
->request_queue
;
489 if (!blk_get_queue(slq
))
491 spin_unlock_irqrestore(shost
->host_lock
, flags
);
493 scsi_kick_queue(slq
);
496 spin_lock_irqsave(shost
->host_lock
, flags
);
498 /* put any unprocessed entries back */
499 list_splice(&starved_list
, &shost
->starved_list
);
500 spin_unlock_irqrestore(shost
->host_lock
, flags
);
504 * Function: scsi_run_queue()
506 * Purpose: Select a proper request queue to serve next
508 * Arguments: q - last request's queue
512 * Notes: The previous command was completely finished, start
513 * a new one if possible.
515 static void scsi_run_queue(struct request_queue
*q
)
517 struct scsi_device
*sdev
= q
->queuedata
;
519 if (scsi_target(sdev
)->single_lun
)
520 scsi_single_lun_run(sdev
);
521 if (!list_empty(&sdev
->host
->starved_list
))
522 scsi_starved_list_run(sdev
->host
);
525 blk_mq_run_hw_queues(q
, false);
530 void scsi_requeue_run_queue(struct work_struct
*work
)
532 struct scsi_device
*sdev
;
533 struct request_queue
*q
;
535 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
536 q
= sdev
->request_queue
;
541 * Function: scsi_requeue_command()
543 * Purpose: Handle post-processing of completed commands.
545 * Arguments: q - queue to operate on
546 * cmd - command that may need to be requeued.
550 * Notes: After command completion, there may be blocks left
551 * over which weren't finished by the previous command
552 * this can be for a number of reasons - the main one is
553 * I/O errors in the middle of the request, in which case
554 * we need to request the blocks that come after the bad
556 * Notes: Upon return, cmd is a stale pointer.
558 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
560 struct scsi_device
*sdev
= cmd
->device
;
561 struct request
*req
= cmd
->request
;
564 spin_lock_irqsave(q
->queue_lock
, flags
);
565 blk_unprep_request(req
);
567 scsi_put_command(cmd
);
568 blk_requeue_request(q
, req
);
569 spin_unlock_irqrestore(q
->queue_lock
, flags
);
573 put_device(&sdev
->sdev_gendev
);
576 void scsi_run_host_queues(struct Scsi_Host
*shost
)
578 struct scsi_device
*sdev
;
580 shost_for_each_device(sdev
, shost
)
581 scsi_run_queue(sdev
->request_queue
);
584 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
586 if (!blk_rq_is_passthrough(cmd
->request
)) {
587 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
589 if (drv
->uninit_command
)
590 drv
->uninit_command(cmd
);
594 static void scsi_mq_free_sgtables(struct scsi_cmnd
*cmd
)
596 struct scsi_data_buffer
*sdb
;
598 if (cmd
->sdb
.table
.nents
)
599 sg_free_table_chained(&cmd
->sdb
.table
, true);
600 if (cmd
->request
->next_rq
) {
601 sdb
= cmd
->request
->next_rq
->special
;
603 sg_free_table_chained(&sdb
->table
, true);
605 if (scsi_prot_sg_count(cmd
))
606 sg_free_table_chained(&cmd
->prot_sdb
->table
, true);
609 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
611 scsi_mq_free_sgtables(cmd
);
612 scsi_uninit_cmd(cmd
);
613 scsi_del_cmd_from_list(cmd
);
617 * Function: scsi_release_buffers()
619 * Purpose: Free resources allocate for a scsi_command.
621 * Arguments: cmd - command that we are bailing.
623 * Lock status: Assumed that no lock is held upon entry.
627 * Notes: In the event that an upper level driver rejects a
628 * command, we must release resources allocated during
629 * the __init_io() function. Primarily this would involve
630 * the scatter-gather table.
632 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
634 if (cmd
->sdb
.table
.nents
)
635 sg_free_table_chained(&cmd
->sdb
.table
, false);
637 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
639 if (scsi_prot_sg_count(cmd
))
640 sg_free_table_chained(&cmd
->prot_sdb
->table
, false);
643 static void scsi_release_bidi_buffers(struct scsi_cmnd
*cmd
)
645 struct scsi_data_buffer
*bidi_sdb
= cmd
->request
->next_rq
->special
;
647 sg_free_table_chained(&bidi_sdb
->table
, false);
648 kmem_cache_free(scsi_sdb_cache
, bidi_sdb
);
649 cmd
->request
->next_rq
->special
= NULL
;
652 static bool scsi_end_request(struct request
*req
, blk_status_t error
,
653 unsigned int bytes
, unsigned int bidi_bytes
)
655 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
656 struct scsi_device
*sdev
= cmd
->device
;
657 struct request_queue
*q
= sdev
->request_queue
;
659 if (blk_update_request(req
, error
, bytes
))
662 /* Bidi request must be completed as a whole */
663 if (unlikely(bidi_bytes
) &&
664 blk_update_request(req
->next_rq
, error
, bidi_bytes
))
667 if (blk_queue_add_random(q
))
668 add_disk_randomness(req
->rq_disk
);
670 if (!blk_rq_is_scsi(req
)) {
671 WARN_ON_ONCE(!(cmd
->flags
& SCMD_INITIALIZED
));
672 cmd
->flags
&= ~SCMD_INITIALIZED
;
673 destroy_rcu_head(&cmd
->rcu
);
678 * In the MQ case the command gets freed by __blk_mq_end_request,
679 * so we have to do all cleanup that depends on it earlier.
681 * We also can't kick the queues from irq context, so we
682 * will have to defer it to a workqueue.
684 scsi_mq_uninit_cmd(cmd
);
687 * queue is still alive, so grab the ref for preventing it
688 * from being cleaned up during running queue.
690 percpu_ref_get(&q
->q_usage_counter
);
692 __blk_mq_end_request(req
, error
);
694 if (scsi_target(sdev
)->single_lun
||
695 !list_empty(&sdev
->host
->starved_list
))
696 kblockd_schedule_work(&sdev
->requeue_work
);
698 blk_mq_run_hw_queues(q
, true);
700 percpu_ref_put(&q
->q_usage_counter
);
705 scsi_release_bidi_buffers(cmd
);
706 scsi_release_buffers(cmd
);
707 scsi_put_command(cmd
);
709 spin_lock_irqsave(q
->queue_lock
, flags
);
710 blk_finish_request(req
, error
);
711 spin_unlock_irqrestore(q
->queue_lock
, flags
);
716 put_device(&sdev
->sdev_gendev
);
721 * __scsi_error_from_host_byte - translate SCSI error code into errno
722 * @cmd: SCSI command (unused)
723 * @result: scsi error code
725 * Translate SCSI error code into block errors.
727 static blk_status_t
__scsi_error_from_host_byte(struct scsi_cmnd
*cmd
,
730 switch (host_byte(result
)) {
731 case DID_TRANSPORT_FAILFAST
:
732 return BLK_STS_TRANSPORT
;
733 case DID_TARGET_FAILURE
:
734 set_host_byte(cmd
, DID_OK
);
735 return BLK_STS_TARGET
;
736 case DID_NEXUS_FAILURE
:
737 set_host_byte(cmd
, DID_OK
);
738 return BLK_STS_NEXUS
;
739 case DID_ALLOC_FAILURE
:
740 set_host_byte(cmd
, DID_OK
);
741 return BLK_STS_NOSPC
;
742 case DID_MEDIUM_ERROR
:
743 set_host_byte(cmd
, DID_OK
);
744 return BLK_STS_MEDIUM
;
746 return BLK_STS_IOERR
;
751 * Function: scsi_io_completion()
753 * Purpose: Completion processing for block device I/O requests.
755 * Arguments: cmd - command that is finished.
757 * Lock status: Assumed that no lock is held upon entry.
761 * Notes: We will finish off the specified number of sectors. If we
762 * are done, the command block will be released and the queue
763 * function will be goosed. If we are not done then we have to
764 * figure out what to do next:
766 * a) We can call scsi_requeue_command(). The request
767 * will be unprepared and put back on the queue. Then
768 * a new command will be created for it. This should
769 * be used if we made forward progress, or if we want
770 * to switch from READ(10) to READ(6) for example.
772 * b) We can call __scsi_queue_insert(). The request will
773 * be put back on the queue and retried using the same
774 * command as before, possibly after a delay.
776 * c) We can call scsi_end_request() with -EIO to fail
777 * the remainder of the request.
779 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
781 int result
= cmd
->result
;
782 struct request_queue
*q
= cmd
->device
->request_queue
;
783 struct request
*req
= cmd
->request
;
784 blk_status_t error
= BLK_STS_OK
;
785 struct scsi_sense_hdr sshdr
;
786 bool sense_valid
= false;
787 int sense_deferred
= 0, level
= 0;
788 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
789 ACTION_DELAYED_RETRY
} action
;
790 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
793 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
795 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
798 if (blk_rq_is_passthrough(req
)) {
802 * SG_IO wants current and deferred errors
804 scsi_req(req
)->sense_len
=
805 min(8 + cmd
->sense_buffer
[7],
806 SCSI_SENSE_BUFFERSIZE
);
809 error
= __scsi_error_from_host_byte(cmd
, result
);
812 * __scsi_error_from_host_byte may have reset the host_byte
814 scsi_req(req
)->result
= cmd
->result
;
815 scsi_req(req
)->resid_len
= scsi_get_resid(cmd
);
817 if (scsi_bidi_cmnd(cmd
)) {
819 * Bidi commands Must be complete as a whole,
820 * both sides at once.
822 scsi_req(req
->next_rq
)->resid_len
= scsi_in(cmd
)->resid
;
823 if (scsi_end_request(req
, BLK_STS_OK
, blk_rq_bytes(req
),
824 blk_rq_bytes(req
->next_rq
)))
828 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
830 * Flush commands do not transfers any data, and thus cannot use
831 * good_bytes != blk_rq_bytes(req) as the signal for an error.
832 * This sets the error explicitly for the problem case.
834 error
= __scsi_error_from_host_byte(cmd
, result
);
837 /* no bidi support for !blk_rq_is_passthrough yet */
838 BUG_ON(blk_bidi_rq(req
));
841 * Next deal with any sectors which we were able to correctly
844 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
845 "%u sectors total, %d bytes done.\n",
846 blk_rq_sectors(req
), good_bytes
));
849 * Recovered errors need reporting, but they're always treated as
850 * success, so fiddle the result code here. For passthrough requests
851 * we already took a copy of the original into sreq->result which
852 * is what gets returned to the user
854 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
855 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
856 * print since caller wants ATA registers. Only occurs on
857 * SCSI ATA PASS_THROUGH commands when CK_COND=1
859 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
861 else if (!(req
->rq_flags
& RQF_QUIET
))
862 scsi_print_sense(cmd
);
864 /* for passthrough error may be set */
868 * Another corner case: the SCSI status byte is non-zero but 'good'.
869 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
870 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
871 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
872 * intermediate statuses (both obsolete in SAM-4) as good.
874 if (status_byte(result
) && scsi_status_is_good(result
)) {
880 * special case: failed zero length commands always need to
881 * drop down into the retry code. Otherwise, if we finished
882 * all bytes in the request we are done now.
884 if (!(blk_rq_bytes(req
) == 0 && error
) &&
885 !scsi_end_request(req
, error
, good_bytes
, 0))
889 * Kill remainder if no retrys.
891 if (error
&& scsi_noretry_cmd(cmd
)) {
892 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
898 * If there had been no error, but we have leftover bytes in the
899 * requeues just queue the command up again.
904 error
= __scsi_error_from_host_byte(cmd
, result
);
906 if (host_byte(result
) == DID_RESET
) {
907 /* Third party bus reset or reset for error recovery
908 * reasons. Just retry the command and see what
911 action
= ACTION_RETRY
;
912 } else if (sense_valid
&& !sense_deferred
) {
913 switch (sshdr
.sense_key
) {
915 if (cmd
->device
->removable
) {
916 /* Detected disc change. Set a bit
917 * and quietly refuse further access.
919 cmd
->device
->changed
= 1;
920 action
= ACTION_FAIL
;
922 /* Must have been a power glitch, or a
923 * bus reset. Could not have been a
924 * media change, so we just retry the
925 * command and see what happens.
927 action
= ACTION_RETRY
;
930 case ILLEGAL_REQUEST
:
931 /* If we had an ILLEGAL REQUEST returned, then
932 * we may have performed an unsupported
933 * command. The only thing this should be
934 * would be a ten byte read where only a six
935 * byte read was supported. Also, on a system
936 * where READ CAPACITY failed, we may have
937 * read past the end of the disk.
939 if ((cmd
->device
->use_10_for_rw
&&
940 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
941 (cmd
->cmnd
[0] == READ_10
||
942 cmd
->cmnd
[0] == WRITE_10
)) {
943 /* This will issue a new 6-byte command. */
944 cmd
->device
->use_10_for_rw
= 0;
945 action
= ACTION_REPREP
;
946 } else if (sshdr
.asc
== 0x10) /* DIX */ {
947 action
= ACTION_FAIL
;
948 error
= BLK_STS_PROTECTION
;
949 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
950 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
951 action
= ACTION_FAIL
;
952 error
= BLK_STS_TARGET
;
954 action
= ACTION_FAIL
;
956 case ABORTED_COMMAND
:
957 action
= ACTION_FAIL
;
958 if (sshdr
.asc
== 0x10) /* DIF */
959 error
= BLK_STS_PROTECTION
;
962 /* If the device is in the process of becoming
963 * ready, or has a temporary blockage, retry.
965 if (sshdr
.asc
== 0x04) {
966 switch (sshdr
.ascq
) {
967 case 0x01: /* becoming ready */
968 case 0x04: /* format in progress */
969 case 0x05: /* rebuild in progress */
970 case 0x06: /* recalculation in progress */
971 case 0x07: /* operation in progress */
972 case 0x08: /* Long write in progress */
973 case 0x09: /* self test in progress */
974 case 0x14: /* space allocation in progress */
975 action
= ACTION_DELAYED_RETRY
;
978 action
= ACTION_FAIL
;
982 action
= ACTION_FAIL
;
984 case VOLUME_OVERFLOW
:
985 /* See SSC3rXX or current. */
986 action
= ACTION_FAIL
;
989 action
= ACTION_FAIL
;
993 action
= ACTION_FAIL
;
995 if (action
!= ACTION_FAIL
&&
996 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
997 action
= ACTION_FAIL
;
1001 /* Give up and fail the remainder of the request */
1002 if (!(req
->rq_flags
& RQF_QUIET
)) {
1003 static DEFINE_RATELIMIT_STATE(_rs
,
1004 DEFAULT_RATELIMIT_INTERVAL
,
1005 DEFAULT_RATELIMIT_BURST
);
1007 if (unlikely(scsi_logging_level
))
1008 level
= SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT
,
1009 SCSI_LOG_MLCOMPLETE_BITS
);
1012 * if logging is enabled the failure will be printed
1013 * in scsi_log_completion(), so avoid duplicate messages
1015 if (!level
&& __ratelimit(&_rs
)) {
1016 scsi_print_result(cmd
, NULL
, FAILED
);
1017 if (driver_byte(result
) & DRIVER_SENSE
)
1018 scsi_print_sense(cmd
);
1019 scsi_print_command(cmd
);
1022 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
1027 /* Unprep the request and put it back at the head of the queue.
1028 * A new command will be prepared and issued.
1031 scsi_mq_requeue_cmd(cmd
);
1033 scsi_release_buffers(cmd
);
1034 scsi_requeue_command(q
, cmd
);
1038 /* Retry the same command immediately */
1039 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
1041 case ACTION_DELAYED_RETRY
:
1042 /* Retry the same command after a delay */
1043 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
1048 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
)
1053 * If sg table allocation fails, requeue request later.
1055 if (unlikely(sg_alloc_table_chained(&sdb
->table
,
1056 blk_rq_nr_phys_segments(req
), sdb
->table
.sgl
)))
1057 return BLKPREP_DEFER
;
1060 * Next, walk the list, and fill in the addresses and sizes of
1063 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1064 BUG_ON(count
> sdb
->table
.nents
);
1065 sdb
->table
.nents
= count
;
1066 sdb
->length
= blk_rq_payload_bytes(req
);
1071 * Function: scsi_init_io()
1073 * Purpose: SCSI I/O initialize function.
1075 * Arguments: cmd - Command descriptor we wish to initialize
1077 * Returns: 0 on success
1078 * BLKPREP_DEFER if the failure is retryable
1079 * BLKPREP_KILL if the failure is fatal
1081 int scsi_init_io(struct scsi_cmnd
*cmd
)
1083 struct scsi_device
*sdev
= cmd
->device
;
1084 struct request
*rq
= cmd
->request
;
1085 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1086 int error
= BLKPREP_KILL
;
1088 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq
)))
1091 error
= scsi_init_sgtable(rq
, &cmd
->sdb
);
1095 if (blk_bidi_rq(rq
)) {
1096 if (!rq
->q
->mq_ops
) {
1097 struct scsi_data_buffer
*bidi_sdb
=
1098 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1100 error
= BLKPREP_DEFER
;
1104 rq
->next_rq
->special
= bidi_sdb
;
1107 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
);
1112 if (blk_integrity_rq(rq
)) {
1113 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1116 if (prot_sdb
== NULL
) {
1118 * This can happen if someone (e.g. multipath)
1119 * queues a command to a device on an adapter
1120 * that does not support DIX.
1123 error
= BLKPREP_KILL
;
1127 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1129 if (sg_alloc_table_chained(&prot_sdb
->table
, ivecs
,
1130 prot_sdb
->table
.sgl
)) {
1131 error
= BLKPREP_DEFER
;
1135 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1136 prot_sdb
->table
.sgl
);
1137 BUG_ON(unlikely(count
> ivecs
));
1138 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1140 cmd
->prot_sdb
= prot_sdb
;
1141 cmd
->prot_sdb
->table
.nents
= count
;
1147 scsi_mq_free_sgtables(cmd
);
1149 scsi_release_buffers(cmd
);
1150 cmd
->request
->special
= NULL
;
1151 scsi_put_command(cmd
);
1152 put_device(&sdev
->sdev_gendev
);
1156 EXPORT_SYMBOL(scsi_init_io
);
1159 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1160 * @rq: Request associated with the SCSI command to be initialized.
1162 * This function initializes the members of struct scsi_cmnd that must be
1163 * initialized before request processing starts and that won't be
1164 * reinitialized if a SCSI command is requeued.
1166 * Called from inside blk_get_request() for pass-through requests and from
1167 * inside scsi_init_command() for filesystem requests.
1169 void scsi_initialize_rq(struct request
*rq
)
1171 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1173 scsi_req_init(&cmd
->req
);
1174 init_rcu_head(&cmd
->rcu
);
1175 cmd
->jiffies_at_alloc
= jiffies
;
1178 EXPORT_SYMBOL(scsi_initialize_rq
);
1180 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1181 void scsi_add_cmd_to_list(struct scsi_cmnd
*cmd
)
1183 struct scsi_device
*sdev
= cmd
->device
;
1184 struct Scsi_Host
*shost
= sdev
->host
;
1185 unsigned long flags
;
1187 if (shost
->use_cmd_list
) {
1188 spin_lock_irqsave(&sdev
->list_lock
, flags
);
1189 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1190 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
1194 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1195 void scsi_del_cmd_from_list(struct scsi_cmnd
*cmd
)
1197 struct scsi_device
*sdev
= cmd
->device
;
1198 struct Scsi_Host
*shost
= sdev
->host
;
1199 unsigned long flags
;
1201 if (shost
->use_cmd_list
) {
1202 spin_lock_irqsave(&sdev
->list_lock
, flags
);
1203 BUG_ON(list_empty(&cmd
->list
));
1204 list_del_init(&cmd
->list
);
1205 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
1209 /* Called after a request has been started. */
1210 void scsi_init_command(struct scsi_device
*dev
, struct scsi_cmnd
*cmd
)
1212 void *buf
= cmd
->sense_buffer
;
1213 void *prot
= cmd
->prot_sdb
;
1214 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1215 unsigned int flags
= cmd
->flags
& SCMD_PRESERVED_FLAGS
;
1216 unsigned long jiffies_at_alloc
;
1219 if (!blk_rq_is_scsi(rq
) && !(flags
& SCMD_INITIALIZED
)) {
1220 flags
|= SCMD_INITIALIZED
;
1221 scsi_initialize_rq(rq
);
1224 jiffies_at_alloc
= cmd
->jiffies_at_alloc
;
1225 retries
= cmd
->retries
;
1226 /* zero out the cmd, except for the embedded scsi_request */
1227 memset((char *)cmd
+ sizeof(cmd
->req
), 0,
1228 sizeof(*cmd
) - sizeof(cmd
->req
) + dev
->host
->hostt
->cmd_size
);
1231 cmd
->sense_buffer
= buf
;
1232 cmd
->prot_sdb
= prot
;
1234 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1235 cmd
->jiffies_at_alloc
= jiffies_at_alloc
;
1236 cmd
->retries
= retries
;
1238 scsi_add_cmd_to_list(cmd
);
1241 static int scsi_setup_scsi_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1243 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1246 * Passthrough requests may transfer data, in which case they must
1247 * a bio attached to them. Or they might contain a SCSI command
1248 * that does not transfer data, in which case they may optionally
1249 * submit a request without an attached bio.
1252 int ret
= scsi_init_io(cmd
);
1256 BUG_ON(blk_rq_bytes(req
));
1258 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1261 cmd
->cmd_len
= scsi_req(req
)->cmd_len
;
1262 cmd
->cmnd
= scsi_req(req
)->cmd
;
1263 cmd
->transfersize
= blk_rq_bytes(req
);
1264 cmd
->allowed
= scsi_req(req
)->retries
;
1269 * Setup a normal block command. These are simple request from filesystems
1270 * that still need to be translated to SCSI CDBs from the ULD.
1272 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1274 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1276 if (unlikely(sdev
->handler
&& sdev
->handler
->prep_fn
)) {
1277 int ret
= sdev
->handler
->prep_fn(sdev
, req
);
1278 if (ret
!= BLKPREP_OK
)
1282 cmd
->cmnd
= scsi_req(req
)->cmd
= scsi_req(req
)->__cmd
;
1283 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1284 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1287 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1289 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1291 if (!blk_rq_bytes(req
))
1292 cmd
->sc_data_direction
= DMA_NONE
;
1293 else if (rq_data_dir(req
) == WRITE
)
1294 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1296 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1298 if (blk_rq_is_scsi(req
))
1299 return scsi_setup_scsi_cmnd(sdev
, req
);
1301 return scsi_setup_fs_cmnd(sdev
, req
);
1305 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1307 int ret
= BLKPREP_OK
;
1310 * If the device is not in running state we will reject some
1313 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1314 switch (sdev
->sdev_state
) {
1316 case SDEV_TRANSPORT_OFFLINE
:
1318 * If the device is offline we refuse to process any
1319 * commands. The device must be brought online
1320 * before trying any recovery commands.
1322 sdev_printk(KERN_ERR
, sdev
,
1323 "rejecting I/O to offline device\n");
1328 * If the device is fully deleted, we refuse to
1329 * process any commands as well.
1331 sdev_printk(KERN_ERR
, sdev
,
1332 "rejecting I/O to dead device\n");
1336 case SDEV_CREATED_BLOCK
:
1337 ret
= BLKPREP_DEFER
;
1341 * If the devices is blocked we defer normal commands.
1343 if (req
&& !(req
->rq_flags
& RQF_PREEMPT
))
1344 ret
= BLKPREP_DEFER
;
1348 * For any other not fully online state we only allow
1349 * special commands. In particular any user initiated
1350 * command is not allowed.
1352 if (req
&& !(req
->rq_flags
& RQF_PREEMPT
))
1361 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1363 struct scsi_device
*sdev
= q
->queuedata
;
1367 case BLKPREP_INVALID
:
1368 scsi_req(req
)->result
= DID_NO_CONNECT
<< 16;
1369 /* release the command and kill it */
1371 struct scsi_cmnd
*cmd
= req
->special
;
1372 scsi_release_buffers(cmd
);
1373 scsi_put_command(cmd
);
1374 put_device(&sdev
->sdev_gendev
);
1375 req
->special
= NULL
;
1380 * If we defer, the blk_peek_request() returns NULL, but the
1381 * queue must be restarted, so we schedule a callback to happen
1384 if (atomic_read(&sdev
->device_busy
) == 0)
1385 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1388 req
->rq_flags
|= RQF_DONTPREP
;
1394 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1396 struct scsi_device
*sdev
= q
->queuedata
;
1397 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1400 ret
= scsi_prep_state_check(sdev
, req
);
1401 if (ret
!= BLKPREP_OK
)
1404 if (!req
->special
) {
1405 /* Bail if we can't get a reference to the device */
1406 if (unlikely(!get_device(&sdev
->sdev_gendev
))) {
1407 ret
= BLKPREP_DEFER
;
1411 scsi_init_command(sdev
, cmd
);
1415 cmd
->tag
= req
->tag
;
1417 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1419 ret
= scsi_setup_cmnd(sdev
, req
);
1421 return scsi_prep_return(q
, req
, ret
);
1424 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1426 scsi_uninit_cmd(blk_mq_rq_to_pdu(req
));
1430 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1433 * Called with the queue_lock held.
1435 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1436 struct scsi_device
*sdev
)
1440 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1441 if (atomic_read(&sdev
->device_blocked
)) {
1446 * unblock after device_blocked iterates to zero
1448 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1450 * For the MQ case we take care of this in the caller.
1453 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1456 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1457 "unblocking device at zero depth\n"));
1460 if (busy
>= sdev
->queue_depth
)
1465 atomic_dec(&sdev
->device_busy
);
1470 * scsi_target_queue_ready: checks if there we can send commands to target
1471 * @sdev: scsi device on starget to check.
1473 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1474 struct scsi_device
*sdev
)
1476 struct scsi_target
*starget
= scsi_target(sdev
);
1479 if (starget
->single_lun
) {
1480 spin_lock_irq(shost
->host_lock
);
1481 if (starget
->starget_sdev_user
&&
1482 starget
->starget_sdev_user
!= sdev
) {
1483 spin_unlock_irq(shost
->host_lock
);
1486 starget
->starget_sdev_user
= sdev
;
1487 spin_unlock_irq(shost
->host_lock
);
1490 if (starget
->can_queue
<= 0)
1493 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1494 if (atomic_read(&starget
->target_blocked
) > 0) {
1499 * unblock after target_blocked iterates to zero
1501 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1504 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1505 "unblocking target at zero depth\n"));
1508 if (busy
>= starget
->can_queue
)
1514 spin_lock_irq(shost
->host_lock
);
1515 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1516 spin_unlock_irq(shost
->host_lock
);
1518 if (starget
->can_queue
> 0)
1519 atomic_dec(&starget
->target_busy
);
1524 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1525 * return 0. We must end up running the queue again whenever 0 is
1526 * returned, else IO can hang.
1528 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1529 struct Scsi_Host
*shost
,
1530 struct scsi_device
*sdev
)
1534 if (scsi_host_in_recovery(shost
))
1537 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1538 if (atomic_read(&shost
->host_blocked
) > 0) {
1543 * unblock after host_blocked iterates to zero
1545 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1549 shost_printk(KERN_INFO
, shost
,
1550 "unblocking host at zero depth\n"));
1553 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1555 if (shost
->host_self_blocked
)
1558 /* We're OK to process the command, so we can't be starved */
1559 if (!list_empty(&sdev
->starved_entry
)) {
1560 spin_lock_irq(shost
->host_lock
);
1561 if (!list_empty(&sdev
->starved_entry
))
1562 list_del_init(&sdev
->starved_entry
);
1563 spin_unlock_irq(shost
->host_lock
);
1569 spin_lock_irq(shost
->host_lock
);
1570 if (list_empty(&sdev
->starved_entry
))
1571 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1572 spin_unlock_irq(shost
->host_lock
);
1574 scsi_dec_host_busy(shost
);
1579 * Busy state exporting function for request stacking drivers.
1581 * For efficiency, no lock is taken to check the busy state of
1582 * shost/starget/sdev, since the returned value is not guaranteed and
1583 * may be changed after request stacking drivers call the function,
1584 * regardless of taking lock or not.
1586 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1587 * needs to return 'not busy'. Otherwise, request stacking drivers
1588 * may hold requests forever.
1590 static int scsi_lld_busy(struct request_queue
*q
)
1592 struct scsi_device
*sdev
= q
->queuedata
;
1593 struct Scsi_Host
*shost
;
1595 if (blk_queue_dying(q
))
1601 * Ignore host/starget busy state.
1602 * Since block layer does not have a concept of fairness across
1603 * multiple queues, congestion of host/starget needs to be handled
1606 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1613 * Kill a request for a dead device
1615 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1617 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1618 struct scsi_device
*sdev
;
1619 struct scsi_target
*starget
;
1620 struct Scsi_Host
*shost
;
1622 blk_start_request(req
);
1624 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1627 starget
= scsi_target(sdev
);
1629 scsi_init_cmd_errh(cmd
);
1630 cmd
->result
= DID_NO_CONNECT
<< 16;
1631 atomic_inc(&cmd
->device
->iorequest_cnt
);
1634 * SCSI request completion path will do scsi_device_unbusy(),
1635 * bump busy counts. To bump the counters, we need to dance
1636 * with the locks as normal issue path does.
1638 atomic_inc(&sdev
->device_busy
);
1639 atomic_inc(&shost
->host_busy
);
1640 if (starget
->can_queue
> 0)
1641 atomic_inc(&starget
->target_busy
);
1643 blk_complete_request(req
);
1646 static void scsi_softirq_done(struct request
*rq
)
1648 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1649 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1652 INIT_LIST_HEAD(&cmd
->eh_entry
);
1654 atomic_inc(&cmd
->device
->iodone_cnt
);
1656 atomic_inc(&cmd
->device
->ioerr_cnt
);
1658 disposition
= scsi_decide_disposition(cmd
);
1659 if (disposition
!= SUCCESS
&&
1660 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1661 sdev_printk(KERN_ERR
, cmd
->device
,
1662 "timing out command, waited %lus\n",
1664 disposition
= SUCCESS
;
1667 scsi_log_completion(cmd
, disposition
);
1669 switch (disposition
) {
1671 scsi_finish_command(cmd
);
1674 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1676 case ADD_TO_MLQUEUE
:
1677 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1680 scsi_eh_scmd_add(cmd
);
1686 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1687 * @cmd: command block we are dispatching.
1689 * Return: nonzero return request was rejected and device's queue needs to be
1692 static int scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
1694 struct Scsi_Host
*host
= cmd
->device
->host
;
1697 atomic_inc(&cmd
->device
->iorequest_cnt
);
1699 /* check if the device is still usable */
1700 if (unlikely(cmd
->device
->sdev_state
== SDEV_DEL
)) {
1701 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1702 * returns an immediate error upwards, and signals
1703 * that the device is no longer present */
1704 cmd
->result
= DID_NO_CONNECT
<< 16;
1708 /* Check to see if the scsi lld made this device blocked. */
1709 if (unlikely(scsi_device_blocked(cmd
->device
))) {
1711 * in blocked state, the command is just put back on
1712 * the device queue. The suspend state has already
1713 * blocked the queue so future requests should not
1714 * occur until the device transitions out of the
1717 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1718 "queuecommand : device blocked\n"));
1719 return SCSI_MLQUEUE_DEVICE_BUSY
;
1722 /* Store the LUN value in cmnd, if needed. */
1723 if (cmd
->device
->lun_in_cdb
)
1724 cmd
->cmnd
[1] = (cmd
->cmnd
[1] & 0x1f) |
1725 (cmd
->device
->lun
<< 5 & 0xe0);
1730 * Before we queue this command, check if the command
1731 * length exceeds what the host adapter can handle.
1733 if (cmd
->cmd_len
> cmd
->device
->host
->max_cmd_len
) {
1734 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1735 "queuecommand : command too long. "
1736 "cdb_size=%d host->max_cmd_len=%d\n",
1737 cmd
->cmd_len
, cmd
->device
->host
->max_cmd_len
));
1738 cmd
->result
= (DID_ABORT
<< 16);
1742 if (unlikely(host
->shost_state
== SHOST_DEL
)) {
1743 cmd
->result
= (DID_NO_CONNECT
<< 16);
1748 trace_scsi_dispatch_cmd_start(cmd
);
1749 rtn
= host
->hostt
->queuecommand(host
, cmd
);
1751 trace_scsi_dispatch_cmd_error(cmd
, rtn
);
1752 if (rtn
!= SCSI_MLQUEUE_DEVICE_BUSY
&&
1753 rtn
!= SCSI_MLQUEUE_TARGET_BUSY
)
1754 rtn
= SCSI_MLQUEUE_HOST_BUSY
;
1756 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1757 "queuecommand : request rejected\n"));
1762 cmd
->scsi_done(cmd
);
1767 * scsi_done - Invoke completion on finished SCSI command.
1768 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1769 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1771 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1772 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1773 * calls blk_complete_request() for further processing.
1775 * This function is interrupt context safe.
1777 static void scsi_done(struct scsi_cmnd
*cmd
)
1779 trace_scsi_dispatch_cmd_done(cmd
);
1780 blk_complete_request(cmd
->request
);
1784 * Function: scsi_request_fn()
1786 * Purpose: Main strategy routine for SCSI.
1788 * Arguments: q - Pointer to actual queue.
1792 * Lock status: request queue lock assumed to be held when called.
1794 * Note: See sd_zbc.c sd_zbc_write_lock_zone() for write order
1795 * protection for ZBC disks.
1797 static void scsi_request_fn(struct request_queue
*q
)
1798 __releases(q
->queue_lock
)
1799 __acquires(q
->queue_lock
)
1801 struct scsi_device
*sdev
= q
->queuedata
;
1802 struct Scsi_Host
*shost
;
1803 struct scsi_cmnd
*cmd
;
1804 struct request
*req
;
1807 * To start with, we keep looping until the queue is empty, or until
1808 * the host is no longer able to accept any more requests.
1814 * get next queueable request. We do this early to make sure
1815 * that the request is fully prepared even if we cannot
1818 req
= blk_peek_request(q
);
1822 if (unlikely(!scsi_device_online(sdev
))) {
1823 sdev_printk(KERN_ERR
, sdev
,
1824 "rejecting I/O to offline device\n");
1825 scsi_kill_request(req
, q
);
1829 if (!scsi_dev_queue_ready(q
, sdev
))
1833 * Remove the request from the request list.
1835 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1836 blk_start_request(req
);
1838 spin_unlock_irq(q
->queue_lock
);
1839 cmd
= blk_mq_rq_to_pdu(req
);
1840 if (cmd
!= req
->special
) {
1841 printk(KERN_CRIT
"impossible request in %s.\n"
1842 "please mail a stack trace to "
1843 "linux-scsi@vger.kernel.org\n",
1845 blk_dump_rq_flags(req
, "foo");
1850 * We hit this when the driver is using a host wide
1851 * tag map. For device level tag maps the queue_depth check
1852 * in the device ready fn would prevent us from trying
1853 * to allocate a tag. Since the map is a shared host resource
1854 * we add the dev to the starved list so it eventually gets
1855 * a run when a tag is freed.
1857 if (blk_queue_tagged(q
) && !(req
->rq_flags
& RQF_QUEUED
)) {
1858 spin_lock_irq(shost
->host_lock
);
1859 if (list_empty(&sdev
->starved_entry
))
1860 list_add_tail(&sdev
->starved_entry
,
1861 &shost
->starved_list
);
1862 spin_unlock_irq(shost
->host_lock
);
1866 if (!scsi_target_queue_ready(shost
, sdev
))
1869 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1870 goto host_not_ready
;
1872 if (sdev
->simple_tags
)
1873 cmd
->flags
|= SCMD_TAGGED
;
1875 cmd
->flags
&= ~SCMD_TAGGED
;
1878 * Finally, initialize any error handling parameters, and set up
1879 * the timers for timeouts.
1881 scsi_init_cmd_errh(cmd
);
1884 * Dispatch the command to the low-level driver.
1886 cmd
->scsi_done
= scsi_done
;
1887 rtn
= scsi_dispatch_cmd(cmd
);
1889 scsi_queue_insert(cmd
, rtn
);
1890 spin_lock_irq(q
->queue_lock
);
1893 spin_lock_irq(q
->queue_lock
);
1899 if (scsi_target(sdev
)->can_queue
> 0)
1900 atomic_dec(&scsi_target(sdev
)->target_busy
);
1903 * lock q, handle tag, requeue req, and decrement device_busy. We
1904 * must return with queue_lock held.
1906 * Decrementing device_busy without checking it is OK, as all such
1907 * cases (host limits or settings) should run the queue at some
1910 spin_lock_irq(q
->queue_lock
);
1911 blk_requeue_request(q
, req
);
1912 atomic_dec(&sdev
->device_busy
);
1914 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1915 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1918 static inline blk_status_t
prep_to_mq(int ret
)
1924 return BLK_STS_RESOURCE
;
1926 return BLK_STS_IOERR
;
1930 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1931 static unsigned int scsi_mq_sgl_size(struct Scsi_Host
*shost
)
1933 return min_t(unsigned int, shost
->sg_tablesize
, SG_CHUNK_SIZE
) *
1934 sizeof(struct scatterlist
);
1937 static int scsi_mq_prep_fn(struct request
*req
)
1939 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1940 struct scsi_device
*sdev
= req
->q
->queuedata
;
1941 struct Scsi_Host
*shost
= sdev
->host
;
1942 struct scatterlist
*sg
;
1944 scsi_init_command(sdev
, cmd
);
1950 cmd
->tag
= req
->tag
;
1951 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1953 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1954 cmd
->sdb
.table
.sgl
= sg
;
1956 if (scsi_host_get_prot(shost
)) {
1957 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1959 cmd
->prot_sdb
->table
.sgl
=
1960 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1963 if (blk_bidi_rq(req
)) {
1964 struct request
*next_rq
= req
->next_rq
;
1965 struct scsi_data_buffer
*bidi_sdb
= blk_mq_rq_to_pdu(next_rq
);
1967 memset(bidi_sdb
, 0, sizeof(struct scsi_data_buffer
));
1968 bidi_sdb
->table
.sgl
=
1969 (struct scatterlist
*)(bidi_sdb
+ 1);
1971 next_rq
->special
= bidi_sdb
;
1974 blk_mq_start_request(req
);
1976 return scsi_setup_cmnd(sdev
, req
);
1979 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1981 trace_scsi_dispatch_cmd_done(cmd
);
1982 blk_mq_complete_request(cmd
->request
);
1985 static void scsi_mq_put_budget(struct blk_mq_hw_ctx
*hctx
)
1987 struct request_queue
*q
= hctx
->queue
;
1988 struct scsi_device
*sdev
= q
->queuedata
;
1990 atomic_dec(&sdev
->device_busy
);
1991 put_device(&sdev
->sdev_gendev
);
1994 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx
*hctx
)
1996 struct request_queue
*q
= hctx
->queue
;
1997 struct scsi_device
*sdev
= q
->queuedata
;
1999 if (!get_device(&sdev
->sdev_gendev
))
2001 if (!scsi_dev_queue_ready(q
, sdev
))
2002 goto out_put_device
;
2007 put_device(&sdev
->sdev_gendev
);
2009 if (atomic_read(&sdev
->device_busy
) == 0 && !scsi_device_blocked(sdev
))
2010 blk_mq_delay_run_hw_queue(hctx
, SCSI_QUEUE_DELAY
);
2014 static blk_status_t
scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2015 const struct blk_mq_queue_data
*bd
)
2017 struct request
*req
= bd
->rq
;
2018 struct request_queue
*q
= req
->q
;
2019 struct scsi_device
*sdev
= q
->queuedata
;
2020 struct Scsi_Host
*shost
= sdev
->host
;
2021 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
2025 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
2026 if (ret
!= BLK_STS_OK
)
2027 goto out_put_budget
;
2029 ret
= BLK_STS_RESOURCE
;
2030 if (!scsi_target_queue_ready(shost
, sdev
))
2031 goto out_put_budget
;
2032 if (!scsi_host_queue_ready(q
, shost
, sdev
))
2033 goto out_dec_target_busy
;
2035 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
2036 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
2037 if (ret
!= BLK_STS_OK
)
2038 goto out_dec_host_busy
;
2039 req
->rq_flags
|= RQF_DONTPREP
;
2041 blk_mq_start_request(req
);
2044 if (sdev
->simple_tags
)
2045 cmd
->flags
|= SCMD_TAGGED
;
2047 cmd
->flags
&= ~SCMD_TAGGED
;
2049 scsi_init_cmd_errh(cmd
);
2050 cmd
->scsi_done
= scsi_mq_done
;
2052 reason
= scsi_dispatch_cmd(cmd
);
2054 scsi_set_blocked(cmd
, reason
);
2055 ret
= BLK_STS_RESOURCE
;
2056 goto out_dec_host_busy
;
2062 scsi_dec_host_busy(shost
);
2063 out_dec_target_busy
:
2064 if (scsi_target(sdev
)->can_queue
> 0)
2065 atomic_dec(&scsi_target(sdev
)->target_busy
);
2067 scsi_mq_put_budget(hctx
);
2071 case BLK_STS_RESOURCE
:
2072 if (atomic_read(&sdev
->device_busy
) ||
2073 scsi_device_blocked(sdev
))
2074 ret
= BLK_STS_DEV_RESOURCE
;
2077 if (unlikely(!scsi_device_online(sdev
)))
2078 scsi_req(req
)->result
= DID_NO_CONNECT
<< 16;
2080 scsi_req(req
)->result
= DID_ERROR
<< 16;
2082 * Make sure to release all allocated resources when
2083 * we hit an error, as we will never see this command
2086 if (req
->rq_flags
& RQF_DONTPREP
)
2087 scsi_mq_uninit_cmd(cmd
);
2093 static enum blk_eh_timer_return
scsi_timeout(struct request
*req
,
2097 return BLK_EH_RESET_TIMER
;
2098 return scsi_times_out(req
);
2101 static int scsi_mq_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
2102 unsigned int hctx_idx
, unsigned int numa_node
)
2104 struct Scsi_Host
*shost
= set
->driver_data
;
2105 const bool unchecked_isa_dma
= shost
->unchecked_isa_dma
;
2106 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
2107 struct scatterlist
*sg
;
2109 if (unchecked_isa_dma
)
2110 cmd
->flags
|= SCMD_UNCHECKED_ISA_DMA
;
2111 cmd
->sense_buffer
= scsi_alloc_sense_buffer(unchecked_isa_dma
,
2112 GFP_KERNEL
, numa_node
);
2113 if (!cmd
->sense_buffer
)
2115 cmd
->req
.sense
= cmd
->sense_buffer
;
2117 if (scsi_host_get_prot(shost
)) {
2118 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) +
2119 shost
->hostt
->cmd_size
;
2120 cmd
->prot_sdb
= (void *)sg
+ scsi_mq_sgl_size(shost
);
2126 static void scsi_mq_exit_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
2127 unsigned int hctx_idx
)
2129 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
2131 scsi_free_sense_buffer(cmd
->flags
& SCMD_UNCHECKED_ISA_DMA
,
2135 static int scsi_map_queues(struct blk_mq_tag_set
*set
)
2137 struct Scsi_Host
*shost
= container_of(set
, struct Scsi_Host
, tag_set
);
2139 if (shost
->hostt
->map_queues
)
2140 return shost
->hostt
->map_queues(shost
);
2141 return blk_mq_map_queues(set
);
2144 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
2146 struct device
*host_dev
;
2147 u64 bounce_limit
= 0xffffffff;
2149 if (shost
->unchecked_isa_dma
)
2150 return BLK_BOUNCE_ISA
;
2152 * Platforms with virtual-DMA translation
2153 * hardware have no practical limit.
2155 if (!PCI_DMA_BUS_IS_PHYS
)
2156 return BLK_BOUNCE_ANY
;
2158 host_dev
= scsi_get_device(shost
);
2159 if (host_dev
&& host_dev
->dma_mask
)
2160 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
2162 return bounce_limit
;
2165 void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
2167 struct device
*dev
= shost
->dma_dev
;
2169 queue_flag_set_unlocked(QUEUE_FLAG_SCSI_PASSTHROUGH
, q
);
2172 * this limit is imposed by hardware restrictions
2174 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
2177 if (scsi_host_prot_dma(shost
)) {
2178 shost
->sg_prot_tablesize
=
2179 min_not_zero(shost
->sg_prot_tablesize
,
2180 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
2181 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
2182 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
2185 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
2186 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
2187 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
2188 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
2190 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2192 if (!shost
->use_clustering
)
2193 q
->limits
.cluster
= 0;
2196 * Set a reasonable default alignment: The larger of 32-byte (dword),
2197 * which is a common minimum for HBAs, and the minimum DMA alignment,
2198 * which is set by the platform.
2200 * Devices that require a bigger alignment can increase it later.
2202 blk_queue_dma_alignment(q
, max(4, dma_get_cache_alignment()) - 1);
2204 EXPORT_SYMBOL_GPL(__scsi_init_queue
);
2206 static int scsi_old_init_rq(struct request_queue
*q
, struct request
*rq
,
2209 struct Scsi_Host
*shost
= q
->rq_alloc_data
;
2210 const bool unchecked_isa_dma
= shost
->unchecked_isa_dma
;
2211 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
2213 memset(cmd
, 0, sizeof(*cmd
));
2215 if (unchecked_isa_dma
)
2216 cmd
->flags
|= SCMD_UNCHECKED_ISA_DMA
;
2217 cmd
->sense_buffer
= scsi_alloc_sense_buffer(unchecked_isa_dma
, gfp
,
2219 if (!cmd
->sense_buffer
)
2221 cmd
->req
.sense
= cmd
->sense_buffer
;
2223 if (scsi_host_get_prot(shost
) >= SHOST_DIX_TYPE0_PROTECTION
) {
2224 cmd
->prot_sdb
= kmem_cache_zalloc(scsi_sdb_cache
, gfp
);
2226 goto fail_free_sense
;
2232 scsi_free_sense_buffer(unchecked_isa_dma
, cmd
->sense_buffer
);
2237 static void scsi_old_exit_rq(struct request_queue
*q
, struct request
*rq
)
2239 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
2242 kmem_cache_free(scsi_sdb_cache
, cmd
->prot_sdb
);
2243 scsi_free_sense_buffer(cmd
->flags
& SCMD_UNCHECKED_ISA_DMA
,
2247 struct request_queue
*scsi_old_alloc_queue(struct scsi_device
*sdev
)
2249 struct Scsi_Host
*shost
= sdev
->host
;
2250 struct request_queue
*q
;
2252 q
= blk_alloc_queue_node(GFP_KERNEL
, NUMA_NO_NODE
);
2255 q
->cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
2256 q
->rq_alloc_data
= shost
;
2257 q
->request_fn
= scsi_request_fn
;
2258 q
->init_rq_fn
= scsi_old_init_rq
;
2259 q
->exit_rq_fn
= scsi_old_exit_rq
;
2260 q
->initialize_rq_fn
= scsi_initialize_rq
;
2262 if (blk_init_allocated_queue(q
) < 0) {
2263 blk_cleanup_queue(q
);
2267 __scsi_init_queue(shost
, q
);
2268 blk_queue_prep_rq(q
, scsi_prep_fn
);
2269 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2270 blk_queue_softirq_done(q
, scsi_softirq_done
);
2271 blk_queue_rq_timed_out(q
, scsi_times_out
);
2272 blk_queue_lld_busy(q
, scsi_lld_busy
);
2276 static const struct blk_mq_ops scsi_mq_ops
= {
2277 .get_budget
= scsi_mq_get_budget
,
2278 .put_budget
= scsi_mq_put_budget
,
2279 .queue_rq
= scsi_queue_rq
,
2280 .complete
= scsi_softirq_done
,
2281 .timeout
= scsi_timeout
,
2282 #ifdef CONFIG_BLK_DEBUG_FS
2283 .show_rq
= scsi_show_rq
,
2285 .init_request
= scsi_mq_init_request
,
2286 .exit_request
= scsi_mq_exit_request
,
2287 .initialize_rq_fn
= scsi_initialize_rq
,
2288 .map_queues
= scsi_map_queues
,
2291 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2293 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2294 if (IS_ERR(sdev
->request_queue
))
2297 sdev
->request_queue
->queuedata
= sdev
;
2298 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2299 return sdev
->request_queue
;
2302 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2304 unsigned int cmd_size
, sgl_size
;
2306 sgl_size
= max_t(unsigned int, sizeof(struct scatterlist
),
2307 scsi_mq_sgl_size(shost
));
2308 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2309 if (scsi_host_get_prot(shost
))
2310 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2312 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2313 shost
->tag_set
.ops
= &scsi_mq_ops
;
2314 shost
->tag_set
.nr_hw_queues
= shost
->nr_hw_queues
? : 1;
2315 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2316 shost
->tag_set
.cmd_size
= cmd_size
;
2317 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2318 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2319 shost
->tag_set
.flags
|=
2320 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost
->hostt
->tag_alloc_policy
);
2321 shost
->tag_set
.driver_data
= shost
;
2323 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2326 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2328 blk_mq_free_tag_set(&shost
->tag_set
);
2332 * scsi_device_from_queue - return sdev associated with a request_queue
2333 * @q: The request queue to return the sdev from
2335 * Return the sdev associated with a request queue or NULL if the
2336 * request_queue does not reference a SCSI device.
2338 struct scsi_device
*scsi_device_from_queue(struct request_queue
*q
)
2340 struct scsi_device
*sdev
= NULL
;
2343 if (q
->mq_ops
== &scsi_mq_ops
)
2344 sdev
= q
->queuedata
;
2345 } else if (q
->request_fn
== scsi_request_fn
)
2346 sdev
= q
->queuedata
;
2347 if (!sdev
|| !get_device(&sdev
->sdev_gendev
))
2352 EXPORT_SYMBOL_GPL(scsi_device_from_queue
);
2355 * Function: scsi_block_requests()
2357 * Purpose: Utility function used by low-level drivers to prevent further
2358 * commands from being queued to the device.
2360 * Arguments: shost - Host in question
2364 * Lock status: No locks are assumed held.
2366 * Notes: There is no timer nor any other means by which the requests
2367 * get unblocked other than the low-level driver calling
2368 * scsi_unblock_requests().
2370 void scsi_block_requests(struct Scsi_Host
*shost
)
2372 shost
->host_self_blocked
= 1;
2374 EXPORT_SYMBOL(scsi_block_requests
);
2377 * Function: scsi_unblock_requests()
2379 * Purpose: Utility function used by low-level drivers to allow further
2380 * commands from being queued to the device.
2382 * Arguments: shost - Host in question
2386 * Lock status: No locks are assumed held.
2388 * Notes: There is no timer nor any other means by which the requests
2389 * get unblocked other than the low-level driver calling
2390 * scsi_unblock_requests().
2392 * This is done as an API function so that changes to the
2393 * internals of the scsi mid-layer won't require wholesale
2394 * changes to drivers that use this feature.
2396 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2398 shost
->host_self_blocked
= 0;
2399 scsi_run_host_queues(shost
);
2401 EXPORT_SYMBOL(scsi_unblock_requests
);
2403 int __init
scsi_init_queue(void)
2405 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2406 sizeof(struct scsi_data_buffer
),
2408 if (!scsi_sdb_cache
) {
2409 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2416 void scsi_exit_queue(void)
2418 kmem_cache_destroy(scsi_sense_cache
);
2419 kmem_cache_destroy(scsi_sense_isadma_cache
);
2420 kmem_cache_destroy(scsi_sdb_cache
);
2424 * scsi_mode_select - issue a mode select
2425 * @sdev: SCSI device to be queried
2426 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2427 * @sp: Save page bit (0 == don't save, 1 == save)
2428 * @modepage: mode page being requested
2429 * @buffer: request buffer (may not be smaller than eight bytes)
2430 * @len: length of request buffer.
2431 * @timeout: command timeout
2432 * @retries: number of retries before failing
2433 * @data: returns a structure abstracting the mode header data
2434 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2435 * must be SCSI_SENSE_BUFFERSIZE big.
2437 * Returns zero if successful; negative error number or scsi
2442 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2443 unsigned char *buffer
, int len
, int timeout
, int retries
,
2444 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2446 unsigned char cmd
[10];
2447 unsigned char *real_buffer
;
2450 memset(cmd
, 0, sizeof(cmd
));
2451 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2453 if (sdev
->use_10_for_ms
) {
2456 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2459 memcpy(real_buffer
+ 8, buffer
, len
);
2463 real_buffer
[2] = data
->medium_type
;
2464 real_buffer
[3] = data
->device_specific
;
2465 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2467 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2468 real_buffer
[7] = data
->block_descriptor_length
;
2470 cmd
[0] = MODE_SELECT_10
;
2474 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2478 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2481 memcpy(real_buffer
+ 4, buffer
, len
);
2484 real_buffer
[1] = data
->medium_type
;
2485 real_buffer
[2] = data
->device_specific
;
2486 real_buffer
[3] = data
->block_descriptor_length
;
2489 cmd
[0] = MODE_SELECT
;
2493 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2494 sshdr
, timeout
, retries
, NULL
);
2498 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2501 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2502 * @sdev: SCSI device to be queried
2503 * @dbd: set if mode sense will allow block descriptors to be returned
2504 * @modepage: mode page being requested
2505 * @buffer: request buffer (may not be smaller than eight bytes)
2506 * @len: length of request buffer.
2507 * @timeout: command timeout
2508 * @retries: number of retries before failing
2509 * @data: returns a structure abstracting the mode header data
2510 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2511 * must be SCSI_SENSE_BUFFERSIZE big.
2513 * Returns zero if unsuccessful, or the header offset (either 4
2514 * or 8 depending on whether a six or ten byte command was
2515 * issued) if successful.
2518 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2519 unsigned char *buffer
, int len
, int timeout
, int retries
,
2520 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2522 unsigned char cmd
[12];
2525 int result
, retry_count
= retries
;
2526 struct scsi_sense_hdr my_sshdr
;
2528 memset(data
, 0, sizeof(*data
));
2529 memset(&cmd
[0], 0, 12);
2530 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2533 /* caller might not be interested in sense, but we need it */
2538 use_10_for_ms
= sdev
->use_10_for_ms
;
2540 if (use_10_for_ms
) {
2544 cmd
[0] = MODE_SENSE_10
;
2551 cmd
[0] = MODE_SENSE
;
2556 memset(buffer
, 0, len
);
2558 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2559 sshdr
, timeout
, retries
, NULL
);
2561 /* This code looks awful: what it's doing is making sure an
2562 * ILLEGAL REQUEST sense return identifies the actual command
2563 * byte as the problem. MODE_SENSE commands can return
2564 * ILLEGAL REQUEST if the code page isn't supported */
2566 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2567 (driver_byte(result
) & DRIVER_SENSE
)) {
2568 if (scsi_sense_valid(sshdr
)) {
2569 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2570 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2572 * Invalid command operation code
2574 sdev
->use_10_for_ms
= 0;
2580 if(scsi_status_is_good(result
)) {
2581 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2582 (modepage
== 6 || modepage
== 8))) {
2583 /* Initio breakage? */
2586 data
->medium_type
= 0;
2587 data
->device_specific
= 0;
2589 data
->block_descriptor_length
= 0;
2590 } else if(use_10_for_ms
) {
2591 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2592 data
->medium_type
= buffer
[2];
2593 data
->device_specific
= buffer
[3];
2594 data
->longlba
= buffer
[4] & 0x01;
2595 data
->block_descriptor_length
= buffer
[6]*256
2598 data
->length
= buffer
[0] + 1;
2599 data
->medium_type
= buffer
[1];
2600 data
->device_specific
= buffer
[2];
2601 data
->block_descriptor_length
= buffer
[3];
2603 data
->header_length
= header_length
;
2604 } else if ((status_byte(result
) == CHECK_CONDITION
) &&
2605 scsi_sense_valid(sshdr
) &&
2606 sshdr
->sense_key
== UNIT_ATTENTION
&& retry_count
) {
2613 EXPORT_SYMBOL(scsi_mode_sense
);
2616 * scsi_test_unit_ready - test if unit is ready
2617 * @sdev: scsi device to change the state of.
2618 * @timeout: command timeout
2619 * @retries: number of retries before failing
2620 * @sshdr: outpout pointer for decoded sense information.
2622 * Returns zero if unsuccessful or an error if TUR failed. For
2623 * removable media, UNIT_ATTENTION sets ->changed flag.
2626 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2627 struct scsi_sense_hdr
*sshdr
)
2630 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2634 /* try to eat the UNIT_ATTENTION if there are enough retries */
2636 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2637 timeout
, retries
, NULL
);
2638 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2639 sshdr
->sense_key
== UNIT_ATTENTION
)
2641 } while (scsi_sense_valid(sshdr
) &&
2642 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2646 EXPORT_SYMBOL(scsi_test_unit_ready
);
2649 * scsi_device_set_state - Take the given device through the device state model.
2650 * @sdev: scsi device to change the state of.
2651 * @state: state to change to.
2653 * Returns zero if successful or an error if the requested
2654 * transition is illegal.
2657 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2659 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2661 if (state
== oldstate
)
2667 case SDEV_CREATED_BLOCK
:
2678 case SDEV_TRANSPORT_OFFLINE
:
2691 case SDEV_TRANSPORT_OFFLINE
:
2699 case SDEV_TRANSPORT_OFFLINE
:
2714 case SDEV_CREATED_BLOCK
:
2721 case SDEV_CREATED_BLOCK
:
2736 case SDEV_TRANSPORT_OFFLINE
:
2748 case SDEV_TRANSPORT_OFFLINE
:
2751 case SDEV_CREATED_BLOCK
:
2759 sdev
->sdev_state
= state
;
2763 SCSI_LOG_ERROR_RECOVERY(1,
2764 sdev_printk(KERN_ERR
, sdev
,
2765 "Illegal state transition %s->%s",
2766 scsi_device_state_name(oldstate
),
2767 scsi_device_state_name(state
))
2771 EXPORT_SYMBOL(scsi_device_set_state
);
2774 * sdev_evt_emit - emit a single SCSI device uevent
2775 * @sdev: associated SCSI device
2776 * @evt: event to emit
2778 * Send a single uevent (scsi_event) to the associated scsi_device.
2780 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2785 switch (evt
->evt_type
) {
2786 case SDEV_EVT_MEDIA_CHANGE
:
2787 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2789 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2790 scsi_rescan_device(&sdev
->sdev_gendev
);
2791 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2793 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2794 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2796 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2797 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2799 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2800 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2802 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2803 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2805 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2806 envp
[idx
++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2808 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2809 envp
[idx
++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2818 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2822 * sdev_evt_thread - send a uevent for each scsi event
2823 * @work: work struct for scsi_device
2825 * Dispatch queued events to their associated scsi_device kobjects
2828 void scsi_evt_thread(struct work_struct
*work
)
2830 struct scsi_device
*sdev
;
2831 enum scsi_device_event evt_type
;
2832 LIST_HEAD(event_list
);
2834 sdev
= container_of(work
, struct scsi_device
, event_work
);
2836 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2837 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2838 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2841 struct scsi_event
*evt
;
2842 struct list_head
*this, *tmp
;
2843 unsigned long flags
;
2845 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2846 list_splice_init(&sdev
->event_list
, &event_list
);
2847 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2849 if (list_empty(&event_list
))
2852 list_for_each_safe(this, tmp
, &event_list
) {
2853 evt
= list_entry(this, struct scsi_event
, node
);
2854 list_del(&evt
->node
);
2855 scsi_evt_emit(sdev
, evt
);
2862 * sdev_evt_send - send asserted event to uevent thread
2863 * @sdev: scsi_device event occurred on
2864 * @evt: event to send
2866 * Assert scsi device event asynchronously.
2868 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2870 unsigned long flags
;
2873 /* FIXME: currently this check eliminates all media change events
2874 * for polled devices. Need to update to discriminate between AN
2875 * and polled events */
2876 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2882 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2883 list_add_tail(&evt
->node
, &sdev
->event_list
);
2884 schedule_work(&sdev
->event_work
);
2885 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2887 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2890 * sdev_evt_alloc - allocate a new scsi event
2891 * @evt_type: type of event to allocate
2892 * @gfpflags: GFP flags for allocation
2894 * Allocates and returns a new scsi_event.
2896 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2899 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2903 evt
->evt_type
= evt_type
;
2904 INIT_LIST_HEAD(&evt
->node
);
2906 /* evt_type-specific initialization, if any */
2908 case SDEV_EVT_MEDIA_CHANGE
:
2909 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2910 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2911 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2912 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2913 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2914 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2915 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2923 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2926 * sdev_evt_send_simple - send asserted event to uevent thread
2927 * @sdev: scsi_device event occurred on
2928 * @evt_type: type of event to send
2929 * @gfpflags: GFP flags for allocation
2931 * Assert scsi device event asynchronously, given an event type.
2933 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2934 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2936 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2938 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2943 sdev_evt_send(sdev
, evt
);
2945 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2948 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2949 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2951 static int scsi_request_fn_active(struct scsi_device
*sdev
)
2953 struct request_queue
*q
= sdev
->request_queue
;
2954 int request_fn_active
;
2956 WARN_ON_ONCE(sdev
->host
->use_blk_mq
);
2958 spin_lock_irq(q
->queue_lock
);
2959 request_fn_active
= q
->request_fn_active
;
2960 spin_unlock_irq(q
->queue_lock
);
2962 return request_fn_active
;
2966 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2967 * @sdev: SCSI device pointer.
2969 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2970 * invoked from scsi_request_fn() have finished.
2972 static void scsi_wait_for_queuecommand(struct scsi_device
*sdev
)
2974 WARN_ON_ONCE(sdev
->host
->use_blk_mq
);
2976 while (scsi_request_fn_active(sdev
))
2981 * scsi_device_quiesce - Block user issued commands.
2982 * @sdev: scsi device to quiesce.
2984 * This works by trying to transition to the SDEV_QUIESCE state
2985 * (which must be a legal transition). When the device is in this
2986 * state, only special requests will be accepted, all others will
2987 * be deferred. Since special requests may also be requeued requests,
2988 * a successful return doesn't guarantee the device will be
2989 * totally quiescent.
2991 * Must be called with user context, may sleep.
2993 * Returns zero if unsuccessful or an error if not.
2996 scsi_device_quiesce(struct scsi_device
*sdev
)
2998 struct request_queue
*q
= sdev
->request_queue
;
3002 * It is allowed to call scsi_device_quiesce() multiple times from
3003 * the same context but concurrent scsi_device_quiesce() calls are
3006 WARN_ON_ONCE(sdev
->quiesced_by
&& sdev
->quiesced_by
!= current
);
3008 blk_set_preempt_only(q
);
3010 blk_mq_freeze_queue(q
);
3012 * Ensure that the effect of blk_set_preempt_only() will be visible
3013 * for percpu_ref_tryget() callers that occur after the queue
3014 * unfreeze even if the queue was already frozen before this function
3015 * was called. See also https://lwn.net/Articles/573497/.
3018 blk_mq_unfreeze_queue(q
);
3020 mutex_lock(&sdev
->state_mutex
);
3021 err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
3023 sdev
->quiesced_by
= current
;
3025 blk_clear_preempt_only(q
);
3026 mutex_unlock(&sdev
->state_mutex
);
3030 EXPORT_SYMBOL(scsi_device_quiesce
);
3033 * scsi_device_resume - Restart user issued commands to a quiesced device.
3034 * @sdev: scsi device to resume.
3036 * Moves the device from quiesced back to running and restarts the
3039 * Must be called with user context, may sleep.
3041 void scsi_device_resume(struct scsi_device
*sdev
)
3043 /* check if the device state was mutated prior to resume, and if
3044 * so assume the state is being managed elsewhere (for example
3045 * device deleted during suspend)
3047 mutex_lock(&sdev
->state_mutex
);
3048 sdev
->quiesced_by
= NULL
;
3049 blk_clear_preempt_only(sdev
->request_queue
);
3050 if (sdev
->sdev_state
== SDEV_QUIESCE
)
3051 scsi_device_set_state(sdev
, SDEV_RUNNING
);
3052 mutex_unlock(&sdev
->state_mutex
);
3054 EXPORT_SYMBOL(scsi_device_resume
);
3057 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
3059 scsi_device_quiesce(sdev
);
3063 scsi_target_quiesce(struct scsi_target
*starget
)
3065 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
3067 EXPORT_SYMBOL(scsi_target_quiesce
);
3070 device_resume_fn(struct scsi_device
*sdev
, void *data
)
3072 scsi_device_resume(sdev
);
3076 scsi_target_resume(struct scsi_target
*starget
)
3078 starget_for_each_device(starget
, NULL
, device_resume_fn
);
3080 EXPORT_SYMBOL(scsi_target_resume
);
3083 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
3084 * @sdev: device to block
3086 * Pause SCSI command processing on the specified device. Does not sleep.
3088 * Returns zero if successful or a negative error code upon failure.
3091 * This routine transitions the device to the SDEV_BLOCK state (which must be
3092 * a legal transition). When the device is in this state, command processing
3093 * is paused until the device leaves the SDEV_BLOCK state. See also
3094 * scsi_internal_device_unblock_nowait().
3096 int scsi_internal_device_block_nowait(struct scsi_device
*sdev
)
3098 struct request_queue
*q
= sdev
->request_queue
;
3099 unsigned long flags
;
3102 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
3104 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
3111 * The device has transitioned to SDEV_BLOCK. Stop the
3112 * block layer from calling the midlayer with this device's
3116 blk_mq_quiesce_queue_nowait(q
);
3118 spin_lock_irqsave(q
->queue_lock
, flags
);
3120 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3125 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait
);
3128 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
3129 * @sdev: device to block
3131 * Pause SCSI command processing on the specified device and wait until all
3132 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
3134 * Returns zero if successful or a negative error code upon failure.
3137 * This routine transitions the device to the SDEV_BLOCK state (which must be
3138 * a legal transition). When the device is in this state, command processing
3139 * is paused until the device leaves the SDEV_BLOCK state. See also
3140 * scsi_internal_device_unblock().
3142 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
3143 * scsi_internal_device_block() has blocked a SCSI device and also
3144 * remove the rport mutex lock and unlock calls from srp_queuecommand().
3146 static int scsi_internal_device_block(struct scsi_device
*sdev
)
3148 struct request_queue
*q
= sdev
->request_queue
;
3151 mutex_lock(&sdev
->state_mutex
);
3152 err
= scsi_internal_device_block_nowait(sdev
);
3155 blk_mq_quiesce_queue(q
);
3157 scsi_wait_for_queuecommand(sdev
);
3159 mutex_unlock(&sdev
->state_mutex
);
3164 void scsi_start_queue(struct scsi_device
*sdev
)
3166 struct request_queue
*q
= sdev
->request_queue
;
3167 unsigned long flags
;
3170 blk_mq_unquiesce_queue(q
);
3172 spin_lock_irqsave(q
->queue_lock
, flags
);
3174 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3179 * scsi_internal_device_unblock_nowait - resume a device after a block request
3180 * @sdev: device to resume
3181 * @new_state: state to set the device to after unblocking
3183 * Restart the device queue for a previously suspended SCSI device. Does not
3186 * Returns zero if successful or a negative error code upon failure.
3189 * This routine transitions the device to the SDEV_RUNNING state or to one of
3190 * the offline states (which must be a legal transition) allowing the midlayer
3191 * to goose the queue for this device.
3193 int scsi_internal_device_unblock_nowait(struct scsi_device
*sdev
,
3194 enum scsi_device_state new_state
)
3197 * Try to transition the scsi device to SDEV_RUNNING or one of the
3198 * offlined states and goose the device queue if successful.
3200 switch (sdev
->sdev_state
) {
3202 case SDEV_TRANSPORT_OFFLINE
:
3203 sdev
->sdev_state
= new_state
;
3205 case SDEV_CREATED_BLOCK
:
3206 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
3207 new_state
== SDEV_OFFLINE
)
3208 sdev
->sdev_state
= new_state
;
3210 sdev
->sdev_state
= SDEV_CREATED
;
3218 scsi_start_queue(sdev
);
3222 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait
);
3225 * scsi_internal_device_unblock - resume a device after a block request
3226 * @sdev: device to resume
3227 * @new_state: state to set the device to after unblocking
3229 * Restart the device queue for a previously suspended SCSI device. May sleep.
3231 * Returns zero if successful or a negative error code upon failure.
3234 * This routine transitions the device to the SDEV_RUNNING state or to one of
3235 * the offline states (which must be a legal transition) allowing the midlayer
3236 * to goose the queue for this device.
3238 static int scsi_internal_device_unblock(struct scsi_device
*sdev
,
3239 enum scsi_device_state new_state
)
3243 mutex_lock(&sdev
->state_mutex
);
3244 ret
= scsi_internal_device_unblock_nowait(sdev
, new_state
);
3245 mutex_unlock(&sdev
->state_mutex
);
3251 device_block(struct scsi_device
*sdev
, void *data
)
3253 scsi_internal_device_block(sdev
);
3257 target_block(struct device
*dev
, void *data
)
3259 if (scsi_is_target_device(dev
))
3260 starget_for_each_device(to_scsi_target(dev
), NULL
,
3266 scsi_target_block(struct device
*dev
)
3268 if (scsi_is_target_device(dev
))
3269 starget_for_each_device(to_scsi_target(dev
), NULL
,
3272 device_for_each_child(dev
, NULL
, target_block
);
3274 EXPORT_SYMBOL_GPL(scsi_target_block
);
3277 device_unblock(struct scsi_device
*sdev
, void *data
)
3279 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
3283 target_unblock(struct device
*dev
, void *data
)
3285 if (scsi_is_target_device(dev
))
3286 starget_for_each_device(to_scsi_target(dev
), data
,
3292 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
3294 if (scsi_is_target_device(dev
))
3295 starget_for_each_device(to_scsi_target(dev
), &new_state
,
3298 device_for_each_child(dev
, &new_state
, target_unblock
);
3300 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
3303 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3304 * @sgl: scatter-gather list
3305 * @sg_count: number of segments in sg
3306 * @offset: offset in bytes into sg, on return offset into the mapped area
3307 * @len: bytes to map, on return number of bytes mapped
3309 * Returns virtual address of the start of the mapped page
3311 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
3312 size_t *offset
, size_t *len
)
3315 size_t sg_len
= 0, len_complete
= 0;
3316 struct scatterlist
*sg
;
3319 WARN_ON(!irqs_disabled());
3321 for_each_sg(sgl
, sg
, sg_count
, i
) {
3322 len_complete
= sg_len
; /* Complete sg-entries */
3323 sg_len
+= sg
->length
;
3324 if (sg_len
> *offset
)
3328 if (unlikely(i
== sg_count
)) {
3329 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
3331 __func__
, sg_len
, *offset
, sg_count
);
3336 /* Offset starting from the beginning of first page in this sg-entry */
3337 *offset
= *offset
- len_complete
+ sg
->offset
;
3339 /* Assumption: contiguous pages can be accessed as "page + i" */
3340 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
3341 *offset
&= ~PAGE_MASK
;
3343 /* Bytes in this sg-entry from *offset to the end of the page */
3344 sg_len
= PAGE_SIZE
- *offset
;
3348 return kmap_atomic(page
);
3350 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3353 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3354 * @virt: virtual address to be unmapped
3356 void scsi_kunmap_atomic_sg(void *virt
)
3358 kunmap_atomic(virt
);
3360 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3362 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3364 atomic_inc(&sdev
->disk_events_disable_depth
);
3366 EXPORT_SYMBOL(sdev_disable_disk_events
);
3368 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3370 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3372 atomic_dec(&sdev
->disk_events_disable_depth
);
3374 EXPORT_SYMBOL(sdev_enable_disk_events
);
3377 * scsi_vpd_lun_id - return a unique device identification
3378 * @sdev: SCSI device
3379 * @id: buffer for the identification
3380 * @id_len: length of the buffer
3382 * Copies a unique device identification into @id based
3383 * on the information in the VPD page 0x83 of the device.
3384 * The string will be formatted as a SCSI name string.
3386 * Returns the length of the identification or error on failure.
3387 * If the identifier is longer than the supplied buffer the actual
3388 * identifier length is returned and the buffer is not zero-padded.
3390 int scsi_vpd_lun_id(struct scsi_device
*sdev
, char *id
, size_t id_len
)
3392 u8 cur_id_type
= 0xff;
3394 const unsigned char *d
, *cur_id_str
;
3395 const struct scsi_vpd
*vpd_pg83
;
3396 int id_size
= -EINVAL
;
3399 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3406 * Look for the correct descriptor.
3407 * Order of preference for lun descriptor:
3408 * - SCSI name string
3409 * - NAA IEEE Registered Extended
3410 * - EUI-64 based 16-byte
3411 * - EUI-64 based 12-byte
3412 * - NAA IEEE Registered
3413 * - NAA IEEE Extended
3415 * as longer descriptors reduce the likelyhood
3416 * of identification clashes.
3419 /* The id string must be at least 20 bytes + terminating NULL byte */
3425 memset(id
, 0, id_len
);
3426 d
= vpd_pg83
->data
+ 4;
3427 while (d
< vpd_pg83
->data
+ vpd_pg83
->len
) {
3428 /* Skip designators not referring to the LUN */
3429 if ((d
[1] & 0x30) != 0x00)
3432 switch (d
[1] & 0xf) {
3435 if (cur_id_size
> d
[3])
3437 /* Prefer anything */
3438 if (cur_id_type
> 0x01 && cur_id_type
!= 0xff)
3441 if (cur_id_size
+ 4 > id_len
)
3442 cur_id_size
= id_len
- 4;
3444 cur_id_type
= d
[1] & 0xf;
3445 id_size
= snprintf(id
, id_len
, "t10.%*pE",
3446 cur_id_size
, cur_id_str
);
3450 if (cur_id_size
> d
[3])
3452 /* Prefer NAA IEEE Registered Extended */
3453 if (cur_id_type
== 0x3 &&
3454 cur_id_size
== d
[3])
3458 cur_id_type
= d
[1] & 0xf;
3459 switch (cur_id_size
) {
3461 id_size
= snprintf(id
, id_len
,
3466 id_size
= snprintf(id
, id_len
,
3471 id_size
= snprintf(id
, id_len
,
3482 if (cur_id_size
> d
[3])
3486 cur_id_type
= d
[1] & 0xf;
3487 switch (cur_id_size
) {
3489 id_size
= snprintf(id
, id_len
,
3494 id_size
= snprintf(id
, id_len
,
3504 /* SCSI name string */
3505 if (cur_id_size
+ 4 > d
[3])
3507 /* Prefer others for truncated descriptor */
3508 if (cur_id_size
&& d
[3] > id_len
)
3510 cur_id_size
= id_size
= d
[3];
3512 cur_id_type
= d
[1] & 0xf;
3513 if (cur_id_size
>= id_len
)
3514 cur_id_size
= id_len
- 1;
3515 memcpy(id
, cur_id_str
, cur_id_size
);
3516 /* Decrease priority for truncated descriptor */
3517 if (cur_id_size
!= id_size
)
3530 EXPORT_SYMBOL(scsi_vpd_lun_id
);
3533 * scsi_vpd_tpg_id - return a target port group identifier
3534 * @sdev: SCSI device
3536 * Returns the Target Port Group identifier from the information
3537 * froom VPD page 0x83 of the device.
3539 * Returns the identifier or error on failure.
3541 int scsi_vpd_tpg_id(struct scsi_device
*sdev
, int *rel_id
)
3543 const unsigned char *d
;
3544 const struct scsi_vpd
*vpd_pg83
;
3545 int group_id
= -EAGAIN
, rel_port
= -1;
3548 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3554 d
= vpd_pg83
->data
+ 4;
3555 while (d
< vpd_pg83
->data
+ vpd_pg83
->len
) {
3556 switch (d
[1] & 0xf) {
3558 /* Relative target port */
3559 rel_port
= get_unaligned_be16(&d
[6]);
3562 /* Target port group */
3563 group_id
= get_unaligned_be16(&d
[6]);
3572 if (group_id
>= 0 && rel_id
&& rel_port
!= -1)
3577 EXPORT_SYMBOL(scsi_vpd_tpg_id
);