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
;
650 BUG_ON(list_empty(&cmd
->list
));
652 scsi_mq_free_sgtables(cmd
);
653 scsi_uninit_cmd(cmd
);
655 spin_lock_irqsave(&sdev
->list_lock
, flags
);
656 list_del_init(&cmd
->list
);
657 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
661 * Function: scsi_release_buffers()
663 * Purpose: Free resources allocate for a scsi_command.
665 * Arguments: cmd - command that we are bailing.
667 * Lock status: Assumed that no lock is held upon entry.
671 * Notes: In the event that an upper level driver rejects a
672 * command, we must release resources allocated during
673 * the __init_io() function. Primarily this would involve
674 * the scatter-gather table.
676 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
678 if (cmd
->sdb
.table
.nents
)
679 scsi_free_sgtable(&cmd
->sdb
, false);
681 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
683 if (scsi_prot_sg_count(cmd
))
684 scsi_free_sgtable(cmd
->prot_sdb
, false);
687 static void scsi_release_bidi_buffers(struct scsi_cmnd
*cmd
)
689 struct scsi_data_buffer
*bidi_sdb
= cmd
->request
->next_rq
->special
;
691 scsi_free_sgtable(bidi_sdb
, false);
692 kmem_cache_free(scsi_sdb_cache
, bidi_sdb
);
693 cmd
->request
->next_rq
->special
= NULL
;
696 static bool scsi_end_request(struct request
*req
, int error
,
697 unsigned int bytes
, unsigned int bidi_bytes
)
699 struct scsi_cmnd
*cmd
= req
->special
;
700 struct scsi_device
*sdev
= cmd
->device
;
701 struct request_queue
*q
= sdev
->request_queue
;
703 if (blk_update_request(req
, error
, bytes
))
706 /* Bidi request must be completed as a whole */
707 if (unlikely(bidi_bytes
) &&
708 blk_update_request(req
->next_rq
, error
, bidi_bytes
))
711 if (blk_queue_add_random(q
))
712 add_disk_randomness(req
->rq_disk
);
716 * In the MQ case the command gets freed by __blk_mq_end_io,
717 * so we have to do all cleanup that depends on it earlier.
719 * We also can't kick the queues from irq context, so we
720 * will have to defer it to a workqueue.
722 scsi_mq_uninit_cmd(cmd
);
724 __blk_mq_end_io(req
, error
);
726 if (scsi_target(sdev
)->single_lun
||
727 !list_empty(&sdev
->host
->starved_list
))
728 kblockd_schedule_work(&sdev
->requeue_work
);
730 blk_mq_start_stopped_hw_queues(q
, true);
732 put_device(&sdev
->sdev_gendev
);
736 spin_lock_irqsave(q
->queue_lock
, flags
);
737 blk_finish_request(req
, error
);
738 spin_unlock_irqrestore(q
->queue_lock
, flags
);
741 scsi_release_bidi_buffers(cmd
);
742 scsi_release_buffers(cmd
);
743 scsi_next_command(cmd
);
750 * __scsi_error_from_host_byte - translate SCSI error code into errno
751 * @cmd: SCSI command (unused)
752 * @result: scsi error code
754 * Translate SCSI error code into standard UNIX errno.
756 * -ENOLINK temporary transport failure
757 * -EREMOTEIO permanent target failure, do not retry
758 * -EBADE permanent nexus failure, retry on other path
759 * -ENOSPC No write space available
760 * -ENODATA Medium error
761 * -EIO unspecified I/O error
763 static int __scsi_error_from_host_byte(struct scsi_cmnd
*cmd
, int result
)
767 switch(host_byte(result
)) {
768 case DID_TRANSPORT_FAILFAST
:
771 case DID_TARGET_FAILURE
:
772 set_host_byte(cmd
, DID_OK
);
775 case DID_NEXUS_FAILURE
:
776 set_host_byte(cmd
, DID_OK
);
779 case DID_ALLOC_FAILURE
:
780 set_host_byte(cmd
, DID_OK
);
783 case DID_MEDIUM_ERROR
:
784 set_host_byte(cmd
, DID_OK
);
796 * Function: scsi_io_completion()
798 * Purpose: Completion processing for block device I/O requests.
800 * Arguments: cmd - command that is finished.
802 * Lock status: Assumed that no lock is held upon entry.
806 * Notes: We will finish off the specified number of sectors. If we
807 * are done, the command block will be released and the queue
808 * function will be goosed. If we are not done then we have to
809 * figure out what to do next:
811 * a) We can call scsi_requeue_command(). The request
812 * will be unprepared and put back on the queue. Then
813 * a new command will be created for it. This should
814 * be used if we made forward progress, or if we want
815 * to switch from READ(10) to READ(6) for example.
817 * b) We can call __scsi_queue_insert(). The request will
818 * be put back on the queue and retried using the same
819 * command as before, possibly after a delay.
821 * c) We can call scsi_end_request() with -EIO to fail
822 * the remainder of the request.
824 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
826 int result
= cmd
->result
;
827 struct request_queue
*q
= cmd
->device
->request_queue
;
828 struct request
*req
= cmd
->request
;
830 struct scsi_sense_hdr sshdr
;
832 int sense_deferred
= 0;
833 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
834 ACTION_DELAYED_RETRY
} action
;
835 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
838 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
840 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
843 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
845 if (sense_valid
&& req
->sense
) {
847 * SG_IO wants current and deferred errors
849 int len
= 8 + cmd
->sense_buffer
[7];
851 if (len
> SCSI_SENSE_BUFFERSIZE
)
852 len
= SCSI_SENSE_BUFFERSIZE
;
853 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
854 req
->sense_len
= len
;
857 error
= __scsi_error_from_host_byte(cmd
, result
);
860 * __scsi_error_from_host_byte may have reset the host_byte
862 req
->errors
= cmd
->result
;
864 req
->resid_len
= scsi_get_resid(cmd
);
866 if (scsi_bidi_cmnd(cmd
)) {
868 * Bidi commands Must be complete as a whole,
869 * both sides at once.
871 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
872 if (scsi_end_request(req
, 0, blk_rq_bytes(req
),
873 blk_rq_bytes(req
->next_rq
)))
877 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
879 * Certain non BLOCK_PC requests are commands that don't
880 * actually transfer anything (FLUSH), so cannot use
881 * good_bytes != blk_rq_bytes(req) as the signal for an error.
882 * This sets the error explicitly for the problem case.
884 error
= __scsi_error_from_host_byte(cmd
, result
);
887 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
888 BUG_ON(blk_bidi_rq(req
));
891 * Next deal with any sectors which we were able to correctly
894 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
895 "%u sectors total, %d bytes done.\n",
896 blk_rq_sectors(req
), good_bytes
));
899 * Recovered errors need reporting, but they're always treated
900 * as success, so fiddle the result code here. For BLOCK_PC
901 * we already took a copy of the original into rq->errors which
902 * is what gets returned to the user
904 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
905 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
906 * print since caller wants ATA registers. Only occurs on
907 * SCSI ATA PASS_THROUGH commands when CK_COND=1
909 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
911 else if (!(req
->cmd_flags
& REQ_QUIET
))
912 scsi_print_sense("", cmd
);
914 /* BLOCK_PC may have set error */
919 * If we finished all bytes in the request we are done now.
921 if (!scsi_end_request(req
, error
, good_bytes
, 0))
925 * Kill remainder if no retrys.
927 if (error
&& scsi_noretry_cmd(cmd
)) {
928 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
934 * If there had been no error, but we have leftover bytes in the
935 * requeues just queue the command up again.
940 error
= __scsi_error_from_host_byte(cmd
, result
);
942 if (host_byte(result
) == DID_RESET
) {
943 /* Third party bus reset or reset for error recovery
944 * reasons. Just retry the command and see what
947 action
= ACTION_RETRY
;
948 } else if (sense_valid
&& !sense_deferred
) {
949 switch (sshdr
.sense_key
) {
951 if (cmd
->device
->removable
) {
952 /* Detected disc change. Set a bit
953 * and quietly refuse further access.
955 cmd
->device
->changed
= 1;
956 action
= ACTION_FAIL
;
958 /* Must have been a power glitch, or a
959 * bus reset. Could not have been a
960 * media change, so we just retry the
961 * command and see what happens.
963 action
= ACTION_RETRY
;
966 case ILLEGAL_REQUEST
:
967 /* If we had an ILLEGAL REQUEST returned, then
968 * we may have performed an unsupported
969 * command. The only thing this should be
970 * would be a ten byte read where only a six
971 * byte read was supported. Also, on a system
972 * where READ CAPACITY failed, we may have
973 * read past the end of the disk.
975 if ((cmd
->device
->use_10_for_rw
&&
976 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
977 (cmd
->cmnd
[0] == READ_10
||
978 cmd
->cmnd
[0] == WRITE_10
)) {
979 /* This will issue a new 6-byte command. */
980 cmd
->device
->use_10_for_rw
= 0;
981 action
= ACTION_REPREP
;
982 } else if (sshdr
.asc
== 0x10) /* DIX */ {
983 action
= ACTION_FAIL
;
985 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
986 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
987 action
= ACTION_FAIL
;
990 action
= ACTION_FAIL
;
992 case ABORTED_COMMAND
:
993 action
= ACTION_FAIL
;
994 if (sshdr
.asc
== 0x10) /* DIF */
998 /* If the device is in the process of becoming
999 * ready, or has a temporary blockage, retry.
1001 if (sshdr
.asc
== 0x04) {
1002 switch (sshdr
.ascq
) {
1003 case 0x01: /* becoming ready */
1004 case 0x04: /* format in progress */
1005 case 0x05: /* rebuild in progress */
1006 case 0x06: /* recalculation in progress */
1007 case 0x07: /* operation in progress */
1008 case 0x08: /* Long write in progress */
1009 case 0x09: /* self test in progress */
1010 case 0x14: /* space allocation in progress */
1011 action
= ACTION_DELAYED_RETRY
;
1014 action
= ACTION_FAIL
;
1018 action
= ACTION_FAIL
;
1020 case VOLUME_OVERFLOW
:
1021 /* See SSC3rXX or current. */
1022 action
= ACTION_FAIL
;
1025 action
= ACTION_FAIL
;
1029 action
= ACTION_FAIL
;
1031 if (action
!= ACTION_FAIL
&&
1032 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
1033 action
= ACTION_FAIL
;
1037 /* Give up and fail the remainder of the request */
1038 if (!(req
->cmd_flags
& REQ_QUIET
)) {
1039 scsi_print_result(cmd
);
1040 if (driver_byte(result
) & DRIVER_SENSE
)
1041 scsi_print_sense("", cmd
);
1042 scsi_print_command(cmd
);
1044 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
1049 /* Unprep the request and put it back at the head of the queue.
1050 * A new command will be prepared and issued.
1053 cmd
->request
->cmd_flags
&= ~REQ_DONTPREP
;
1054 scsi_mq_uninit_cmd(cmd
);
1055 scsi_mq_requeue_cmd(cmd
);
1057 scsi_release_buffers(cmd
);
1058 scsi_requeue_command(q
, cmd
);
1062 /* Retry the same command immediately */
1063 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
1065 case ACTION_DELAYED_RETRY
:
1066 /* Retry the same command after a delay */
1067 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
1072 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
1078 * If sg table allocation fails, requeue request later.
1080 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
1081 gfp_mask
, req
->mq_ctx
!= NULL
)))
1082 return BLKPREP_DEFER
;
1085 * Next, walk the list, and fill in the addresses and sizes of
1088 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1089 BUG_ON(count
> sdb
->table
.nents
);
1090 sdb
->table
.nents
= count
;
1091 sdb
->length
= blk_rq_bytes(req
);
1096 * Function: scsi_init_io()
1098 * Purpose: SCSI I/O initialize function.
1100 * Arguments: cmd - Command descriptor we wish to initialize
1102 * Returns: 0 on success
1103 * BLKPREP_DEFER if the failure is retryable
1104 * BLKPREP_KILL if the failure is fatal
1106 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
1108 struct scsi_device
*sdev
= cmd
->device
;
1109 struct request
*rq
= cmd
->request
;
1110 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1113 BUG_ON(!rq
->nr_phys_segments
);
1115 error
= scsi_init_sgtable(rq
, &cmd
->sdb
, gfp_mask
);
1119 if (blk_bidi_rq(rq
)) {
1120 if (!rq
->q
->mq_ops
) {
1121 struct scsi_data_buffer
*bidi_sdb
=
1122 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1124 error
= BLKPREP_DEFER
;
1128 rq
->next_rq
->special
= bidi_sdb
;
1131 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
,
1137 if (blk_integrity_rq(rq
)) {
1138 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1141 BUG_ON(prot_sdb
== NULL
);
1142 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1144 if (scsi_alloc_sgtable(prot_sdb
, ivecs
, gfp_mask
, is_mq
)) {
1145 error
= BLKPREP_DEFER
;
1149 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1150 prot_sdb
->table
.sgl
);
1151 BUG_ON(unlikely(count
> ivecs
));
1152 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1154 cmd
->prot_sdb
= prot_sdb
;
1155 cmd
->prot_sdb
->table
.nents
= count
;
1161 scsi_mq_free_sgtables(cmd
);
1163 scsi_release_buffers(cmd
);
1164 cmd
->request
->special
= NULL
;
1165 scsi_put_command(cmd
);
1166 put_device(&sdev
->sdev_gendev
);
1170 EXPORT_SYMBOL(scsi_init_io
);
1172 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1173 struct request
*req
)
1175 struct scsi_cmnd
*cmd
;
1177 if (!req
->special
) {
1178 /* Bail if we can't get a reference to the device */
1179 if (!get_device(&sdev
->sdev_gendev
))
1182 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1183 if (unlikely(!cmd
)) {
1184 put_device(&sdev
->sdev_gendev
);
1192 /* pull a tag out of the request if we have one */
1193 cmd
->tag
= req
->tag
;
1196 cmd
->cmnd
= req
->cmd
;
1197 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1202 static int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1204 struct scsi_cmnd
*cmd
= req
->special
;
1207 * BLOCK_PC requests may transfer data, in which case they must
1208 * a bio attached to them. Or they might contain a SCSI command
1209 * that does not transfer data, in which case they may optionally
1210 * submit a request without an attached bio.
1213 int ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1217 BUG_ON(blk_rq_bytes(req
));
1219 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1222 cmd
->cmd_len
= req
->cmd_len
;
1223 cmd
->transfersize
= blk_rq_bytes(req
);
1224 cmd
->allowed
= req
->retries
;
1229 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1230 * that still need to be translated to SCSI CDBs from the ULD.
1232 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1234 struct scsi_cmnd
*cmd
= req
->special
;
1236 if (unlikely(sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
1237 && sdev
->scsi_dh_data
->scsi_dh
->prep_fn
)) {
1238 int ret
= sdev
->scsi_dh_data
->scsi_dh
->prep_fn(sdev
, req
);
1239 if (ret
!= BLKPREP_OK
)
1243 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1244 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1247 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1249 struct scsi_cmnd
*cmd
= req
->special
;
1251 if (!blk_rq_bytes(req
))
1252 cmd
->sc_data_direction
= DMA_NONE
;
1253 else if (rq_data_dir(req
) == WRITE
)
1254 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1256 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1258 switch (req
->cmd_type
) {
1260 return scsi_setup_fs_cmnd(sdev
, req
);
1261 case REQ_TYPE_BLOCK_PC
:
1262 return scsi_setup_blk_pc_cmnd(sdev
, req
);
1264 return BLKPREP_KILL
;
1269 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1271 int ret
= BLKPREP_OK
;
1274 * If the device is not in running state we will reject some
1277 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1278 switch (sdev
->sdev_state
) {
1280 case SDEV_TRANSPORT_OFFLINE
:
1282 * If the device is offline we refuse to process any
1283 * commands. The device must be brought online
1284 * before trying any recovery commands.
1286 sdev_printk(KERN_ERR
, sdev
,
1287 "rejecting I/O to offline device\n");
1292 * If the device is fully deleted, we refuse to
1293 * process any commands as well.
1295 sdev_printk(KERN_ERR
, sdev
,
1296 "rejecting I/O to dead device\n");
1301 case SDEV_CREATED_BLOCK
:
1303 * If the devices is blocked we defer normal commands.
1305 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1306 ret
= BLKPREP_DEFER
;
1310 * For any other not fully online state we only allow
1311 * special commands. In particular any user initiated
1312 * command is not allowed.
1314 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1323 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1325 struct scsi_device
*sdev
= q
->queuedata
;
1329 req
->errors
= DID_NO_CONNECT
<< 16;
1330 /* release the command and kill it */
1332 struct scsi_cmnd
*cmd
= req
->special
;
1333 scsi_release_buffers(cmd
);
1334 scsi_put_command(cmd
);
1335 put_device(&sdev
->sdev_gendev
);
1336 req
->special
= NULL
;
1341 * If we defer, the blk_peek_request() returns NULL, but the
1342 * queue must be restarted, so we schedule a callback to happen
1345 if (atomic_read(&sdev
->device_busy
) == 0)
1346 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1349 req
->cmd_flags
|= REQ_DONTPREP
;
1355 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1357 struct scsi_device
*sdev
= q
->queuedata
;
1358 struct scsi_cmnd
*cmd
;
1361 ret
= scsi_prep_state_check(sdev
, req
);
1362 if (ret
!= BLKPREP_OK
)
1365 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1366 if (unlikely(!cmd
)) {
1367 ret
= BLKPREP_DEFER
;
1371 ret
= scsi_setup_cmnd(sdev
, req
);
1373 return scsi_prep_return(q
, req
, ret
);
1376 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1378 scsi_uninit_cmd(req
->special
);
1382 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1385 * Called with the queue_lock held.
1387 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1388 struct scsi_device
*sdev
)
1392 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1393 if (atomic_read(&sdev
->device_blocked
)) {
1398 * unblock after device_blocked iterates to zero
1400 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1402 * For the MQ case we take care of this in the caller.
1405 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1408 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1409 "unblocking device at zero depth\n"));
1412 if (busy
>= sdev
->queue_depth
)
1417 atomic_dec(&sdev
->device_busy
);
1422 * scsi_target_queue_ready: checks if there we can send commands to target
1423 * @sdev: scsi device on starget to check.
1425 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1426 struct scsi_device
*sdev
)
1428 struct scsi_target
*starget
= scsi_target(sdev
);
1431 if (starget
->single_lun
) {
1432 spin_lock_irq(shost
->host_lock
);
1433 if (starget
->starget_sdev_user
&&
1434 starget
->starget_sdev_user
!= sdev
) {
1435 spin_unlock_irq(shost
->host_lock
);
1438 starget
->starget_sdev_user
= sdev
;
1439 spin_unlock_irq(shost
->host_lock
);
1442 if (starget
->can_queue
<= 0)
1445 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1446 if (atomic_read(&starget
->target_blocked
) > 0) {
1451 * unblock after target_blocked iterates to zero
1453 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1456 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1457 "unblocking target at zero depth\n"));
1460 if (busy
>= starget
->can_queue
)
1466 spin_lock_irq(shost
->host_lock
);
1467 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1468 spin_unlock_irq(shost
->host_lock
);
1470 if (starget
->can_queue
> 0)
1471 atomic_dec(&starget
->target_busy
);
1476 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1477 * return 0. We must end up running the queue again whenever 0 is
1478 * returned, else IO can hang.
1480 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1481 struct Scsi_Host
*shost
,
1482 struct scsi_device
*sdev
)
1486 if (scsi_host_in_recovery(shost
))
1489 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1490 if (atomic_read(&shost
->host_blocked
) > 0) {
1495 * unblock after host_blocked iterates to zero
1497 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1501 shost_printk(KERN_INFO
, shost
,
1502 "unblocking host at zero depth\n"));
1505 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1507 if (shost
->host_self_blocked
)
1510 /* We're OK to process the command, so we can't be starved */
1511 if (!list_empty(&sdev
->starved_entry
)) {
1512 spin_lock_irq(shost
->host_lock
);
1513 if (!list_empty(&sdev
->starved_entry
))
1514 list_del_init(&sdev
->starved_entry
);
1515 spin_unlock_irq(shost
->host_lock
);
1521 spin_lock_irq(shost
->host_lock
);
1522 if (list_empty(&sdev
->starved_entry
))
1523 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1524 spin_unlock_irq(shost
->host_lock
);
1526 atomic_dec(&shost
->host_busy
);
1531 * Busy state exporting function for request stacking drivers.
1533 * For efficiency, no lock is taken to check the busy state of
1534 * shost/starget/sdev, since the returned value is not guaranteed and
1535 * may be changed after request stacking drivers call the function,
1536 * regardless of taking lock or not.
1538 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1539 * needs to return 'not busy'. Otherwise, request stacking drivers
1540 * may hold requests forever.
1542 static int scsi_lld_busy(struct request_queue
*q
)
1544 struct scsi_device
*sdev
= q
->queuedata
;
1545 struct Scsi_Host
*shost
;
1547 if (blk_queue_dying(q
))
1553 * Ignore host/starget busy state.
1554 * Since block layer does not have a concept of fairness across
1555 * multiple queues, congestion of host/starget needs to be handled
1558 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1565 * Kill a request for a dead device
1567 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1569 struct scsi_cmnd
*cmd
= req
->special
;
1570 struct scsi_device
*sdev
;
1571 struct scsi_target
*starget
;
1572 struct Scsi_Host
*shost
;
1574 blk_start_request(req
);
1576 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1579 starget
= scsi_target(sdev
);
1581 scsi_init_cmd_errh(cmd
);
1582 cmd
->result
= DID_NO_CONNECT
<< 16;
1583 atomic_inc(&cmd
->device
->iorequest_cnt
);
1586 * SCSI request completion path will do scsi_device_unbusy(),
1587 * bump busy counts. To bump the counters, we need to dance
1588 * with the locks as normal issue path does.
1590 atomic_inc(&sdev
->device_busy
);
1591 atomic_inc(&shost
->host_busy
);
1592 if (starget
->can_queue
> 0)
1593 atomic_inc(&starget
->target_busy
);
1595 blk_complete_request(req
);
1598 static void scsi_softirq_done(struct request
*rq
)
1600 struct scsi_cmnd
*cmd
= rq
->special
;
1601 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1604 INIT_LIST_HEAD(&cmd
->eh_entry
);
1606 atomic_inc(&cmd
->device
->iodone_cnt
);
1608 atomic_inc(&cmd
->device
->ioerr_cnt
);
1610 disposition
= scsi_decide_disposition(cmd
);
1611 if (disposition
!= SUCCESS
&&
1612 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1613 sdev_printk(KERN_ERR
, cmd
->device
,
1614 "timing out command, waited %lus\n",
1616 disposition
= SUCCESS
;
1619 scsi_log_completion(cmd
, disposition
);
1621 switch (disposition
) {
1623 scsi_finish_command(cmd
);
1626 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1628 case ADD_TO_MLQUEUE
:
1629 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1632 if (!scsi_eh_scmd_add(cmd
, 0))
1633 scsi_finish_command(cmd
);
1638 * scsi_done - Invoke completion on finished SCSI command.
1639 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1640 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1642 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1643 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1644 * calls blk_complete_request() for further processing.
1646 * This function is interrupt context safe.
1648 static void scsi_done(struct scsi_cmnd
*cmd
)
1650 trace_scsi_dispatch_cmd_done(cmd
);
1651 blk_complete_request(cmd
->request
);
1655 * Function: scsi_request_fn()
1657 * Purpose: Main strategy routine for SCSI.
1659 * Arguments: q - Pointer to actual queue.
1663 * Lock status: IO request lock assumed to be held when called.
1665 static void scsi_request_fn(struct request_queue
*q
)
1666 __releases(q
->queue_lock
)
1667 __acquires(q
->queue_lock
)
1669 struct scsi_device
*sdev
= q
->queuedata
;
1670 struct Scsi_Host
*shost
;
1671 struct scsi_cmnd
*cmd
;
1672 struct request
*req
;
1675 * To start with, we keep looping until the queue is empty, or until
1676 * the host is no longer able to accept any more requests.
1682 * get next queueable request. We do this early to make sure
1683 * that the request is fully prepared even if we cannot
1686 req
= blk_peek_request(q
);
1690 if (unlikely(!scsi_device_online(sdev
))) {
1691 sdev_printk(KERN_ERR
, sdev
,
1692 "rejecting I/O to offline device\n");
1693 scsi_kill_request(req
, q
);
1697 if (!scsi_dev_queue_ready(q
, sdev
))
1701 * Remove the request from the request list.
1703 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1704 blk_start_request(req
);
1706 spin_unlock_irq(q
->queue_lock
);
1708 if (unlikely(cmd
== NULL
)) {
1709 printk(KERN_CRIT
"impossible request in %s.\n"
1710 "please mail a stack trace to "
1711 "linux-scsi@vger.kernel.org\n",
1713 blk_dump_rq_flags(req
, "foo");
1718 * We hit this when the driver is using a host wide
1719 * tag map. For device level tag maps the queue_depth check
1720 * in the device ready fn would prevent us from trying
1721 * to allocate a tag. Since the map is a shared host resource
1722 * we add the dev to the starved list so it eventually gets
1723 * a run when a tag is freed.
1725 if (blk_queue_tagged(q
) && !blk_rq_tagged(req
)) {
1726 spin_lock_irq(shost
->host_lock
);
1727 if (list_empty(&sdev
->starved_entry
))
1728 list_add_tail(&sdev
->starved_entry
,
1729 &shost
->starved_list
);
1730 spin_unlock_irq(shost
->host_lock
);
1734 if (!scsi_target_queue_ready(shost
, sdev
))
1737 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1738 goto host_not_ready
;
1741 * Finally, initialize any error handling parameters, and set up
1742 * the timers for timeouts.
1744 scsi_init_cmd_errh(cmd
);
1747 * Dispatch the command to the low-level driver.
1749 cmd
->scsi_done
= scsi_done
;
1750 rtn
= scsi_dispatch_cmd(cmd
);
1752 scsi_queue_insert(cmd
, rtn
);
1753 spin_lock_irq(q
->queue_lock
);
1756 spin_lock_irq(q
->queue_lock
);
1762 if (scsi_target(sdev
)->can_queue
> 0)
1763 atomic_dec(&scsi_target(sdev
)->target_busy
);
1766 * lock q, handle tag, requeue req, and decrement device_busy. We
1767 * must return with queue_lock held.
1769 * Decrementing device_busy without checking it is OK, as all such
1770 * cases (host limits or settings) should run the queue at some
1773 spin_lock_irq(q
->queue_lock
);
1774 blk_requeue_request(q
, req
);
1775 atomic_dec(&sdev
->device_busy
);
1777 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1778 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1781 static inline int prep_to_mq(int ret
)
1787 return BLK_MQ_RQ_QUEUE_BUSY
;
1789 return BLK_MQ_RQ_QUEUE_ERROR
;
1793 static int scsi_mq_prep_fn(struct request
*req
)
1795 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1796 struct scsi_device
*sdev
= req
->q
->queuedata
;
1797 struct Scsi_Host
*shost
= sdev
->host
;
1798 unsigned char *sense_buf
= cmd
->sense_buffer
;
1799 struct scatterlist
*sg
;
1801 memset(cmd
, 0, sizeof(struct scsi_cmnd
));
1807 cmd
->sense_buffer
= sense_buf
;
1809 cmd
->tag
= req
->tag
;
1811 req
->cmd
= req
->__cmd
;
1812 cmd
->cmnd
= req
->cmd
;
1813 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1815 INIT_LIST_HEAD(&cmd
->list
);
1816 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1817 cmd
->jiffies_at_alloc
= jiffies
;
1820 * XXX: cmd_list lookups are only used by two drivers, try to get
1821 * rid of this list in common code.
1823 spin_lock_irq(&sdev
->list_lock
);
1824 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1825 spin_unlock_irq(&sdev
->list_lock
);
1827 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1828 cmd
->sdb
.table
.sgl
= sg
;
1830 if (scsi_host_get_prot(shost
)) {
1831 cmd
->prot_sdb
= (void *)sg
+
1832 shost
->sg_tablesize
* sizeof(struct scatterlist
);
1833 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1835 cmd
->prot_sdb
->table
.sgl
=
1836 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1839 if (blk_bidi_rq(req
)) {
1840 struct request
*next_rq
= req
->next_rq
;
1841 struct scsi_data_buffer
*bidi_sdb
= blk_mq_rq_to_pdu(next_rq
);
1843 memset(bidi_sdb
, 0, sizeof(struct scsi_data_buffer
));
1844 bidi_sdb
->table
.sgl
=
1845 (struct scatterlist
*)(bidi_sdb
+ 1);
1847 next_rq
->special
= bidi_sdb
;
1850 return scsi_setup_cmnd(sdev
, req
);
1853 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1855 trace_scsi_dispatch_cmd_done(cmd
);
1856 blk_mq_complete_request(cmd
->request
);
1859 static int scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
, struct request
*req
)
1861 struct request_queue
*q
= req
->q
;
1862 struct scsi_device
*sdev
= q
->queuedata
;
1863 struct Scsi_Host
*shost
= sdev
->host
;
1864 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1868 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
1872 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1873 if (!get_device(&sdev
->sdev_gendev
))
1876 if (!scsi_dev_queue_ready(q
, sdev
))
1877 goto out_put_device
;
1878 if (!scsi_target_queue_ready(shost
, sdev
))
1879 goto out_dec_device_busy
;
1880 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1881 goto out_dec_target_busy
;
1883 if (!(req
->cmd_flags
& REQ_DONTPREP
)) {
1884 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
1886 goto out_dec_host_busy
;
1887 req
->cmd_flags
|= REQ_DONTPREP
;
1890 scsi_init_cmd_errh(cmd
);
1891 cmd
->scsi_done
= scsi_mq_done
;
1893 reason
= scsi_dispatch_cmd(cmd
);
1895 scsi_set_blocked(cmd
, reason
);
1896 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1897 goto out_dec_host_busy
;
1900 return BLK_MQ_RQ_QUEUE_OK
;
1903 atomic_dec(&shost
->host_busy
);
1904 out_dec_target_busy
:
1905 if (scsi_target(sdev
)->can_queue
> 0)
1906 atomic_dec(&scsi_target(sdev
)->target_busy
);
1907 out_dec_device_busy
:
1908 atomic_dec(&sdev
->device_busy
);
1910 put_device(&sdev
->sdev_gendev
);
1913 case BLK_MQ_RQ_QUEUE_BUSY
:
1914 blk_mq_stop_hw_queue(hctx
);
1915 if (atomic_read(&sdev
->device_busy
) == 0 &&
1916 !scsi_device_blocked(sdev
))
1917 blk_mq_delay_queue(hctx
, SCSI_QUEUE_DELAY
);
1919 case BLK_MQ_RQ_QUEUE_ERROR
:
1921 * Make sure to release all allocated ressources when
1922 * we hit an error, as we will never see this command
1925 if (req
->cmd_flags
& REQ_DONTPREP
)
1926 scsi_mq_uninit_cmd(cmd
);
1934 static int scsi_init_request(void *data
, struct request
*rq
,
1935 unsigned int hctx_idx
, unsigned int request_idx
,
1936 unsigned int numa_node
)
1938 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1940 cmd
->sense_buffer
= kzalloc_node(SCSI_SENSE_BUFFERSIZE
, GFP_KERNEL
,
1942 if (!cmd
->sense_buffer
)
1947 static void scsi_exit_request(void *data
, struct request
*rq
,
1948 unsigned int hctx_idx
, unsigned int request_idx
)
1950 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1952 kfree(cmd
->sense_buffer
);
1955 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1957 struct device
*host_dev
;
1958 u64 bounce_limit
= 0xffffffff;
1960 if (shost
->unchecked_isa_dma
)
1961 return BLK_BOUNCE_ISA
;
1963 * Platforms with virtual-DMA translation
1964 * hardware have no practical limit.
1966 if (!PCI_DMA_BUS_IS_PHYS
)
1967 return BLK_BOUNCE_ANY
;
1969 host_dev
= scsi_get_device(shost
);
1970 if (host_dev
&& host_dev
->dma_mask
)
1971 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
1973 return bounce_limit
;
1976 static void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
1978 struct device
*dev
= shost
->dma_dev
;
1981 * this limit is imposed by hardware restrictions
1983 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
1984 SCSI_MAX_SG_CHAIN_SEGMENTS
));
1986 if (scsi_host_prot_dma(shost
)) {
1987 shost
->sg_prot_tablesize
=
1988 min_not_zero(shost
->sg_prot_tablesize
,
1989 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
1990 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
1991 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
1994 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
1995 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1996 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1997 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1999 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2001 if (!shost
->use_clustering
)
2002 q
->limits
.cluster
= 0;
2005 * set a reasonable default alignment on word boundaries: the
2006 * host and device may alter it using
2007 * blk_queue_update_dma_alignment() later.
2009 blk_queue_dma_alignment(q
, 0x03);
2012 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
2013 request_fn_proc
*request_fn
)
2015 struct request_queue
*q
;
2017 q
= blk_init_queue(request_fn
, NULL
);
2020 __scsi_init_queue(shost
, q
);
2023 EXPORT_SYMBOL(__scsi_alloc_queue
);
2025 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
2027 struct request_queue
*q
;
2029 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
2033 blk_queue_prep_rq(q
, scsi_prep_fn
);
2034 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2035 blk_queue_softirq_done(q
, scsi_softirq_done
);
2036 blk_queue_rq_timed_out(q
, scsi_times_out
);
2037 blk_queue_lld_busy(q
, scsi_lld_busy
);
2041 static struct blk_mq_ops scsi_mq_ops
= {
2042 .map_queue
= blk_mq_map_queue
,
2043 .queue_rq
= scsi_queue_rq
,
2044 .complete
= scsi_softirq_done
,
2045 .timeout
= scsi_times_out
,
2046 .init_request
= scsi_init_request
,
2047 .exit_request
= scsi_exit_request
,
2050 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2052 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2053 if (IS_ERR(sdev
->request_queue
))
2056 sdev
->request_queue
->queuedata
= sdev
;
2057 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2058 return sdev
->request_queue
;
2061 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2063 unsigned int cmd_size
, sgl_size
, tbl_size
;
2065 tbl_size
= shost
->sg_tablesize
;
2066 if (tbl_size
> SCSI_MAX_SG_SEGMENTS
)
2067 tbl_size
= SCSI_MAX_SG_SEGMENTS
;
2068 sgl_size
= tbl_size
* sizeof(struct scatterlist
);
2069 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2070 if (scsi_host_get_prot(shost
))
2071 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2073 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2074 shost
->tag_set
.ops
= &scsi_mq_ops
;
2075 shost
->tag_set
.nr_hw_queues
= 1;
2076 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2077 shost
->tag_set
.cmd_size
= cmd_size
;
2078 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2079 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2080 shost
->tag_set
.driver_data
= shost
;
2082 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2085 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2087 blk_mq_free_tag_set(&shost
->tag_set
);
2091 * Function: scsi_block_requests()
2093 * Purpose: Utility function used by low-level drivers to prevent further
2094 * commands from being queued to the device.
2096 * Arguments: shost - Host in question
2100 * Lock status: No locks are assumed held.
2102 * Notes: There is no timer nor any other means by which the requests
2103 * get unblocked other than the low-level driver calling
2104 * scsi_unblock_requests().
2106 void scsi_block_requests(struct Scsi_Host
*shost
)
2108 shost
->host_self_blocked
= 1;
2110 EXPORT_SYMBOL(scsi_block_requests
);
2113 * Function: scsi_unblock_requests()
2115 * Purpose: Utility function used by low-level drivers to allow further
2116 * commands from being queued to the device.
2118 * Arguments: shost - Host in question
2122 * Lock status: No locks are assumed held.
2124 * Notes: There is no timer nor any other means by which the requests
2125 * get unblocked other than the low-level driver calling
2126 * scsi_unblock_requests().
2128 * This is done as an API function so that changes to the
2129 * internals of the scsi mid-layer won't require wholesale
2130 * changes to drivers that use this feature.
2132 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2134 shost
->host_self_blocked
= 0;
2135 scsi_run_host_queues(shost
);
2137 EXPORT_SYMBOL(scsi_unblock_requests
);
2139 int __init
scsi_init_queue(void)
2143 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2144 sizeof(struct scsi_data_buffer
),
2146 if (!scsi_sdb_cache
) {
2147 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2151 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2152 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2153 int size
= sgp
->size
* sizeof(struct scatterlist
);
2155 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
2156 SLAB_HWCACHE_ALIGN
, NULL
);
2158 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
2163 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
2166 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
2175 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2176 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2178 mempool_destroy(sgp
->pool
);
2180 kmem_cache_destroy(sgp
->slab
);
2182 kmem_cache_destroy(scsi_sdb_cache
);
2187 void scsi_exit_queue(void)
2191 kmem_cache_destroy(scsi_sdb_cache
);
2193 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2194 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2195 mempool_destroy(sgp
->pool
);
2196 kmem_cache_destroy(sgp
->slab
);
2201 * scsi_mode_select - issue a mode select
2202 * @sdev: SCSI device to be queried
2203 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2204 * @sp: Save page bit (0 == don't save, 1 == save)
2205 * @modepage: mode page being requested
2206 * @buffer: request buffer (may not be smaller than eight bytes)
2207 * @len: length of request buffer.
2208 * @timeout: command timeout
2209 * @retries: number of retries before failing
2210 * @data: returns a structure abstracting the mode header data
2211 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2212 * must be SCSI_SENSE_BUFFERSIZE big.
2214 * Returns zero if successful; negative error number or scsi
2219 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2220 unsigned char *buffer
, int len
, int timeout
, int retries
,
2221 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2223 unsigned char cmd
[10];
2224 unsigned char *real_buffer
;
2227 memset(cmd
, 0, sizeof(cmd
));
2228 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2230 if (sdev
->use_10_for_ms
) {
2233 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2236 memcpy(real_buffer
+ 8, buffer
, len
);
2240 real_buffer
[2] = data
->medium_type
;
2241 real_buffer
[3] = data
->device_specific
;
2242 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2244 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2245 real_buffer
[7] = data
->block_descriptor_length
;
2247 cmd
[0] = MODE_SELECT_10
;
2251 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2255 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2258 memcpy(real_buffer
+ 4, buffer
, len
);
2261 real_buffer
[1] = data
->medium_type
;
2262 real_buffer
[2] = data
->device_specific
;
2263 real_buffer
[3] = data
->block_descriptor_length
;
2266 cmd
[0] = MODE_SELECT
;
2270 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2271 sshdr
, timeout
, retries
, NULL
);
2275 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2278 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2279 * @sdev: SCSI device to be queried
2280 * @dbd: set if mode sense will allow block descriptors to be returned
2281 * @modepage: mode page being requested
2282 * @buffer: request buffer (may not be smaller than eight bytes)
2283 * @len: length of request buffer.
2284 * @timeout: command timeout
2285 * @retries: number of retries before failing
2286 * @data: returns a structure abstracting the mode header data
2287 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2288 * must be SCSI_SENSE_BUFFERSIZE big.
2290 * Returns zero if unsuccessful, or the header offset (either 4
2291 * or 8 depending on whether a six or ten byte command was
2292 * issued) if successful.
2295 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2296 unsigned char *buffer
, int len
, int timeout
, int retries
,
2297 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2299 unsigned char cmd
[12];
2303 struct scsi_sense_hdr my_sshdr
;
2305 memset(data
, 0, sizeof(*data
));
2306 memset(&cmd
[0], 0, 12);
2307 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2310 /* caller might not be interested in sense, but we need it */
2315 use_10_for_ms
= sdev
->use_10_for_ms
;
2317 if (use_10_for_ms
) {
2321 cmd
[0] = MODE_SENSE_10
;
2328 cmd
[0] = MODE_SENSE
;
2333 memset(buffer
, 0, len
);
2335 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2336 sshdr
, timeout
, retries
, NULL
);
2338 /* This code looks awful: what it's doing is making sure an
2339 * ILLEGAL REQUEST sense return identifies the actual command
2340 * byte as the problem. MODE_SENSE commands can return
2341 * ILLEGAL REQUEST if the code page isn't supported */
2343 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2344 (driver_byte(result
) & DRIVER_SENSE
)) {
2345 if (scsi_sense_valid(sshdr
)) {
2346 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2347 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2349 * Invalid command operation code
2351 sdev
->use_10_for_ms
= 0;
2357 if(scsi_status_is_good(result
)) {
2358 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2359 (modepage
== 6 || modepage
== 8))) {
2360 /* Initio breakage? */
2363 data
->medium_type
= 0;
2364 data
->device_specific
= 0;
2366 data
->block_descriptor_length
= 0;
2367 } else if(use_10_for_ms
) {
2368 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2369 data
->medium_type
= buffer
[2];
2370 data
->device_specific
= buffer
[3];
2371 data
->longlba
= buffer
[4] & 0x01;
2372 data
->block_descriptor_length
= buffer
[6]*256
2375 data
->length
= buffer
[0] + 1;
2376 data
->medium_type
= buffer
[1];
2377 data
->device_specific
= buffer
[2];
2378 data
->block_descriptor_length
= buffer
[3];
2380 data
->header_length
= header_length
;
2385 EXPORT_SYMBOL(scsi_mode_sense
);
2388 * scsi_test_unit_ready - test if unit is ready
2389 * @sdev: scsi device to change the state of.
2390 * @timeout: command timeout
2391 * @retries: number of retries before failing
2392 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2393 * returning sense. Make sure that this is cleared before passing
2396 * Returns zero if unsuccessful or an error if TUR failed. For
2397 * removable media, UNIT_ATTENTION sets ->changed flag.
2400 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2401 struct scsi_sense_hdr
*sshdr_external
)
2404 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2406 struct scsi_sense_hdr
*sshdr
;
2409 if (!sshdr_external
)
2410 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
2412 sshdr
= sshdr_external
;
2414 /* try to eat the UNIT_ATTENTION if there are enough retries */
2416 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2417 timeout
, retries
, NULL
);
2418 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2419 sshdr
->sense_key
== UNIT_ATTENTION
)
2421 } while (scsi_sense_valid(sshdr
) &&
2422 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2424 if (!sshdr_external
)
2428 EXPORT_SYMBOL(scsi_test_unit_ready
);
2431 * scsi_device_set_state - Take the given device through the device state model.
2432 * @sdev: scsi device to change the state of.
2433 * @state: state to change to.
2435 * Returns zero if unsuccessful or an error if the requested
2436 * transition is illegal.
2439 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2441 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2443 if (state
== oldstate
)
2449 case SDEV_CREATED_BLOCK
:
2460 case SDEV_TRANSPORT_OFFLINE
:
2473 case SDEV_TRANSPORT_OFFLINE
:
2481 case SDEV_TRANSPORT_OFFLINE
:
2496 case SDEV_CREATED_BLOCK
:
2503 case SDEV_CREATED_BLOCK
:
2518 case SDEV_TRANSPORT_OFFLINE
:
2531 case SDEV_TRANSPORT_OFFLINE
:
2533 case SDEV_CREATED_BLOCK
:
2541 sdev
->sdev_state
= state
;
2545 SCSI_LOG_ERROR_RECOVERY(1,
2546 sdev_printk(KERN_ERR
, sdev
,
2547 "Illegal state transition %s->%s",
2548 scsi_device_state_name(oldstate
),
2549 scsi_device_state_name(state
))
2553 EXPORT_SYMBOL(scsi_device_set_state
);
2556 * sdev_evt_emit - emit a single SCSI device uevent
2557 * @sdev: associated SCSI device
2558 * @evt: event to emit
2560 * Send a single uevent (scsi_event) to the associated scsi_device.
2562 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2567 switch (evt
->evt_type
) {
2568 case SDEV_EVT_MEDIA_CHANGE
:
2569 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2571 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2572 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2574 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2575 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2577 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2578 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2580 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2581 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2583 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2584 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2593 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2597 * sdev_evt_thread - send a uevent for each scsi event
2598 * @work: work struct for scsi_device
2600 * Dispatch queued events to their associated scsi_device kobjects
2603 void scsi_evt_thread(struct work_struct
*work
)
2605 struct scsi_device
*sdev
;
2606 enum scsi_device_event evt_type
;
2607 LIST_HEAD(event_list
);
2609 sdev
= container_of(work
, struct scsi_device
, event_work
);
2611 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2612 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2613 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2616 struct scsi_event
*evt
;
2617 struct list_head
*this, *tmp
;
2618 unsigned long flags
;
2620 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2621 list_splice_init(&sdev
->event_list
, &event_list
);
2622 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2624 if (list_empty(&event_list
))
2627 list_for_each_safe(this, tmp
, &event_list
) {
2628 evt
= list_entry(this, struct scsi_event
, node
);
2629 list_del(&evt
->node
);
2630 scsi_evt_emit(sdev
, evt
);
2637 * sdev_evt_send - send asserted event to uevent thread
2638 * @sdev: scsi_device event occurred on
2639 * @evt: event to send
2641 * Assert scsi device event asynchronously.
2643 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2645 unsigned long flags
;
2648 /* FIXME: currently this check eliminates all media change events
2649 * for polled devices. Need to update to discriminate between AN
2650 * and polled events */
2651 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2657 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2658 list_add_tail(&evt
->node
, &sdev
->event_list
);
2659 schedule_work(&sdev
->event_work
);
2660 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2662 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2665 * sdev_evt_alloc - allocate a new scsi event
2666 * @evt_type: type of event to allocate
2667 * @gfpflags: GFP flags for allocation
2669 * Allocates and returns a new scsi_event.
2671 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2674 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2678 evt
->evt_type
= evt_type
;
2679 INIT_LIST_HEAD(&evt
->node
);
2681 /* evt_type-specific initialization, if any */
2683 case SDEV_EVT_MEDIA_CHANGE
:
2684 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2685 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2686 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2687 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2688 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2696 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2699 * sdev_evt_send_simple - send asserted event to uevent thread
2700 * @sdev: scsi_device event occurred on
2701 * @evt_type: type of event to send
2702 * @gfpflags: GFP flags for allocation
2704 * Assert scsi device event asynchronously, given an event type.
2706 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2707 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2709 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2711 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2716 sdev_evt_send(sdev
, evt
);
2718 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2721 * scsi_device_quiesce - Block user issued commands.
2722 * @sdev: scsi device to quiesce.
2724 * This works by trying to transition to the SDEV_QUIESCE state
2725 * (which must be a legal transition). When the device is in this
2726 * state, only special requests will be accepted, all others will
2727 * be deferred. Since special requests may also be requeued requests,
2728 * a successful return doesn't guarantee the device will be
2729 * totally quiescent.
2731 * Must be called with user context, may sleep.
2733 * Returns zero if unsuccessful or an error if not.
2736 scsi_device_quiesce(struct scsi_device
*sdev
)
2738 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2742 scsi_run_queue(sdev
->request_queue
);
2743 while (atomic_read(&sdev
->device_busy
)) {
2744 msleep_interruptible(200);
2745 scsi_run_queue(sdev
->request_queue
);
2749 EXPORT_SYMBOL(scsi_device_quiesce
);
2752 * scsi_device_resume - Restart user issued commands to a quiesced device.
2753 * @sdev: scsi device to resume.
2755 * Moves the device from quiesced back to running and restarts the
2758 * Must be called with user context, may sleep.
2760 void scsi_device_resume(struct scsi_device
*sdev
)
2762 /* check if the device state was mutated prior to resume, and if
2763 * so assume the state is being managed elsewhere (for example
2764 * device deleted during suspend)
2766 if (sdev
->sdev_state
!= SDEV_QUIESCE
||
2767 scsi_device_set_state(sdev
, SDEV_RUNNING
))
2769 scsi_run_queue(sdev
->request_queue
);
2771 EXPORT_SYMBOL(scsi_device_resume
);
2774 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2776 scsi_device_quiesce(sdev
);
2780 scsi_target_quiesce(struct scsi_target
*starget
)
2782 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2784 EXPORT_SYMBOL(scsi_target_quiesce
);
2787 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2789 scsi_device_resume(sdev
);
2793 scsi_target_resume(struct scsi_target
*starget
)
2795 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2797 EXPORT_SYMBOL(scsi_target_resume
);
2800 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2801 * @sdev: device to block
2803 * Block request made by scsi lld's to temporarily stop all
2804 * scsi commands on the specified device. Called from interrupt
2805 * or normal process context.
2807 * Returns zero if successful or error if not
2810 * This routine transitions the device to the SDEV_BLOCK state
2811 * (which must be a legal transition). When the device is in this
2812 * state, all commands are deferred until the scsi lld reenables
2813 * the device with scsi_device_unblock or device_block_tmo fires.
2816 scsi_internal_device_block(struct scsi_device
*sdev
)
2818 struct request_queue
*q
= sdev
->request_queue
;
2819 unsigned long flags
;
2822 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2824 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2831 * The device has transitioned to SDEV_BLOCK. Stop the
2832 * block layer from calling the midlayer with this device's
2836 blk_mq_stop_hw_queues(q
);
2838 spin_lock_irqsave(q
->queue_lock
, flags
);
2840 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2845 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2848 * scsi_internal_device_unblock - resume a device after a block request
2849 * @sdev: device to resume
2850 * @new_state: state to set devices to after unblocking
2852 * Called by scsi lld's or the midlayer to restart the device queue
2853 * for the previously suspended scsi device. Called from interrupt or
2854 * normal process context.
2856 * Returns zero if successful or error if not.
2859 * This routine transitions the device to the SDEV_RUNNING state
2860 * or to one of the offline states (which must be a legal transition)
2861 * allowing the midlayer to goose the queue for this device.
2864 scsi_internal_device_unblock(struct scsi_device
*sdev
,
2865 enum scsi_device_state new_state
)
2867 struct request_queue
*q
= sdev
->request_queue
;
2868 unsigned long flags
;
2871 * Try to transition the scsi device to SDEV_RUNNING or one of the
2872 * offlined states and goose the device queue if successful.
2874 if ((sdev
->sdev_state
== SDEV_BLOCK
) ||
2875 (sdev
->sdev_state
== SDEV_TRANSPORT_OFFLINE
))
2876 sdev
->sdev_state
= new_state
;
2877 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
) {
2878 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2879 new_state
== SDEV_OFFLINE
)
2880 sdev
->sdev_state
= new_state
;
2882 sdev
->sdev_state
= SDEV_CREATED
;
2883 } else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2884 sdev
->sdev_state
!= SDEV_OFFLINE
)
2888 blk_mq_start_stopped_hw_queues(q
, false);
2890 spin_lock_irqsave(q
->queue_lock
, flags
);
2892 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2897 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2900 device_block(struct scsi_device
*sdev
, void *data
)
2902 scsi_internal_device_block(sdev
);
2906 target_block(struct device
*dev
, void *data
)
2908 if (scsi_is_target_device(dev
))
2909 starget_for_each_device(to_scsi_target(dev
), NULL
,
2915 scsi_target_block(struct device
*dev
)
2917 if (scsi_is_target_device(dev
))
2918 starget_for_each_device(to_scsi_target(dev
), NULL
,
2921 device_for_each_child(dev
, NULL
, target_block
);
2923 EXPORT_SYMBOL_GPL(scsi_target_block
);
2926 device_unblock(struct scsi_device
*sdev
, void *data
)
2928 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2932 target_unblock(struct device
*dev
, void *data
)
2934 if (scsi_is_target_device(dev
))
2935 starget_for_each_device(to_scsi_target(dev
), data
,
2941 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2943 if (scsi_is_target_device(dev
))
2944 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2947 device_for_each_child(dev
, &new_state
, target_unblock
);
2949 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2952 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2953 * @sgl: scatter-gather list
2954 * @sg_count: number of segments in sg
2955 * @offset: offset in bytes into sg, on return offset into the mapped area
2956 * @len: bytes to map, on return number of bytes mapped
2958 * Returns virtual address of the start of the mapped page
2960 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2961 size_t *offset
, size_t *len
)
2964 size_t sg_len
= 0, len_complete
= 0;
2965 struct scatterlist
*sg
;
2968 WARN_ON(!irqs_disabled());
2970 for_each_sg(sgl
, sg
, sg_count
, i
) {
2971 len_complete
= sg_len
; /* Complete sg-entries */
2972 sg_len
+= sg
->length
;
2973 if (sg_len
> *offset
)
2977 if (unlikely(i
== sg_count
)) {
2978 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2980 __func__
, sg_len
, *offset
, sg_count
);
2985 /* Offset starting from the beginning of first page in this sg-entry */
2986 *offset
= *offset
- len_complete
+ sg
->offset
;
2988 /* Assumption: contiguous pages can be accessed as "page + i" */
2989 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2990 *offset
&= ~PAGE_MASK
;
2992 /* Bytes in this sg-entry from *offset to the end of the page */
2993 sg_len
= PAGE_SIZE
- *offset
;
2997 return kmap_atomic(page
);
2999 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3002 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3003 * @virt: virtual address to be unmapped
3005 void scsi_kunmap_atomic_sg(void *virt
)
3007 kunmap_atomic(virt
);
3009 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3011 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3013 atomic_inc(&sdev
->disk_events_disable_depth
);
3015 EXPORT_SYMBOL(sdev_disable_disk_events
);
3017 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3019 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3021 atomic_dec(&sdev
->disk_events_disable_depth
);
3023 EXPORT_SYMBOL(sdev_enable_disk_events
);