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) { .size = 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_request,
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_request(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
);
739 scsi_release_bidi_buffers(cmd
);
741 spin_lock_irqsave(q
->queue_lock
, flags
);
742 blk_finish_request(req
, error
);
743 spin_unlock_irqrestore(q
->queue_lock
, flags
);
745 scsi_release_buffers(cmd
);
746 scsi_next_command(cmd
);
753 * __scsi_error_from_host_byte - translate SCSI error code into errno
754 * @cmd: SCSI command (unused)
755 * @result: scsi error code
757 * Translate SCSI error code into standard UNIX errno.
759 * -ENOLINK temporary transport failure
760 * -EREMOTEIO permanent target failure, do not retry
761 * -EBADE permanent nexus failure, retry on other path
762 * -ENOSPC No write space available
763 * -ENODATA Medium error
764 * -EIO unspecified I/O error
766 static int __scsi_error_from_host_byte(struct scsi_cmnd
*cmd
, int result
)
770 switch(host_byte(result
)) {
771 case DID_TRANSPORT_FAILFAST
:
774 case DID_TARGET_FAILURE
:
775 set_host_byte(cmd
, DID_OK
);
778 case DID_NEXUS_FAILURE
:
779 set_host_byte(cmd
, DID_OK
);
782 case DID_ALLOC_FAILURE
:
783 set_host_byte(cmd
, DID_OK
);
786 case DID_MEDIUM_ERROR
:
787 set_host_byte(cmd
, DID_OK
);
799 * Function: scsi_io_completion()
801 * Purpose: Completion processing for block device I/O requests.
803 * Arguments: cmd - command that is finished.
805 * Lock status: Assumed that no lock is held upon entry.
809 * Notes: We will finish off the specified number of sectors. If we
810 * are done, the command block will be released and the queue
811 * function will be goosed. If we are not done then we have to
812 * figure out what to do next:
814 * a) We can call scsi_requeue_command(). The request
815 * will be unprepared and put back on the queue. Then
816 * a new command will be created for it. This should
817 * be used if we made forward progress, or if we want
818 * to switch from READ(10) to READ(6) for example.
820 * b) We can call __scsi_queue_insert(). The request will
821 * be put back on the queue and retried using the same
822 * command as before, possibly after a delay.
824 * c) We can call scsi_end_request() with -EIO to fail
825 * the remainder of the request.
827 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
829 int result
= cmd
->result
;
830 struct request_queue
*q
= cmd
->device
->request_queue
;
831 struct request
*req
= cmd
->request
;
833 struct scsi_sense_hdr sshdr
;
835 int sense_deferred
= 0;
836 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
837 ACTION_DELAYED_RETRY
} action
;
838 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
841 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
843 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
846 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
848 if (sense_valid
&& req
->sense
) {
850 * SG_IO wants current and deferred errors
852 int len
= 8 + cmd
->sense_buffer
[7];
854 if (len
> SCSI_SENSE_BUFFERSIZE
)
855 len
= SCSI_SENSE_BUFFERSIZE
;
856 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
857 req
->sense_len
= len
;
860 error
= __scsi_error_from_host_byte(cmd
, result
);
863 * __scsi_error_from_host_byte may have reset the host_byte
865 req
->errors
= cmd
->result
;
867 req
->resid_len
= scsi_get_resid(cmd
);
869 if (scsi_bidi_cmnd(cmd
)) {
871 * Bidi commands Must be complete as a whole,
872 * both sides at once.
874 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
875 if (scsi_end_request(req
, 0, blk_rq_bytes(req
),
876 blk_rq_bytes(req
->next_rq
)))
880 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
882 * Certain non BLOCK_PC requests are commands that don't
883 * actually transfer anything (FLUSH), so cannot use
884 * good_bytes != blk_rq_bytes(req) as the signal for an error.
885 * This sets the error explicitly for the problem case.
887 error
= __scsi_error_from_host_byte(cmd
, result
);
890 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
891 BUG_ON(blk_bidi_rq(req
));
894 * Next deal with any sectors which we were able to correctly
897 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
898 "%u sectors total, %d bytes done.\n",
899 blk_rq_sectors(req
), good_bytes
));
902 * Recovered errors need reporting, but they're always treated
903 * as success, so fiddle the result code here. For BLOCK_PC
904 * we already took a copy of the original into rq->errors which
905 * is what gets returned to the user
907 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
908 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
909 * print since caller wants ATA registers. Only occurs on
910 * SCSI ATA PASS_THROUGH commands when CK_COND=1
912 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
914 else if (!(req
->cmd_flags
& REQ_QUIET
))
915 scsi_print_sense("", cmd
);
917 /* BLOCK_PC may have set error */
922 * If we finished all bytes in the request we are done now.
924 if (!scsi_end_request(req
, error
, good_bytes
, 0))
928 * Kill remainder if no retrys.
930 if (error
&& scsi_noretry_cmd(cmd
)) {
931 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
937 * If there had been no error, but we have leftover bytes in the
938 * requeues just queue the command up again.
943 error
= __scsi_error_from_host_byte(cmd
, result
);
945 if (host_byte(result
) == DID_RESET
) {
946 /* Third party bus reset or reset for error recovery
947 * reasons. Just retry the command and see what
950 action
= ACTION_RETRY
;
951 } else if (sense_valid
&& !sense_deferred
) {
952 switch (sshdr
.sense_key
) {
954 if (cmd
->device
->removable
) {
955 /* Detected disc change. Set a bit
956 * and quietly refuse further access.
958 cmd
->device
->changed
= 1;
959 action
= ACTION_FAIL
;
961 /* Must have been a power glitch, or a
962 * bus reset. Could not have been a
963 * media change, so we just retry the
964 * command and see what happens.
966 action
= ACTION_RETRY
;
969 case ILLEGAL_REQUEST
:
970 /* If we had an ILLEGAL REQUEST returned, then
971 * we may have performed an unsupported
972 * command. The only thing this should be
973 * would be a ten byte read where only a six
974 * byte read was supported. Also, on a system
975 * where READ CAPACITY failed, we may have
976 * read past the end of the disk.
978 if ((cmd
->device
->use_10_for_rw
&&
979 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
980 (cmd
->cmnd
[0] == READ_10
||
981 cmd
->cmnd
[0] == WRITE_10
)) {
982 /* This will issue a new 6-byte command. */
983 cmd
->device
->use_10_for_rw
= 0;
984 action
= ACTION_REPREP
;
985 } else if (sshdr
.asc
== 0x10) /* DIX */ {
986 action
= ACTION_FAIL
;
988 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
989 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
990 action
= ACTION_FAIL
;
993 action
= ACTION_FAIL
;
995 case ABORTED_COMMAND
:
996 action
= ACTION_FAIL
;
997 if (sshdr
.asc
== 0x10) /* DIF */
1001 /* If the device is in the process of becoming
1002 * ready, or has a temporary blockage, retry.
1004 if (sshdr
.asc
== 0x04) {
1005 switch (sshdr
.ascq
) {
1006 case 0x01: /* becoming ready */
1007 case 0x04: /* format in progress */
1008 case 0x05: /* rebuild in progress */
1009 case 0x06: /* recalculation in progress */
1010 case 0x07: /* operation in progress */
1011 case 0x08: /* Long write in progress */
1012 case 0x09: /* self test in progress */
1013 case 0x14: /* space allocation in progress */
1014 action
= ACTION_DELAYED_RETRY
;
1017 action
= ACTION_FAIL
;
1021 action
= ACTION_FAIL
;
1023 case VOLUME_OVERFLOW
:
1024 /* See SSC3rXX or current. */
1025 action
= ACTION_FAIL
;
1028 action
= ACTION_FAIL
;
1032 action
= ACTION_FAIL
;
1034 if (action
!= ACTION_FAIL
&&
1035 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
1036 action
= ACTION_FAIL
;
1040 /* Give up and fail the remainder of the request */
1041 if (!(req
->cmd_flags
& REQ_QUIET
)) {
1042 scsi_print_result(cmd
);
1043 if (driver_byte(result
) & DRIVER_SENSE
)
1044 scsi_print_sense("", cmd
);
1045 scsi_print_command(cmd
);
1047 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
1052 /* Unprep the request and put it back at the head of the queue.
1053 * A new command will be prepared and issued.
1056 cmd
->request
->cmd_flags
&= ~REQ_DONTPREP
;
1057 scsi_mq_uninit_cmd(cmd
);
1058 scsi_mq_requeue_cmd(cmd
);
1060 scsi_release_buffers(cmd
);
1061 scsi_requeue_command(q
, cmd
);
1065 /* Retry the same command immediately */
1066 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
1068 case ACTION_DELAYED_RETRY
:
1069 /* Retry the same command after a delay */
1070 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
1075 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
1081 * If sg table allocation fails, requeue request later.
1083 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
1084 gfp_mask
, req
->mq_ctx
!= NULL
)))
1085 return BLKPREP_DEFER
;
1088 * Next, walk the list, and fill in the addresses and sizes of
1091 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1092 BUG_ON(count
> sdb
->table
.nents
);
1093 sdb
->table
.nents
= count
;
1094 sdb
->length
= blk_rq_bytes(req
);
1099 * Function: scsi_init_io()
1101 * Purpose: SCSI I/O initialize function.
1103 * Arguments: cmd - Command descriptor we wish to initialize
1105 * Returns: 0 on success
1106 * BLKPREP_DEFER if the failure is retryable
1107 * BLKPREP_KILL if the failure is fatal
1109 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
1111 struct scsi_device
*sdev
= cmd
->device
;
1112 struct request
*rq
= cmd
->request
;
1113 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1116 BUG_ON(!rq
->nr_phys_segments
);
1118 error
= scsi_init_sgtable(rq
, &cmd
->sdb
, gfp_mask
);
1122 if (blk_bidi_rq(rq
)) {
1123 if (!rq
->q
->mq_ops
) {
1124 struct scsi_data_buffer
*bidi_sdb
=
1125 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1127 error
= BLKPREP_DEFER
;
1131 rq
->next_rq
->special
= bidi_sdb
;
1134 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
,
1140 if (blk_integrity_rq(rq
)) {
1141 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1144 BUG_ON(prot_sdb
== NULL
);
1145 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1147 if (scsi_alloc_sgtable(prot_sdb
, ivecs
, gfp_mask
, is_mq
)) {
1148 error
= BLKPREP_DEFER
;
1152 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1153 prot_sdb
->table
.sgl
);
1154 BUG_ON(unlikely(count
> ivecs
));
1155 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1157 cmd
->prot_sdb
= prot_sdb
;
1158 cmd
->prot_sdb
->table
.nents
= count
;
1164 scsi_mq_free_sgtables(cmd
);
1166 scsi_release_buffers(cmd
);
1167 cmd
->request
->special
= NULL
;
1168 scsi_put_command(cmd
);
1169 put_device(&sdev
->sdev_gendev
);
1173 EXPORT_SYMBOL(scsi_init_io
);
1175 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1176 struct request
*req
)
1178 struct scsi_cmnd
*cmd
;
1180 if (!req
->special
) {
1181 /* Bail if we can't get a reference to the device */
1182 if (!get_device(&sdev
->sdev_gendev
))
1185 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1186 if (unlikely(!cmd
)) {
1187 put_device(&sdev
->sdev_gendev
);
1195 /* pull a tag out of the request if we have one */
1196 cmd
->tag
= req
->tag
;
1199 cmd
->cmnd
= req
->cmd
;
1200 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1205 static int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1207 struct scsi_cmnd
*cmd
= req
->special
;
1210 * BLOCK_PC requests may transfer data, in which case they must
1211 * a bio attached to them. Or they might contain a SCSI command
1212 * that does not transfer data, in which case they may optionally
1213 * submit a request without an attached bio.
1216 int ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1220 BUG_ON(blk_rq_bytes(req
));
1222 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1225 cmd
->cmd_len
= req
->cmd_len
;
1226 cmd
->transfersize
= blk_rq_bytes(req
);
1227 cmd
->allowed
= req
->retries
;
1232 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1233 * that still need to be translated to SCSI CDBs from the ULD.
1235 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1237 struct scsi_cmnd
*cmd
= req
->special
;
1239 if (unlikely(sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
1240 && sdev
->scsi_dh_data
->scsi_dh
->prep_fn
)) {
1241 int ret
= sdev
->scsi_dh_data
->scsi_dh
->prep_fn(sdev
, req
);
1242 if (ret
!= BLKPREP_OK
)
1246 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1247 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1250 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1252 struct scsi_cmnd
*cmd
= req
->special
;
1254 if (!blk_rq_bytes(req
))
1255 cmd
->sc_data_direction
= DMA_NONE
;
1256 else if (rq_data_dir(req
) == WRITE
)
1257 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1259 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1261 switch (req
->cmd_type
) {
1263 return scsi_setup_fs_cmnd(sdev
, req
);
1264 case REQ_TYPE_BLOCK_PC
:
1265 return scsi_setup_blk_pc_cmnd(sdev
, req
);
1267 return BLKPREP_KILL
;
1272 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1274 int ret
= BLKPREP_OK
;
1277 * If the device is not in running state we will reject some
1280 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1281 switch (sdev
->sdev_state
) {
1283 case SDEV_TRANSPORT_OFFLINE
:
1285 * If the device is offline we refuse to process any
1286 * commands. The device must be brought online
1287 * before trying any recovery commands.
1289 sdev_printk(KERN_ERR
, sdev
,
1290 "rejecting I/O to offline device\n");
1295 * If the device is fully deleted, we refuse to
1296 * process any commands as well.
1298 sdev_printk(KERN_ERR
, sdev
,
1299 "rejecting I/O to dead device\n");
1304 case SDEV_CREATED_BLOCK
:
1306 * If the devices is blocked we defer normal commands.
1308 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1309 ret
= BLKPREP_DEFER
;
1313 * For any other not fully online state we only allow
1314 * special commands. In particular any user initiated
1315 * command is not allowed.
1317 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1326 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1328 struct scsi_device
*sdev
= q
->queuedata
;
1332 req
->errors
= DID_NO_CONNECT
<< 16;
1333 /* release the command and kill it */
1335 struct scsi_cmnd
*cmd
= req
->special
;
1336 scsi_release_buffers(cmd
);
1337 scsi_put_command(cmd
);
1338 put_device(&sdev
->sdev_gendev
);
1339 req
->special
= NULL
;
1344 * If we defer, the blk_peek_request() returns NULL, but the
1345 * queue must be restarted, so we schedule a callback to happen
1348 if (atomic_read(&sdev
->device_busy
) == 0)
1349 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1352 req
->cmd_flags
|= REQ_DONTPREP
;
1358 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1360 struct scsi_device
*sdev
= q
->queuedata
;
1361 struct scsi_cmnd
*cmd
;
1364 ret
= scsi_prep_state_check(sdev
, req
);
1365 if (ret
!= BLKPREP_OK
)
1368 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1369 if (unlikely(!cmd
)) {
1370 ret
= BLKPREP_DEFER
;
1374 ret
= scsi_setup_cmnd(sdev
, req
);
1376 return scsi_prep_return(q
, req
, ret
);
1379 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1381 scsi_uninit_cmd(req
->special
);
1385 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1388 * Called with the queue_lock held.
1390 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1391 struct scsi_device
*sdev
)
1395 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1396 if (atomic_read(&sdev
->device_blocked
)) {
1401 * unblock after device_blocked iterates to zero
1403 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1405 * For the MQ case we take care of this in the caller.
1408 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1411 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1412 "unblocking device at zero depth\n"));
1415 if (busy
>= sdev
->queue_depth
)
1420 atomic_dec(&sdev
->device_busy
);
1425 * scsi_target_queue_ready: checks if there we can send commands to target
1426 * @sdev: scsi device on starget to check.
1428 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1429 struct scsi_device
*sdev
)
1431 struct scsi_target
*starget
= scsi_target(sdev
);
1434 if (starget
->single_lun
) {
1435 spin_lock_irq(shost
->host_lock
);
1436 if (starget
->starget_sdev_user
&&
1437 starget
->starget_sdev_user
!= sdev
) {
1438 spin_unlock_irq(shost
->host_lock
);
1441 starget
->starget_sdev_user
= sdev
;
1442 spin_unlock_irq(shost
->host_lock
);
1445 if (starget
->can_queue
<= 0)
1448 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1449 if (atomic_read(&starget
->target_blocked
) > 0) {
1454 * unblock after target_blocked iterates to zero
1456 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1459 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1460 "unblocking target at zero depth\n"));
1463 if (busy
>= starget
->can_queue
)
1469 spin_lock_irq(shost
->host_lock
);
1470 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1471 spin_unlock_irq(shost
->host_lock
);
1473 if (starget
->can_queue
> 0)
1474 atomic_dec(&starget
->target_busy
);
1479 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1480 * return 0. We must end up running the queue again whenever 0 is
1481 * returned, else IO can hang.
1483 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1484 struct Scsi_Host
*shost
,
1485 struct scsi_device
*sdev
)
1489 if (scsi_host_in_recovery(shost
))
1492 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1493 if (atomic_read(&shost
->host_blocked
) > 0) {
1498 * unblock after host_blocked iterates to zero
1500 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1504 shost_printk(KERN_INFO
, shost
,
1505 "unblocking host at zero depth\n"));
1508 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1510 if (shost
->host_self_blocked
)
1513 /* We're OK to process the command, so we can't be starved */
1514 if (!list_empty(&sdev
->starved_entry
)) {
1515 spin_lock_irq(shost
->host_lock
);
1516 if (!list_empty(&sdev
->starved_entry
))
1517 list_del_init(&sdev
->starved_entry
);
1518 spin_unlock_irq(shost
->host_lock
);
1524 spin_lock_irq(shost
->host_lock
);
1525 if (list_empty(&sdev
->starved_entry
))
1526 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1527 spin_unlock_irq(shost
->host_lock
);
1529 atomic_dec(&shost
->host_busy
);
1534 * Busy state exporting function for request stacking drivers.
1536 * For efficiency, no lock is taken to check the busy state of
1537 * shost/starget/sdev, since the returned value is not guaranteed and
1538 * may be changed after request stacking drivers call the function,
1539 * regardless of taking lock or not.
1541 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1542 * needs to return 'not busy'. Otherwise, request stacking drivers
1543 * may hold requests forever.
1545 static int scsi_lld_busy(struct request_queue
*q
)
1547 struct scsi_device
*sdev
= q
->queuedata
;
1548 struct Scsi_Host
*shost
;
1550 if (blk_queue_dying(q
))
1556 * Ignore host/starget busy state.
1557 * Since block layer does not have a concept of fairness across
1558 * multiple queues, congestion of host/starget needs to be handled
1561 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1568 * Kill a request for a dead device
1570 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1572 struct scsi_cmnd
*cmd
= req
->special
;
1573 struct scsi_device
*sdev
;
1574 struct scsi_target
*starget
;
1575 struct Scsi_Host
*shost
;
1577 blk_start_request(req
);
1579 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1582 starget
= scsi_target(sdev
);
1584 scsi_init_cmd_errh(cmd
);
1585 cmd
->result
= DID_NO_CONNECT
<< 16;
1586 atomic_inc(&cmd
->device
->iorequest_cnt
);
1589 * SCSI request completion path will do scsi_device_unbusy(),
1590 * bump busy counts. To bump the counters, we need to dance
1591 * with the locks as normal issue path does.
1593 atomic_inc(&sdev
->device_busy
);
1594 atomic_inc(&shost
->host_busy
);
1595 if (starget
->can_queue
> 0)
1596 atomic_inc(&starget
->target_busy
);
1598 blk_complete_request(req
);
1601 static void scsi_softirq_done(struct request
*rq
)
1603 struct scsi_cmnd
*cmd
= rq
->special
;
1604 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1607 INIT_LIST_HEAD(&cmd
->eh_entry
);
1609 atomic_inc(&cmd
->device
->iodone_cnt
);
1611 atomic_inc(&cmd
->device
->ioerr_cnt
);
1613 disposition
= scsi_decide_disposition(cmd
);
1614 if (disposition
!= SUCCESS
&&
1615 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1616 sdev_printk(KERN_ERR
, cmd
->device
,
1617 "timing out command, waited %lus\n",
1619 disposition
= SUCCESS
;
1622 scsi_log_completion(cmd
, disposition
);
1624 switch (disposition
) {
1626 scsi_finish_command(cmd
);
1629 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1631 case ADD_TO_MLQUEUE
:
1632 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1635 if (!scsi_eh_scmd_add(cmd
, 0))
1636 scsi_finish_command(cmd
);
1641 * scsi_done - Invoke completion on finished SCSI command.
1642 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1643 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1645 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1646 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1647 * calls blk_complete_request() for further processing.
1649 * This function is interrupt context safe.
1651 static void scsi_done(struct scsi_cmnd
*cmd
)
1653 trace_scsi_dispatch_cmd_done(cmd
);
1654 blk_complete_request(cmd
->request
);
1658 * Function: scsi_request_fn()
1660 * Purpose: Main strategy routine for SCSI.
1662 * Arguments: q - Pointer to actual queue.
1666 * Lock status: IO request lock assumed to be held when called.
1668 static void scsi_request_fn(struct request_queue
*q
)
1669 __releases(q
->queue_lock
)
1670 __acquires(q
->queue_lock
)
1672 struct scsi_device
*sdev
= q
->queuedata
;
1673 struct Scsi_Host
*shost
;
1674 struct scsi_cmnd
*cmd
;
1675 struct request
*req
;
1678 * To start with, we keep looping until the queue is empty, or until
1679 * the host is no longer able to accept any more requests.
1685 * get next queueable request. We do this early to make sure
1686 * that the request is fully prepared even if we cannot
1689 req
= blk_peek_request(q
);
1693 if (unlikely(!scsi_device_online(sdev
))) {
1694 sdev_printk(KERN_ERR
, sdev
,
1695 "rejecting I/O to offline device\n");
1696 scsi_kill_request(req
, q
);
1700 if (!scsi_dev_queue_ready(q
, sdev
))
1704 * Remove the request from the request list.
1706 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1707 blk_start_request(req
);
1709 spin_unlock_irq(q
->queue_lock
);
1711 if (unlikely(cmd
== NULL
)) {
1712 printk(KERN_CRIT
"impossible request in %s.\n"
1713 "please mail a stack trace to "
1714 "linux-scsi@vger.kernel.org\n",
1716 blk_dump_rq_flags(req
, "foo");
1721 * We hit this when the driver is using a host wide
1722 * tag map. For device level tag maps the queue_depth check
1723 * in the device ready fn would prevent us from trying
1724 * to allocate a tag. Since the map is a shared host resource
1725 * we add the dev to the starved list so it eventually gets
1726 * a run when a tag is freed.
1728 if (blk_queue_tagged(q
) && !blk_rq_tagged(req
)) {
1729 spin_lock_irq(shost
->host_lock
);
1730 if (list_empty(&sdev
->starved_entry
))
1731 list_add_tail(&sdev
->starved_entry
,
1732 &shost
->starved_list
);
1733 spin_unlock_irq(shost
->host_lock
);
1737 if (!scsi_target_queue_ready(shost
, sdev
))
1740 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1741 goto host_not_ready
;
1744 * Finally, initialize any error handling parameters, and set up
1745 * the timers for timeouts.
1747 scsi_init_cmd_errh(cmd
);
1750 * Dispatch the command to the low-level driver.
1752 cmd
->scsi_done
= scsi_done
;
1753 rtn
= scsi_dispatch_cmd(cmd
);
1755 scsi_queue_insert(cmd
, rtn
);
1756 spin_lock_irq(q
->queue_lock
);
1759 spin_lock_irq(q
->queue_lock
);
1765 if (scsi_target(sdev
)->can_queue
> 0)
1766 atomic_dec(&scsi_target(sdev
)->target_busy
);
1769 * lock q, handle tag, requeue req, and decrement device_busy. We
1770 * must return with queue_lock held.
1772 * Decrementing device_busy without checking it is OK, as all such
1773 * cases (host limits or settings) should run the queue at some
1776 spin_lock_irq(q
->queue_lock
);
1777 blk_requeue_request(q
, req
);
1778 atomic_dec(&sdev
->device_busy
);
1780 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1781 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1784 static inline int prep_to_mq(int ret
)
1790 return BLK_MQ_RQ_QUEUE_BUSY
;
1792 return BLK_MQ_RQ_QUEUE_ERROR
;
1796 static int scsi_mq_prep_fn(struct request
*req
)
1798 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1799 struct scsi_device
*sdev
= req
->q
->queuedata
;
1800 struct Scsi_Host
*shost
= sdev
->host
;
1801 unsigned char *sense_buf
= cmd
->sense_buffer
;
1802 struct scatterlist
*sg
;
1804 memset(cmd
, 0, sizeof(struct scsi_cmnd
));
1810 cmd
->sense_buffer
= sense_buf
;
1812 cmd
->tag
= req
->tag
;
1814 cmd
->cmnd
= req
->cmd
;
1815 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1817 INIT_LIST_HEAD(&cmd
->list
);
1818 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1819 cmd
->jiffies_at_alloc
= jiffies
;
1821 if (shost
->use_cmd_list
) {
1822 spin_lock_irq(&sdev
->list_lock
);
1823 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1824 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 blk_mq_start_request(req
);
1852 return scsi_setup_cmnd(sdev
, req
);
1855 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1857 trace_scsi_dispatch_cmd_done(cmd
);
1858 blk_mq_complete_request(cmd
->request
);
1861 static int scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
, struct request
*req
,
1864 struct request_queue
*q
= req
->q
;
1865 struct scsi_device
*sdev
= q
->queuedata
;
1866 struct Scsi_Host
*shost
= sdev
->host
;
1867 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1871 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
1875 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1876 if (!get_device(&sdev
->sdev_gendev
))
1879 if (!scsi_dev_queue_ready(q
, sdev
))
1880 goto out_put_device
;
1881 if (!scsi_target_queue_ready(shost
, sdev
))
1882 goto out_dec_device_busy
;
1883 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1884 goto out_dec_target_busy
;
1887 if (!(req
->cmd_flags
& REQ_DONTPREP
)) {
1888 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
1890 goto out_dec_host_busy
;
1891 req
->cmd_flags
|= REQ_DONTPREP
;
1893 blk_mq_start_request(req
);
1896 if (blk_queue_tagged(q
))
1897 req
->cmd_flags
|= REQ_QUEUED
;
1899 req
->cmd_flags
&= ~REQ_QUEUED
;
1901 scsi_init_cmd_errh(cmd
);
1902 cmd
->scsi_done
= scsi_mq_done
;
1904 reason
= scsi_dispatch_cmd(cmd
);
1906 scsi_set_blocked(cmd
, reason
);
1907 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1908 goto out_dec_host_busy
;
1911 return BLK_MQ_RQ_QUEUE_OK
;
1914 atomic_dec(&shost
->host_busy
);
1915 out_dec_target_busy
:
1916 if (scsi_target(sdev
)->can_queue
> 0)
1917 atomic_dec(&scsi_target(sdev
)->target_busy
);
1918 out_dec_device_busy
:
1919 atomic_dec(&sdev
->device_busy
);
1921 put_device(&sdev
->sdev_gendev
);
1924 case BLK_MQ_RQ_QUEUE_BUSY
:
1925 blk_mq_stop_hw_queue(hctx
);
1926 if (atomic_read(&sdev
->device_busy
) == 0 &&
1927 !scsi_device_blocked(sdev
))
1928 blk_mq_delay_queue(hctx
, SCSI_QUEUE_DELAY
);
1930 case BLK_MQ_RQ_QUEUE_ERROR
:
1932 * Make sure to release all allocated ressources when
1933 * we hit an error, as we will never see this command
1936 if (req
->cmd_flags
& REQ_DONTPREP
)
1937 scsi_mq_uninit_cmd(cmd
);
1945 static enum blk_eh_timer_return
scsi_timeout(struct request
*req
,
1949 return BLK_EH_RESET_TIMER
;
1950 return scsi_times_out(req
);
1953 static int scsi_init_request(void *data
, struct request
*rq
,
1954 unsigned int hctx_idx
, unsigned int request_idx
,
1955 unsigned int numa_node
)
1957 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1959 cmd
->sense_buffer
= kzalloc_node(SCSI_SENSE_BUFFERSIZE
, GFP_KERNEL
,
1961 if (!cmd
->sense_buffer
)
1966 static void scsi_exit_request(void *data
, struct request
*rq
,
1967 unsigned int hctx_idx
, unsigned int request_idx
)
1969 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1971 kfree(cmd
->sense_buffer
);
1974 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1976 struct device
*host_dev
;
1977 u64 bounce_limit
= 0xffffffff;
1979 if (shost
->unchecked_isa_dma
)
1980 return BLK_BOUNCE_ISA
;
1982 * Platforms with virtual-DMA translation
1983 * hardware have no practical limit.
1985 if (!PCI_DMA_BUS_IS_PHYS
)
1986 return BLK_BOUNCE_ANY
;
1988 host_dev
= scsi_get_device(shost
);
1989 if (host_dev
&& host_dev
->dma_mask
)
1990 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
1992 return bounce_limit
;
1995 static void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
1997 struct device
*dev
= shost
->dma_dev
;
2000 * this limit is imposed by hardware restrictions
2002 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
2003 SCSI_MAX_SG_CHAIN_SEGMENTS
));
2005 if (scsi_host_prot_dma(shost
)) {
2006 shost
->sg_prot_tablesize
=
2007 min_not_zero(shost
->sg_prot_tablesize
,
2008 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
2009 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
2010 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
2013 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
2014 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
2015 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
2016 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
2018 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2020 if (!shost
->use_clustering
)
2021 q
->limits
.cluster
= 0;
2024 * set a reasonable default alignment on word boundaries: the
2025 * host and device may alter it using
2026 * blk_queue_update_dma_alignment() later.
2028 blk_queue_dma_alignment(q
, 0x03);
2031 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
2032 request_fn_proc
*request_fn
)
2034 struct request_queue
*q
;
2036 q
= blk_init_queue(request_fn
, NULL
);
2039 __scsi_init_queue(shost
, q
);
2042 EXPORT_SYMBOL(__scsi_alloc_queue
);
2044 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
2046 struct request_queue
*q
;
2048 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
2052 blk_queue_prep_rq(q
, scsi_prep_fn
);
2053 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2054 blk_queue_softirq_done(q
, scsi_softirq_done
);
2055 blk_queue_rq_timed_out(q
, scsi_times_out
);
2056 blk_queue_lld_busy(q
, scsi_lld_busy
);
2060 static struct blk_mq_ops scsi_mq_ops
= {
2061 .map_queue
= blk_mq_map_queue
,
2062 .queue_rq
= scsi_queue_rq
,
2063 .complete
= scsi_softirq_done
,
2064 .timeout
= scsi_timeout
,
2065 .init_request
= scsi_init_request
,
2066 .exit_request
= scsi_exit_request
,
2069 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2071 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2072 if (IS_ERR(sdev
->request_queue
))
2075 sdev
->request_queue
->queuedata
= sdev
;
2076 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2077 return sdev
->request_queue
;
2080 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2082 unsigned int cmd_size
, sgl_size
, tbl_size
;
2084 tbl_size
= shost
->sg_tablesize
;
2085 if (tbl_size
> SCSI_MAX_SG_SEGMENTS
)
2086 tbl_size
= SCSI_MAX_SG_SEGMENTS
;
2087 sgl_size
= tbl_size
* sizeof(struct scatterlist
);
2088 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2089 if (scsi_host_get_prot(shost
))
2090 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2092 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2093 shost
->tag_set
.ops
= &scsi_mq_ops
;
2094 shost
->tag_set
.nr_hw_queues
= 1;
2095 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2096 shost
->tag_set
.cmd_size
= cmd_size
;
2097 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2098 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2099 shost
->tag_set
.driver_data
= shost
;
2101 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2104 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2106 blk_mq_free_tag_set(&shost
->tag_set
);
2110 * Function: scsi_block_requests()
2112 * Purpose: Utility function used by low-level drivers to prevent further
2113 * commands from being queued to the device.
2115 * Arguments: shost - Host in question
2119 * Lock status: No locks are assumed held.
2121 * Notes: There is no timer nor any other means by which the requests
2122 * get unblocked other than the low-level driver calling
2123 * scsi_unblock_requests().
2125 void scsi_block_requests(struct Scsi_Host
*shost
)
2127 shost
->host_self_blocked
= 1;
2129 EXPORT_SYMBOL(scsi_block_requests
);
2132 * Function: scsi_unblock_requests()
2134 * Purpose: Utility function used by low-level drivers to allow further
2135 * commands from being queued to the device.
2137 * Arguments: shost - Host in question
2141 * Lock status: No locks are assumed held.
2143 * Notes: There is no timer nor any other means by which the requests
2144 * get unblocked other than the low-level driver calling
2145 * scsi_unblock_requests().
2147 * This is done as an API function so that changes to the
2148 * internals of the scsi mid-layer won't require wholesale
2149 * changes to drivers that use this feature.
2151 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2153 shost
->host_self_blocked
= 0;
2154 scsi_run_host_queues(shost
);
2156 EXPORT_SYMBOL(scsi_unblock_requests
);
2158 int __init
scsi_init_queue(void)
2162 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2163 sizeof(struct scsi_data_buffer
),
2165 if (!scsi_sdb_cache
) {
2166 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2170 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2171 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2172 int size
= sgp
->size
* sizeof(struct scatterlist
);
2174 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
2175 SLAB_HWCACHE_ALIGN
, NULL
);
2177 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
2182 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
2185 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
2194 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2195 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2197 mempool_destroy(sgp
->pool
);
2199 kmem_cache_destroy(sgp
->slab
);
2201 kmem_cache_destroy(scsi_sdb_cache
);
2206 void scsi_exit_queue(void)
2210 kmem_cache_destroy(scsi_sdb_cache
);
2212 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
2213 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
2214 mempool_destroy(sgp
->pool
);
2215 kmem_cache_destroy(sgp
->slab
);
2220 * scsi_mode_select - issue a mode select
2221 * @sdev: SCSI device to be queried
2222 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2223 * @sp: Save page bit (0 == don't save, 1 == save)
2224 * @modepage: mode page being requested
2225 * @buffer: request buffer (may not be smaller than eight bytes)
2226 * @len: length of request buffer.
2227 * @timeout: command timeout
2228 * @retries: number of retries before failing
2229 * @data: returns a structure abstracting the mode header data
2230 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2231 * must be SCSI_SENSE_BUFFERSIZE big.
2233 * Returns zero if successful; negative error number or scsi
2238 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2239 unsigned char *buffer
, int len
, int timeout
, int retries
,
2240 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2242 unsigned char cmd
[10];
2243 unsigned char *real_buffer
;
2246 memset(cmd
, 0, sizeof(cmd
));
2247 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2249 if (sdev
->use_10_for_ms
) {
2252 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2255 memcpy(real_buffer
+ 8, buffer
, len
);
2259 real_buffer
[2] = data
->medium_type
;
2260 real_buffer
[3] = data
->device_specific
;
2261 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2263 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2264 real_buffer
[7] = data
->block_descriptor_length
;
2266 cmd
[0] = MODE_SELECT_10
;
2270 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2274 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2277 memcpy(real_buffer
+ 4, buffer
, len
);
2280 real_buffer
[1] = data
->medium_type
;
2281 real_buffer
[2] = data
->device_specific
;
2282 real_buffer
[3] = data
->block_descriptor_length
;
2285 cmd
[0] = MODE_SELECT
;
2289 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2290 sshdr
, timeout
, retries
, NULL
);
2294 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2297 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2298 * @sdev: SCSI device to be queried
2299 * @dbd: set if mode sense will allow block descriptors to be returned
2300 * @modepage: mode page being requested
2301 * @buffer: request buffer (may not be smaller than eight bytes)
2302 * @len: length of request buffer.
2303 * @timeout: command timeout
2304 * @retries: number of retries before failing
2305 * @data: returns a structure abstracting the mode header data
2306 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2307 * must be SCSI_SENSE_BUFFERSIZE big.
2309 * Returns zero if unsuccessful, or the header offset (either 4
2310 * or 8 depending on whether a six or ten byte command was
2311 * issued) if successful.
2314 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2315 unsigned char *buffer
, int len
, int timeout
, int retries
,
2316 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2318 unsigned char cmd
[12];
2322 struct scsi_sense_hdr my_sshdr
;
2324 memset(data
, 0, sizeof(*data
));
2325 memset(&cmd
[0], 0, 12);
2326 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2329 /* caller might not be interested in sense, but we need it */
2334 use_10_for_ms
= sdev
->use_10_for_ms
;
2336 if (use_10_for_ms
) {
2340 cmd
[0] = MODE_SENSE_10
;
2347 cmd
[0] = MODE_SENSE
;
2352 memset(buffer
, 0, len
);
2354 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2355 sshdr
, timeout
, retries
, NULL
);
2357 /* This code looks awful: what it's doing is making sure an
2358 * ILLEGAL REQUEST sense return identifies the actual command
2359 * byte as the problem. MODE_SENSE commands can return
2360 * ILLEGAL REQUEST if the code page isn't supported */
2362 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2363 (driver_byte(result
) & DRIVER_SENSE
)) {
2364 if (scsi_sense_valid(sshdr
)) {
2365 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2366 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2368 * Invalid command operation code
2370 sdev
->use_10_for_ms
= 0;
2376 if(scsi_status_is_good(result
)) {
2377 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2378 (modepage
== 6 || modepage
== 8))) {
2379 /* Initio breakage? */
2382 data
->medium_type
= 0;
2383 data
->device_specific
= 0;
2385 data
->block_descriptor_length
= 0;
2386 } else if(use_10_for_ms
) {
2387 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2388 data
->medium_type
= buffer
[2];
2389 data
->device_specific
= buffer
[3];
2390 data
->longlba
= buffer
[4] & 0x01;
2391 data
->block_descriptor_length
= buffer
[6]*256
2394 data
->length
= buffer
[0] + 1;
2395 data
->medium_type
= buffer
[1];
2396 data
->device_specific
= buffer
[2];
2397 data
->block_descriptor_length
= buffer
[3];
2399 data
->header_length
= header_length
;
2404 EXPORT_SYMBOL(scsi_mode_sense
);
2407 * scsi_test_unit_ready - test if unit is ready
2408 * @sdev: scsi device to change the state of.
2409 * @timeout: command timeout
2410 * @retries: number of retries before failing
2411 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2412 * returning sense. Make sure that this is cleared before passing
2415 * Returns zero if unsuccessful or an error if TUR failed. For
2416 * removable media, UNIT_ATTENTION sets ->changed flag.
2419 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2420 struct scsi_sense_hdr
*sshdr_external
)
2423 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2425 struct scsi_sense_hdr
*sshdr
;
2428 if (!sshdr_external
)
2429 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
2431 sshdr
= sshdr_external
;
2433 /* try to eat the UNIT_ATTENTION if there are enough retries */
2435 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2436 timeout
, retries
, NULL
);
2437 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2438 sshdr
->sense_key
== UNIT_ATTENTION
)
2440 } while (scsi_sense_valid(sshdr
) &&
2441 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2443 if (!sshdr_external
)
2447 EXPORT_SYMBOL(scsi_test_unit_ready
);
2450 * scsi_device_set_state - Take the given device through the device state model.
2451 * @sdev: scsi device to change the state of.
2452 * @state: state to change to.
2454 * Returns zero if unsuccessful or an error if the requested
2455 * transition is illegal.
2458 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2460 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2462 if (state
== oldstate
)
2468 case SDEV_CREATED_BLOCK
:
2479 case SDEV_TRANSPORT_OFFLINE
:
2492 case SDEV_TRANSPORT_OFFLINE
:
2500 case SDEV_TRANSPORT_OFFLINE
:
2515 case SDEV_CREATED_BLOCK
:
2522 case SDEV_CREATED_BLOCK
:
2537 case SDEV_TRANSPORT_OFFLINE
:
2550 case SDEV_TRANSPORT_OFFLINE
:
2552 case SDEV_CREATED_BLOCK
:
2560 sdev
->sdev_state
= state
;
2564 SCSI_LOG_ERROR_RECOVERY(1,
2565 sdev_printk(KERN_ERR
, sdev
,
2566 "Illegal state transition %s->%s",
2567 scsi_device_state_name(oldstate
),
2568 scsi_device_state_name(state
))
2572 EXPORT_SYMBOL(scsi_device_set_state
);
2575 * sdev_evt_emit - emit a single SCSI device uevent
2576 * @sdev: associated SCSI device
2577 * @evt: event to emit
2579 * Send a single uevent (scsi_event) to the associated scsi_device.
2581 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2586 switch (evt
->evt_type
) {
2587 case SDEV_EVT_MEDIA_CHANGE
:
2588 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2590 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2591 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2593 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2594 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2596 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2597 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2599 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2600 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2602 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2603 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2612 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2616 * sdev_evt_thread - send a uevent for each scsi event
2617 * @work: work struct for scsi_device
2619 * Dispatch queued events to their associated scsi_device kobjects
2622 void scsi_evt_thread(struct work_struct
*work
)
2624 struct scsi_device
*sdev
;
2625 enum scsi_device_event evt_type
;
2626 LIST_HEAD(event_list
);
2628 sdev
= container_of(work
, struct scsi_device
, event_work
);
2630 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2631 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2632 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2635 struct scsi_event
*evt
;
2636 struct list_head
*this, *tmp
;
2637 unsigned long flags
;
2639 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2640 list_splice_init(&sdev
->event_list
, &event_list
);
2641 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2643 if (list_empty(&event_list
))
2646 list_for_each_safe(this, tmp
, &event_list
) {
2647 evt
= list_entry(this, struct scsi_event
, node
);
2648 list_del(&evt
->node
);
2649 scsi_evt_emit(sdev
, evt
);
2656 * sdev_evt_send - send asserted event to uevent thread
2657 * @sdev: scsi_device event occurred on
2658 * @evt: event to send
2660 * Assert scsi device event asynchronously.
2662 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2664 unsigned long flags
;
2667 /* FIXME: currently this check eliminates all media change events
2668 * for polled devices. Need to update to discriminate between AN
2669 * and polled events */
2670 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2676 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2677 list_add_tail(&evt
->node
, &sdev
->event_list
);
2678 schedule_work(&sdev
->event_work
);
2679 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2681 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2684 * sdev_evt_alloc - allocate a new scsi event
2685 * @evt_type: type of event to allocate
2686 * @gfpflags: GFP flags for allocation
2688 * Allocates and returns a new scsi_event.
2690 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2693 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2697 evt
->evt_type
= evt_type
;
2698 INIT_LIST_HEAD(&evt
->node
);
2700 /* evt_type-specific initialization, if any */
2702 case SDEV_EVT_MEDIA_CHANGE
:
2703 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2704 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2705 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2706 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2707 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2715 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2718 * sdev_evt_send_simple - send asserted event to uevent thread
2719 * @sdev: scsi_device event occurred on
2720 * @evt_type: type of event to send
2721 * @gfpflags: GFP flags for allocation
2723 * Assert scsi device event asynchronously, given an event type.
2725 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2726 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2728 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2730 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2735 sdev_evt_send(sdev
, evt
);
2737 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2740 * scsi_device_quiesce - Block user issued commands.
2741 * @sdev: scsi device to quiesce.
2743 * This works by trying to transition to the SDEV_QUIESCE state
2744 * (which must be a legal transition). When the device is in this
2745 * state, only special requests will be accepted, all others will
2746 * be deferred. Since special requests may also be requeued requests,
2747 * a successful return doesn't guarantee the device will be
2748 * totally quiescent.
2750 * Must be called with user context, may sleep.
2752 * Returns zero if unsuccessful or an error if not.
2755 scsi_device_quiesce(struct scsi_device
*sdev
)
2757 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2761 scsi_run_queue(sdev
->request_queue
);
2762 while (atomic_read(&sdev
->device_busy
)) {
2763 msleep_interruptible(200);
2764 scsi_run_queue(sdev
->request_queue
);
2768 EXPORT_SYMBOL(scsi_device_quiesce
);
2771 * scsi_device_resume - Restart user issued commands to a quiesced device.
2772 * @sdev: scsi device to resume.
2774 * Moves the device from quiesced back to running and restarts the
2777 * Must be called with user context, may sleep.
2779 void scsi_device_resume(struct scsi_device
*sdev
)
2781 /* check if the device state was mutated prior to resume, and if
2782 * so assume the state is being managed elsewhere (for example
2783 * device deleted during suspend)
2785 if (sdev
->sdev_state
!= SDEV_QUIESCE
||
2786 scsi_device_set_state(sdev
, SDEV_RUNNING
))
2788 scsi_run_queue(sdev
->request_queue
);
2790 EXPORT_SYMBOL(scsi_device_resume
);
2793 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2795 scsi_device_quiesce(sdev
);
2799 scsi_target_quiesce(struct scsi_target
*starget
)
2801 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2803 EXPORT_SYMBOL(scsi_target_quiesce
);
2806 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2808 scsi_device_resume(sdev
);
2812 scsi_target_resume(struct scsi_target
*starget
)
2814 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2816 EXPORT_SYMBOL(scsi_target_resume
);
2819 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2820 * @sdev: device to block
2822 * Block request made by scsi lld's to temporarily stop all
2823 * scsi commands on the specified device. Called from interrupt
2824 * or normal process context.
2826 * Returns zero if successful or error if not
2829 * This routine transitions the device to the SDEV_BLOCK state
2830 * (which must be a legal transition). When the device is in this
2831 * state, all commands are deferred until the scsi lld reenables
2832 * the device with scsi_device_unblock or device_block_tmo fires.
2835 scsi_internal_device_block(struct scsi_device
*sdev
)
2837 struct request_queue
*q
= sdev
->request_queue
;
2838 unsigned long flags
;
2841 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2843 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2850 * The device has transitioned to SDEV_BLOCK. Stop the
2851 * block layer from calling the midlayer with this device's
2855 blk_mq_stop_hw_queues(q
);
2857 spin_lock_irqsave(q
->queue_lock
, flags
);
2859 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2864 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2867 * scsi_internal_device_unblock - resume a device after a block request
2868 * @sdev: device to resume
2869 * @new_state: state to set devices to after unblocking
2871 * Called by scsi lld's or the midlayer to restart the device queue
2872 * for the previously suspended scsi device. Called from interrupt or
2873 * normal process context.
2875 * Returns zero if successful or error if not.
2878 * This routine transitions the device to the SDEV_RUNNING state
2879 * or to one of the offline states (which must be a legal transition)
2880 * allowing the midlayer to goose the queue for this device.
2883 scsi_internal_device_unblock(struct scsi_device
*sdev
,
2884 enum scsi_device_state new_state
)
2886 struct request_queue
*q
= sdev
->request_queue
;
2887 unsigned long flags
;
2890 * Try to transition the scsi device to SDEV_RUNNING or one of the
2891 * offlined states and goose the device queue if successful.
2893 if ((sdev
->sdev_state
== SDEV_BLOCK
) ||
2894 (sdev
->sdev_state
== SDEV_TRANSPORT_OFFLINE
))
2895 sdev
->sdev_state
= new_state
;
2896 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
) {
2897 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2898 new_state
== SDEV_OFFLINE
)
2899 sdev
->sdev_state
= new_state
;
2901 sdev
->sdev_state
= SDEV_CREATED
;
2902 } else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2903 sdev
->sdev_state
!= SDEV_OFFLINE
)
2907 blk_mq_start_stopped_hw_queues(q
, false);
2909 spin_lock_irqsave(q
->queue_lock
, flags
);
2911 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2916 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2919 device_block(struct scsi_device
*sdev
, void *data
)
2921 scsi_internal_device_block(sdev
);
2925 target_block(struct device
*dev
, void *data
)
2927 if (scsi_is_target_device(dev
))
2928 starget_for_each_device(to_scsi_target(dev
), NULL
,
2934 scsi_target_block(struct device
*dev
)
2936 if (scsi_is_target_device(dev
))
2937 starget_for_each_device(to_scsi_target(dev
), NULL
,
2940 device_for_each_child(dev
, NULL
, target_block
);
2942 EXPORT_SYMBOL_GPL(scsi_target_block
);
2945 device_unblock(struct scsi_device
*sdev
, void *data
)
2947 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2951 target_unblock(struct device
*dev
, void *data
)
2953 if (scsi_is_target_device(dev
))
2954 starget_for_each_device(to_scsi_target(dev
), data
,
2960 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2962 if (scsi_is_target_device(dev
))
2963 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2966 device_for_each_child(dev
, &new_state
, target_unblock
);
2968 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2971 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2972 * @sgl: scatter-gather list
2973 * @sg_count: number of segments in sg
2974 * @offset: offset in bytes into sg, on return offset into the mapped area
2975 * @len: bytes to map, on return number of bytes mapped
2977 * Returns virtual address of the start of the mapped page
2979 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2980 size_t *offset
, size_t *len
)
2983 size_t sg_len
= 0, len_complete
= 0;
2984 struct scatterlist
*sg
;
2987 WARN_ON(!irqs_disabled());
2989 for_each_sg(sgl
, sg
, sg_count
, i
) {
2990 len_complete
= sg_len
; /* Complete sg-entries */
2991 sg_len
+= sg
->length
;
2992 if (sg_len
> *offset
)
2996 if (unlikely(i
== sg_count
)) {
2997 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2999 __func__
, sg_len
, *offset
, sg_count
);
3004 /* Offset starting from the beginning of first page in this sg-entry */
3005 *offset
= *offset
- len_complete
+ sg
->offset
;
3007 /* Assumption: contiguous pages can be accessed as "page + i" */
3008 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
3009 *offset
&= ~PAGE_MASK
;
3011 /* Bytes in this sg-entry from *offset to the end of the page */
3012 sg_len
= PAGE_SIZE
- *offset
;
3016 return kmap_atomic(page
);
3018 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3021 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3022 * @virt: virtual address to be unmapped
3024 void scsi_kunmap_atomic_sg(void *virt
)
3026 kunmap_atomic(virt
);
3028 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3030 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3032 atomic_inc(&sdev
->disk_events_disable_depth
);
3034 EXPORT_SYMBOL(sdev_disable_disk_events
);
3036 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3038 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3040 atomic_dec(&sdev
->disk_events_disable_depth
);
3042 EXPORT_SYMBOL(sdev_enable_disk_events
);