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/ratelimit.h>
25 #include <asm/unaligned.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
37 #include <trace/events/scsi.h>
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
44 * Size of integrity metadata is usually small, 1 inline sg should
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define SCSI_INLINE_PROT_SG_CNT 0
49 #define SCSI_INLINE_SG_CNT 0
51 #define SCSI_INLINE_PROT_SG_CNT 1
52 #define SCSI_INLINE_SG_CNT 2
55 static struct kmem_cache
*scsi_sdb_cache
;
56 static struct kmem_cache
*scsi_sense_cache
;
57 static struct kmem_cache
*scsi_sense_isadma_cache
;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex
);
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
);
62 static inline struct kmem_cache
*
63 scsi_select_sense_cache(bool unchecked_isa_dma
)
65 return unchecked_isa_dma
? scsi_sense_isadma_cache
: scsi_sense_cache
;
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma
,
69 unsigned char *sense_buffer
)
71 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma
),
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma
,
76 gfp_t gfp_mask
, int numa_node
)
78 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma
),
82 int scsi_init_sense_cache(struct Scsi_Host
*shost
)
84 struct kmem_cache
*cache
;
87 mutex_lock(&scsi_sense_cache_mutex
);
88 cache
= scsi_select_sense_cache(shost
->unchecked_isa_dma
);
92 if (shost
->unchecked_isa_dma
) {
93 scsi_sense_isadma_cache
=
94 kmem_cache_create("scsi_sense_cache(DMA)",
95 SCSI_SENSE_BUFFERSIZE
, 0,
96 SLAB_HWCACHE_ALIGN
| SLAB_CACHE_DMA
, NULL
);
97 if (!scsi_sense_isadma_cache
)
101 kmem_cache_create_usercopy("scsi_sense_cache",
102 SCSI_SENSE_BUFFERSIZE
, 0, SLAB_HWCACHE_ALIGN
,
103 0, SCSI_SENSE_BUFFERSIZE
, NULL
);
104 if (!scsi_sense_cache
)
108 mutex_unlock(&scsi_sense_cache_mutex
);
113 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114 * not change behaviour from the previous unplug mechanism, experimentation
115 * may prove this needs changing.
117 #define SCSI_QUEUE_DELAY 3
120 scsi_set_blocked(struct scsi_cmnd
*cmd
, int reason
)
122 struct Scsi_Host
*host
= cmd
->device
->host
;
123 struct scsi_device
*device
= cmd
->device
;
124 struct scsi_target
*starget
= scsi_target(device
);
127 * Set the appropriate busy bit for the device/host.
129 * If the host/device isn't busy, assume that something actually
130 * completed, and that we should be able to queue a command now.
132 * Note that the prior mid-layer assumption that any host could
133 * always queue at least one command is now broken. The mid-layer
134 * will implement a user specifiable stall (see
135 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136 * if a command is requeued with no other commands outstanding
137 * either for the device or for the host.
140 case SCSI_MLQUEUE_HOST_BUSY
:
141 atomic_set(&host
->host_blocked
, host
->max_host_blocked
);
143 case SCSI_MLQUEUE_DEVICE_BUSY
:
144 case SCSI_MLQUEUE_EH_RETRY
:
145 atomic_set(&device
->device_blocked
,
146 device
->max_device_blocked
);
148 case SCSI_MLQUEUE_TARGET_BUSY
:
149 atomic_set(&starget
->target_blocked
,
150 starget
->max_target_blocked
);
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd
*cmd
)
157 if (cmd
->request
->rq_flags
& RQF_DONTPREP
) {
158 cmd
->request
->rq_flags
&= ~RQF_DONTPREP
;
159 scsi_mq_uninit_cmd(cmd
);
163 blk_mq_requeue_request(cmd
->request
, true);
167 * __scsi_queue_insert - private queue insertion
168 * @cmd: The SCSI command being requeued
169 * @reason: The reason for the requeue
170 * @unbusy: Whether the queue should be unbusied
172 * This is a private queue insertion. The public interface
173 * scsi_queue_insert() always assumes the queue should be unbusied
174 * because it's always called before the completion. This function is
175 * for a requeue after completion, which should only occur in this
178 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, bool unbusy
)
180 struct scsi_device
*device
= cmd
->device
;
182 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
183 "Inserting command %p into mlqueue\n", cmd
));
185 scsi_set_blocked(cmd
, reason
);
188 * Decrement the counters, since these commands are no longer
189 * active on the host/device.
192 scsi_device_unbusy(device
);
195 * Requeue this command. It will go before all other commands
196 * that are already in the queue. Schedule requeue work under
197 * lock such that the kblockd_schedule_work() call happens
198 * before blk_cleanup_queue() finishes.
202 blk_mq_requeue_request(cmd
->request
, true);
206 * Function: scsi_queue_insert()
208 * Purpose: Insert a command in the midlevel queue.
210 * Arguments: cmd - command that we are adding to queue.
211 * reason - why we are inserting command to queue.
213 * Lock status: Assumed that lock is not held upon entry.
217 * Notes: We do this for one of two cases. Either the host is busy
218 * and it cannot accept any more commands for the time being,
219 * or the device returned QUEUE_FULL and can accept no more
221 * Notes: This could be called either from an interrupt context or a
222 * normal process context.
224 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
226 __scsi_queue_insert(cmd
, reason
, true);
231 * __scsi_execute - insert request and wait for the result
234 * @data_direction: data direction
235 * @buffer: data buffer
236 * @bufflen: len of buffer
237 * @sense: optional sense buffer
238 * @sshdr: optional decoded sense header
239 * @timeout: request timeout in seconds
240 * @retries: number of times to retry request
241 * @flags: flags for ->cmd_flags
242 * @rq_flags: flags for ->rq_flags
243 * @resid: optional residual length
245 * Returns the scsi_cmnd result field if a command was executed, or a negative
246 * Linux error code if we didn't get that far.
248 int __scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
249 int data_direction
, void *buffer
, unsigned bufflen
,
250 unsigned char *sense
, struct scsi_sense_hdr
*sshdr
,
251 int timeout
, int retries
, u64 flags
, req_flags_t rq_flags
,
255 struct scsi_request
*rq
;
256 int ret
= DRIVER_ERROR
<< 24;
258 req
= blk_get_request(sdev
->request_queue
,
259 data_direction
== DMA_TO_DEVICE
?
260 REQ_OP_SCSI_OUT
: REQ_OP_SCSI_IN
, BLK_MQ_REQ_PREEMPT
);
265 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
266 buffer
, bufflen
, GFP_NOIO
))
269 rq
->cmd_len
= COMMAND_SIZE(cmd
[0]);
270 memcpy(rq
->cmd
, cmd
, rq
->cmd_len
);
271 rq
->retries
= retries
;
272 req
->timeout
= timeout
;
273 req
->cmd_flags
|= flags
;
274 req
->rq_flags
|= rq_flags
| RQF_QUIET
;
277 * head injection *required* here otherwise quiesce won't work
279 blk_execute_rq(req
->q
, NULL
, req
, 1);
282 * Some devices (USB mass-storage in particular) may transfer
283 * garbage data together with a residue indicating that the data
284 * is invalid. Prevent the garbage from being misinterpreted
285 * and prevent security leaks by zeroing out the excess data.
287 if (unlikely(rq
->resid_len
> 0 && rq
->resid_len
<= bufflen
))
288 memset(buffer
+ (bufflen
- rq
->resid_len
), 0, rq
->resid_len
);
291 *resid
= rq
->resid_len
;
292 if (sense
&& rq
->sense_len
)
293 memcpy(sense
, rq
->sense
, SCSI_SENSE_BUFFERSIZE
);
295 scsi_normalize_sense(rq
->sense
, rq
->sense_len
, sshdr
);
298 blk_put_request(req
);
302 EXPORT_SYMBOL(__scsi_execute
);
305 * Function: scsi_init_cmd_errh()
307 * Purpose: Initialize cmd fields related to error handling.
309 * Arguments: cmd - command that is ready to be queued.
311 * Notes: This function has the job of initializing a number of
312 * fields related to error handling. Typically this will
313 * be called once for each command, as required.
315 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
317 scsi_set_resid(cmd
, 0);
318 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
319 if (cmd
->cmd_len
== 0)
320 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
324 * Decrement the host_busy counter and wake up the error handler if necessary.
325 * Avoid as follows that the error handler is not woken up if shost->host_busy
326 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327 * with an RCU read lock in this function to ensure that this function in its
328 * entirety either finishes before scsi_eh_scmd_add() increases the
329 * host_failed counter or that it notices the shost state change made by
330 * scsi_eh_scmd_add().
332 static void scsi_dec_host_busy(struct Scsi_Host
*shost
)
337 atomic_dec(&shost
->host_busy
);
338 if (unlikely(scsi_host_in_recovery(shost
))) {
339 spin_lock_irqsave(shost
->host_lock
, flags
);
340 if (shost
->host_failed
|| shost
->host_eh_scheduled
)
341 scsi_eh_wakeup(shost
);
342 spin_unlock_irqrestore(shost
->host_lock
, flags
);
347 void scsi_device_unbusy(struct scsi_device
*sdev
)
349 struct Scsi_Host
*shost
= sdev
->host
;
350 struct scsi_target
*starget
= scsi_target(sdev
);
352 scsi_dec_host_busy(shost
);
354 if (starget
->can_queue
> 0)
355 atomic_dec(&starget
->target_busy
);
357 atomic_dec(&sdev
->device_busy
);
360 static void scsi_kick_queue(struct request_queue
*q
)
362 blk_mq_run_hw_queues(q
, false);
366 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367 * and call blk_run_queue for all the scsi_devices on the target -
368 * including current_sdev first.
370 * Called with *no* scsi locks held.
372 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
374 struct Scsi_Host
*shost
= current_sdev
->host
;
375 struct scsi_device
*sdev
, *tmp
;
376 struct scsi_target
*starget
= scsi_target(current_sdev
);
379 spin_lock_irqsave(shost
->host_lock
, flags
);
380 starget
->starget_sdev_user
= NULL
;
381 spin_unlock_irqrestore(shost
->host_lock
, flags
);
384 * Call blk_run_queue for all LUNs on the target, starting with
385 * current_sdev. We race with others (to set starget_sdev_user),
386 * but in most cases, we will be first. Ideally, each LU on the
387 * target would get some limited time or requests on the target.
389 scsi_kick_queue(current_sdev
->request_queue
);
391 spin_lock_irqsave(shost
->host_lock
, flags
);
392 if (starget
->starget_sdev_user
)
394 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
395 same_target_siblings
) {
396 if (sdev
== current_sdev
)
398 if (scsi_device_get(sdev
))
401 spin_unlock_irqrestore(shost
->host_lock
, flags
);
402 scsi_kick_queue(sdev
->request_queue
);
403 spin_lock_irqsave(shost
->host_lock
, flags
);
405 scsi_device_put(sdev
);
408 spin_unlock_irqrestore(shost
->host_lock
, flags
);
411 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
413 if (atomic_read(&sdev
->device_busy
) >= sdev
->queue_depth
)
415 if (atomic_read(&sdev
->device_blocked
) > 0)
420 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
422 if (starget
->can_queue
> 0) {
423 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
425 if (atomic_read(&starget
->target_blocked
) > 0)
431 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
433 if (shost
->can_queue
> 0 &&
434 atomic_read(&shost
->host_busy
) >= shost
->can_queue
)
436 if (atomic_read(&shost
->host_blocked
) > 0)
438 if (shost
->host_self_blocked
)
443 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
445 LIST_HEAD(starved_list
);
446 struct scsi_device
*sdev
;
449 spin_lock_irqsave(shost
->host_lock
, flags
);
450 list_splice_init(&shost
->starved_list
, &starved_list
);
452 while (!list_empty(&starved_list
)) {
453 struct request_queue
*slq
;
456 * As long as shost is accepting commands and we have
457 * starved queues, call blk_run_queue. scsi_request_fn
458 * drops the queue_lock and can add us back to the
461 * host_lock protects the starved_list and starved_entry.
462 * scsi_request_fn must get the host_lock before checking
463 * or modifying starved_list or starved_entry.
465 if (scsi_host_is_busy(shost
))
468 sdev
= list_entry(starved_list
.next
,
469 struct scsi_device
, starved_entry
);
470 list_del_init(&sdev
->starved_entry
);
471 if (scsi_target_is_busy(scsi_target(sdev
))) {
472 list_move_tail(&sdev
->starved_entry
,
473 &shost
->starved_list
);
478 * Once we drop the host lock, a racing scsi_remove_device()
479 * call may remove the sdev from the starved list and destroy
480 * it and the queue. Mitigate by taking a reference to the
481 * queue and never touching the sdev again after we drop the
482 * host lock. Note: if __scsi_remove_device() invokes
483 * blk_cleanup_queue() before the queue is run from this
484 * function then blk_run_queue() will return immediately since
485 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
487 slq
= sdev
->request_queue
;
488 if (!blk_get_queue(slq
))
490 spin_unlock_irqrestore(shost
->host_lock
, flags
);
492 scsi_kick_queue(slq
);
495 spin_lock_irqsave(shost
->host_lock
, flags
);
497 /* put any unprocessed entries back */
498 list_splice(&starved_list
, &shost
->starved_list
);
499 spin_unlock_irqrestore(shost
->host_lock
, flags
);
503 * Function: scsi_run_queue()
505 * Purpose: Select a proper request queue to serve next
507 * Arguments: q - last request's queue
511 * Notes: The previous command was completely finished, start
512 * a new one if possible.
514 static void scsi_run_queue(struct request_queue
*q
)
516 struct scsi_device
*sdev
= q
->queuedata
;
518 if (scsi_target(sdev
)->single_lun
)
519 scsi_single_lun_run(sdev
);
520 if (!list_empty(&sdev
->host
->starved_list
))
521 scsi_starved_list_run(sdev
->host
);
523 blk_mq_run_hw_queues(q
, false);
526 void scsi_requeue_run_queue(struct work_struct
*work
)
528 struct scsi_device
*sdev
;
529 struct request_queue
*q
;
531 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
532 q
= sdev
->request_queue
;
536 void scsi_run_host_queues(struct Scsi_Host
*shost
)
538 struct scsi_device
*sdev
;
540 shost_for_each_device(sdev
, shost
)
541 scsi_run_queue(sdev
->request_queue
);
544 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
546 if (!blk_rq_is_passthrough(cmd
->request
)) {
547 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
549 if (drv
->uninit_command
)
550 drv
->uninit_command(cmd
);
554 static void scsi_mq_free_sgtables(struct scsi_cmnd
*cmd
)
556 if (cmd
->sdb
.table
.nents
)
557 sg_free_table_chained(&cmd
->sdb
.table
,
559 if (scsi_prot_sg_count(cmd
))
560 sg_free_table_chained(&cmd
->prot_sdb
->table
,
561 SCSI_INLINE_PROT_SG_CNT
);
564 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
566 scsi_mq_free_sgtables(cmd
);
567 scsi_uninit_cmd(cmd
);
568 scsi_del_cmd_from_list(cmd
);
571 /* Returns false when no more bytes to process, true if there are more */
572 static bool scsi_end_request(struct request
*req
, blk_status_t error
,
575 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
576 struct scsi_device
*sdev
= cmd
->device
;
577 struct request_queue
*q
= sdev
->request_queue
;
579 if (blk_update_request(req
, error
, bytes
))
582 if (blk_queue_add_random(q
))
583 add_disk_randomness(req
->rq_disk
);
585 if (!blk_rq_is_scsi(req
)) {
586 WARN_ON_ONCE(!(cmd
->flags
& SCMD_INITIALIZED
));
587 cmd
->flags
&= ~SCMD_INITIALIZED
;
591 * Calling rcu_barrier() is not necessary here because the
592 * SCSI error handler guarantees that the function called by
593 * call_rcu() has been called before scsi_end_request() is
596 destroy_rcu_head(&cmd
->rcu
);
599 * In the MQ case the command gets freed by __blk_mq_end_request,
600 * so we have to do all cleanup that depends on it earlier.
602 * We also can't kick the queues from irq context, so we
603 * will have to defer it to a workqueue.
605 scsi_mq_uninit_cmd(cmd
);
608 * queue is still alive, so grab the ref for preventing it
609 * from being cleaned up during running queue.
611 percpu_ref_get(&q
->q_usage_counter
);
613 __blk_mq_end_request(req
, error
);
615 if (scsi_target(sdev
)->single_lun
||
616 !list_empty(&sdev
->host
->starved_list
))
617 kblockd_schedule_work(&sdev
->requeue_work
);
619 blk_mq_run_hw_queues(q
, true);
621 percpu_ref_put(&q
->q_usage_counter
);
626 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
628 * @result: scsi error code
630 * Translate a SCSI result code into a blk_status_t value. May reset the host
631 * byte of @cmd->result.
633 static blk_status_t
scsi_result_to_blk_status(struct scsi_cmnd
*cmd
, int result
)
635 switch (host_byte(result
)) {
638 * Also check the other bytes than the status byte in result
639 * to handle the case when a SCSI LLD sets result to
640 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
642 if (scsi_status_is_good(result
) && (result
& ~0xff) == 0)
644 return BLK_STS_IOERR
;
645 case DID_TRANSPORT_FAILFAST
:
646 return BLK_STS_TRANSPORT
;
647 case DID_TARGET_FAILURE
:
648 set_host_byte(cmd
, DID_OK
);
649 return BLK_STS_TARGET
;
650 case DID_NEXUS_FAILURE
:
651 set_host_byte(cmd
, DID_OK
);
652 return BLK_STS_NEXUS
;
653 case DID_ALLOC_FAILURE
:
654 set_host_byte(cmd
, DID_OK
);
655 return BLK_STS_NOSPC
;
656 case DID_MEDIUM_ERROR
:
657 set_host_byte(cmd
, DID_OK
);
658 return BLK_STS_MEDIUM
;
660 return BLK_STS_IOERR
;
664 /* Helper for scsi_io_completion() when "reprep" action required. */
665 static void scsi_io_completion_reprep(struct scsi_cmnd
*cmd
,
666 struct request_queue
*q
)
668 /* A new command will be prepared and issued. */
669 scsi_mq_requeue_cmd(cmd
);
672 /* Helper for scsi_io_completion() when special action required. */
673 static void scsi_io_completion_action(struct scsi_cmnd
*cmd
, int result
)
675 struct request_queue
*q
= cmd
->device
->request_queue
;
676 struct request
*req
= cmd
->request
;
678 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
679 ACTION_DELAYED_RETRY
} action
;
680 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
681 struct scsi_sense_hdr sshdr
;
683 bool sense_current
= true; /* false implies "deferred sense" */
684 blk_status_t blk_stat
;
686 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
688 sense_current
= !scsi_sense_is_deferred(&sshdr
);
690 blk_stat
= scsi_result_to_blk_status(cmd
, result
);
692 if (host_byte(result
) == DID_RESET
) {
693 /* Third party bus reset or reset for error recovery
694 * reasons. Just retry the command and see what
697 action
= ACTION_RETRY
;
698 } else if (sense_valid
&& sense_current
) {
699 switch (sshdr
.sense_key
) {
701 if (cmd
->device
->removable
) {
702 /* Detected disc change. Set a bit
703 * and quietly refuse further access.
705 cmd
->device
->changed
= 1;
706 action
= ACTION_FAIL
;
708 /* Must have been a power glitch, or a
709 * bus reset. Could not have been a
710 * media change, so we just retry the
711 * command and see what happens.
713 action
= ACTION_RETRY
;
716 case ILLEGAL_REQUEST
:
717 /* If we had an ILLEGAL REQUEST returned, then
718 * we may have performed an unsupported
719 * command. The only thing this should be
720 * would be a ten byte read where only a six
721 * byte read was supported. Also, on a system
722 * where READ CAPACITY failed, we may have
723 * read past the end of the disk.
725 if ((cmd
->device
->use_10_for_rw
&&
726 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
727 (cmd
->cmnd
[0] == READ_10
||
728 cmd
->cmnd
[0] == WRITE_10
)) {
729 /* This will issue a new 6-byte command. */
730 cmd
->device
->use_10_for_rw
= 0;
731 action
= ACTION_REPREP
;
732 } else if (sshdr
.asc
== 0x10) /* DIX */ {
733 action
= ACTION_FAIL
;
734 blk_stat
= BLK_STS_PROTECTION
;
735 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
736 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
737 action
= ACTION_FAIL
;
738 blk_stat
= BLK_STS_TARGET
;
740 action
= ACTION_FAIL
;
742 case ABORTED_COMMAND
:
743 action
= ACTION_FAIL
;
744 if (sshdr
.asc
== 0x10) /* DIF */
745 blk_stat
= BLK_STS_PROTECTION
;
748 /* If the device is in the process of becoming
749 * ready, or has a temporary blockage, retry.
751 if (sshdr
.asc
== 0x04) {
752 switch (sshdr
.ascq
) {
753 case 0x01: /* becoming ready */
754 case 0x04: /* format in progress */
755 case 0x05: /* rebuild in progress */
756 case 0x06: /* recalculation in progress */
757 case 0x07: /* operation in progress */
758 case 0x08: /* Long write in progress */
759 case 0x09: /* self test in progress */
760 case 0x14: /* space allocation in progress */
761 case 0x1a: /* start stop unit in progress */
762 case 0x1b: /* sanitize in progress */
763 case 0x1d: /* configuration in progress */
764 case 0x24: /* depopulation in progress */
765 action
= ACTION_DELAYED_RETRY
;
768 action
= ACTION_FAIL
;
772 action
= ACTION_FAIL
;
774 case VOLUME_OVERFLOW
:
775 /* See SSC3rXX or current. */
776 action
= ACTION_FAIL
;
779 action
= ACTION_FAIL
;
783 action
= ACTION_FAIL
;
785 if (action
!= ACTION_FAIL
&&
786 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
787 action
= ACTION_FAIL
;
791 /* Give up and fail the remainder of the request */
792 if (!(req
->rq_flags
& RQF_QUIET
)) {
793 static DEFINE_RATELIMIT_STATE(_rs
,
794 DEFAULT_RATELIMIT_INTERVAL
,
795 DEFAULT_RATELIMIT_BURST
);
797 if (unlikely(scsi_logging_level
))
799 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT
,
800 SCSI_LOG_MLCOMPLETE_BITS
);
803 * if logging is enabled the failure will be printed
804 * in scsi_log_completion(), so avoid duplicate messages
806 if (!level
&& __ratelimit(&_rs
)) {
807 scsi_print_result(cmd
, NULL
, FAILED
);
808 if (driver_byte(result
) == DRIVER_SENSE
)
809 scsi_print_sense(cmd
);
810 scsi_print_command(cmd
);
813 if (!scsi_end_request(req
, blk_stat
, blk_rq_err_bytes(req
)))
817 scsi_io_completion_reprep(cmd
, q
);
820 /* Retry the same command immediately */
821 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, false);
823 case ACTION_DELAYED_RETRY
:
824 /* Retry the same command after a delay */
825 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, false);
831 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
832 * new result that may suppress further error checking. Also modifies
833 * *blk_statp in some cases.
835 static int scsi_io_completion_nz_result(struct scsi_cmnd
*cmd
, int result
,
836 blk_status_t
*blk_statp
)
839 bool sense_current
= true; /* false implies "deferred sense" */
840 struct request
*req
= cmd
->request
;
841 struct scsi_sense_hdr sshdr
;
843 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
845 sense_current
= !scsi_sense_is_deferred(&sshdr
);
847 if (blk_rq_is_passthrough(req
)) {
850 * SG_IO wants current and deferred errors
852 scsi_req(req
)->sense_len
=
853 min(8 + cmd
->sense_buffer
[7],
854 SCSI_SENSE_BUFFERSIZE
);
857 *blk_statp
= scsi_result_to_blk_status(cmd
, result
);
858 } else if (blk_rq_bytes(req
) == 0 && sense_current
) {
860 * Flush commands do not transfers any data, and thus cannot use
861 * good_bytes != blk_rq_bytes(req) as the signal for an error.
862 * This sets *blk_statp explicitly for the problem case.
864 *blk_statp
= scsi_result_to_blk_status(cmd
, result
);
867 * Recovered errors need reporting, but they're always treated as
868 * success, so fiddle the result code here. For passthrough requests
869 * we already took a copy of the original into sreq->result which
870 * is what gets returned to the user
872 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
873 bool do_print
= true;
875 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
876 * skip print since caller wants ATA registers. Only occurs
877 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
879 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
881 else if (req
->rq_flags
& RQF_QUIET
)
884 scsi_print_sense(cmd
);
886 /* for passthrough, *blk_statp may be set */
887 *blk_statp
= BLK_STS_OK
;
890 * Another corner case: the SCSI status byte is non-zero but 'good'.
891 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
892 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
893 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
894 * intermediate statuses (both obsolete in SAM-4) as good.
896 if (status_byte(result
) && scsi_status_is_good(result
)) {
898 *blk_statp
= BLK_STS_OK
;
904 * Function: scsi_io_completion()
906 * Purpose: Completion processing for block device I/O requests.
908 * Arguments: cmd - command that is finished.
910 * Lock status: Assumed that no lock is held upon entry.
914 * Notes: We will finish off the specified number of sectors. If we
915 * are done, the command block will be released and the queue
916 * function will be goosed. If we are not done then we have to
917 * figure out what to do next:
919 * a) We can call scsi_requeue_command(). The request
920 * will be unprepared and put back on the queue. Then
921 * a new command will be created for it. This should
922 * be used if we made forward progress, or if we want
923 * to switch from READ(10) to READ(6) for example.
925 * b) We can call __scsi_queue_insert(). The request will
926 * be put back on the queue and retried using the same
927 * command as before, possibly after a delay.
929 * c) We can call scsi_end_request() with blk_stat other than
930 * BLK_STS_OK, to fail the remainder of the request.
932 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
934 int result
= cmd
->result
;
935 struct request_queue
*q
= cmd
->device
->request_queue
;
936 struct request
*req
= cmd
->request
;
937 blk_status_t blk_stat
= BLK_STS_OK
;
939 if (unlikely(result
)) /* a nz result may or may not be an error */
940 result
= scsi_io_completion_nz_result(cmd
, result
, &blk_stat
);
942 if (unlikely(blk_rq_is_passthrough(req
))) {
944 * scsi_result_to_blk_status may have reset the host_byte
946 scsi_req(req
)->result
= cmd
->result
;
950 * Next deal with any sectors which we were able to correctly
953 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
954 "%u sectors total, %d bytes done.\n",
955 blk_rq_sectors(req
), good_bytes
));
958 * Next deal with any sectors which we were able to correctly
959 * handle. Failed, zero length commands always need to drop down
960 * to retry code. Fast path should return in this block.
962 if (likely(blk_rq_bytes(req
) > 0 || blk_stat
== BLK_STS_OK
)) {
963 if (likely(!scsi_end_request(req
, blk_stat
, good_bytes
)))
964 return; /* no bytes remaining */
967 /* Kill remainder if no retries. */
968 if (unlikely(blk_stat
&& scsi_noretry_cmd(cmd
))) {
969 if (scsi_end_request(req
, blk_stat
, blk_rq_bytes(req
)))
971 "Bytes remaining after failed, no-retry command");
976 * If there had been no error, but we have leftover bytes in the
977 * requeues just queue the command up again.
979 if (likely(result
== 0))
980 scsi_io_completion_reprep(cmd
, q
);
982 scsi_io_completion_action(cmd
, result
);
985 static blk_status_t
scsi_init_sgtable(struct request
*req
,
986 struct scsi_data_buffer
*sdb
)
991 * If sg table allocation fails, requeue request later.
993 if (unlikely(sg_alloc_table_chained(&sdb
->table
,
994 blk_rq_nr_phys_segments(req
), sdb
->table
.sgl
,
995 SCSI_INLINE_SG_CNT
)))
996 return BLK_STS_RESOURCE
;
999 * Next, walk the list, and fill in the addresses and sizes of
1002 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1003 BUG_ON(count
> sdb
->table
.nents
);
1004 sdb
->table
.nents
= count
;
1005 sdb
->length
= blk_rq_payload_bytes(req
);
1010 * Function: scsi_init_io()
1012 * Purpose: SCSI I/O initialize function.
1014 * Arguments: cmd - Command descriptor we wish to initialize
1016 * Returns: BLK_STS_OK on success
1017 * BLK_STS_RESOURCE if the failure is retryable
1018 * BLK_STS_IOERR if the failure is fatal
1020 blk_status_t
scsi_init_io(struct scsi_cmnd
*cmd
)
1022 struct request
*rq
= cmd
->request
;
1025 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq
)))
1026 return BLK_STS_IOERR
;
1028 ret
= scsi_init_sgtable(rq
, &cmd
->sdb
);
1032 if (blk_integrity_rq(rq
)) {
1033 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1036 if (WARN_ON_ONCE(!prot_sdb
)) {
1038 * This can happen if someone (e.g. multipath)
1039 * queues a command to a device on an adapter
1040 * that does not support DIX.
1042 ret
= BLK_STS_IOERR
;
1043 goto out_free_sgtables
;
1046 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1048 if (sg_alloc_table_chained(&prot_sdb
->table
, ivecs
,
1049 prot_sdb
->table
.sgl
,
1050 SCSI_INLINE_PROT_SG_CNT
)) {
1051 ret
= BLK_STS_RESOURCE
;
1052 goto out_free_sgtables
;
1055 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1056 prot_sdb
->table
.sgl
);
1057 BUG_ON(count
> ivecs
);
1058 BUG_ON(count
> queue_max_integrity_segments(rq
->q
));
1060 cmd
->prot_sdb
= prot_sdb
;
1061 cmd
->prot_sdb
->table
.nents
= count
;
1066 scsi_mq_free_sgtables(cmd
);
1069 EXPORT_SYMBOL(scsi_init_io
);
1072 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073 * @rq: Request associated with the SCSI command to be initialized.
1075 * This function initializes the members of struct scsi_cmnd that must be
1076 * initialized before request processing starts and that won't be
1077 * reinitialized if a SCSI command is requeued.
1079 * Called from inside blk_get_request() for pass-through requests and from
1080 * inside scsi_init_command() for filesystem requests.
1082 static void scsi_initialize_rq(struct request
*rq
)
1084 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1086 scsi_req_init(&cmd
->req
);
1087 init_rcu_head(&cmd
->rcu
);
1088 cmd
->jiffies_at_alloc
= jiffies
;
1092 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1093 void scsi_add_cmd_to_list(struct scsi_cmnd
*cmd
)
1095 struct scsi_device
*sdev
= cmd
->device
;
1096 struct Scsi_Host
*shost
= sdev
->host
;
1097 unsigned long flags
;
1099 if (shost
->use_cmd_list
) {
1100 spin_lock_irqsave(&sdev
->list_lock
, flags
);
1101 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1102 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
1106 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1107 void scsi_del_cmd_from_list(struct scsi_cmnd
*cmd
)
1109 struct scsi_device
*sdev
= cmd
->device
;
1110 struct Scsi_Host
*shost
= sdev
->host
;
1111 unsigned long flags
;
1113 if (shost
->use_cmd_list
) {
1114 spin_lock_irqsave(&sdev
->list_lock
, flags
);
1115 BUG_ON(list_empty(&cmd
->list
));
1116 list_del_init(&cmd
->list
);
1117 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
1121 /* Called after a request has been started. */
1122 void scsi_init_command(struct scsi_device
*dev
, struct scsi_cmnd
*cmd
)
1124 void *buf
= cmd
->sense_buffer
;
1125 void *prot
= cmd
->prot_sdb
;
1126 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1127 unsigned int flags
= cmd
->flags
& SCMD_PRESERVED_FLAGS
;
1128 unsigned long jiffies_at_alloc
;
1131 if (!blk_rq_is_scsi(rq
) && !(flags
& SCMD_INITIALIZED
)) {
1132 flags
|= SCMD_INITIALIZED
;
1133 scsi_initialize_rq(rq
);
1136 jiffies_at_alloc
= cmd
->jiffies_at_alloc
;
1137 retries
= cmd
->retries
;
1138 /* zero out the cmd, except for the embedded scsi_request */
1139 memset((char *)cmd
+ sizeof(cmd
->req
), 0,
1140 sizeof(*cmd
) - sizeof(cmd
->req
) + dev
->host
->hostt
->cmd_size
);
1143 cmd
->sense_buffer
= buf
;
1144 cmd
->prot_sdb
= prot
;
1146 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1147 cmd
->jiffies_at_alloc
= jiffies_at_alloc
;
1148 cmd
->retries
= retries
;
1150 scsi_add_cmd_to_list(cmd
);
1153 static blk_status_t
scsi_setup_scsi_cmnd(struct scsi_device
*sdev
,
1154 struct request
*req
)
1156 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1159 * Passthrough requests may transfer data, in which case they must
1160 * a bio attached to them. Or they might contain a SCSI command
1161 * that does not transfer data, in which case they may optionally
1162 * submit a request without an attached bio.
1165 blk_status_t ret
= scsi_init_io(cmd
);
1166 if (unlikely(ret
!= BLK_STS_OK
))
1169 BUG_ON(blk_rq_bytes(req
));
1171 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1174 cmd
->cmd_len
= scsi_req(req
)->cmd_len
;
1175 cmd
->cmnd
= scsi_req(req
)->cmd
;
1176 cmd
->transfersize
= blk_rq_bytes(req
);
1177 cmd
->allowed
= scsi_req(req
)->retries
;
1182 * Setup a normal block command. These are simple request from filesystems
1183 * that still need to be translated to SCSI CDBs from the ULD.
1185 static blk_status_t
scsi_setup_fs_cmnd(struct scsi_device
*sdev
,
1186 struct request
*req
)
1188 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1190 if (unlikely(sdev
->handler
&& sdev
->handler
->prep_fn
)) {
1191 blk_status_t ret
= sdev
->handler
->prep_fn(sdev
, req
);
1192 if (ret
!= BLK_STS_OK
)
1196 cmd
->cmnd
= scsi_req(req
)->cmd
= scsi_req(req
)->__cmd
;
1197 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1198 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1201 static blk_status_t
scsi_setup_cmnd(struct scsi_device
*sdev
,
1202 struct request
*req
)
1204 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1206 if (!blk_rq_bytes(req
))
1207 cmd
->sc_data_direction
= DMA_NONE
;
1208 else if (rq_data_dir(req
) == WRITE
)
1209 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1211 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1213 if (blk_rq_is_scsi(req
))
1214 return scsi_setup_scsi_cmnd(sdev
, req
);
1216 return scsi_setup_fs_cmnd(sdev
, req
);
1220 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1222 switch (sdev
->sdev_state
) {
1224 case SDEV_TRANSPORT_OFFLINE
:
1226 * If the device is offline we refuse to process any
1227 * commands. The device must be brought online
1228 * before trying any recovery commands.
1230 sdev_printk(KERN_ERR
, sdev
,
1231 "rejecting I/O to offline device\n");
1232 return BLK_STS_IOERR
;
1235 * If the device is fully deleted, we refuse to
1236 * process any commands as well.
1238 sdev_printk(KERN_ERR
, sdev
,
1239 "rejecting I/O to dead device\n");
1240 return BLK_STS_IOERR
;
1242 case SDEV_CREATED_BLOCK
:
1243 return BLK_STS_RESOURCE
;
1246 * If the devices is blocked we defer normal commands.
1248 if (req
&& !(req
->rq_flags
& RQF_PREEMPT
))
1249 return BLK_STS_RESOURCE
;
1253 * For any other not fully online state we only allow
1254 * special commands. In particular any user initiated
1255 * command is not allowed.
1257 if (req
&& !(req
->rq_flags
& RQF_PREEMPT
))
1258 return BLK_STS_IOERR
;
1264 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1267 * Called with the queue_lock held.
1269 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1270 struct scsi_device
*sdev
)
1274 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1275 if (atomic_read(&sdev
->device_blocked
)) {
1280 * unblock after device_blocked iterates to zero
1282 if (atomic_dec_return(&sdev
->device_blocked
) > 0)
1284 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1285 "unblocking device at zero depth\n"));
1288 if (busy
>= sdev
->queue_depth
)
1293 atomic_dec(&sdev
->device_busy
);
1298 * scsi_target_queue_ready: checks if there we can send commands to target
1299 * @sdev: scsi device on starget to check.
1301 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1302 struct scsi_device
*sdev
)
1304 struct scsi_target
*starget
= scsi_target(sdev
);
1307 if (starget
->single_lun
) {
1308 spin_lock_irq(shost
->host_lock
);
1309 if (starget
->starget_sdev_user
&&
1310 starget
->starget_sdev_user
!= sdev
) {
1311 spin_unlock_irq(shost
->host_lock
);
1314 starget
->starget_sdev_user
= sdev
;
1315 spin_unlock_irq(shost
->host_lock
);
1318 if (starget
->can_queue
<= 0)
1321 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1322 if (atomic_read(&starget
->target_blocked
) > 0) {
1327 * unblock after target_blocked iterates to zero
1329 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1332 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1333 "unblocking target at zero depth\n"));
1336 if (busy
>= starget
->can_queue
)
1342 spin_lock_irq(shost
->host_lock
);
1343 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1344 spin_unlock_irq(shost
->host_lock
);
1346 if (starget
->can_queue
> 0)
1347 atomic_dec(&starget
->target_busy
);
1352 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1353 * return 0. We must end up running the queue again whenever 0 is
1354 * returned, else IO can hang.
1356 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1357 struct Scsi_Host
*shost
,
1358 struct scsi_device
*sdev
)
1362 if (scsi_host_in_recovery(shost
))
1365 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1366 if (atomic_read(&shost
->host_blocked
) > 0) {
1371 * unblock after host_blocked iterates to zero
1373 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1377 shost_printk(KERN_INFO
, shost
,
1378 "unblocking host at zero depth\n"));
1381 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1383 if (shost
->host_self_blocked
)
1386 /* We're OK to process the command, so we can't be starved */
1387 if (!list_empty(&sdev
->starved_entry
)) {
1388 spin_lock_irq(shost
->host_lock
);
1389 if (!list_empty(&sdev
->starved_entry
))
1390 list_del_init(&sdev
->starved_entry
);
1391 spin_unlock_irq(shost
->host_lock
);
1397 spin_lock_irq(shost
->host_lock
);
1398 if (list_empty(&sdev
->starved_entry
))
1399 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1400 spin_unlock_irq(shost
->host_lock
);
1402 scsi_dec_host_busy(shost
);
1407 * Busy state exporting function for request stacking drivers.
1409 * For efficiency, no lock is taken to check the busy state of
1410 * shost/starget/sdev, since the returned value is not guaranteed and
1411 * may be changed after request stacking drivers call the function,
1412 * regardless of taking lock or not.
1414 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1415 * needs to return 'not busy'. Otherwise, request stacking drivers
1416 * may hold requests forever.
1418 static bool scsi_mq_lld_busy(struct request_queue
*q
)
1420 struct scsi_device
*sdev
= q
->queuedata
;
1421 struct Scsi_Host
*shost
;
1423 if (blk_queue_dying(q
))
1429 * Ignore host/starget busy state.
1430 * Since block layer does not have a concept of fairness across
1431 * multiple queues, congestion of host/starget needs to be handled
1434 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1440 static void scsi_softirq_done(struct request
*rq
)
1442 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1443 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1446 INIT_LIST_HEAD(&cmd
->eh_entry
);
1448 atomic_inc(&cmd
->device
->iodone_cnt
);
1450 atomic_inc(&cmd
->device
->ioerr_cnt
);
1452 disposition
= scsi_decide_disposition(cmd
);
1453 if (disposition
!= SUCCESS
&&
1454 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1455 scmd_printk(KERN_ERR
, cmd
,
1456 "timing out command, waited %lus\n",
1458 disposition
= SUCCESS
;
1461 scsi_log_completion(cmd
, disposition
);
1463 switch (disposition
) {
1465 scsi_finish_command(cmd
);
1468 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1470 case ADD_TO_MLQUEUE
:
1471 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1474 scsi_eh_scmd_add(cmd
);
1480 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1481 * @cmd: command block we are dispatching.
1483 * Return: nonzero return request was rejected and device's queue needs to be
1486 static int scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
1488 struct Scsi_Host
*host
= cmd
->device
->host
;
1491 atomic_inc(&cmd
->device
->iorequest_cnt
);
1493 /* check if the device is still usable */
1494 if (unlikely(cmd
->device
->sdev_state
== SDEV_DEL
)) {
1495 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1496 * returns an immediate error upwards, and signals
1497 * that the device is no longer present */
1498 cmd
->result
= DID_NO_CONNECT
<< 16;
1502 /* Check to see if the scsi lld made this device blocked. */
1503 if (unlikely(scsi_device_blocked(cmd
->device
))) {
1505 * in blocked state, the command is just put back on
1506 * the device queue. The suspend state has already
1507 * blocked the queue so future requests should not
1508 * occur until the device transitions out of the
1511 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1512 "queuecommand : device blocked\n"));
1513 return SCSI_MLQUEUE_DEVICE_BUSY
;
1516 /* Store the LUN value in cmnd, if needed. */
1517 if (cmd
->device
->lun_in_cdb
)
1518 cmd
->cmnd
[1] = (cmd
->cmnd
[1] & 0x1f) |
1519 (cmd
->device
->lun
<< 5 & 0xe0);
1524 * Before we queue this command, check if the command
1525 * length exceeds what the host adapter can handle.
1527 if (cmd
->cmd_len
> cmd
->device
->host
->max_cmd_len
) {
1528 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1529 "queuecommand : command too long. "
1530 "cdb_size=%d host->max_cmd_len=%d\n",
1531 cmd
->cmd_len
, cmd
->device
->host
->max_cmd_len
));
1532 cmd
->result
= (DID_ABORT
<< 16);
1536 if (unlikely(host
->shost_state
== SHOST_DEL
)) {
1537 cmd
->result
= (DID_NO_CONNECT
<< 16);
1542 trace_scsi_dispatch_cmd_start(cmd
);
1543 rtn
= host
->hostt
->queuecommand(host
, cmd
);
1545 trace_scsi_dispatch_cmd_error(cmd
, rtn
);
1546 if (rtn
!= SCSI_MLQUEUE_DEVICE_BUSY
&&
1547 rtn
!= SCSI_MLQUEUE_TARGET_BUSY
)
1548 rtn
= SCSI_MLQUEUE_HOST_BUSY
;
1550 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1551 "queuecommand : request rejected\n"));
1556 cmd
->scsi_done(cmd
);
1560 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1561 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host
*shost
)
1563 return min_t(unsigned int, shost
->sg_tablesize
, SCSI_INLINE_SG_CNT
) *
1564 sizeof(struct scatterlist
);
1567 static blk_status_t
scsi_mq_prep_fn(struct request
*req
)
1569 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1570 struct scsi_device
*sdev
= req
->q
->queuedata
;
1571 struct Scsi_Host
*shost
= sdev
->host
;
1572 struct scatterlist
*sg
;
1574 scsi_init_command(sdev
, cmd
);
1577 cmd
->tag
= req
->tag
;
1578 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1580 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1581 cmd
->sdb
.table
.sgl
= sg
;
1583 if (scsi_host_get_prot(shost
)) {
1584 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1586 cmd
->prot_sdb
->table
.sgl
=
1587 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1590 blk_mq_start_request(req
);
1592 return scsi_setup_cmnd(sdev
, req
);
1595 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1597 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE
, &cmd
->state
)))
1599 trace_scsi_dispatch_cmd_done(cmd
);
1602 * If the block layer didn't complete the request due to a timeout
1603 * injection, scsi must clear its internal completed state so that the
1604 * timeout handler will see it needs to escalate its own error
1607 if (unlikely(!blk_mq_complete_request(cmd
->request
)))
1608 clear_bit(SCMD_STATE_COMPLETE
, &cmd
->state
);
1611 static void scsi_mq_put_budget(struct blk_mq_hw_ctx
*hctx
)
1613 struct request_queue
*q
= hctx
->queue
;
1614 struct scsi_device
*sdev
= q
->queuedata
;
1616 atomic_dec(&sdev
->device_busy
);
1619 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx
*hctx
)
1621 struct request_queue
*q
= hctx
->queue
;
1622 struct scsi_device
*sdev
= q
->queuedata
;
1624 if (scsi_dev_queue_ready(q
, sdev
))
1627 if (atomic_read(&sdev
->device_busy
) == 0 && !scsi_device_blocked(sdev
))
1628 blk_mq_delay_run_hw_queue(hctx
, SCSI_QUEUE_DELAY
);
1632 static blk_status_t
scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1633 const struct blk_mq_queue_data
*bd
)
1635 struct request
*req
= bd
->rq
;
1636 struct request_queue
*q
= req
->q
;
1637 struct scsi_device
*sdev
= q
->queuedata
;
1638 struct Scsi_Host
*shost
= sdev
->host
;
1639 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1644 * If the device is not in running state we will reject some or all
1647 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1648 ret
= scsi_prep_state_check(sdev
, req
);
1649 if (ret
!= BLK_STS_OK
)
1650 goto out_put_budget
;
1653 ret
= BLK_STS_RESOURCE
;
1654 if (!scsi_target_queue_ready(shost
, sdev
))
1655 goto out_put_budget
;
1656 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1657 goto out_dec_target_busy
;
1659 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
1660 ret
= scsi_mq_prep_fn(req
);
1661 if (ret
!= BLK_STS_OK
)
1662 goto out_dec_host_busy
;
1663 req
->rq_flags
|= RQF_DONTPREP
;
1665 clear_bit(SCMD_STATE_COMPLETE
, &cmd
->state
);
1666 blk_mq_start_request(req
);
1669 if (sdev
->simple_tags
)
1670 cmd
->flags
|= SCMD_TAGGED
;
1672 cmd
->flags
&= ~SCMD_TAGGED
;
1674 scsi_init_cmd_errh(cmd
);
1675 cmd
->scsi_done
= scsi_mq_done
;
1677 reason
= scsi_dispatch_cmd(cmd
);
1679 scsi_set_blocked(cmd
, reason
);
1680 ret
= BLK_STS_RESOURCE
;
1681 goto out_dec_host_busy
;
1687 scsi_dec_host_busy(shost
);
1688 out_dec_target_busy
:
1689 if (scsi_target(sdev
)->can_queue
> 0)
1690 atomic_dec(&scsi_target(sdev
)->target_busy
);
1692 scsi_mq_put_budget(hctx
);
1696 case BLK_STS_RESOURCE
:
1697 if (atomic_read(&sdev
->device_busy
) ||
1698 scsi_device_blocked(sdev
))
1699 ret
= BLK_STS_DEV_RESOURCE
;
1702 if (unlikely(!scsi_device_online(sdev
)))
1703 scsi_req(req
)->result
= DID_NO_CONNECT
<< 16;
1705 scsi_req(req
)->result
= DID_ERROR
<< 16;
1707 * Make sure to release all allocated resources when
1708 * we hit an error, as we will never see this command
1711 if (req
->rq_flags
& RQF_DONTPREP
)
1712 scsi_mq_uninit_cmd(cmd
);
1718 static enum blk_eh_timer_return
scsi_timeout(struct request
*req
,
1722 return BLK_EH_RESET_TIMER
;
1723 return scsi_times_out(req
);
1726 static int scsi_mq_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1727 unsigned int hctx_idx
, unsigned int numa_node
)
1729 struct Scsi_Host
*shost
= set
->driver_data
;
1730 const bool unchecked_isa_dma
= shost
->unchecked_isa_dma
;
1731 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1732 struct scatterlist
*sg
;
1734 if (unchecked_isa_dma
)
1735 cmd
->flags
|= SCMD_UNCHECKED_ISA_DMA
;
1736 cmd
->sense_buffer
= scsi_alloc_sense_buffer(unchecked_isa_dma
,
1737 GFP_KERNEL
, numa_node
);
1738 if (!cmd
->sense_buffer
)
1740 cmd
->req
.sense
= cmd
->sense_buffer
;
1742 if (scsi_host_get_prot(shost
)) {
1743 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) +
1744 shost
->hostt
->cmd_size
;
1745 cmd
->prot_sdb
= (void *)sg
+ scsi_mq_inline_sgl_size(shost
);
1751 static void scsi_mq_exit_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1752 unsigned int hctx_idx
)
1754 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1756 scsi_free_sense_buffer(cmd
->flags
& SCMD_UNCHECKED_ISA_DMA
,
1760 static int scsi_map_queues(struct blk_mq_tag_set
*set
)
1762 struct Scsi_Host
*shost
= container_of(set
, struct Scsi_Host
, tag_set
);
1764 if (shost
->hostt
->map_queues
)
1765 return shost
->hostt
->map_queues(shost
);
1766 return blk_mq_map_queues(&set
->map
[HCTX_TYPE_DEFAULT
]);
1769 void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
1771 struct device
*dev
= shost
->dma_dev
;
1774 * this limit is imposed by hardware restrictions
1776 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
1779 if (scsi_host_prot_dma(shost
)) {
1780 shost
->sg_prot_tablesize
=
1781 min_not_zero(shost
->sg_prot_tablesize
,
1782 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
1783 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
1784 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
1787 if (dev
->dma_mask
) {
1788 shost
->max_sectors
= min_t(unsigned int, shost
->max_sectors
,
1789 dma_max_mapping_size(dev
) >> SECTOR_SHIFT
);
1791 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
1792 if (shost
->unchecked_isa_dma
)
1793 blk_queue_bounce_limit(q
, BLK_BOUNCE_ISA
);
1794 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1795 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1797 blk_queue_max_segment_size(q
, shost
->max_segment_size
);
1798 blk_queue_virt_boundary(q
, shost
->virt_boundary_mask
);
1799 dma_set_max_seg_size(dev
, queue_max_segment_size(q
));
1802 * Set a reasonable default alignment: The larger of 32-byte (dword),
1803 * which is a common minimum for HBAs, and the minimum DMA alignment,
1804 * which is set by the platform.
1806 * Devices that require a bigger alignment can increase it later.
1808 blk_queue_dma_alignment(q
, max(4, dma_get_cache_alignment()) - 1);
1810 EXPORT_SYMBOL_GPL(__scsi_init_queue
);
1812 static const struct blk_mq_ops scsi_mq_ops
= {
1813 .get_budget
= scsi_mq_get_budget
,
1814 .put_budget
= scsi_mq_put_budget
,
1815 .queue_rq
= scsi_queue_rq
,
1816 .complete
= scsi_softirq_done
,
1817 .timeout
= scsi_timeout
,
1818 #ifdef CONFIG_BLK_DEBUG_FS
1819 .show_rq
= scsi_show_rq
,
1821 .init_request
= scsi_mq_init_request
,
1822 .exit_request
= scsi_mq_exit_request
,
1823 .initialize_rq_fn
= scsi_initialize_rq
,
1824 .busy
= scsi_mq_lld_busy
,
1825 .map_queues
= scsi_map_queues
,
1828 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
1830 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
1831 if (IS_ERR(sdev
->request_queue
))
1834 sdev
->request_queue
->queuedata
= sdev
;
1835 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
1836 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH
, sdev
->request_queue
);
1837 return sdev
->request_queue
;
1840 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
1842 unsigned int cmd_size
, sgl_size
;
1844 sgl_size
= scsi_mq_inline_sgl_size(shost
);
1845 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
1846 if (scsi_host_get_prot(shost
))
1847 cmd_size
+= sizeof(struct scsi_data_buffer
) +
1848 sizeof(struct scatterlist
) * SCSI_INLINE_PROT_SG_CNT
;
1850 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
1851 shost
->tag_set
.ops
= &scsi_mq_ops
;
1852 shost
->tag_set
.nr_hw_queues
= shost
->nr_hw_queues
? : 1;
1853 shost
->tag_set
.queue_depth
= shost
->can_queue
;
1854 shost
->tag_set
.cmd_size
= cmd_size
;
1855 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
1856 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
;
1857 shost
->tag_set
.flags
|=
1858 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost
->hostt
->tag_alloc_policy
);
1859 shost
->tag_set
.driver_data
= shost
;
1861 return blk_mq_alloc_tag_set(&shost
->tag_set
);
1864 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
1866 blk_mq_free_tag_set(&shost
->tag_set
);
1870 * scsi_device_from_queue - return sdev associated with a request_queue
1871 * @q: The request queue to return the sdev from
1873 * Return the sdev associated with a request queue or NULL if the
1874 * request_queue does not reference a SCSI device.
1876 struct scsi_device
*scsi_device_from_queue(struct request_queue
*q
)
1878 struct scsi_device
*sdev
= NULL
;
1880 if (q
->mq_ops
== &scsi_mq_ops
)
1881 sdev
= q
->queuedata
;
1882 if (!sdev
|| !get_device(&sdev
->sdev_gendev
))
1887 EXPORT_SYMBOL_GPL(scsi_device_from_queue
);
1890 * Function: scsi_block_requests()
1892 * Purpose: Utility function used by low-level drivers to prevent further
1893 * commands from being queued to the device.
1895 * Arguments: shost - Host in question
1899 * Lock status: No locks are assumed held.
1901 * Notes: There is no timer nor any other means by which the requests
1902 * get unblocked other than the low-level driver calling
1903 * scsi_unblock_requests().
1905 void scsi_block_requests(struct Scsi_Host
*shost
)
1907 shost
->host_self_blocked
= 1;
1909 EXPORT_SYMBOL(scsi_block_requests
);
1912 * Function: scsi_unblock_requests()
1914 * Purpose: Utility function used by low-level drivers to allow further
1915 * commands from being queued to the device.
1917 * Arguments: shost - Host in question
1921 * Lock status: No locks are assumed held.
1923 * Notes: There is no timer nor any other means by which the requests
1924 * get unblocked other than the low-level driver calling
1925 * scsi_unblock_requests().
1927 * This is done as an API function so that changes to the
1928 * internals of the scsi mid-layer won't require wholesale
1929 * changes to drivers that use this feature.
1931 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1933 shost
->host_self_blocked
= 0;
1934 scsi_run_host_queues(shost
);
1936 EXPORT_SYMBOL(scsi_unblock_requests
);
1938 int __init
scsi_init_queue(void)
1940 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
1941 sizeof(struct scsi_data_buffer
),
1943 if (!scsi_sdb_cache
) {
1944 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
1951 void scsi_exit_queue(void)
1953 kmem_cache_destroy(scsi_sense_cache
);
1954 kmem_cache_destroy(scsi_sense_isadma_cache
);
1955 kmem_cache_destroy(scsi_sdb_cache
);
1959 * scsi_mode_select - issue a mode select
1960 * @sdev: SCSI device to be queried
1961 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1962 * @sp: Save page bit (0 == don't save, 1 == save)
1963 * @modepage: mode page being requested
1964 * @buffer: request buffer (may not be smaller than eight bytes)
1965 * @len: length of request buffer.
1966 * @timeout: command timeout
1967 * @retries: number of retries before failing
1968 * @data: returns a structure abstracting the mode header data
1969 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1970 * must be SCSI_SENSE_BUFFERSIZE big.
1972 * Returns zero if successful; negative error number or scsi
1977 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1978 unsigned char *buffer
, int len
, int timeout
, int retries
,
1979 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1981 unsigned char cmd
[10];
1982 unsigned char *real_buffer
;
1985 memset(cmd
, 0, sizeof(cmd
));
1986 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1988 if (sdev
->use_10_for_ms
) {
1991 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1994 memcpy(real_buffer
+ 8, buffer
, len
);
1998 real_buffer
[2] = data
->medium_type
;
1999 real_buffer
[3] = data
->device_specific
;
2000 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2002 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2003 real_buffer
[7] = data
->block_descriptor_length
;
2005 cmd
[0] = MODE_SELECT_10
;
2009 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2013 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2016 memcpy(real_buffer
+ 4, buffer
, len
);
2019 real_buffer
[1] = data
->medium_type
;
2020 real_buffer
[2] = data
->device_specific
;
2021 real_buffer
[3] = data
->block_descriptor_length
;
2024 cmd
[0] = MODE_SELECT
;
2028 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2029 sshdr
, timeout
, retries
, NULL
);
2033 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2036 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2037 * @sdev: SCSI device to be queried
2038 * @dbd: set if mode sense will allow block descriptors to be returned
2039 * @modepage: mode page being requested
2040 * @buffer: request buffer (may not be smaller than eight bytes)
2041 * @len: length of request buffer.
2042 * @timeout: command timeout
2043 * @retries: number of retries before failing
2044 * @data: returns a structure abstracting the mode header data
2045 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2046 * must be SCSI_SENSE_BUFFERSIZE big.
2048 * Returns zero if unsuccessful, or the header offset (either 4
2049 * or 8 depending on whether a six or ten byte command was
2050 * issued) if successful.
2053 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2054 unsigned char *buffer
, int len
, int timeout
, int retries
,
2055 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2057 unsigned char cmd
[12];
2060 int result
, retry_count
= retries
;
2061 struct scsi_sense_hdr my_sshdr
;
2063 memset(data
, 0, sizeof(*data
));
2064 memset(&cmd
[0], 0, 12);
2065 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2068 /* caller might not be interested in sense, but we need it */
2073 use_10_for_ms
= sdev
->use_10_for_ms
;
2075 if (use_10_for_ms
) {
2079 cmd
[0] = MODE_SENSE_10
;
2086 cmd
[0] = MODE_SENSE
;
2091 memset(buffer
, 0, len
);
2093 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2094 sshdr
, timeout
, retries
, NULL
);
2096 /* This code looks awful: what it's doing is making sure an
2097 * ILLEGAL REQUEST sense return identifies the actual command
2098 * byte as the problem. MODE_SENSE commands can return
2099 * ILLEGAL REQUEST if the code page isn't supported */
2101 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2102 driver_byte(result
) == DRIVER_SENSE
) {
2103 if (scsi_sense_valid(sshdr
)) {
2104 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2105 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2107 * Invalid command operation code
2109 sdev
->use_10_for_ms
= 0;
2115 if(scsi_status_is_good(result
)) {
2116 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2117 (modepage
== 6 || modepage
== 8))) {
2118 /* Initio breakage? */
2121 data
->medium_type
= 0;
2122 data
->device_specific
= 0;
2124 data
->block_descriptor_length
= 0;
2125 } else if(use_10_for_ms
) {
2126 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2127 data
->medium_type
= buffer
[2];
2128 data
->device_specific
= buffer
[3];
2129 data
->longlba
= buffer
[4] & 0x01;
2130 data
->block_descriptor_length
= buffer
[6]*256
2133 data
->length
= buffer
[0] + 1;
2134 data
->medium_type
= buffer
[1];
2135 data
->device_specific
= buffer
[2];
2136 data
->block_descriptor_length
= buffer
[3];
2138 data
->header_length
= header_length
;
2139 } else if ((status_byte(result
) == CHECK_CONDITION
) &&
2140 scsi_sense_valid(sshdr
) &&
2141 sshdr
->sense_key
== UNIT_ATTENTION
&& retry_count
) {
2148 EXPORT_SYMBOL(scsi_mode_sense
);
2151 * scsi_test_unit_ready - test if unit is ready
2152 * @sdev: scsi device to change the state of.
2153 * @timeout: command timeout
2154 * @retries: number of retries before failing
2155 * @sshdr: outpout pointer for decoded sense information.
2157 * Returns zero if unsuccessful or an error if TUR failed. For
2158 * removable media, UNIT_ATTENTION sets ->changed flag.
2161 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2162 struct scsi_sense_hdr
*sshdr
)
2165 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2169 /* try to eat the UNIT_ATTENTION if there are enough retries */
2171 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2173 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2174 sshdr
->sense_key
== UNIT_ATTENTION
)
2176 } while (scsi_sense_valid(sshdr
) &&
2177 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2181 EXPORT_SYMBOL(scsi_test_unit_ready
);
2184 * scsi_device_set_state - Take the given device through the device state model.
2185 * @sdev: scsi device to change the state of.
2186 * @state: state to change to.
2188 * Returns zero if successful or an error if the requested
2189 * transition is illegal.
2192 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2194 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2196 if (state
== oldstate
)
2202 case SDEV_CREATED_BLOCK
:
2213 case SDEV_TRANSPORT_OFFLINE
:
2226 case SDEV_TRANSPORT_OFFLINE
:
2234 case SDEV_TRANSPORT_OFFLINE
:
2249 case SDEV_CREATED_BLOCK
:
2257 case SDEV_CREATED_BLOCK
:
2272 case SDEV_TRANSPORT_OFFLINE
:
2284 case SDEV_TRANSPORT_OFFLINE
:
2287 case SDEV_CREATED_BLOCK
:
2295 sdev
->sdev_state
= state
;
2299 SCSI_LOG_ERROR_RECOVERY(1,
2300 sdev_printk(KERN_ERR
, sdev
,
2301 "Illegal state transition %s->%s",
2302 scsi_device_state_name(oldstate
),
2303 scsi_device_state_name(state
))
2307 EXPORT_SYMBOL(scsi_device_set_state
);
2310 * sdev_evt_emit - emit a single SCSI device uevent
2311 * @sdev: associated SCSI device
2312 * @evt: event to emit
2314 * Send a single uevent (scsi_event) to the associated scsi_device.
2316 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2321 switch (evt
->evt_type
) {
2322 case SDEV_EVT_MEDIA_CHANGE
:
2323 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2325 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2326 scsi_rescan_device(&sdev
->sdev_gendev
);
2327 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2329 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2330 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2332 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2333 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2335 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2336 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2338 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2339 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2341 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2342 envp
[idx
++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2344 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2345 envp
[idx
++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2354 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2358 * sdev_evt_thread - send a uevent for each scsi event
2359 * @work: work struct for scsi_device
2361 * Dispatch queued events to their associated scsi_device kobjects
2364 void scsi_evt_thread(struct work_struct
*work
)
2366 struct scsi_device
*sdev
;
2367 enum scsi_device_event evt_type
;
2368 LIST_HEAD(event_list
);
2370 sdev
= container_of(work
, struct scsi_device
, event_work
);
2372 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2373 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2374 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2377 struct scsi_event
*evt
;
2378 struct list_head
*this, *tmp
;
2379 unsigned long flags
;
2381 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2382 list_splice_init(&sdev
->event_list
, &event_list
);
2383 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2385 if (list_empty(&event_list
))
2388 list_for_each_safe(this, tmp
, &event_list
) {
2389 evt
= list_entry(this, struct scsi_event
, node
);
2390 list_del(&evt
->node
);
2391 scsi_evt_emit(sdev
, evt
);
2398 * sdev_evt_send - send asserted event to uevent thread
2399 * @sdev: scsi_device event occurred on
2400 * @evt: event to send
2402 * Assert scsi device event asynchronously.
2404 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2406 unsigned long flags
;
2409 /* FIXME: currently this check eliminates all media change events
2410 * for polled devices. Need to update to discriminate between AN
2411 * and polled events */
2412 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2418 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2419 list_add_tail(&evt
->node
, &sdev
->event_list
);
2420 schedule_work(&sdev
->event_work
);
2421 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2423 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2426 * sdev_evt_alloc - allocate a new scsi event
2427 * @evt_type: type of event to allocate
2428 * @gfpflags: GFP flags for allocation
2430 * Allocates and returns a new scsi_event.
2432 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2435 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2439 evt
->evt_type
= evt_type
;
2440 INIT_LIST_HEAD(&evt
->node
);
2442 /* evt_type-specific initialization, if any */
2444 case SDEV_EVT_MEDIA_CHANGE
:
2445 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2446 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2447 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2448 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2449 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2450 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2451 case SDEV_EVT_POWER_ON_RESET_OCCURRED
:
2459 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2462 * sdev_evt_send_simple - send asserted event to uevent thread
2463 * @sdev: scsi_device event occurred on
2464 * @evt_type: type of event to send
2465 * @gfpflags: GFP flags for allocation
2467 * Assert scsi device event asynchronously, given an event type.
2469 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2470 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2472 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2474 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2479 sdev_evt_send(sdev
, evt
);
2481 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2484 * scsi_device_quiesce - Block user issued commands.
2485 * @sdev: scsi device to quiesce.
2487 * This works by trying to transition to the SDEV_QUIESCE state
2488 * (which must be a legal transition). When the device is in this
2489 * state, only special requests will be accepted, all others will
2490 * be deferred. Since special requests may also be requeued requests,
2491 * a successful return doesn't guarantee the device will be
2492 * totally quiescent.
2494 * Must be called with user context, may sleep.
2496 * Returns zero if unsuccessful or an error if not.
2499 scsi_device_quiesce(struct scsi_device
*sdev
)
2501 struct request_queue
*q
= sdev
->request_queue
;
2505 * It is allowed to call scsi_device_quiesce() multiple times from
2506 * the same context but concurrent scsi_device_quiesce() calls are
2509 WARN_ON_ONCE(sdev
->quiesced_by
&& sdev
->quiesced_by
!= current
);
2511 if (sdev
->quiesced_by
== current
)
2516 blk_mq_freeze_queue(q
);
2518 * Ensure that the effect of blk_set_pm_only() will be visible
2519 * for percpu_ref_tryget() callers that occur after the queue
2520 * unfreeze even if the queue was already frozen before this function
2521 * was called. See also https://lwn.net/Articles/573497/.
2524 blk_mq_unfreeze_queue(q
);
2526 mutex_lock(&sdev
->state_mutex
);
2527 err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2529 sdev
->quiesced_by
= current
;
2531 blk_clear_pm_only(q
);
2532 mutex_unlock(&sdev
->state_mutex
);
2536 EXPORT_SYMBOL(scsi_device_quiesce
);
2539 * scsi_device_resume - Restart user issued commands to a quiesced device.
2540 * @sdev: scsi device to resume.
2542 * Moves the device from quiesced back to running and restarts the
2545 * Must be called with user context, may sleep.
2547 void scsi_device_resume(struct scsi_device
*sdev
)
2549 /* check if the device state was mutated prior to resume, and if
2550 * so assume the state is being managed elsewhere (for example
2551 * device deleted during suspend)
2553 mutex_lock(&sdev
->state_mutex
);
2554 if (sdev
->quiesced_by
) {
2555 sdev
->quiesced_by
= NULL
;
2556 blk_clear_pm_only(sdev
->request_queue
);
2558 if (sdev
->sdev_state
== SDEV_QUIESCE
)
2559 scsi_device_set_state(sdev
, SDEV_RUNNING
);
2560 mutex_unlock(&sdev
->state_mutex
);
2562 EXPORT_SYMBOL(scsi_device_resume
);
2565 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2567 scsi_device_quiesce(sdev
);
2571 scsi_target_quiesce(struct scsi_target
*starget
)
2573 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2575 EXPORT_SYMBOL(scsi_target_quiesce
);
2578 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2580 scsi_device_resume(sdev
);
2584 scsi_target_resume(struct scsi_target
*starget
)
2586 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2588 EXPORT_SYMBOL(scsi_target_resume
);
2591 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2592 * @sdev: device to block
2594 * Pause SCSI command processing on the specified device. Does not sleep.
2596 * Returns zero if successful or a negative error code upon failure.
2599 * This routine transitions the device to the SDEV_BLOCK state (which must be
2600 * a legal transition). When the device is in this state, command processing
2601 * is paused until the device leaves the SDEV_BLOCK state. See also
2602 * scsi_internal_device_unblock_nowait().
2604 int scsi_internal_device_block_nowait(struct scsi_device
*sdev
)
2606 struct request_queue
*q
= sdev
->request_queue
;
2609 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2611 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2618 * The device has transitioned to SDEV_BLOCK. Stop the
2619 * block layer from calling the midlayer with this device's
2622 blk_mq_quiesce_queue_nowait(q
);
2625 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait
);
2628 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2629 * @sdev: device to block
2631 * Pause SCSI command processing on the specified device and wait until all
2632 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2634 * Returns zero if successful or a negative error code upon failure.
2637 * This routine transitions the device to the SDEV_BLOCK state (which must be
2638 * a legal transition). When the device is in this state, command processing
2639 * is paused until the device leaves the SDEV_BLOCK state. See also
2640 * scsi_internal_device_unblock().
2642 static int scsi_internal_device_block(struct scsi_device
*sdev
)
2644 struct request_queue
*q
= sdev
->request_queue
;
2647 mutex_lock(&sdev
->state_mutex
);
2648 err
= scsi_internal_device_block_nowait(sdev
);
2650 blk_mq_quiesce_queue(q
);
2651 mutex_unlock(&sdev
->state_mutex
);
2656 void scsi_start_queue(struct scsi_device
*sdev
)
2658 struct request_queue
*q
= sdev
->request_queue
;
2660 blk_mq_unquiesce_queue(q
);
2664 * scsi_internal_device_unblock_nowait - resume a device after a block request
2665 * @sdev: device to resume
2666 * @new_state: state to set the device to after unblocking
2668 * Restart the device queue for a previously suspended SCSI device. Does not
2671 * Returns zero if successful or a negative error code upon failure.
2674 * This routine transitions the device to the SDEV_RUNNING state or to one of
2675 * the offline states (which must be a legal transition) allowing the midlayer
2676 * to goose the queue for this device.
2678 int scsi_internal_device_unblock_nowait(struct scsi_device
*sdev
,
2679 enum scsi_device_state new_state
)
2682 * Try to transition the scsi device to SDEV_RUNNING or one of the
2683 * offlined states and goose the device queue if successful.
2685 switch (sdev
->sdev_state
) {
2687 case SDEV_TRANSPORT_OFFLINE
:
2688 sdev
->sdev_state
= new_state
;
2690 case SDEV_CREATED_BLOCK
:
2691 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2692 new_state
== SDEV_OFFLINE
)
2693 sdev
->sdev_state
= new_state
;
2695 sdev
->sdev_state
= SDEV_CREATED
;
2703 scsi_start_queue(sdev
);
2707 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait
);
2710 * scsi_internal_device_unblock - resume a device after a block request
2711 * @sdev: device to resume
2712 * @new_state: state to set the device to after unblocking
2714 * Restart the device queue for a previously suspended SCSI device. May sleep.
2716 * Returns zero if successful or a negative error code upon failure.
2719 * This routine transitions the device to the SDEV_RUNNING state or to one of
2720 * the offline states (which must be a legal transition) allowing the midlayer
2721 * to goose the queue for this device.
2723 static int scsi_internal_device_unblock(struct scsi_device
*sdev
,
2724 enum scsi_device_state new_state
)
2728 mutex_lock(&sdev
->state_mutex
);
2729 ret
= scsi_internal_device_unblock_nowait(sdev
, new_state
);
2730 mutex_unlock(&sdev
->state_mutex
);
2736 device_block(struct scsi_device
*sdev
, void *data
)
2738 scsi_internal_device_block(sdev
);
2742 target_block(struct device
*dev
, void *data
)
2744 if (scsi_is_target_device(dev
))
2745 starget_for_each_device(to_scsi_target(dev
), NULL
,
2751 scsi_target_block(struct device
*dev
)
2753 if (scsi_is_target_device(dev
))
2754 starget_for_each_device(to_scsi_target(dev
), NULL
,
2757 device_for_each_child(dev
, NULL
, target_block
);
2759 EXPORT_SYMBOL_GPL(scsi_target_block
);
2762 device_unblock(struct scsi_device
*sdev
, void *data
)
2764 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2768 target_unblock(struct device
*dev
, void *data
)
2770 if (scsi_is_target_device(dev
))
2771 starget_for_each_device(to_scsi_target(dev
), data
,
2777 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2779 if (scsi_is_target_device(dev
))
2780 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2783 device_for_each_child(dev
, &new_state
, target_unblock
);
2785 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2788 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2789 * @sgl: scatter-gather list
2790 * @sg_count: number of segments in sg
2791 * @offset: offset in bytes into sg, on return offset into the mapped area
2792 * @len: bytes to map, on return number of bytes mapped
2794 * Returns virtual address of the start of the mapped page
2796 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2797 size_t *offset
, size_t *len
)
2800 size_t sg_len
= 0, len_complete
= 0;
2801 struct scatterlist
*sg
;
2804 WARN_ON(!irqs_disabled());
2806 for_each_sg(sgl
, sg
, sg_count
, i
) {
2807 len_complete
= sg_len
; /* Complete sg-entries */
2808 sg_len
+= sg
->length
;
2809 if (sg_len
> *offset
)
2813 if (unlikely(i
== sg_count
)) {
2814 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2816 __func__
, sg_len
, *offset
, sg_count
);
2821 /* Offset starting from the beginning of first page in this sg-entry */
2822 *offset
= *offset
- len_complete
+ sg
->offset
;
2824 /* Assumption: contiguous pages can be accessed as "page + i" */
2825 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2826 *offset
&= ~PAGE_MASK
;
2828 /* Bytes in this sg-entry from *offset to the end of the page */
2829 sg_len
= PAGE_SIZE
- *offset
;
2833 return kmap_atomic(page
);
2835 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2838 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2839 * @virt: virtual address to be unmapped
2841 void scsi_kunmap_atomic_sg(void *virt
)
2843 kunmap_atomic(virt
);
2845 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
2847 void sdev_disable_disk_events(struct scsi_device
*sdev
)
2849 atomic_inc(&sdev
->disk_events_disable_depth
);
2851 EXPORT_SYMBOL(sdev_disable_disk_events
);
2853 void sdev_enable_disk_events(struct scsi_device
*sdev
)
2855 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
2857 atomic_dec(&sdev
->disk_events_disable_depth
);
2859 EXPORT_SYMBOL(sdev_enable_disk_events
);
2862 * scsi_vpd_lun_id - return a unique device identification
2863 * @sdev: SCSI device
2864 * @id: buffer for the identification
2865 * @id_len: length of the buffer
2867 * Copies a unique device identification into @id based
2868 * on the information in the VPD page 0x83 of the device.
2869 * The string will be formatted as a SCSI name string.
2871 * Returns the length of the identification or error on failure.
2872 * If the identifier is longer than the supplied buffer the actual
2873 * identifier length is returned and the buffer is not zero-padded.
2875 int scsi_vpd_lun_id(struct scsi_device
*sdev
, char *id
, size_t id_len
)
2877 u8 cur_id_type
= 0xff;
2879 const unsigned char *d
, *cur_id_str
;
2880 const struct scsi_vpd
*vpd_pg83
;
2881 int id_size
= -EINVAL
;
2884 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
2891 * Look for the correct descriptor.
2892 * Order of preference for lun descriptor:
2893 * - SCSI name string
2894 * - NAA IEEE Registered Extended
2895 * - EUI-64 based 16-byte
2896 * - EUI-64 based 12-byte
2897 * - NAA IEEE Registered
2898 * - NAA IEEE Extended
2900 * as longer descriptors reduce the likelyhood
2901 * of identification clashes.
2904 /* The id string must be at least 20 bytes + terminating NULL byte */
2910 memset(id
, 0, id_len
);
2911 d
= vpd_pg83
->data
+ 4;
2912 while (d
< vpd_pg83
->data
+ vpd_pg83
->len
) {
2913 /* Skip designators not referring to the LUN */
2914 if ((d
[1] & 0x30) != 0x00)
2917 switch (d
[1] & 0xf) {
2920 if (cur_id_size
> d
[3])
2922 /* Prefer anything */
2923 if (cur_id_type
> 0x01 && cur_id_type
!= 0xff)
2926 if (cur_id_size
+ 4 > id_len
)
2927 cur_id_size
= id_len
- 4;
2929 cur_id_type
= d
[1] & 0xf;
2930 id_size
= snprintf(id
, id_len
, "t10.%*pE",
2931 cur_id_size
, cur_id_str
);
2935 if (cur_id_size
> d
[3])
2937 /* Prefer NAA IEEE Registered Extended */
2938 if (cur_id_type
== 0x3 &&
2939 cur_id_size
== d
[3])
2943 cur_id_type
= d
[1] & 0xf;
2944 switch (cur_id_size
) {
2946 id_size
= snprintf(id
, id_len
,
2951 id_size
= snprintf(id
, id_len
,
2956 id_size
= snprintf(id
, id_len
,
2967 if (cur_id_size
> d
[3])
2971 cur_id_type
= d
[1] & 0xf;
2972 switch (cur_id_size
) {
2974 id_size
= snprintf(id
, id_len
,
2979 id_size
= snprintf(id
, id_len
,
2989 /* SCSI name string */
2990 if (cur_id_size
+ 4 > d
[3])
2992 /* Prefer others for truncated descriptor */
2993 if (cur_id_size
&& d
[3] > id_len
)
2995 cur_id_size
= id_size
= d
[3];
2997 cur_id_type
= d
[1] & 0xf;
2998 if (cur_id_size
>= id_len
)
2999 cur_id_size
= id_len
- 1;
3000 memcpy(id
, cur_id_str
, cur_id_size
);
3001 /* Decrease priority for truncated descriptor */
3002 if (cur_id_size
!= id_size
)
3015 EXPORT_SYMBOL(scsi_vpd_lun_id
);
3018 * scsi_vpd_tpg_id - return a target port group identifier
3019 * @sdev: SCSI device
3021 * Returns the Target Port Group identifier from the information
3022 * froom VPD page 0x83 of the device.
3024 * Returns the identifier or error on failure.
3026 int scsi_vpd_tpg_id(struct scsi_device
*sdev
, int *rel_id
)
3028 const unsigned char *d
;
3029 const struct scsi_vpd
*vpd_pg83
;
3030 int group_id
= -EAGAIN
, rel_port
= -1;
3033 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3039 d
= vpd_pg83
->data
+ 4;
3040 while (d
< vpd_pg83
->data
+ vpd_pg83
->len
) {
3041 switch (d
[1] & 0xf) {
3043 /* Relative target port */
3044 rel_port
= get_unaligned_be16(&d
[6]);
3047 /* Target port group */
3048 group_id
= get_unaligned_be16(&d
[6]);
3057 if (group_id
>= 0 && rel_id
&& rel_port
!= -1)
3062 EXPORT_SYMBOL(scsi_vpd_tpg_id
);