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/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/init.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/hardirq.h>
23 #include <linux/scatterlist.h>
24 #include <linux/blk-mq.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>
34 #include <trace/events/scsi.h>
36 #include "scsi_priv.h"
37 #include "scsi_logging.h"
40 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
41 #define SG_MEMPOOL_SIZE 2
43 struct scsi_host_sg_pool
{
46 struct kmem_cache
*slab
;
50 #define SP(x) { x, "sgpool-" __stringify(x) }
51 #if (SCSI_MAX_SG_SEGMENTS < 32)
52 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
54 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
57 #if (SCSI_MAX_SG_SEGMENTS > 32)
59 #if (SCSI_MAX_SG_SEGMENTS > 64)
61 #if (SCSI_MAX_SG_SEGMENTS > 128)
63 #if (SCSI_MAX_SG_SEGMENTS > 256)
64 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
69 SP(SCSI_MAX_SG_SEGMENTS
)
73 struct kmem_cache
*scsi_sdb_cache
;
76 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
77 * not change behaviour from the previous unplug mechanism, experimentation
78 * may prove this needs changing.
80 #define SCSI_QUEUE_DELAY 3
83 scsi_set_blocked(struct scsi_cmnd
*cmd
, int reason
)
85 struct Scsi_Host
*host
= cmd
->device
->host
;
86 struct scsi_device
*device
= cmd
->device
;
87 struct scsi_target
*starget
= scsi_target(device
);
90 * Set the appropriate busy bit for the device/host.
92 * If the host/device isn't busy, assume that something actually
93 * completed, and that we should be able to queue a command now.
95 * Note that the prior mid-layer assumption that any host could
96 * always queue at least one command is now broken. The mid-layer
97 * will implement a user specifiable stall (see
98 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
99 * if a command is requeued with no other commands outstanding
100 * either for the device or for the host.
103 case SCSI_MLQUEUE_HOST_BUSY
:
104 atomic_set(&host
->host_blocked
, host
->max_host_blocked
);
106 case SCSI_MLQUEUE_DEVICE_BUSY
:
107 case SCSI_MLQUEUE_EH_RETRY
:
108 atomic_set(&device
->device_blocked
,
109 device
->max_device_blocked
);
111 case SCSI_MLQUEUE_TARGET_BUSY
:
112 atomic_set(&starget
->target_blocked
,
113 starget
->max_target_blocked
);
118 static void scsi_mq_requeue_cmd(struct scsi_cmnd
*cmd
)
120 struct scsi_device
*sdev
= cmd
->device
;
121 struct request_queue
*q
= cmd
->request
->q
;
123 blk_mq_requeue_request(cmd
->request
);
124 blk_mq_kick_requeue_list(q
);
125 put_device(&sdev
->sdev_gendev
);
129 * __scsi_queue_insert - private queue insertion
130 * @cmd: The SCSI command being requeued
131 * @reason: The reason for the requeue
132 * @unbusy: Whether the queue should be unbusied
134 * This is a private queue insertion. The public interface
135 * scsi_queue_insert() always assumes the queue should be unbusied
136 * because it's always called before the completion. This function is
137 * for a requeue after completion, which should only occur in this
140 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, int unbusy
)
142 struct scsi_device
*device
= cmd
->device
;
143 struct request_queue
*q
= device
->request_queue
;
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
147 "Inserting command %p into mlqueue\n", cmd
));
149 scsi_set_blocked(cmd
, reason
);
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
156 scsi_device_unbusy(device
);
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_cleanup_queue() finishes.
166 scsi_mq_requeue_cmd(cmd
);
169 spin_lock_irqsave(q
->queue_lock
, flags
);
170 blk_requeue_request(q
, cmd
->request
);
171 kblockd_schedule_work(&device
->requeue_work
);
172 spin_unlock_irqrestore(q
->queue_lock
, flags
);
176 * Function: scsi_queue_insert()
178 * Purpose: Insert a command in the midlevel queue.
180 * Arguments: cmd - command that we are adding to queue.
181 * reason - why we are inserting command to queue.
183 * Lock status: Assumed that lock is not held upon entry.
187 * Notes: We do this for one of two cases. Either the host is busy
188 * and it cannot accept any more commands for the time being,
189 * or the device returned QUEUE_FULL and can accept no more
191 * Notes: This could be called either from an interrupt context or a
192 * normal process context.
194 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
196 __scsi_queue_insert(cmd
, reason
, 1);
199 * scsi_execute - insert request and wait for the result
202 * @data_direction: data direction
203 * @buffer: data buffer
204 * @bufflen: len of buffer
205 * @sense: optional sense buffer
206 * @timeout: request timeout in seconds
207 * @retries: number of times to retry request
208 * @flags: or into request flags;
209 * @resid: optional residual length
211 * returns the req->errors value which is the scsi_cmnd result
214 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
215 int data_direction
, void *buffer
, unsigned bufflen
,
216 unsigned char *sense
, int timeout
, int retries
, u64 flags
,
220 int write
= (data_direction
== DMA_TO_DEVICE
);
221 int ret
= DRIVER_ERROR
<< 24;
223 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
226 blk_rq_set_block_pc(req
);
228 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
229 buffer
, bufflen
, __GFP_WAIT
))
232 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
233 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
236 req
->retries
= retries
;
237 req
->timeout
= timeout
;
238 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
241 * head injection *required* here otherwise quiesce won't work
243 blk_execute_rq(req
->q
, NULL
, req
, 1);
246 * Some devices (USB mass-storage in particular) may transfer
247 * garbage data together with a residue indicating that the data
248 * is invalid. Prevent the garbage from being misinterpreted
249 * and prevent security leaks by zeroing out the excess data.
251 if (unlikely(req
->resid_len
> 0 && req
->resid_len
<= bufflen
))
252 memset(buffer
+ (bufflen
- req
->resid_len
), 0, req
->resid_len
);
255 *resid
= req
->resid_len
;
258 blk_put_request(req
);
262 EXPORT_SYMBOL(scsi_execute
);
264 int scsi_execute_req_flags(struct scsi_device
*sdev
, const unsigned char *cmd
,
265 int data_direction
, void *buffer
, unsigned bufflen
,
266 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
,
267 int *resid
, u64 flags
)
273 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
275 return DRIVER_ERROR
<< 24;
277 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
278 sense
, timeout
, retries
, flags
, resid
);
280 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
285 EXPORT_SYMBOL(scsi_execute_req_flags
);
288 * Function: scsi_init_cmd_errh()
290 * Purpose: Initialize cmd fields related to error handling.
292 * Arguments: cmd - command that is ready to be queued.
294 * Notes: This function has the job of initializing a number of
295 * fields related to error handling. Typically this will
296 * be called once for each command, as required.
298 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
300 cmd
->serial_number
= 0;
301 scsi_set_resid(cmd
, 0);
302 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
303 if (cmd
->cmd_len
== 0)
304 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
307 void scsi_device_unbusy(struct scsi_device
*sdev
)
309 struct Scsi_Host
*shost
= sdev
->host
;
310 struct scsi_target
*starget
= scsi_target(sdev
);
313 atomic_dec(&shost
->host_busy
);
314 if (starget
->can_queue
> 0)
315 atomic_dec(&starget
->target_busy
);
317 if (unlikely(scsi_host_in_recovery(shost
) &&
318 (shost
->host_failed
|| shost
->host_eh_scheduled
))) {
319 spin_lock_irqsave(shost
->host_lock
, flags
);
320 scsi_eh_wakeup(shost
);
321 spin_unlock_irqrestore(shost
->host_lock
, flags
);
324 atomic_dec(&sdev
->device_busy
);
327 static void scsi_kick_queue(struct request_queue
*q
)
330 blk_mq_start_hw_queues(q
);
336 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
337 * and call blk_run_queue for all the scsi_devices on the target -
338 * including current_sdev first.
340 * Called with *no* scsi locks held.
342 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
344 struct Scsi_Host
*shost
= current_sdev
->host
;
345 struct scsi_device
*sdev
, *tmp
;
346 struct scsi_target
*starget
= scsi_target(current_sdev
);
349 spin_lock_irqsave(shost
->host_lock
, flags
);
350 starget
->starget_sdev_user
= NULL
;
351 spin_unlock_irqrestore(shost
->host_lock
, flags
);
354 * Call blk_run_queue for all LUNs on the target, starting with
355 * current_sdev. We race with others (to set starget_sdev_user),
356 * but in most cases, we will be first. Ideally, each LU on the
357 * target would get some limited time or requests on the target.
359 scsi_kick_queue(current_sdev
->request_queue
);
361 spin_lock_irqsave(shost
->host_lock
, flags
);
362 if (starget
->starget_sdev_user
)
364 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
365 same_target_siblings
) {
366 if (sdev
== current_sdev
)
368 if (scsi_device_get(sdev
))
371 spin_unlock_irqrestore(shost
->host_lock
, flags
);
372 scsi_kick_queue(sdev
->request_queue
);
373 spin_lock_irqsave(shost
->host_lock
, flags
);
375 scsi_device_put(sdev
);
378 spin_unlock_irqrestore(shost
->host_lock
, flags
);
381 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
383 if (atomic_read(&sdev
->device_busy
) >= sdev
->queue_depth
)
385 if (atomic_read(&sdev
->device_blocked
) > 0)
390 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
392 if (starget
->can_queue
> 0) {
393 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
395 if (atomic_read(&starget
->target_blocked
) > 0)
401 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
403 if (shost
->can_queue
> 0 &&
404 atomic_read(&shost
->host_busy
) >= shost
->can_queue
)
406 if (atomic_read(&shost
->host_blocked
) > 0)
408 if (shost
->host_self_blocked
)
413 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
415 LIST_HEAD(starved_list
);
416 struct scsi_device
*sdev
;
419 spin_lock_irqsave(shost
->host_lock
, flags
);
420 list_splice_init(&shost
->starved_list
, &starved_list
);
422 while (!list_empty(&starved_list
)) {
423 struct request_queue
*slq
;
426 * As long as shost is accepting commands and we have
427 * starved queues, call blk_run_queue. scsi_request_fn
428 * drops the queue_lock and can add us back to the
431 * host_lock protects the starved_list and starved_entry.
432 * scsi_request_fn must get the host_lock before checking
433 * or modifying starved_list or starved_entry.
435 if (scsi_host_is_busy(shost
))
438 sdev
= list_entry(starved_list
.next
,
439 struct scsi_device
, starved_entry
);
440 list_del_init(&sdev
->starved_entry
);
441 if (scsi_target_is_busy(scsi_target(sdev
))) {
442 list_move_tail(&sdev
->starved_entry
,
443 &shost
->starved_list
);
448 * Once we drop the host lock, a racing scsi_remove_device()
449 * call may remove the sdev from the starved list and destroy
450 * it and the queue. Mitigate by taking a reference to the
451 * queue and never touching the sdev again after we drop the
452 * host lock. Note: if __scsi_remove_device() invokes
453 * blk_cleanup_queue() before the queue is run from this
454 * function then blk_run_queue() will return immediately since
455 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
457 slq
= sdev
->request_queue
;
458 if (!blk_get_queue(slq
))
460 spin_unlock_irqrestore(shost
->host_lock
, flags
);
462 scsi_kick_queue(slq
);
465 spin_lock_irqsave(shost
->host_lock
, flags
);
467 /* put any unprocessed entries back */
468 list_splice(&starved_list
, &shost
->starved_list
);
469 spin_unlock_irqrestore(shost
->host_lock
, flags
);
473 * Function: scsi_run_queue()
475 * Purpose: Select a proper request queue to serve next
477 * Arguments: q - last request's queue
481 * Notes: The previous command was completely finished, start
482 * a new one if possible.
484 static void scsi_run_queue(struct request_queue
*q
)
486 struct scsi_device
*sdev
= q
->queuedata
;
488 if (scsi_target(sdev
)->single_lun
)
489 scsi_single_lun_run(sdev
);
490 if (!list_empty(&sdev
->host
->starved_list
))
491 scsi_starved_list_run(sdev
->host
);
494 blk_mq_start_stopped_hw_queues(q
, false);
499 void scsi_requeue_run_queue(struct work_struct
*work
)
501 struct scsi_device
*sdev
;
502 struct request_queue
*q
;
504 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
505 q
= sdev
->request_queue
;
510 * Function: scsi_requeue_command()
512 * Purpose: Handle post-processing of completed commands.
514 * Arguments: q - queue to operate on
515 * cmd - command that may need to be requeued.
519 * Notes: After command completion, there may be blocks left
520 * over which weren't finished by the previous command
521 * this can be for a number of reasons - the main one is
522 * I/O errors in the middle of the request, in which case
523 * we need to request the blocks that come after the bad
525 * Notes: Upon return, cmd is a stale pointer.
527 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
529 struct scsi_device
*sdev
= cmd
->device
;
530 struct request
*req
= cmd
->request
;
533 spin_lock_irqsave(q
->queue_lock
, flags
);
534 blk_unprep_request(req
);
536 scsi_put_command(cmd
);
537 blk_requeue_request(q
, req
);
538 spin_unlock_irqrestore(q
->queue_lock
, flags
);
542 put_device(&sdev
->sdev_gendev
);
545 void scsi_next_command(struct scsi_cmnd
*cmd
)
547 struct scsi_device
*sdev
= cmd
->device
;
548 struct request_queue
*q
= sdev
->request_queue
;
550 scsi_put_command(cmd
);
553 put_device(&sdev
->sdev_gendev
);
556 void scsi_run_host_queues(struct Scsi_Host
*shost
)
558 struct scsi_device
*sdev
;
560 shost_for_each_device(sdev
, shost
)
561 scsi_run_queue(sdev
->request_queue
);
564 static inline unsigned int scsi_sgtable_index(unsigned short nents
)
568 BUG_ON(nents
> SCSI_MAX_SG_SEGMENTS
);
573 index
= get_count_order(nents
) - 3;
578 static void scsi_sg_free(struct scatterlist
*sgl
, unsigned int nents
)
580 struct scsi_host_sg_pool
*sgp
;
582 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
583 mempool_free(sgl
, sgp
->pool
);
586 static struct scatterlist
*scsi_sg_alloc(unsigned int nents
, gfp_t gfp_mask
)
588 struct scsi_host_sg_pool
*sgp
;
590 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
591 return mempool_alloc(sgp
->pool
, gfp_mask
);
594 static void scsi_free_sgtable(struct scsi_data_buffer
*sdb
, bool mq
)
596 if (mq
&& sdb
->table
.nents
<= SCSI_MAX_SG_SEGMENTS
)
598 __sg_free_table(&sdb
->table
, SCSI_MAX_SG_SEGMENTS
, mq
, scsi_sg_free
);
601 static int scsi_alloc_sgtable(struct scsi_data_buffer
*sdb
, int nents
,
602 gfp_t gfp_mask
, bool mq
)
604 struct scatterlist
*first_chunk
= NULL
;
610 if (nents
<= SCSI_MAX_SG_SEGMENTS
) {
611 sdb
->table
.nents
= nents
;
612 sg_init_table(sdb
->table
.sgl
, sdb
->table
.nents
);
615 first_chunk
= sdb
->table
.sgl
;
618 ret
= __sg_alloc_table(&sdb
->table
, nents
, SCSI_MAX_SG_SEGMENTS
,
619 first_chunk
, gfp_mask
, scsi_sg_alloc
);
621 scsi_free_sgtable(sdb
, mq
);
625 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
627 if (cmd
->request
->cmd_type
== REQ_TYPE_FS
) {
628 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
630 if (drv
->uninit_command
)
631 drv
->uninit_command(cmd
);
635 static void scsi_mq_free_sgtables(struct scsi_cmnd
*cmd
)
637 if (cmd
->sdb
.table
.nents
)
638 scsi_free_sgtable(&cmd
->sdb
, true);
639 if (cmd
->request
->next_rq
&& cmd
->request
->next_rq
->special
)
640 scsi_free_sgtable(cmd
->request
->next_rq
->special
, true);
641 if (scsi_prot_sg_count(cmd
))
642 scsi_free_sgtable(cmd
->prot_sdb
, true);
645 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
647 struct scsi_device
*sdev
= cmd
->device
;
648 struct Scsi_Host
*shost
= sdev
->host
;
651 scsi_mq_free_sgtables(cmd
);
652 scsi_uninit_cmd(cmd
);
654 if (shost
->use_cmd_list
) {
655 BUG_ON(list_empty(&cmd
->list
));
656 spin_lock_irqsave(&sdev
->list_lock
, flags
);
657 list_del_init(&cmd
->list
);
658 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
663 * Function: scsi_release_buffers()
665 * Purpose: Free resources allocate for a scsi_command.
667 * Arguments: cmd - command that we are bailing.
669 * Lock status: Assumed that no lock is held upon entry.
673 * Notes: In the event that an upper level driver rejects a
674 * command, we must release resources allocated during
675 * the __init_io() function. Primarily this would involve
676 * the scatter-gather table.
678 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
680 if (cmd
->sdb
.table
.nents
)
681 scsi_free_sgtable(&cmd
->sdb
, false);
683 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
685 if (scsi_prot_sg_count(cmd
))
686 scsi_free_sgtable(cmd
->prot_sdb
, false);
689 static void scsi_release_bidi_buffers(struct scsi_cmnd
*cmd
)
691 struct scsi_data_buffer
*bidi_sdb
= cmd
->request
->next_rq
->special
;
693 scsi_free_sgtable(bidi_sdb
, false);
694 kmem_cache_free(scsi_sdb_cache
, bidi_sdb
);
695 cmd
->request
->next_rq
->special
= NULL
;
698 static bool scsi_end_request(struct request
*req
, int error
,
699 unsigned int bytes
, unsigned int bidi_bytes
)
701 struct scsi_cmnd
*cmd
= req
->special
;
702 struct scsi_device
*sdev
= cmd
->device
;
703 struct request_queue
*q
= sdev
->request_queue
;
705 if (blk_update_request(req
, error
, bytes
))
708 /* Bidi request must be completed as a whole */
709 if (unlikely(bidi_bytes
) &&
710 blk_update_request(req
->next_rq
, error
, bidi_bytes
))
713 if (blk_queue_add_random(q
))
714 add_disk_randomness(req
->rq_disk
);
718 * In the MQ case the command gets freed by __blk_mq_end_io,
719 * so we have to do all cleanup that depends on it earlier.
721 * We also can't kick the queues from irq context, so we
722 * will have to defer it to a workqueue.
724 scsi_mq_uninit_cmd(cmd
);
726 __blk_mq_end_io(req
, error
);
728 if (scsi_target(sdev
)->single_lun
||
729 !list_empty(&sdev
->host
->starved_list
))
730 kblockd_schedule_work(&sdev
->requeue_work
);
732 blk_mq_start_stopped_hw_queues(q
, true);
734 put_device(&sdev
->sdev_gendev
);
738 spin_lock_irqsave(q
->queue_lock
, flags
);
739 blk_finish_request(req
, error
);
740 spin_unlock_irqrestore(q
->queue_lock
, flags
);
743 scsi_release_bidi_buffers(cmd
);
744 scsi_release_buffers(cmd
);
745 scsi_next_command(cmd
);
752 * __scsi_error_from_host_byte - translate SCSI error code into errno
753 * @cmd: SCSI command (unused)
754 * @result: scsi error code
756 * Translate SCSI error code into standard UNIX errno.
758 * -ENOLINK temporary transport failure
759 * -EREMOTEIO permanent target failure, do not retry
760 * -EBADE permanent nexus failure, retry on other path
761 * -ENOSPC No write space available
762 * -ENODATA Medium error
763 * -EIO unspecified I/O error
765 static int __scsi_error_from_host_byte(struct scsi_cmnd
*cmd
, int result
)
769 switch(host_byte(result
)) {
770 case DID_TRANSPORT_FAILFAST
:
773 case DID_TARGET_FAILURE
:
774 set_host_byte(cmd
, DID_OK
);
777 case DID_NEXUS_FAILURE
:
778 set_host_byte(cmd
, DID_OK
);
781 case DID_ALLOC_FAILURE
:
782 set_host_byte(cmd
, DID_OK
);
785 case DID_MEDIUM_ERROR
:
786 set_host_byte(cmd
, DID_OK
);
798 * Function: scsi_io_completion()
800 * Purpose: Completion processing for block device I/O requests.
802 * Arguments: cmd - command that is finished.
804 * Lock status: Assumed that no lock is held upon entry.
808 * Notes: We will finish off the specified number of sectors. If we
809 * are done, the command block will be released and the queue
810 * function will be goosed. If we are not done then we have to
811 * figure out what to do next:
813 * a) We can call scsi_requeue_command(). The request
814 * will be unprepared and put back on the queue. Then
815 * a new command will be created for it. This should
816 * be used if we made forward progress, or if we want
817 * to switch from READ(10) to READ(6) for example.
819 * b) We can call __scsi_queue_insert(). The request will
820 * be put back on the queue and retried using the same
821 * command as before, possibly after a delay.
823 * c) We can call scsi_end_request() with -EIO to fail
824 * the remainder of the request.
826 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
828 int result
= cmd
->result
;
829 struct request_queue
*q
= cmd
->device
->request_queue
;
830 struct request
*req
= cmd
->request
;
832 struct scsi_sense_hdr sshdr
;
834 int sense_deferred
= 0;
835 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
836 ACTION_DELAYED_RETRY
} action
;
837 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
840 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
842 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
845 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
847 if (sense_valid
&& req
->sense
) {
849 * SG_IO wants current and deferred errors
851 int len
= 8 + cmd
->sense_buffer
[7];
853 if (len
> SCSI_SENSE_BUFFERSIZE
)
854 len
= SCSI_SENSE_BUFFERSIZE
;
855 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
856 req
->sense_len
= len
;
859 error
= __scsi_error_from_host_byte(cmd
, result
);
862 * __scsi_error_from_host_byte may have reset the host_byte
864 req
->errors
= cmd
->result
;
866 req
->resid_len
= scsi_get_resid(cmd
);
868 if (scsi_bidi_cmnd(cmd
)) {
870 * Bidi commands Must be complete as a whole,
871 * both sides at once.
873 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
874 if (scsi_end_request(req
, 0, blk_rq_bytes(req
),
875 blk_rq_bytes(req
->next_rq
)))
879 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
881 * Certain non BLOCK_PC requests are commands that don't
882 * actually transfer anything (FLUSH), so cannot use
883 * good_bytes != blk_rq_bytes(req) as the signal for an error.
884 * This sets the error explicitly for the problem case.
886 error
= __scsi_error_from_host_byte(cmd
, result
);
889 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
890 BUG_ON(blk_bidi_rq(req
));
893 * Next deal with any sectors which we were able to correctly
896 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
897 "%u sectors total, %d bytes done.\n",
898 blk_rq_sectors(req
), good_bytes
));
901 * Recovered errors need reporting, but they're always treated
902 * as success, so fiddle the result code here. For BLOCK_PC
903 * we already took a copy of the original into rq->errors which
904 * is what gets returned to the user
906 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
907 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
908 * print since caller wants ATA registers. Only occurs on
909 * SCSI ATA PASS_THROUGH commands when CK_COND=1
911 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
913 else if (!(req
->cmd_flags
& REQ_QUIET
))
914 scsi_print_sense("", cmd
);
916 /* BLOCK_PC may have set error */
921 * If we finished all bytes in the request we are done now.
923 if (!scsi_end_request(req
, error
, good_bytes
, 0))
927 * Kill remainder if no retrys.
929 if (error
&& scsi_noretry_cmd(cmd
)) {
930 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
936 * If there had been no error, but we have leftover bytes in the
937 * requeues just queue the command up again.
942 error
= __scsi_error_from_host_byte(cmd
, result
);
944 if (host_byte(result
) == DID_RESET
) {
945 /* Third party bus reset or reset for error recovery
946 * reasons. Just retry the command and see what
949 action
= ACTION_RETRY
;
950 } else if (sense_valid
&& !sense_deferred
) {
951 switch (sshdr
.sense_key
) {
953 if (cmd
->device
->removable
) {
954 /* Detected disc change. Set a bit
955 * and quietly refuse further access.
957 cmd
->device
->changed
= 1;
958 action
= ACTION_FAIL
;
960 /* Must have been a power glitch, or a
961 * bus reset. Could not have been a
962 * media change, so we just retry the
963 * command and see what happens.
965 action
= ACTION_RETRY
;
968 case ILLEGAL_REQUEST
:
969 /* If we had an ILLEGAL REQUEST returned, then
970 * we may have performed an unsupported
971 * command. The only thing this should be
972 * would be a ten byte read where only a six
973 * byte read was supported. Also, on a system
974 * where READ CAPACITY failed, we may have
975 * read past the end of the disk.
977 if ((cmd
->device
->use_10_for_rw
&&
978 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
979 (cmd
->cmnd
[0] == READ_10
||
980 cmd
->cmnd
[0] == WRITE_10
)) {
981 /* This will issue a new 6-byte command. */
982 cmd
->device
->use_10_for_rw
= 0;
983 action
= ACTION_REPREP
;
984 } else if (sshdr
.asc
== 0x10) /* DIX */ {
985 action
= ACTION_FAIL
;
987 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
988 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
989 action
= ACTION_FAIL
;
992 action
= ACTION_FAIL
;
994 case ABORTED_COMMAND
:
995 action
= ACTION_FAIL
;
996 if (sshdr
.asc
== 0x10) /* DIF */
1000 /* If the device is in the process of becoming
1001 * ready, or has a temporary blockage, retry.
1003 if (sshdr
.asc
== 0x04) {
1004 switch (sshdr
.ascq
) {
1005 case 0x01: /* becoming ready */
1006 case 0x04: /* format in progress */
1007 case 0x05: /* rebuild in progress */
1008 case 0x06: /* recalculation in progress */
1009 case 0x07: /* operation in progress */
1010 case 0x08: /* Long write in progress */
1011 case 0x09: /* self test in progress */
1012 case 0x14: /* space allocation in progress */
1013 action
= ACTION_DELAYED_RETRY
;
1016 action
= ACTION_FAIL
;
1020 action
= ACTION_FAIL
;
1022 case VOLUME_OVERFLOW
:
1023 /* See SSC3rXX or current. */
1024 action
= ACTION_FAIL
;
1027 action
= ACTION_FAIL
;
1031 action
= ACTION_FAIL
;
1033 if (action
!= ACTION_FAIL
&&
1034 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
1035 action
= ACTION_FAIL
;
1039 /* Give up and fail the remainder of the request */
1040 if (!(req
->cmd_flags
& REQ_QUIET
)) {
1041 scsi_print_result(cmd
);
1042 if (driver_byte(result
) & DRIVER_SENSE
)
1043 scsi_print_sense("", cmd
);
1044 scsi_print_command(cmd
);
1046 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
1051 /* Unprep the request and put it back at the head of the queue.
1052 * A new command will be prepared and issued.
1055 cmd
->request
->cmd_flags
&= ~REQ_DONTPREP
;
1056 scsi_mq_uninit_cmd(cmd
);
1057 scsi_mq_requeue_cmd(cmd
);
1059 scsi_release_buffers(cmd
);
1060 scsi_requeue_command(q
, cmd
);
1064 /* Retry the same command immediately */
1065 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
1067 case ACTION_DELAYED_RETRY
:
1068 /* Retry the same command after a delay */
1069 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
1074 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
1080 * If sg table allocation fails, requeue request later.
1082 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
1083 gfp_mask
, req
->mq_ctx
!= NULL
)))
1084 return BLKPREP_DEFER
;
1087 * Next, walk the list, and fill in the addresses and sizes of
1090 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1091 BUG_ON(count
> sdb
->table
.nents
);
1092 sdb
->table
.nents
= count
;
1093 sdb
->length
= blk_rq_bytes(req
);
1098 * Function: scsi_init_io()
1100 * Purpose: SCSI I/O initialize function.
1102 * Arguments: cmd - Command descriptor we wish to initialize
1104 * Returns: 0 on success
1105 * BLKPREP_DEFER if the failure is retryable
1106 * BLKPREP_KILL if the failure is fatal
1108 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
1110 struct scsi_device
*sdev
= cmd
->device
;
1111 struct request
*rq
= cmd
->request
;
1112 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1115 BUG_ON(!rq
->nr_phys_segments
);
1117 error
= scsi_init_sgtable(rq
, &cmd
->sdb
, gfp_mask
);
1121 if (blk_bidi_rq(rq
)) {
1122 if (!rq
->q
->mq_ops
) {
1123 struct scsi_data_buffer
*bidi_sdb
=
1124 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1126 error
= BLKPREP_DEFER
;
1130 rq
->next_rq
->special
= bidi_sdb
;
1133 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
,
1139 if (blk_integrity_rq(rq
)) {
1140 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1143 BUG_ON(prot_sdb
== NULL
);
1144 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1146 if (scsi_alloc_sgtable(prot_sdb
, ivecs
, gfp_mask
, is_mq
)) {
1147 error
= BLKPREP_DEFER
;
1151 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1152 prot_sdb
->table
.sgl
);
1153 BUG_ON(unlikely(count
> ivecs
));
1154 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1156 cmd
->prot_sdb
= prot_sdb
;
1157 cmd
->prot_sdb
->table
.nents
= count
;
1163 scsi_mq_free_sgtables(cmd
);
1165 scsi_release_buffers(cmd
);
1166 cmd
->request
->special
= NULL
;
1167 scsi_put_command(cmd
);
1168 put_device(&sdev
->sdev_gendev
);
1172 EXPORT_SYMBOL(scsi_init_io
);
1174 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1175 struct request
*req
)
1177 struct scsi_cmnd
*cmd
;
1179 if (!req
->special
) {
1180 /* Bail if we can't get a reference to the device */
1181 if (!get_device(&sdev
->sdev_gendev
))
1184 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1185 if (unlikely(!cmd
)) {
1186 put_device(&sdev
->sdev_gendev
);
1194 /* pull a tag out of the request if we have one */
1195 cmd
->tag
= req
->tag
;
1198 cmd
->cmnd
= req
->cmd
;
1199 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1204 static int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1206 struct scsi_cmnd
*cmd
= req
->special
;
1209 * BLOCK_PC requests may transfer data, in which case they must
1210 * a bio attached to them. Or they might contain a SCSI command
1211 * that does not transfer data, in which case they may optionally
1212 * submit a request without an attached bio.
1215 int ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1219 BUG_ON(blk_rq_bytes(req
));
1221 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1224 cmd
->cmd_len
= req
->cmd_len
;
1225 cmd
->transfersize
= blk_rq_bytes(req
);
1226 cmd
->allowed
= req
->retries
;
1231 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1232 * that still need to be translated to SCSI CDBs from the ULD.
1234 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1236 struct scsi_cmnd
*cmd
= req
->special
;
1238 if (unlikely(sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
1239 && sdev
->scsi_dh_data
->scsi_dh
->prep_fn
)) {
1240 int ret
= sdev
->scsi_dh_data
->scsi_dh
->prep_fn(sdev
, req
);
1241 if (ret
!= BLKPREP_OK
)
1245 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1246 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1249 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1251 struct scsi_cmnd
*cmd
= req
->special
;
1253 if (!blk_rq_bytes(req
))
1254 cmd
->sc_data_direction
= DMA_NONE
;
1255 else if (rq_data_dir(req
) == WRITE
)
1256 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1258 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1260 switch (req
->cmd_type
) {
1262 return scsi_setup_fs_cmnd(sdev
, req
);
1263 case REQ_TYPE_BLOCK_PC
:
1264 return scsi_setup_blk_pc_cmnd(sdev
, req
);
1266 return BLKPREP_KILL
;
1271 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1273 int ret
= BLKPREP_OK
;
1276 * If the device is not in running state we will reject some
1279 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1280 switch (sdev
->sdev_state
) {
1282 case SDEV_TRANSPORT_OFFLINE
:
1284 * If the device is offline we refuse to process any
1285 * commands. The device must be brought online
1286 * before trying any recovery commands.
1288 sdev_printk(KERN_ERR
, sdev
,
1289 "rejecting I/O to offline device\n");
1294 * If the device is fully deleted, we refuse to
1295 * process any commands as well.
1297 sdev_printk(KERN_ERR
, sdev
,
1298 "rejecting I/O to dead device\n");
1303 case SDEV_CREATED_BLOCK
:
1305 * If the devices is blocked we defer normal commands.
1307 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1308 ret
= BLKPREP_DEFER
;
1312 * For any other not fully online state we only allow
1313 * special commands. In particular any user initiated
1314 * command is not allowed.
1316 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1325 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1327 struct scsi_device
*sdev
= q
->queuedata
;
1331 req
->errors
= DID_NO_CONNECT
<< 16;
1332 /* release the command and kill it */
1334 struct scsi_cmnd
*cmd
= req
->special
;
1335 scsi_release_buffers(cmd
);
1336 scsi_put_command(cmd
);
1337 put_device(&sdev
->sdev_gendev
);
1338 req
->special
= NULL
;
1343 * If we defer, the blk_peek_request() returns NULL, but the
1344 * queue must be restarted, so we schedule a callback to happen
1347 if (atomic_read(&sdev
->device_busy
) == 0)
1348 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1351 req
->cmd_flags
|= REQ_DONTPREP
;
1357 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1359 struct scsi_device
*sdev
= q
->queuedata
;
1360 struct scsi_cmnd
*cmd
;
1363 ret
= scsi_prep_state_check(sdev
, req
);
1364 if (ret
!= BLKPREP_OK
)
1367 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1368 if (unlikely(!cmd
)) {
1369 ret
= BLKPREP_DEFER
;
1373 ret
= scsi_setup_cmnd(sdev
, req
);
1375 return scsi_prep_return(q
, req
, ret
);
1378 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1380 scsi_uninit_cmd(req
->special
);
1384 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1387 * Called with the queue_lock held.
1389 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1390 struct scsi_device
*sdev
)
1394 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1395 if (atomic_read(&sdev
->device_blocked
)) {
1400 * unblock after device_blocked iterates to zero
1402 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1404 * For the MQ case we take care of this in the caller.
1407 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1410 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1411 "unblocking device at zero depth\n"));
1414 if (busy
>= sdev
->queue_depth
)
1419 atomic_dec(&sdev
->device_busy
);
1424 * scsi_target_queue_ready: checks if there we can send commands to target
1425 * @sdev: scsi device on starget to check.
1427 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1428 struct scsi_device
*sdev
)
1430 struct scsi_target
*starget
= scsi_target(sdev
);
1433 if (starget
->single_lun
) {
1434 spin_lock_irq(shost
->host_lock
);
1435 if (starget
->starget_sdev_user
&&
1436 starget
->starget_sdev_user
!= sdev
) {
1437 spin_unlock_irq(shost
->host_lock
);
1440 starget
->starget_sdev_user
= sdev
;
1441 spin_unlock_irq(shost
->host_lock
);
1444 if (starget
->can_queue
<= 0)
1447 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1448 if (atomic_read(&starget
->target_blocked
) > 0) {
1453 * unblock after target_blocked iterates to zero
1455 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1458 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1459 "unblocking target at zero depth\n"));
1462 if (busy
>= starget
->can_queue
)
1468 spin_lock_irq(shost
->host_lock
);
1469 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1470 spin_unlock_irq(shost
->host_lock
);
1472 if (starget
->can_queue
> 0)
1473 atomic_dec(&starget
->target_busy
);
1478 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1479 * return 0. We must end up running the queue again whenever 0 is
1480 * returned, else IO can hang.
1482 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1483 struct Scsi_Host
*shost
,
1484 struct scsi_device
*sdev
)
1488 if (scsi_host_in_recovery(shost
))
1491 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1492 if (atomic_read(&shost
->host_blocked
) > 0) {
1497 * unblock after host_blocked iterates to zero
1499 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1503 shost_printk(KERN_INFO
, shost
,
1504 "unblocking host at zero depth\n"));
1507 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1509 if (shost
->host_self_blocked
)
1512 /* We're OK to process the command, so we can't be starved */
1513 if (!list_empty(&sdev
->starved_entry
)) {
1514 spin_lock_irq(shost
->host_lock
);
1515 if (!list_empty(&sdev
->starved_entry
))
1516 list_del_init(&sdev
->starved_entry
);
1517 spin_unlock_irq(shost
->host_lock
);
1523 spin_lock_irq(shost
->host_lock
);
1524 if (list_empty(&sdev
->starved_entry
))
1525 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1526 spin_unlock_irq(shost
->host_lock
);
1528 atomic_dec(&shost
->host_busy
);
1533 * Busy state exporting function for request stacking drivers.
1535 * For efficiency, no lock is taken to check the busy state of
1536 * shost/starget/sdev, since the returned value is not guaranteed and
1537 * may be changed after request stacking drivers call the function,
1538 * regardless of taking lock or not.
1540 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1541 * needs to return 'not busy'. Otherwise, request stacking drivers
1542 * may hold requests forever.
1544 static int scsi_lld_busy(struct request_queue
*q
)
1546 struct scsi_device
*sdev
= q
->queuedata
;
1547 struct Scsi_Host
*shost
;
1549 if (blk_queue_dying(q
))
1555 * Ignore host/starget busy state.
1556 * Since block layer does not have a concept of fairness across
1557 * multiple queues, congestion of host/starget needs to be handled
1560 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1567 * Kill a request for a dead device
1569 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1571 struct scsi_cmnd
*cmd
= req
->special
;
1572 struct scsi_device
*sdev
;
1573 struct scsi_target
*starget
;
1574 struct Scsi_Host
*shost
;
1576 blk_start_request(req
);
1578 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1581 starget
= scsi_target(sdev
);
1583 scsi_init_cmd_errh(cmd
);
1584 cmd
->result
= DID_NO_CONNECT
<< 16;
1585 atomic_inc(&cmd
->device
->iorequest_cnt
);
1588 * SCSI request completion path will do scsi_device_unbusy(),
1589 * bump busy counts. To bump the counters, we need to dance
1590 * with the locks as normal issue path does.
1592 atomic_inc(&sdev
->device_busy
);
1593 atomic_inc(&shost
->host_busy
);
1594 if (starget
->can_queue
> 0)
1595 atomic_inc(&starget
->target_busy
);
1597 blk_complete_request(req
);
1600 static void scsi_softirq_done(struct request
*rq
)
1602 struct scsi_cmnd
*cmd
= rq
->special
;
1603 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1606 INIT_LIST_HEAD(&cmd
->eh_entry
);
1608 atomic_inc(&cmd
->device
->iodone_cnt
);
1610 atomic_inc(&cmd
->device
->ioerr_cnt
);
1612 disposition
= scsi_decide_disposition(cmd
);
1613 if (disposition
!= SUCCESS
&&
1614 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1615 sdev_printk(KERN_ERR
, cmd
->device
,
1616 "timing out command, waited %lus\n",
1618 disposition
= SUCCESS
;
1621 scsi_log_completion(cmd
, disposition
);
1623 switch (disposition
) {
1625 scsi_finish_command(cmd
);
1628 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1630 case ADD_TO_MLQUEUE
:
1631 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1634 if (!scsi_eh_scmd_add(cmd
, 0))
1635 scsi_finish_command(cmd
);
1640 * scsi_done - Invoke completion on finished SCSI command.
1641 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1642 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1644 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1645 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1646 * calls blk_complete_request() for further processing.
1648 * This function is interrupt context safe.
1650 static void scsi_done(struct scsi_cmnd
*cmd
)
1652 trace_scsi_dispatch_cmd_done(cmd
);
1653 blk_complete_request(cmd
->request
);
1657 * Function: scsi_request_fn()
1659 * Purpose: Main strategy routine for SCSI.
1661 * Arguments: q - Pointer to actual queue.
1665 * Lock status: IO request lock assumed to be held when called.
1667 static void scsi_request_fn(struct request_queue
*q
)
1668 __releases(q
->queue_lock
)
1669 __acquires(q
->queue_lock
)
1671 struct scsi_device
*sdev
= q
->queuedata
;
1672 struct Scsi_Host
*shost
;
1673 struct scsi_cmnd
*cmd
;
1674 struct request
*req
;
1677 * To start with, we keep looping until the queue is empty, or until
1678 * the host is no longer able to accept any more requests.
1684 * get next queueable request. We do this early to make sure
1685 * that the request is fully prepared even if we cannot
1688 req
= blk_peek_request(q
);
1692 if (unlikely(!scsi_device_online(sdev
))) {
1693 sdev_printk(KERN_ERR
, sdev
,
1694 "rejecting I/O to offline device\n");
1695 scsi_kill_request(req
, q
);
1699 if (!scsi_dev_queue_ready(q
, sdev
))
1703 * Remove the request from the request list.
1705 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1706 blk_start_request(req
);
1708 spin_unlock_irq(q
->queue_lock
);
1710 if (unlikely(cmd
== NULL
)) {
1711 printk(KERN_CRIT
"impossible request in %s.\n"
1712 "please mail a stack trace to "
1713 "linux-scsi@vger.kernel.org\n",
1715 blk_dump_rq_flags(req
, "foo");
1720 * We hit this when the driver is using a host wide
1721 * tag map. For device level tag maps the queue_depth check
1722 * in the device ready fn would prevent us from trying
1723 * to allocate a tag. Since the map is a shared host resource
1724 * we add the dev to the starved list so it eventually gets
1725 * a run when a tag is freed.
1727 if (blk_queue_tagged(q
) && !blk_rq_tagged(req
)) {
1728 spin_lock_irq(shost
->host_lock
);
1729 if (list_empty(&sdev
->starved_entry
))
1730 list_add_tail(&sdev
->starved_entry
,
1731 &shost
->starved_list
);
1732 spin_unlock_irq(shost
->host_lock
);
1736 if (!scsi_target_queue_ready(shost
, sdev
))
1739 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1740 goto host_not_ready
;
1743 * Finally, initialize any error handling parameters, and set up
1744 * the timers for timeouts.
1746 scsi_init_cmd_errh(cmd
);
1749 * Dispatch the command to the low-level driver.
1751 cmd
->scsi_done
= scsi_done
;
1752 rtn
= scsi_dispatch_cmd(cmd
);
1754 scsi_queue_insert(cmd
, rtn
);
1755 spin_lock_irq(q
->queue_lock
);
1758 spin_lock_irq(q
->queue_lock
);
1764 if (scsi_target(sdev
)->can_queue
> 0)
1765 atomic_dec(&scsi_target(sdev
)->target_busy
);
1768 * lock q, handle tag, requeue req, and decrement device_busy. We
1769 * must return with queue_lock held.
1771 * Decrementing device_busy without checking it is OK, as all such
1772 * cases (host limits or settings) should run the queue at some
1775 spin_lock_irq(q
->queue_lock
);
1776 blk_requeue_request(q
, req
);
1777 atomic_dec(&sdev
->device_busy
);
1779 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1780 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1783 static inline int prep_to_mq(int ret
)
1789 return BLK_MQ_RQ_QUEUE_BUSY
;
1791 return BLK_MQ_RQ_QUEUE_ERROR
;
1795 static int scsi_mq_prep_fn(struct request
*req
)
1797 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1798 struct scsi_device
*sdev
= req
->q
->queuedata
;
1799 struct Scsi_Host
*shost
= sdev
->host
;
1800 unsigned char *sense_buf
= cmd
->sense_buffer
;
1801 struct scatterlist
*sg
;
1803 memset(cmd
, 0, sizeof(struct scsi_cmnd
));
1809 cmd
->sense_buffer
= sense_buf
;
1811 cmd
->tag
= req
->tag
;
1813 cmd
->cmnd
= req
->cmd
;
1814 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1816 INIT_LIST_HEAD(&cmd
->list
);
1817 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1818 cmd
->jiffies_at_alloc
= jiffies
;
1820 if (shost
->use_cmd_list
) {
1821 spin_lock_irq(&sdev
->list_lock
);
1822 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1823 spin_unlock_irq(&sdev
->list_lock
);
1826 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1827 cmd
->sdb
.table
.sgl
= sg
;
1829 if (scsi_host_get_prot(shost
)) {
1830 cmd
->prot_sdb
= (void *)sg
+
1831 shost
->sg_tablesize
* sizeof(struct scatterlist
);
1832 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1834 cmd
->prot_sdb
->table
.sgl
=
1835 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1838 if (blk_bidi_rq(req
)) {
1839 struct request
*next_rq
= req
->next_rq
;
1840 struct scsi_data_buffer
*bidi_sdb
= blk_mq_rq_to_pdu(next_rq
);
1842 memset(bidi_sdb
, 0, sizeof(struct scsi_data_buffer
));
1843 bidi_sdb
->table
.sgl
=
1844 (struct scatterlist
*)(bidi_sdb
+ 1);
1846 next_rq
->special
= bidi_sdb
;
1849 return scsi_setup_cmnd(sdev
, req
);
1852 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1854 trace_scsi_dispatch_cmd_done(cmd
);
1855 blk_mq_complete_request(cmd
->request
);
1858 static int scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
, struct request
*req
)
1860 struct request_queue
*q
= req
->q
;
1861 struct scsi_device
*sdev
= q
->queuedata
;
1862 struct Scsi_Host
*shost
= sdev
->host
;
1863 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1867 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
1871 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1872 if (!get_device(&sdev
->sdev_gendev
))
1875 if (!scsi_dev_queue_ready(q
, sdev
))
1876 goto out_put_device
;
1877 if (!scsi_target_queue_ready(shost
, sdev
))
1878 goto out_dec_device_busy
;
1879 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1880 goto out_dec_target_busy
;
1882 if (!(req
->cmd_flags
& REQ_DONTPREP
)) {
1883 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
1885 goto out_dec_host_busy
;
1886 req
->cmd_flags
|= REQ_DONTPREP
;
1889 scsi_init_cmd_errh(cmd
);
1890 cmd
->scsi_done
= scsi_mq_done
;
1892 reason
= scsi_dispatch_cmd(cmd
);
1894 scsi_set_blocked(cmd
, reason
);
1895 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1896 goto out_dec_host_busy
;
1899 return BLK_MQ_RQ_QUEUE_OK
;
1902 atomic_dec(&shost
->host_busy
);
1903 out_dec_target_busy
:
1904 if (scsi_target(sdev
)->can_queue
> 0)
1905 atomic_dec(&scsi_target(sdev
)->target_busy
);
1906 out_dec_device_busy
:
1907 atomic_dec(&sdev
->device_busy
);
1909 put_device(&sdev
->sdev_gendev
);
1912 case BLK_MQ_RQ_QUEUE_BUSY
:
1913 blk_mq_stop_hw_queue(hctx
);
1914 if (atomic_read(&sdev
->device_busy
) == 0 &&
1915 !scsi_device_blocked(sdev
))
1916 blk_mq_delay_queue(hctx
, SCSI_QUEUE_DELAY
);
1918 case BLK_MQ_RQ_QUEUE_ERROR
:
1920 * Make sure to release all allocated ressources when
1921 * we hit an error, as we will never see this command
1924 if (req
->cmd_flags
& REQ_DONTPREP
)
1925 scsi_mq_uninit_cmd(cmd
);
1933 static int scsi_init_request(void *data
, struct request
*rq
,
1934 unsigned int hctx_idx
, unsigned int request_idx
,
1935 unsigned int numa_node
)
1937 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1939 cmd
->sense_buffer
= kzalloc_node(SCSI_SENSE_BUFFERSIZE
, GFP_KERNEL
,
1941 if (!cmd
->sense_buffer
)
1946 static void scsi_exit_request(void *data
, struct request
*rq
,
1947 unsigned int hctx_idx
, unsigned int request_idx
)
1949 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1951 kfree(cmd
->sense_buffer
);
1954 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1956 struct device
*host_dev
;
1957 u64 bounce_limit
= 0xffffffff;
1959 if (shost
->unchecked_isa_dma
)
1960 return BLK_BOUNCE_ISA
;
1962 * Platforms with virtual-DMA translation
1963 * hardware have no practical limit.
1965 if (!PCI_DMA_BUS_IS_PHYS
)
1966 return BLK_BOUNCE_ANY
;
1968 host_dev
= scsi_get_device(shost
);
1969 if (host_dev
&& host_dev
->dma_mask
)
1970 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
1972 return bounce_limit
;
1975 static void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
1977 struct device
*dev
= shost
->dma_dev
;
1980 * this limit is imposed by hardware restrictions
1982 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
1983 SCSI_MAX_SG_CHAIN_SEGMENTS
));
1985 if (scsi_host_prot_dma(shost
)) {
1986 shost
->sg_prot_tablesize
=
1987 min_not_zero(shost
->sg_prot_tablesize
,
1988 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
1989 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
1990 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
1993 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
1994 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1995 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1996 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1998 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2000 if (!shost
->use_clustering
)
2001 q
->limits
.cluster
= 0;
2004 * set a reasonable default alignment on word boundaries: the
2005 * host and device may alter it using
2006 * blk_queue_update_dma_alignment() later.
2008 blk_queue_dma_alignment(q
, 0x03);
2011 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
2012 request_fn_proc
*request_fn
)
2014 struct request_queue
*q
;
2016 q
= blk_init_queue(request_fn
, NULL
);
2019 __scsi_init_queue(shost
, q
);
2022 EXPORT_SYMBOL(__scsi_alloc_queue
);
2024 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
2026 struct request_queue
*q
;
2028 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
2032 blk_queue_prep_rq(q
, scsi_prep_fn
);
2033 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2034 blk_queue_softirq_done(q
, scsi_softirq_done
);
2035 blk_queue_rq_timed_out(q
, scsi_times_out
);
2036 blk_queue_lld_busy(q
, scsi_lld_busy
);
2040 static struct blk_mq_ops scsi_mq_ops
= {
2041 .map_queue
= blk_mq_map_queue
,
2042 .queue_rq
= scsi_queue_rq
,
2043 .complete
= scsi_softirq_done
,
2044 .timeout
= scsi_times_out
,
2045 .init_request
= scsi_init_request
,
2046 .exit_request
= scsi_exit_request
,
2049 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2051 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2052 if (IS_ERR(sdev
->request_queue
))
2055 sdev
->request_queue
->queuedata
= sdev
;
2056 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2057 return sdev
->request_queue
;
2060 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2062 unsigned int cmd_size
, sgl_size
, tbl_size
;
2064 tbl_size
= shost
->sg_tablesize
;
2065 if (tbl_size
> SCSI_MAX_SG_SEGMENTS
)
2066 tbl_size
= SCSI_MAX_SG_SEGMENTS
;
2067 sgl_size
= tbl_size
* sizeof(struct scatterlist
);
2068 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2069 if (scsi_host_get_prot(shost
))
2070 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2072 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2073 shost
->tag_set
.ops
= &scsi_mq_ops
;
2074 shost
->tag_set
.nr_hw_queues
= 1;
2075 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2076 shost
->tag_set
.cmd_size
= cmd_size
;
2077 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2078 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2079 shost
->tag_set
.driver_data
= shost
;
2081 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2084 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2086 blk_mq_free_tag_set(&shost
->tag_set
);
2090 * Function: scsi_block_requests()
2092 * Purpose: Utility function used by low-level drivers to prevent further
2093 * commands from being queued to the device.
2095 * Arguments: shost - Host in question
2099 * Lock status: No locks are assumed held.
2101 * Notes: There is no timer nor any other means by which the requests
2102 * get unblocked other than the low-level driver calling
2103 * scsi_unblock_requests().
2105 void scsi_block_requests(struct Scsi_Host
*shost
)
2107 shost
->host_self_blocked
= 1;
2109 EXPORT_SYMBOL(scsi_block_requests
);
2112 * Function: scsi_unblock_requests()
2114 * Purpose: Utility function used by low-level drivers to allow further
2115 * commands from being queued to the device.
2117 * Arguments: shost - Host in question
2121 * Lock status: No locks are assumed held.
2123 * Notes: There is no timer nor any other means by which the requests
2124 * get unblocked other than the low-level driver calling
2125 * scsi_unblock_requests().
2127 * This is done as an API function so that changes to the
2128 * internals of the scsi mid-layer won't require wholesale
2129 * changes to drivers that use this feature.
2131 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2133 shost
->host_self_blocked
= 0;
2134 scsi_run_host_queues(shost
);
2136 EXPORT_SYMBOL(scsi_unblock_requests
);
2138 int __init
scsi_init_queue(void)
2142 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2143 sizeof(struct scsi_data_buffer
),
2145 if (!scsi_sdb_cache
) {
2146 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2150 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2151 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2152 int size
= sgp
->size
* sizeof(struct scatterlist
);
2154 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
2155 SLAB_HWCACHE_ALIGN
, NULL
);
2157 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
2162 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
2165 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
2174 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2175 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2177 mempool_destroy(sgp
->pool
);
2179 kmem_cache_destroy(sgp
->slab
);
2181 kmem_cache_destroy(scsi_sdb_cache
);
2186 void scsi_exit_queue(void)
2190 kmem_cache_destroy(scsi_sdb_cache
);
2192 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2193 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2194 mempool_destroy(sgp
->pool
);
2195 kmem_cache_destroy(sgp
->slab
);
2200 * scsi_mode_select - issue a mode select
2201 * @sdev: SCSI device to be queried
2202 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2203 * @sp: Save page bit (0 == don't save, 1 == save)
2204 * @modepage: mode page being requested
2205 * @buffer: request buffer (may not be smaller than eight bytes)
2206 * @len: length of request buffer.
2207 * @timeout: command timeout
2208 * @retries: number of retries before failing
2209 * @data: returns a structure abstracting the mode header data
2210 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2211 * must be SCSI_SENSE_BUFFERSIZE big.
2213 * Returns zero if successful; negative error number or scsi
2218 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2219 unsigned char *buffer
, int len
, int timeout
, int retries
,
2220 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2222 unsigned char cmd
[10];
2223 unsigned char *real_buffer
;
2226 memset(cmd
, 0, sizeof(cmd
));
2227 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2229 if (sdev
->use_10_for_ms
) {
2232 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2235 memcpy(real_buffer
+ 8, buffer
, len
);
2239 real_buffer
[2] = data
->medium_type
;
2240 real_buffer
[3] = data
->device_specific
;
2241 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2243 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2244 real_buffer
[7] = data
->block_descriptor_length
;
2246 cmd
[0] = MODE_SELECT_10
;
2250 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2254 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2257 memcpy(real_buffer
+ 4, buffer
, len
);
2260 real_buffer
[1] = data
->medium_type
;
2261 real_buffer
[2] = data
->device_specific
;
2262 real_buffer
[3] = data
->block_descriptor_length
;
2265 cmd
[0] = MODE_SELECT
;
2269 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2270 sshdr
, timeout
, retries
, NULL
);
2274 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2277 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2278 * @sdev: SCSI device to be queried
2279 * @dbd: set if mode sense will allow block descriptors to be returned
2280 * @modepage: mode page being requested
2281 * @buffer: request buffer (may not be smaller than eight bytes)
2282 * @len: length of request buffer.
2283 * @timeout: command timeout
2284 * @retries: number of retries before failing
2285 * @data: returns a structure abstracting the mode header data
2286 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2287 * must be SCSI_SENSE_BUFFERSIZE big.
2289 * Returns zero if unsuccessful, or the header offset (either 4
2290 * or 8 depending on whether a six or ten byte command was
2291 * issued) if successful.
2294 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2295 unsigned char *buffer
, int len
, int timeout
, int retries
,
2296 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2298 unsigned char cmd
[12];
2302 struct scsi_sense_hdr my_sshdr
;
2304 memset(data
, 0, sizeof(*data
));
2305 memset(&cmd
[0], 0, 12);
2306 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2309 /* caller might not be interested in sense, but we need it */
2314 use_10_for_ms
= sdev
->use_10_for_ms
;
2316 if (use_10_for_ms
) {
2320 cmd
[0] = MODE_SENSE_10
;
2327 cmd
[0] = MODE_SENSE
;
2332 memset(buffer
, 0, len
);
2334 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2335 sshdr
, timeout
, retries
, NULL
);
2337 /* This code looks awful: what it's doing is making sure an
2338 * ILLEGAL REQUEST sense return identifies the actual command
2339 * byte as the problem. MODE_SENSE commands can return
2340 * ILLEGAL REQUEST if the code page isn't supported */
2342 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2343 (driver_byte(result
) & DRIVER_SENSE
)) {
2344 if (scsi_sense_valid(sshdr
)) {
2345 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2346 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2348 * Invalid command operation code
2350 sdev
->use_10_for_ms
= 0;
2356 if(scsi_status_is_good(result
)) {
2357 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2358 (modepage
== 6 || modepage
== 8))) {
2359 /* Initio breakage? */
2362 data
->medium_type
= 0;
2363 data
->device_specific
= 0;
2365 data
->block_descriptor_length
= 0;
2366 } else if(use_10_for_ms
) {
2367 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2368 data
->medium_type
= buffer
[2];
2369 data
->device_specific
= buffer
[3];
2370 data
->longlba
= buffer
[4] & 0x01;
2371 data
->block_descriptor_length
= buffer
[6]*256
2374 data
->length
= buffer
[0] + 1;
2375 data
->medium_type
= buffer
[1];
2376 data
->device_specific
= buffer
[2];
2377 data
->block_descriptor_length
= buffer
[3];
2379 data
->header_length
= header_length
;
2384 EXPORT_SYMBOL(scsi_mode_sense
);
2387 * scsi_test_unit_ready - test if unit is ready
2388 * @sdev: scsi device to change the state of.
2389 * @timeout: command timeout
2390 * @retries: number of retries before failing
2391 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2392 * returning sense. Make sure that this is cleared before passing
2395 * Returns zero if unsuccessful or an error if TUR failed. For
2396 * removable media, UNIT_ATTENTION sets ->changed flag.
2399 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2400 struct scsi_sense_hdr
*sshdr_external
)
2403 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2405 struct scsi_sense_hdr
*sshdr
;
2408 if (!sshdr_external
)
2409 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
2411 sshdr
= sshdr_external
;
2413 /* try to eat the UNIT_ATTENTION if there are enough retries */
2415 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2416 timeout
, retries
, NULL
);
2417 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2418 sshdr
->sense_key
== UNIT_ATTENTION
)
2420 } while (scsi_sense_valid(sshdr
) &&
2421 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2423 if (!sshdr_external
)
2427 EXPORT_SYMBOL(scsi_test_unit_ready
);
2430 * scsi_device_set_state - Take the given device through the device state model.
2431 * @sdev: scsi device to change the state of.
2432 * @state: state to change to.
2434 * Returns zero if unsuccessful or an error if the requested
2435 * transition is illegal.
2438 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2440 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2442 if (state
== oldstate
)
2448 case SDEV_CREATED_BLOCK
:
2459 case SDEV_TRANSPORT_OFFLINE
:
2472 case SDEV_TRANSPORT_OFFLINE
:
2480 case SDEV_TRANSPORT_OFFLINE
:
2495 case SDEV_CREATED_BLOCK
:
2502 case SDEV_CREATED_BLOCK
:
2517 case SDEV_TRANSPORT_OFFLINE
:
2530 case SDEV_TRANSPORT_OFFLINE
:
2532 case SDEV_CREATED_BLOCK
:
2540 sdev
->sdev_state
= state
;
2544 SCSI_LOG_ERROR_RECOVERY(1,
2545 sdev_printk(KERN_ERR
, sdev
,
2546 "Illegal state transition %s->%s",
2547 scsi_device_state_name(oldstate
),
2548 scsi_device_state_name(state
))
2552 EXPORT_SYMBOL(scsi_device_set_state
);
2555 * sdev_evt_emit - emit a single SCSI device uevent
2556 * @sdev: associated SCSI device
2557 * @evt: event to emit
2559 * Send a single uevent (scsi_event) to the associated scsi_device.
2561 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2566 switch (evt
->evt_type
) {
2567 case SDEV_EVT_MEDIA_CHANGE
:
2568 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2570 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2571 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2573 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2574 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2576 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2577 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2579 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2580 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2582 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2583 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2592 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2596 * sdev_evt_thread - send a uevent for each scsi event
2597 * @work: work struct for scsi_device
2599 * Dispatch queued events to their associated scsi_device kobjects
2602 void scsi_evt_thread(struct work_struct
*work
)
2604 struct scsi_device
*sdev
;
2605 enum scsi_device_event evt_type
;
2606 LIST_HEAD(event_list
);
2608 sdev
= container_of(work
, struct scsi_device
, event_work
);
2610 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2611 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2612 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2615 struct scsi_event
*evt
;
2616 struct list_head
*this, *tmp
;
2617 unsigned long flags
;
2619 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2620 list_splice_init(&sdev
->event_list
, &event_list
);
2621 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2623 if (list_empty(&event_list
))
2626 list_for_each_safe(this, tmp
, &event_list
) {
2627 evt
= list_entry(this, struct scsi_event
, node
);
2628 list_del(&evt
->node
);
2629 scsi_evt_emit(sdev
, evt
);
2636 * sdev_evt_send - send asserted event to uevent thread
2637 * @sdev: scsi_device event occurred on
2638 * @evt: event to send
2640 * Assert scsi device event asynchronously.
2642 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2644 unsigned long flags
;
2647 /* FIXME: currently this check eliminates all media change events
2648 * for polled devices. Need to update to discriminate between AN
2649 * and polled events */
2650 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2656 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2657 list_add_tail(&evt
->node
, &sdev
->event_list
);
2658 schedule_work(&sdev
->event_work
);
2659 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2661 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2664 * sdev_evt_alloc - allocate a new scsi event
2665 * @evt_type: type of event to allocate
2666 * @gfpflags: GFP flags for allocation
2668 * Allocates and returns a new scsi_event.
2670 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2673 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2677 evt
->evt_type
= evt_type
;
2678 INIT_LIST_HEAD(&evt
->node
);
2680 /* evt_type-specific initialization, if any */
2682 case SDEV_EVT_MEDIA_CHANGE
:
2683 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2684 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2685 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2686 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2687 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2695 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2698 * sdev_evt_send_simple - send asserted event to uevent thread
2699 * @sdev: scsi_device event occurred on
2700 * @evt_type: type of event to send
2701 * @gfpflags: GFP flags for allocation
2703 * Assert scsi device event asynchronously, given an event type.
2705 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2706 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2708 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2710 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2715 sdev_evt_send(sdev
, evt
);
2717 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2720 * scsi_device_quiesce - Block user issued commands.
2721 * @sdev: scsi device to quiesce.
2723 * This works by trying to transition to the SDEV_QUIESCE state
2724 * (which must be a legal transition). When the device is in this
2725 * state, only special requests will be accepted, all others will
2726 * be deferred. Since special requests may also be requeued requests,
2727 * a successful return doesn't guarantee the device will be
2728 * totally quiescent.
2730 * Must be called with user context, may sleep.
2732 * Returns zero if unsuccessful or an error if not.
2735 scsi_device_quiesce(struct scsi_device
*sdev
)
2737 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2741 scsi_run_queue(sdev
->request_queue
);
2742 while (atomic_read(&sdev
->device_busy
)) {
2743 msleep_interruptible(200);
2744 scsi_run_queue(sdev
->request_queue
);
2748 EXPORT_SYMBOL(scsi_device_quiesce
);
2751 * scsi_device_resume - Restart user issued commands to a quiesced device.
2752 * @sdev: scsi device to resume.
2754 * Moves the device from quiesced back to running and restarts the
2757 * Must be called with user context, may sleep.
2759 void scsi_device_resume(struct scsi_device
*sdev
)
2761 /* check if the device state was mutated prior to resume, and if
2762 * so assume the state is being managed elsewhere (for example
2763 * device deleted during suspend)
2765 if (sdev
->sdev_state
!= SDEV_QUIESCE
||
2766 scsi_device_set_state(sdev
, SDEV_RUNNING
))
2768 scsi_run_queue(sdev
->request_queue
);
2770 EXPORT_SYMBOL(scsi_device_resume
);
2773 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2775 scsi_device_quiesce(sdev
);
2779 scsi_target_quiesce(struct scsi_target
*starget
)
2781 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2783 EXPORT_SYMBOL(scsi_target_quiesce
);
2786 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2788 scsi_device_resume(sdev
);
2792 scsi_target_resume(struct scsi_target
*starget
)
2794 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2796 EXPORT_SYMBOL(scsi_target_resume
);
2799 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2800 * @sdev: device to block
2802 * Block request made by scsi lld's to temporarily stop all
2803 * scsi commands on the specified device. Called from interrupt
2804 * or normal process context.
2806 * Returns zero if successful or error if not
2809 * This routine transitions the device to the SDEV_BLOCK state
2810 * (which must be a legal transition). When the device is in this
2811 * state, all commands are deferred until the scsi lld reenables
2812 * the device with scsi_device_unblock or device_block_tmo fires.
2815 scsi_internal_device_block(struct scsi_device
*sdev
)
2817 struct request_queue
*q
= sdev
->request_queue
;
2818 unsigned long flags
;
2821 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2823 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2830 * The device has transitioned to SDEV_BLOCK. Stop the
2831 * block layer from calling the midlayer with this device's
2835 blk_mq_stop_hw_queues(q
);
2837 spin_lock_irqsave(q
->queue_lock
, flags
);
2839 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2844 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2847 * scsi_internal_device_unblock - resume a device after a block request
2848 * @sdev: device to resume
2849 * @new_state: state to set devices to after unblocking
2851 * Called by scsi lld's or the midlayer to restart the device queue
2852 * for the previously suspended scsi device. Called from interrupt or
2853 * normal process context.
2855 * Returns zero if successful or error if not.
2858 * This routine transitions the device to the SDEV_RUNNING state
2859 * or to one of the offline states (which must be a legal transition)
2860 * allowing the midlayer to goose the queue for this device.
2863 scsi_internal_device_unblock(struct scsi_device
*sdev
,
2864 enum scsi_device_state new_state
)
2866 struct request_queue
*q
= sdev
->request_queue
;
2867 unsigned long flags
;
2870 * Try to transition the scsi device to SDEV_RUNNING or one of the
2871 * offlined states and goose the device queue if successful.
2873 if ((sdev
->sdev_state
== SDEV_BLOCK
) ||
2874 (sdev
->sdev_state
== SDEV_TRANSPORT_OFFLINE
))
2875 sdev
->sdev_state
= new_state
;
2876 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
) {
2877 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2878 new_state
== SDEV_OFFLINE
)
2879 sdev
->sdev_state
= new_state
;
2881 sdev
->sdev_state
= SDEV_CREATED
;
2882 } else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2883 sdev
->sdev_state
!= SDEV_OFFLINE
)
2887 blk_mq_start_stopped_hw_queues(q
, false);
2889 spin_lock_irqsave(q
->queue_lock
, flags
);
2891 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2896 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2899 device_block(struct scsi_device
*sdev
, void *data
)
2901 scsi_internal_device_block(sdev
);
2905 target_block(struct device
*dev
, void *data
)
2907 if (scsi_is_target_device(dev
))
2908 starget_for_each_device(to_scsi_target(dev
), NULL
,
2914 scsi_target_block(struct device
*dev
)
2916 if (scsi_is_target_device(dev
))
2917 starget_for_each_device(to_scsi_target(dev
), NULL
,
2920 device_for_each_child(dev
, NULL
, target_block
);
2922 EXPORT_SYMBOL_GPL(scsi_target_block
);
2925 device_unblock(struct scsi_device
*sdev
, void *data
)
2927 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2931 target_unblock(struct device
*dev
, void *data
)
2933 if (scsi_is_target_device(dev
))
2934 starget_for_each_device(to_scsi_target(dev
), data
,
2940 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2942 if (scsi_is_target_device(dev
))
2943 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2946 device_for_each_child(dev
, &new_state
, target_unblock
);
2948 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2951 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2952 * @sgl: scatter-gather list
2953 * @sg_count: number of segments in sg
2954 * @offset: offset in bytes into sg, on return offset into the mapped area
2955 * @len: bytes to map, on return number of bytes mapped
2957 * Returns virtual address of the start of the mapped page
2959 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2960 size_t *offset
, size_t *len
)
2963 size_t sg_len
= 0, len_complete
= 0;
2964 struct scatterlist
*sg
;
2967 WARN_ON(!irqs_disabled());
2969 for_each_sg(sgl
, sg
, sg_count
, i
) {
2970 len_complete
= sg_len
; /* Complete sg-entries */
2971 sg_len
+= sg
->length
;
2972 if (sg_len
> *offset
)
2976 if (unlikely(i
== sg_count
)) {
2977 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2979 __func__
, sg_len
, *offset
, sg_count
);
2984 /* Offset starting from the beginning of first page in this sg-entry */
2985 *offset
= *offset
- len_complete
+ sg
->offset
;
2987 /* Assumption: contiguous pages can be accessed as "page + i" */
2988 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2989 *offset
&= ~PAGE_MASK
;
2991 /* Bytes in this sg-entry from *offset to the end of the page */
2992 sg_len
= PAGE_SIZE
- *offset
;
2996 return kmap_atomic(page
);
2998 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3001 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3002 * @virt: virtual address to be unmapped
3004 void scsi_kunmap_atomic_sg(void *virt
)
3006 kunmap_atomic(virt
);
3008 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3010 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3012 atomic_inc(&sdev
->disk_events_disable_depth
);
3014 EXPORT_SYMBOL(sdev_disable_disk_events
);
3016 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3018 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3020 atomic_dec(&sdev
->disk_events_disable_depth
);
3022 EXPORT_SYMBOL(sdev_enable_disk_events
);