2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_dh.h>
35 #include <trace/events/scsi.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
41 struct kmem_cache
*scsi_sdb_cache
;
44 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
45 * not change behaviour from the previous unplug mechanism, experimentation
46 * may prove this needs changing.
48 #define SCSI_QUEUE_DELAY 3
51 scsi_set_blocked(struct scsi_cmnd
*cmd
, int reason
)
53 struct Scsi_Host
*host
= cmd
->device
->host
;
54 struct scsi_device
*device
= cmd
->device
;
55 struct scsi_target
*starget
= scsi_target(device
);
58 * Set the appropriate busy bit for the device/host.
60 * If the host/device isn't busy, assume that something actually
61 * completed, and that we should be able to queue a command now.
63 * Note that the prior mid-layer assumption that any host could
64 * always queue at least one command is now broken. The mid-layer
65 * will implement a user specifiable stall (see
66 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
67 * if a command is requeued with no other commands outstanding
68 * either for the device or for the host.
71 case SCSI_MLQUEUE_HOST_BUSY
:
72 atomic_set(&host
->host_blocked
, host
->max_host_blocked
);
74 case SCSI_MLQUEUE_DEVICE_BUSY
:
75 case SCSI_MLQUEUE_EH_RETRY
:
76 atomic_set(&device
->device_blocked
,
77 device
->max_device_blocked
);
79 case SCSI_MLQUEUE_TARGET_BUSY
:
80 atomic_set(&starget
->target_blocked
,
81 starget
->max_target_blocked
);
86 static void scsi_mq_requeue_cmd(struct scsi_cmnd
*cmd
)
88 struct scsi_device
*sdev
= cmd
->device
;
90 blk_mq_requeue_request(cmd
->request
, true);
91 put_device(&sdev
->sdev_gendev
);
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
106 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, int unbusy
)
108 struct scsi_device
*device
= cmd
->device
;
109 struct request_queue
*q
= device
->request_queue
;
112 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
113 "Inserting command %p into mlqueue\n", cmd
));
115 scsi_set_blocked(cmd
, reason
);
118 * Decrement the counters, since these commands are no longer
119 * active on the host/device.
122 scsi_device_unbusy(device
);
125 * Requeue this command. It will go before all other commands
126 * that are already in the queue. Schedule requeue work under
127 * lock such that the kblockd_schedule_work() call happens
128 * before blk_cleanup_queue() finishes.
132 scsi_mq_requeue_cmd(cmd
);
135 spin_lock_irqsave(q
->queue_lock
, flags
);
136 blk_requeue_request(q
, cmd
->request
);
137 kblockd_schedule_work(&device
->requeue_work
);
138 spin_unlock_irqrestore(q
->queue_lock
, flags
);
142 * Function: scsi_queue_insert()
144 * Purpose: Insert a command in the midlevel queue.
146 * Arguments: cmd - command that we are adding to queue.
147 * reason - why we are inserting command to queue.
149 * Lock status: Assumed that lock is not held upon entry.
153 * Notes: We do this for one of two cases. Either the host is busy
154 * and it cannot accept any more commands for the time being,
155 * or the device returned QUEUE_FULL and can accept no more
157 * Notes: This could be called either from an interrupt context or a
158 * normal process context.
160 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
162 __scsi_queue_insert(cmd
, reason
, 1);
165 static int __scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
166 int data_direction
, void *buffer
, unsigned bufflen
,
167 unsigned char *sense
, int timeout
, int retries
, u64 flags
,
168 req_flags_t rq_flags
, int *resid
)
171 int write
= (data_direction
== DMA_TO_DEVICE
);
172 int ret
= DRIVER_ERROR
<< 24;
174 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_RECLAIM
);
177 blk_rq_set_block_pc(req
);
179 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
180 buffer
, bufflen
, __GFP_RECLAIM
))
183 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
184 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
187 req
->retries
= retries
;
188 req
->timeout
= timeout
;
189 req
->cmd_flags
|= flags
;
190 req
->rq_flags
|= rq_flags
| RQF_QUIET
| RQF_PREEMPT
;
193 * head injection *required* here otherwise quiesce won't work
195 blk_execute_rq(req
->q
, NULL
, req
, 1);
198 * Some devices (USB mass-storage in particular) may transfer
199 * garbage data together with a residue indicating that the data
200 * is invalid. Prevent the garbage from being misinterpreted
201 * and prevent security leaks by zeroing out the excess data.
203 if (unlikely(req
->resid_len
> 0 && req
->resid_len
<= bufflen
))
204 memset(buffer
+ (bufflen
- req
->resid_len
), 0, req
->resid_len
);
207 *resid
= req
->resid_len
;
210 blk_put_request(req
);
216 * scsi_execute - insert request and wait for the result
219 * @data_direction: data direction
220 * @buffer: data buffer
221 * @bufflen: len of buffer
222 * @sense: optional sense buffer
223 * @timeout: request timeout in seconds
224 * @retries: number of times to retry request
225 * @flags: or into request flags;
226 * @resid: optional residual length
228 * returns the req->errors value which is the scsi_cmnd result
231 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
232 int data_direction
, void *buffer
, unsigned bufflen
,
233 unsigned char *sense
, int timeout
, int retries
, u64 flags
,
236 return __scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
, sense
,
237 timeout
, retries
, flags
, 0, resid
);
239 EXPORT_SYMBOL(scsi_execute
);
241 int scsi_execute_req_flags(struct scsi_device
*sdev
, const unsigned char *cmd
,
242 int data_direction
, void *buffer
, unsigned bufflen
,
243 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
,
244 int *resid
, u64 flags
, req_flags_t rq_flags
)
250 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
252 return DRIVER_ERROR
<< 24;
254 result
= __scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
255 sense
, timeout
, retries
, flags
, rq_flags
, resid
);
257 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
262 EXPORT_SYMBOL(scsi_execute_req_flags
);
265 * Function: scsi_init_cmd_errh()
267 * Purpose: Initialize cmd fields related to error handling.
269 * Arguments: cmd - command that is ready to be queued.
271 * Notes: This function has the job of initializing a number of
272 * fields related to error handling. Typically this will
273 * be called once for each command, as required.
275 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
277 cmd
->serial_number
= 0;
278 scsi_set_resid(cmd
, 0);
279 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
280 if (cmd
->cmd_len
== 0)
281 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
284 void scsi_device_unbusy(struct scsi_device
*sdev
)
286 struct Scsi_Host
*shost
= sdev
->host
;
287 struct scsi_target
*starget
= scsi_target(sdev
);
290 atomic_dec(&shost
->host_busy
);
291 if (starget
->can_queue
> 0)
292 atomic_dec(&starget
->target_busy
);
294 if (unlikely(scsi_host_in_recovery(shost
) &&
295 (shost
->host_failed
|| shost
->host_eh_scheduled
))) {
296 spin_lock_irqsave(shost
->host_lock
, flags
);
297 scsi_eh_wakeup(shost
);
298 spin_unlock_irqrestore(shost
->host_lock
, flags
);
301 atomic_dec(&sdev
->device_busy
);
304 static void scsi_kick_queue(struct request_queue
*q
)
307 blk_mq_start_hw_queues(q
);
313 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
314 * and call blk_run_queue for all the scsi_devices on the target -
315 * including current_sdev first.
317 * Called with *no* scsi locks held.
319 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
321 struct Scsi_Host
*shost
= current_sdev
->host
;
322 struct scsi_device
*sdev
, *tmp
;
323 struct scsi_target
*starget
= scsi_target(current_sdev
);
326 spin_lock_irqsave(shost
->host_lock
, flags
);
327 starget
->starget_sdev_user
= NULL
;
328 spin_unlock_irqrestore(shost
->host_lock
, flags
);
331 * Call blk_run_queue for all LUNs on the target, starting with
332 * current_sdev. We race with others (to set starget_sdev_user),
333 * but in most cases, we will be first. Ideally, each LU on the
334 * target would get some limited time or requests on the target.
336 scsi_kick_queue(current_sdev
->request_queue
);
338 spin_lock_irqsave(shost
->host_lock
, flags
);
339 if (starget
->starget_sdev_user
)
341 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
342 same_target_siblings
) {
343 if (sdev
== current_sdev
)
345 if (scsi_device_get(sdev
))
348 spin_unlock_irqrestore(shost
->host_lock
, flags
);
349 scsi_kick_queue(sdev
->request_queue
);
350 spin_lock_irqsave(shost
->host_lock
, flags
);
352 scsi_device_put(sdev
);
355 spin_unlock_irqrestore(shost
->host_lock
, flags
);
358 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
360 if (atomic_read(&sdev
->device_busy
) >= sdev
->queue_depth
)
362 if (atomic_read(&sdev
->device_blocked
) > 0)
367 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
369 if (starget
->can_queue
> 0) {
370 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
372 if (atomic_read(&starget
->target_blocked
) > 0)
378 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
380 if (shost
->can_queue
> 0 &&
381 atomic_read(&shost
->host_busy
) >= shost
->can_queue
)
383 if (atomic_read(&shost
->host_blocked
) > 0)
385 if (shost
->host_self_blocked
)
390 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
392 LIST_HEAD(starved_list
);
393 struct scsi_device
*sdev
;
396 spin_lock_irqsave(shost
->host_lock
, flags
);
397 list_splice_init(&shost
->starved_list
, &starved_list
);
399 while (!list_empty(&starved_list
)) {
400 struct request_queue
*slq
;
403 * As long as shost is accepting commands and we have
404 * starved queues, call blk_run_queue. scsi_request_fn
405 * drops the queue_lock and can add us back to the
408 * host_lock protects the starved_list and starved_entry.
409 * scsi_request_fn must get the host_lock before checking
410 * or modifying starved_list or starved_entry.
412 if (scsi_host_is_busy(shost
))
415 sdev
= list_entry(starved_list
.next
,
416 struct scsi_device
, starved_entry
);
417 list_del_init(&sdev
->starved_entry
);
418 if (scsi_target_is_busy(scsi_target(sdev
))) {
419 list_move_tail(&sdev
->starved_entry
,
420 &shost
->starved_list
);
425 * Once we drop the host lock, a racing scsi_remove_device()
426 * call may remove the sdev from the starved list and destroy
427 * it and the queue. Mitigate by taking a reference to the
428 * queue and never touching the sdev again after we drop the
429 * host lock. Note: if __scsi_remove_device() invokes
430 * blk_cleanup_queue() before the queue is run from this
431 * function then blk_run_queue() will return immediately since
432 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
434 slq
= sdev
->request_queue
;
435 if (!blk_get_queue(slq
))
437 spin_unlock_irqrestore(shost
->host_lock
, flags
);
439 scsi_kick_queue(slq
);
442 spin_lock_irqsave(shost
->host_lock
, flags
);
444 /* put any unprocessed entries back */
445 list_splice(&starved_list
, &shost
->starved_list
);
446 spin_unlock_irqrestore(shost
->host_lock
, flags
);
450 * Function: scsi_run_queue()
452 * Purpose: Select a proper request queue to serve next
454 * Arguments: q - last request's queue
458 * Notes: The previous command was completely finished, start
459 * a new one if possible.
461 static void scsi_run_queue(struct request_queue
*q
)
463 struct scsi_device
*sdev
= q
->queuedata
;
465 if (scsi_target(sdev
)->single_lun
)
466 scsi_single_lun_run(sdev
);
467 if (!list_empty(&sdev
->host
->starved_list
))
468 scsi_starved_list_run(sdev
->host
);
471 blk_mq_start_stopped_hw_queues(q
, false);
476 void scsi_requeue_run_queue(struct work_struct
*work
)
478 struct scsi_device
*sdev
;
479 struct request_queue
*q
;
481 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
482 q
= sdev
->request_queue
;
487 * Function: scsi_requeue_command()
489 * Purpose: Handle post-processing of completed commands.
491 * Arguments: q - queue to operate on
492 * cmd - command that may need to be requeued.
496 * Notes: After command completion, there may be blocks left
497 * over which weren't finished by the previous command
498 * this can be for a number of reasons - the main one is
499 * I/O errors in the middle of the request, in which case
500 * we need to request the blocks that come after the bad
502 * Notes: Upon return, cmd is a stale pointer.
504 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
506 struct scsi_device
*sdev
= cmd
->device
;
507 struct request
*req
= cmd
->request
;
510 spin_lock_irqsave(q
->queue_lock
, flags
);
511 blk_unprep_request(req
);
513 scsi_put_command(cmd
);
514 blk_requeue_request(q
, req
);
515 spin_unlock_irqrestore(q
->queue_lock
, flags
);
519 put_device(&sdev
->sdev_gendev
);
522 void scsi_run_host_queues(struct Scsi_Host
*shost
)
524 struct scsi_device
*sdev
;
526 shost_for_each_device(sdev
, shost
)
527 scsi_run_queue(sdev
->request_queue
);
530 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
532 if (cmd
->request
->cmd_type
== REQ_TYPE_FS
) {
533 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
535 if (drv
->uninit_command
)
536 drv
->uninit_command(cmd
);
540 static void scsi_mq_free_sgtables(struct scsi_cmnd
*cmd
)
542 struct scsi_data_buffer
*sdb
;
544 if (cmd
->sdb
.table
.nents
)
545 sg_free_table_chained(&cmd
->sdb
.table
, true);
546 if (cmd
->request
->next_rq
) {
547 sdb
= cmd
->request
->next_rq
->special
;
549 sg_free_table_chained(&sdb
->table
, true);
551 if (scsi_prot_sg_count(cmd
))
552 sg_free_table_chained(&cmd
->prot_sdb
->table
, true);
555 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
557 struct scsi_device
*sdev
= cmd
->device
;
558 struct Scsi_Host
*shost
= sdev
->host
;
561 scsi_mq_free_sgtables(cmd
);
562 scsi_uninit_cmd(cmd
);
564 if (shost
->use_cmd_list
) {
565 BUG_ON(list_empty(&cmd
->list
));
566 spin_lock_irqsave(&sdev
->list_lock
, flags
);
567 list_del_init(&cmd
->list
);
568 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
573 * Function: scsi_release_buffers()
575 * Purpose: Free resources allocate for a scsi_command.
577 * Arguments: cmd - command that we are bailing.
579 * Lock status: Assumed that no lock is held upon entry.
583 * Notes: In the event that an upper level driver rejects a
584 * command, we must release resources allocated during
585 * the __init_io() function. Primarily this would involve
586 * the scatter-gather table.
588 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
590 if (cmd
->sdb
.table
.nents
)
591 sg_free_table_chained(&cmd
->sdb
.table
, false);
593 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
595 if (scsi_prot_sg_count(cmd
))
596 sg_free_table_chained(&cmd
->prot_sdb
->table
, false);
599 static void scsi_release_bidi_buffers(struct scsi_cmnd
*cmd
)
601 struct scsi_data_buffer
*bidi_sdb
= cmd
->request
->next_rq
->special
;
603 sg_free_table_chained(&bidi_sdb
->table
, false);
604 kmem_cache_free(scsi_sdb_cache
, bidi_sdb
);
605 cmd
->request
->next_rq
->special
= NULL
;
608 static bool scsi_end_request(struct request
*req
, int error
,
609 unsigned int bytes
, unsigned int bidi_bytes
)
611 struct scsi_cmnd
*cmd
= req
->special
;
612 struct scsi_device
*sdev
= cmd
->device
;
613 struct request_queue
*q
= sdev
->request_queue
;
615 if (blk_update_request(req
, error
, bytes
))
618 /* Bidi request must be completed as a whole */
619 if (unlikely(bidi_bytes
) &&
620 blk_update_request(req
->next_rq
, error
, bidi_bytes
))
623 if (blk_queue_add_random(q
))
624 add_disk_randomness(req
->rq_disk
);
628 * In the MQ case the command gets freed by __blk_mq_end_request,
629 * so we have to do all cleanup that depends on it earlier.
631 * We also can't kick the queues from irq context, so we
632 * will have to defer it to a workqueue.
634 scsi_mq_uninit_cmd(cmd
);
636 __blk_mq_end_request(req
, error
);
638 if (scsi_target(sdev
)->single_lun
||
639 !list_empty(&sdev
->host
->starved_list
))
640 kblockd_schedule_work(&sdev
->requeue_work
);
642 blk_mq_start_stopped_hw_queues(q
, true);
647 scsi_release_bidi_buffers(cmd
);
649 spin_lock_irqsave(q
->queue_lock
, flags
);
650 blk_finish_request(req
, error
);
651 spin_unlock_irqrestore(q
->queue_lock
, flags
);
653 scsi_release_buffers(cmd
);
655 scsi_put_command(cmd
);
659 put_device(&sdev
->sdev_gendev
);
664 * __scsi_error_from_host_byte - translate SCSI error code into errno
665 * @cmd: SCSI command (unused)
666 * @result: scsi error code
668 * Translate SCSI error code into standard UNIX errno.
670 * -ENOLINK temporary transport failure
671 * -EREMOTEIO permanent target failure, do not retry
672 * -EBADE permanent nexus failure, retry on other path
673 * -ENOSPC No write space available
674 * -ENODATA Medium error
675 * -EIO unspecified I/O error
677 static int __scsi_error_from_host_byte(struct scsi_cmnd
*cmd
, int result
)
681 switch(host_byte(result
)) {
682 case DID_TRANSPORT_FAILFAST
:
685 case DID_TARGET_FAILURE
:
686 set_host_byte(cmd
, DID_OK
);
689 case DID_NEXUS_FAILURE
:
690 set_host_byte(cmd
, DID_OK
);
693 case DID_ALLOC_FAILURE
:
694 set_host_byte(cmd
, DID_OK
);
697 case DID_MEDIUM_ERROR
:
698 set_host_byte(cmd
, DID_OK
);
710 * Function: scsi_io_completion()
712 * Purpose: Completion processing for block device I/O requests.
714 * Arguments: cmd - command that is finished.
716 * Lock status: Assumed that no lock is held upon entry.
720 * Notes: We will finish off the specified number of sectors. If we
721 * are done, the command block will be released and the queue
722 * function will be goosed. If we are not done then we have to
723 * figure out what to do next:
725 * a) We can call scsi_requeue_command(). The request
726 * will be unprepared and put back on the queue. Then
727 * a new command will be created for it. This should
728 * be used if we made forward progress, or if we want
729 * to switch from READ(10) to READ(6) for example.
731 * b) We can call __scsi_queue_insert(). The request will
732 * be put back on the queue and retried using the same
733 * command as before, possibly after a delay.
735 * c) We can call scsi_end_request() with -EIO to fail
736 * the remainder of the request.
738 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
740 int result
= cmd
->result
;
741 struct request_queue
*q
= cmd
->device
->request_queue
;
742 struct request
*req
= cmd
->request
;
744 struct scsi_sense_hdr sshdr
;
745 bool sense_valid
= false;
746 int sense_deferred
= 0, level
= 0;
747 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
748 ACTION_DELAYED_RETRY
} action
;
749 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
752 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
754 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
757 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
759 if (sense_valid
&& req
->sense
) {
761 * SG_IO wants current and deferred errors
763 int len
= 8 + cmd
->sense_buffer
[7];
765 if (len
> SCSI_SENSE_BUFFERSIZE
)
766 len
= SCSI_SENSE_BUFFERSIZE
;
767 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
768 req
->sense_len
= len
;
771 error
= __scsi_error_from_host_byte(cmd
, result
);
774 * __scsi_error_from_host_byte may have reset the host_byte
776 req
->errors
= cmd
->result
;
778 req
->resid_len
= scsi_get_resid(cmd
);
780 if (scsi_bidi_cmnd(cmd
)) {
782 * Bidi commands Must be complete as a whole,
783 * both sides at once.
785 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
786 if (scsi_end_request(req
, 0, blk_rq_bytes(req
),
787 blk_rq_bytes(req
->next_rq
)))
791 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
793 * Certain non BLOCK_PC requests are commands that don't
794 * actually transfer anything (FLUSH), so cannot use
795 * good_bytes != blk_rq_bytes(req) as the signal for an error.
796 * This sets the error explicitly for the problem case.
798 error
= __scsi_error_from_host_byte(cmd
, result
);
801 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
802 BUG_ON(blk_bidi_rq(req
));
805 * Next deal with any sectors which we were able to correctly
808 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
809 "%u sectors total, %d bytes done.\n",
810 blk_rq_sectors(req
), good_bytes
));
813 * Recovered errors need reporting, but they're always treated
814 * as success, so fiddle the result code here. For BLOCK_PC
815 * we already took a copy of the original into rq->errors which
816 * is what gets returned to the user
818 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
819 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
820 * print since caller wants ATA registers. Only occurs on
821 * SCSI ATA PASS_THROUGH commands when CK_COND=1
823 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
825 else if (!(req
->rq_flags
& RQF_QUIET
))
826 scsi_print_sense(cmd
);
828 /* BLOCK_PC may have set error */
833 * special case: failed zero length commands always need to
834 * drop down into the retry code. Otherwise, if we finished
835 * all bytes in the request we are done now.
837 if (!(blk_rq_bytes(req
) == 0 && error
) &&
838 !scsi_end_request(req
, error
, good_bytes
, 0))
842 * Kill remainder if no retrys.
844 if (error
&& scsi_noretry_cmd(cmd
)) {
845 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
851 * If there had been no error, but we have leftover bytes in the
852 * requeues just queue the command up again.
857 error
= __scsi_error_from_host_byte(cmd
, result
);
859 if (host_byte(result
) == DID_RESET
) {
860 /* Third party bus reset or reset for error recovery
861 * reasons. Just retry the command and see what
864 action
= ACTION_RETRY
;
865 } else if (sense_valid
&& !sense_deferred
) {
866 switch (sshdr
.sense_key
) {
868 if (cmd
->device
->removable
) {
869 /* Detected disc change. Set a bit
870 * and quietly refuse further access.
872 cmd
->device
->changed
= 1;
873 action
= ACTION_FAIL
;
875 /* Must have been a power glitch, or a
876 * bus reset. Could not have been a
877 * media change, so we just retry the
878 * command and see what happens.
880 action
= ACTION_RETRY
;
883 case ILLEGAL_REQUEST
:
884 /* If we had an ILLEGAL REQUEST returned, then
885 * we may have performed an unsupported
886 * command. The only thing this should be
887 * would be a ten byte read where only a six
888 * byte read was supported. Also, on a system
889 * where READ CAPACITY failed, we may have
890 * read past the end of the disk.
892 if ((cmd
->device
->use_10_for_rw
&&
893 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
894 (cmd
->cmnd
[0] == READ_10
||
895 cmd
->cmnd
[0] == WRITE_10
)) {
896 /* This will issue a new 6-byte command. */
897 cmd
->device
->use_10_for_rw
= 0;
898 action
= ACTION_REPREP
;
899 } else if (sshdr
.asc
== 0x10) /* DIX */ {
900 action
= ACTION_FAIL
;
902 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
903 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
904 action
= ACTION_FAIL
;
907 action
= ACTION_FAIL
;
909 case ABORTED_COMMAND
:
910 action
= ACTION_FAIL
;
911 if (sshdr
.asc
== 0x10) /* DIF */
915 /* If the device is in the process of becoming
916 * ready, or has a temporary blockage, retry.
918 if (sshdr
.asc
== 0x04) {
919 switch (sshdr
.ascq
) {
920 case 0x01: /* becoming ready */
921 case 0x04: /* format in progress */
922 case 0x05: /* rebuild in progress */
923 case 0x06: /* recalculation in progress */
924 case 0x07: /* operation in progress */
925 case 0x08: /* Long write in progress */
926 case 0x09: /* self test in progress */
927 case 0x14: /* space allocation in progress */
928 action
= ACTION_DELAYED_RETRY
;
931 action
= ACTION_FAIL
;
935 action
= ACTION_FAIL
;
937 case VOLUME_OVERFLOW
:
938 /* See SSC3rXX or current. */
939 action
= ACTION_FAIL
;
942 action
= ACTION_FAIL
;
946 action
= ACTION_FAIL
;
948 if (action
!= ACTION_FAIL
&&
949 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
950 action
= ACTION_FAIL
;
954 /* Give up and fail the remainder of the request */
955 if (!(req
->rq_flags
& RQF_QUIET
)) {
956 static DEFINE_RATELIMIT_STATE(_rs
,
957 DEFAULT_RATELIMIT_INTERVAL
,
958 DEFAULT_RATELIMIT_BURST
);
960 if (unlikely(scsi_logging_level
))
961 level
= SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT
,
962 SCSI_LOG_MLCOMPLETE_BITS
);
965 * if logging is enabled the failure will be printed
966 * in scsi_log_completion(), so avoid duplicate messages
968 if (!level
&& __ratelimit(&_rs
)) {
969 scsi_print_result(cmd
, NULL
, FAILED
);
970 if (driver_byte(result
) & DRIVER_SENSE
)
971 scsi_print_sense(cmd
);
972 scsi_print_command(cmd
);
975 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
980 /* Unprep the request and put it back at the head of the queue.
981 * A new command will be prepared and issued.
984 cmd
->request
->rq_flags
&= ~RQF_DONTPREP
;
985 scsi_mq_uninit_cmd(cmd
);
986 scsi_mq_requeue_cmd(cmd
);
988 scsi_release_buffers(cmd
);
989 scsi_requeue_command(q
, cmd
);
993 /* Retry the same command immediately */
994 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
996 case ACTION_DELAYED_RETRY
:
997 /* Retry the same command after a delay */
998 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
1003 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
)
1008 * If sg table allocation fails, requeue request later.
1010 if (unlikely(sg_alloc_table_chained(&sdb
->table
, req
->nr_phys_segments
,
1012 return BLKPREP_DEFER
;
1015 * Next, walk the list, and fill in the addresses and sizes of
1018 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1019 BUG_ON(count
> sdb
->table
.nents
);
1020 sdb
->table
.nents
= count
;
1021 sdb
->length
= blk_rq_bytes(req
);
1026 * Function: scsi_init_io()
1028 * Purpose: SCSI I/O initialize function.
1030 * Arguments: cmd - Command descriptor we wish to initialize
1032 * Returns: 0 on success
1033 * BLKPREP_DEFER if the failure is retryable
1034 * BLKPREP_KILL if the failure is fatal
1036 int scsi_init_io(struct scsi_cmnd
*cmd
)
1038 struct scsi_device
*sdev
= cmd
->device
;
1039 struct request
*rq
= cmd
->request
;
1040 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1043 BUG_ON(!rq
->nr_phys_segments
);
1045 error
= scsi_init_sgtable(rq
, &cmd
->sdb
);
1049 if (blk_bidi_rq(rq
)) {
1050 if (!rq
->q
->mq_ops
) {
1051 struct scsi_data_buffer
*bidi_sdb
=
1052 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1054 error
= BLKPREP_DEFER
;
1058 rq
->next_rq
->special
= bidi_sdb
;
1061 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
);
1066 if (blk_integrity_rq(rq
)) {
1067 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1070 if (prot_sdb
== NULL
) {
1072 * This can happen if someone (e.g. multipath)
1073 * queues a command to a device on an adapter
1074 * that does not support DIX.
1077 error
= BLKPREP_KILL
;
1081 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1083 if (sg_alloc_table_chained(&prot_sdb
->table
, ivecs
,
1084 prot_sdb
->table
.sgl
)) {
1085 error
= BLKPREP_DEFER
;
1089 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1090 prot_sdb
->table
.sgl
);
1091 BUG_ON(unlikely(count
> ivecs
));
1092 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1094 cmd
->prot_sdb
= prot_sdb
;
1095 cmd
->prot_sdb
->table
.nents
= count
;
1101 scsi_mq_free_sgtables(cmd
);
1103 scsi_release_buffers(cmd
);
1104 cmd
->request
->special
= NULL
;
1105 scsi_put_command(cmd
);
1106 put_device(&sdev
->sdev_gendev
);
1110 EXPORT_SYMBOL(scsi_init_io
);
1112 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1113 struct request
*req
)
1115 struct scsi_cmnd
*cmd
;
1117 if (!req
->special
) {
1118 /* Bail if we can't get a reference to the device */
1119 if (!get_device(&sdev
->sdev_gendev
))
1122 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1123 if (unlikely(!cmd
)) {
1124 put_device(&sdev
->sdev_gendev
);
1132 /* pull a tag out of the request if we have one */
1133 cmd
->tag
= req
->tag
;
1136 cmd
->cmnd
= req
->cmd
;
1137 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1142 static int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1144 struct scsi_cmnd
*cmd
= req
->special
;
1147 * BLOCK_PC requests may transfer data, in which case they must
1148 * a bio attached to them. Or they might contain a SCSI command
1149 * that does not transfer data, in which case they may optionally
1150 * submit a request without an attached bio.
1153 int ret
= scsi_init_io(cmd
);
1157 BUG_ON(blk_rq_bytes(req
));
1159 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1162 cmd
->cmd_len
= req
->cmd_len
;
1163 cmd
->transfersize
= blk_rq_bytes(req
);
1164 cmd
->allowed
= req
->retries
;
1169 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1170 * that still need to be translated to SCSI CDBs from the ULD.
1172 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1174 struct scsi_cmnd
*cmd
= req
->special
;
1176 if (unlikely(sdev
->handler
&& sdev
->handler
->prep_fn
)) {
1177 int ret
= sdev
->handler
->prep_fn(sdev
, req
);
1178 if (ret
!= BLKPREP_OK
)
1182 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1183 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1186 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1188 struct scsi_cmnd
*cmd
= req
->special
;
1190 if (!blk_rq_bytes(req
))
1191 cmd
->sc_data_direction
= DMA_NONE
;
1192 else if (rq_data_dir(req
) == WRITE
)
1193 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1195 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1197 switch (req
->cmd_type
) {
1199 return scsi_setup_fs_cmnd(sdev
, req
);
1200 case REQ_TYPE_BLOCK_PC
:
1201 return scsi_setup_blk_pc_cmnd(sdev
, req
);
1203 return BLKPREP_KILL
;
1208 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1210 int ret
= BLKPREP_OK
;
1213 * If the device is not in running state we will reject some
1216 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1217 switch (sdev
->sdev_state
) {
1219 case SDEV_TRANSPORT_OFFLINE
:
1221 * If the device is offline we refuse to process any
1222 * commands. The device must be brought online
1223 * before trying any recovery commands.
1225 sdev_printk(KERN_ERR
, sdev
,
1226 "rejecting I/O to offline device\n");
1231 * If the device is fully deleted, we refuse to
1232 * process any commands as well.
1234 sdev_printk(KERN_ERR
, sdev
,
1235 "rejecting I/O to dead device\n");
1239 case SDEV_CREATED_BLOCK
:
1240 ret
= BLKPREP_DEFER
;
1244 * If the devices is blocked we defer normal commands.
1246 if (!(req
->rq_flags
& RQF_PREEMPT
))
1247 ret
= BLKPREP_DEFER
;
1251 * For any other not fully online state we only allow
1252 * special commands. In particular any user initiated
1253 * command is not allowed.
1255 if (!(req
->rq_flags
& RQF_PREEMPT
))
1264 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1266 struct scsi_device
*sdev
= q
->queuedata
;
1270 case BLKPREP_INVALID
:
1271 req
->errors
= DID_NO_CONNECT
<< 16;
1272 /* release the command and kill it */
1274 struct scsi_cmnd
*cmd
= req
->special
;
1275 scsi_release_buffers(cmd
);
1276 scsi_put_command(cmd
);
1277 put_device(&sdev
->sdev_gendev
);
1278 req
->special
= NULL
;
1283 * If we defer, the blk_peek_request() returns NULL, but the
1284 * queue must be restarted, so we schedule a callback to happen
1287 if (atomic_read(&sdev
->device_busy
) == 0)
1288 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1291 req
->rq_flags
|= RQF_DONTPREP
;
1297 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1299 struct scsi_device
*sdev
= q
->queuedata
;
1300 struct scsi_cmnd
*cmd
;
1303 ret
= scsi_prep_state_check(sdev
, req
);
1304 if (ret
!= BLKPREP_OK
)
1307 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1308 if (unlikely(!cmd
)) {
1309 ret
= BLKPREP_DEFER
;
1313 ret
= scsi_setup_cmnd(sdev
, req
);
1315 return scsi_prep_return(q
, req
, ret
);
1318 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1320 scsi_uninit_cmd(req
->special
);
1324 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1327 * Called with the queue_lock held.
1329 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1330 struct scsi_device
*sdev
)
1334 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1335 if (atomic_read(&sdev
->device_blocked
)) {
1340 * unblock after device_blocked iterates to zero
1342 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1344 * For the MQ case we take care of this in the caller.
1347 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1350 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1351 "unblocking device at zero depth\n"));
1354 if (busy
>= sdev
->queue_depth
)
1359 atomic_dec(&sdev
->device_busy
);
1364 * scsi_target_queue_ready: checks if there we can send commands to target
1365 * @sdev: scsi device on starget to check.
1367 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1368 struct scsi_device
*sdev
)
1370 struct scsi_target
*starget
= scsi_target(sdev
);
1373 if (starget
->single_lun
) {
1374 spin_lock_irq(shost
->host_lock
);
1375 if (starget
->starget_sdev_user
&&
1376 starget
->starget_sdev_user
!= sdev
) {
1377 spin_unlock_irq(shost
->host_lock
);
1380 starget
->starget_sdev_user
= sdev
;
1381 spin_unlock_irq(shost
->host_lock
);
1384 if (starget
->can_queue
<= 0)
1387 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1388 if (atomic_read(&starget
->target_blocked
) > 0) {
1393 * unblock after target_blocked iterates to zero
1395 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1398 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1399 "unblocking target at zero depth\n"));
1402 if (busy
>= starget
->can_queue
)
1408 spin_lock_irq(shost
->host_lock
);
1409 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1410 spin_unlock_irq(shost
->host_lock
);
1412 if (starget
->can_queue
> 0)
1413 atomic_dec(&starget
->target_busy
);
1418 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1419 * return 0. We must end up running the queue again whenever 0 is
1420 * returned, else IO can hang.
1422 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1423 struct Scsi_Host
*shost
,
1424 struct scsi_device
*sdev
)
1428 if (scsi_host_in_recovery(shost
))
1431 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1432 if (atomic_read(&shost
->host_blocked
) > 0) {
1437 * unblock after host_blocked iterates to zero
1439 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1443 shost_printk(KERN_INFO
, shost
,
1444 "unblocking host at zero depth\n"));
1447 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1449 if (shost
->host_self_blocked
)
1452 /* We're OK to process the command, so we can't be starved */
1453 if (!list_empty(&sdev
->starved_entry
)) {
1454 spin_lock_irq(shost
->host_lock
);
1455 if (!list_empty(&sdev
->starved_entry
))
1456 list_del_init(&sdev
->starved_entry
);
1457 spin_unlock_irq(shost
->host_lock
);
1463 spin_lock_irq(shost
->host_lock
);
1464 if (list_empty(&sdev
->starved_entry
))
1465 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1466 spin_unlock_irq(shost
->host_lock
);
1468 atomic_dec(&shost
->host_busy
);
1473 * Busy state exporting function for request stacking drivers.
1475 * For efficiency, no lock is taken to check the busy state of
1476 * shost/starget/sdev, since the returned value is not guaranteed and
1477 * may be changed after request stacking drivers call the function,
1478 * regardless of taking lock or not.
1480 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1481 * needs to return 'not busy'. Otherwise, request stacking drivers
1482 * may hold requests forever.
1484 static int scsi_lld_busy(struct request_queue
*q
)
1486 struct scsi_device
*sdev
= q
->queuedata
;
1487 struct Scsi_Host
*shost
;
1489 if (blk_queue_dying(q
))
1495 * Ignore host/starget busy state.
1496 * Since block layer does not have a concept of fairness across
1497 * multiple queues, congestion of host/starget needs to be handled
1500 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1507 * Kill a request for a dead device
1509 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1511 struct scsi_cmnd
*cmd
= req
->special
;
1512 struct scsi_device
*sdev
;
1513 struct scsi_target
*starget
;
1514 struct Scsi_Host
*shost
;
1516 blk_start_request(req
);
1518 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1521 starget
= scsi_target(sdev
);
1523 scsi_init_cmd_errh(cmd
);
1524 cmd
->result
= DID_NO_CONNECT
<< 16;
1525 atomic_inc(&cmd
->device
->iorequest_cnt
);
1528 * SCSI request completion path will do scsi_device_unbusy(),
1529 * bump busy counts. To bump the counters, we need to dance
1530 * with the locks as normal issue path does.
1532 atomic_inc(&sdev
->device_busy
);
1533 atomic_inc(&shost
->host_busy
);
1534 if (starget
->can_queue
> 0)
1535 atomic_inc(&starget
->target_busy
);
1537 blk_complete_request(req
);
1540 static void scsi_softirq_done(struct request
*rq
)
1542 struct scsi_cmnd
*cmd
= rq
->special
;
1543 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1546 INIT_LIST_HEAD(&cmd
->eh_entry
);
1548 atomic_inc(&cmd
->device
->iodone_cnt
);
1550 atomic_inc(&cmd
->device
->ioerr_cnt
);
1552 disposition
= scsi_decide_disposition(cmd
);
1553 if (disposition
!= SUCCESS
&&
1554 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1555 sdev_printk(KERN_ERR
, cmd
->device
,
1556 "timing out command, waited %lus\n",
1558 disposition
= SUCCESS
;
1561 scsi_log_completion(cmd
, disposition
);
1563 switch (disposition
) {
1565 scsi_finish_command(cmd
);
1568 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1570 case ADD_TO_MLQUEUE
:
1571 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1574 if (!scsi_eh_scmd_add(cmd
, 0))
1575 scsi_finish_command(cmd
);
1580 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1581 * @cmd: command block we are dispatching.
1583 * Return: nonzero return request was rejected and device's queue needs to be
1586 static int scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
1588 struct Scsi_Host
*host
= cmd
->device
->host
;
1591 atomic_inc(&cmd
->device
->iorequest_cnt
);
1593 /* check if the device is still usable */
1594 if (unlikely(cmd
->device
->sdev_state
== SDEV_DEL
)) {
1595 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1596 * returns an immediate error upwards, and signals
1597 * that the device is no longer present */
1598 cmd
->result
= DID_NO_CONNECT
<< 16;
1602 /* Check to see if the scsi lld made this device blocked. */
1603 if (unlikely(scsi_device_blocked(cmd
->device
))) {
1605 * in blocked state, the command is just put back on
1606 * the device queue. The suspend state has already
1607 * blocked the queue so future requests should not
1608 * occur until the device transitions out of the
1611 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1612 "queuecommand : device blocked\n"));
1613 return SCSI_MLQUEUE_DEVICE_BUSY
;
1616 /* Store the LUN value in cmnd, if needed. */
1617 if (cmd
->device
->lun_in_cdb
)
1618 cmd
->cmnd
[1] = (cmd
->cmnd
[1] & 0x1f) |
1619 (cmd
->device
->lun
<< 5 & 0xe0);
1624 * Before we queue this command, check if the command
1625 * length exceeds what the host adapter can handle.
1627 if (cmd
->cmd_len
> cmd
->device
->host
->max_cmd_len
) {
1628 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1629 "queuecommand : command too long. "
1630 "cdb_size=%d host->max_cmd_len=%d\n",
1631 cmd
->cmd_len
, cmd
->device
->host
->max_cmd_len
));
1632 cmd
->result
= (DID_ABORT
<< 16);
1636 if (unlikely(host
->shost_state
== SHOST_DEL
)) {
1637 cmd
->result
= (DID_NO_CONNECT
<< 16);
1642 trace_scsi_dispatch_cmd_start(cmd
);
1643 rtn
= host
->hostt
->queuecommand(host
, cmd
);
1645 trace_scsi_dispatch_cmd_error(cmd
, rtn
);
1646 if (rtn
!= SCSI_MLQUEUE_DEVICE_BUSY
&&
1647 rtn
!= SCSI_MLQUEUE_TARGET_BUSY
)
1648 rtn
= SCSI_MLQUEUE_HOST_BUSY
;
1650 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1651 "queuecommand : request rejected\n"));
1656 cmd
->scsi_done(cmd
);
1661 * scsi_done - Invoke completion on finished SCSI command.
1662 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1663 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1665 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1666 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1667 * calls blk_complete_request() for further processing.
1669 * This function is interrupt context safe.
1671 static void scsi_done(struct scsi_cmnd
*cmd
)
1673 trace_scsi_dispatch_cmd_done(cmd
);
1674 blk_complete_request(cmd
->request
);
1678 * Function: scsi_request_fn()
1680 * Purpose: Main strategy routine for SCSI.
1682 * Arguments: q - Pointer to actual queue.
1686 * Lock status: IO request lock assumed to be held when called.
1688 static void scsi_request_fn(struct request_queue
*q
)
1689 __releases(q
->queue_lock
)
1690 __acquires(q
->queue_lock
)
1692 struct scsi_device
*sdev
= q
->queuedata
;
1693 struct Scsi_Host
*shost
;
1694 struct scsi_cmnd
*cmd
;
1695 struct request
*req
;
1698 * To start with, we keep looping until the queue is empty, or until
1699 * the host is no longer able to accept any more requests.
1705 * get next queueable request. We do this early to make sure
1706 * that the request is fully prepared even if we cannot
1709 req
= blk_peek_request(q
);
1713 if (unlikely(!scsi_device_online(sdev
))) {
1714 sdev_printk(KERN_ERR
, sdev
,
1715 "rejecting I/O to offline device\n");
1716 scsi_kill_request(req
, q
);
1720 if (!scsi_dev_queue_ready(q
, sdev
))
1724 * Remove the request from the request list.
1726 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1727 blk_start_request(req
);
1729 spin_unlock_irq(q
->queue_lock
);
1731 if (unlikely(cmd
== NULL
)) {
1732 printk(KERN_CRIT
"impossible request in %s.\n"
1733 "please mail a stack trace to "
1734 "linux-scsi@vger.kernel.org\n",
1736 blk_dump_rq_flags(req
, "foo");
1741 * We hit this when the driver is using a host wide
1742 * tag map. For device level tag maps the queue_depth check
1743 * in the device ready fn would prevent us from trying
1744 * to allocate a tag. Since the map is a shared host resource
1745 * we add the dev to the starved list so it eventually gets
1746 * a run when a tag is freed.
1748 if (blk_queue_tagged(q
) && !(req
->rq_flags
& RQF_QUEUED
)) {
1749 spin_lock_irq(shost
->host_lock
);
1750 if (list_empty(&sdev
->starved_entry
))
1751 list_add_tail(&sdev
->starved_entry
,
1752 &shost
->starved_list
);
1753 spin_unlock_irq(shost
->host_lock
);
1757 if (!scsi_target_queue_ready(shost
, sdev
))
1760 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1761 goto host_not_ready
;
1763 if (sdev
->simple_tags
)
1764 cmd
->flags
|= SCMD_TAGGED
;
1766 cmd
->flags
&= ~SCMD_TAGGED
;
1769 * Finally, initialize any error handling parameters, and set up
1770 * the timers for timeouts.
1772 scsi_init_cmd_errh(cmd
);
1775 * Dispatch the command to the low-level driver.
1777 cmd
->scsi_done
= scsi_done
;
1778 rtn
= scsi_dispatch_cmd(cmd
);
1780 scsi_queue_insert(cmd
, rtn
);
1781 spin_lock_irq(q
->queue_lock
);
1784 spin_lock_irq(q
->queue_lock
);
1790 if (scsi_target(sdev
)->can_queue
> 0)
1791 atomic_dec(&scsi_target(sdev
)->target_busy
);
1794 * lock q, handle tag, requeue req, and decrement device_busy. We
1795 * must return with queue_lock held.
1797 * Decrementing device_busy without checking it is OK, as all such
1798 * cases (host limits or settings) should run the queue at some
1801 spin_lock_irq(q
->queue_lock
);
1802 blk_requeue_request(q
, req
);
1803 atomic_dec(&sdev
->device_busy
);
1805 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1806 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1809 static inline int prep_to_mq(int ret
)
1815 return BLK_MQ_RQ_QUEUE_BUSY
;
1817 return BLK_MQ_RQ_QUEUE_ERROR
;
1821 static int scsi_mq_prep_fn(struct request
*req
)
1823 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1824 struct scsi_device
*sdev
= req
->q
->queuedata
;
1825 struct Scsi_Host
*shost
= sdev
->host
;
1826 unsigned char *sense_buf
= cmd
->sense_buffer
;
1827 struct scatterlist
*sg
;
1829 memset(cmd
, 0, sizeof(struct scsi_cmnd
));
1835 cmd
->sense_buffer
= sense_buf
;
1837 cmd
->tag
= req
->tag
;
1839 cmd
->cmnd
= req
->cmd
;
1840 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1842 INIT_LIST_HEAD(&cmd
->list
);
1843 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1844 cmd
->jiffies_at_alloc
= jiffies
;
1846 if (shost
->use_cmd_list
) {
1847 spin_lock_irq(&sdev
->list_lock
);
1848 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1849 spin_unlock_irq(&sdev
->list_lock
);
1852 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1853 cmd
->sdb
.table
.sgl
= sg
;
1855 if (scsi_host_get_prot(shost
)) {
1856 cmd
->prot_sdb
= (void *)sg
+
1858 shost
->sg_tablesize
, SG_CHUNK_SIZE
) *
1859 sizeof(struct scatterlist
);
1860 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1862 cmd
->prot_sdb
->table
.sgl
=
1863 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1866 if (blk_bidi_rq(req
)) {
1867 struct request
*next_rq
= req
->next_rq
;
1868 struct scsi_data_buffer
*bidi_sdb
= blk_mq_rq_to_pdu(next_rq
);
1870 memset(bidi_sdb
, 0, sizeof(struct scsi_data_buffer
));
1871 bidi_sdb
->table
.sgl
=
1872 (struct scatterlist
*)(bidi_sdb
+ 1);
1874 next_rq
->special
= bidi_sdb
;
1877 blk_mq_start_request(req
);
1879 return scsi_setup_cmnd(sdev
, req
);
1882 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1884 trace_scsi_dispatch_cmd_done(cmd
);
1885 blk_mq_complete_request(cmd
->request
, cmd
->request
->errors
);
1888 static int scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1889 const struct blk_mq_queue_data
*bd
)
1891 struct request
*req
= bd
->rq
;
1892 struct request_queue
*q
= req
->q
;
1893 struct scsi_device
*sdev
= q
->queuedata
;
1894 struct Scsi_Host
*shost
= sdev
->host
;
1895 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1899 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
1903 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1904 if (!get_device(&sdev
->sdev_gendev
))
1907 if (!scsi_dev_queue_ready(q
, sdev
))
1908 goto out_put_device
;
1909 if (!scsi_target_queue_ready(shost
, sdev
))
1910 goto out_dec_device_busy
;
1911 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1912 goto out_dec_target_busy
;
1915 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
1916 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
1918 goto out_dec_host_busy
;
1919 req
->rq_flags
|= RQF_DONTPREP
;
1921 blk_mq_start_request(req
);
1924 if (sdev
->simple_tags
)
1925 cmd
->flags
|= SCMD_TAGGED
;
1927 cmd
->flags
&= ~SCMD_TAGGED
;
1929 scsi_init_cmd_errh(cmd
);
1930 cmd
->scsi_done
= scsi_mq_done
;
1932 reason
= scsi_dispatch_cmd(cmd
);
1934 scsi_set_blocked(cmd
, reason
);
1935 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1936 goto out_dec_host_busy
;
1939 return BLK_MQ_RQ_QUEUE_OK
;
1942 atomic_dec(&shost
->host_busy
);
1943 out_dec_target_busy
:
1944 if (scsi_target(sdev
)->can_queue
> 0)
1945 atomic_dec(&scsi_target(sdev
)->target_busy
);
1946 out_dec_device_busy
:
1947 atomic_dec(&sdev
->device_busy
);
1949 put_device(&sdev
->sdev_gendev
);
1952 case BLK_MQ_RQ_QUEUE_BUSY
:
1953 if (atomic_read(&sdev
->device_busy
) == 0 &&
1954 !scsi_device_blocked(sdev
))
1955 blk_mq_delay_queue(hctx
, SCSI_QUEUE_DELAY
);
1957 case BLK_MQ_RQ_QUEUE_ERROR
:
1959 * Make sure to release all allocated ressources when
1960 * we hit an error, as we will never see this command
1963 if (req
->rq_flags
& RQF_DONTPREP
)
1964 scsi_mq_uninit_cmd(cmd
);
1972 static enum blk_eh_timer_return
scsi_timeout(struct request
*req
,
1976 return BLK_EH_RESET_TIMER
;
1977 return scsi_times_out(req
);
1980 static int scsi_init_request(void *data
, struct request
*rq
,
1981 unsigned int hctx_idx
, unsigned int request_idx
,
1982 unsigned int numa_node
)
1984 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1986 cmd
->sense_buffer
= kzalloc_node(SCSI_SENSE_BUFFERSIZE
, GFP_KERNEL
,
1988 if (!cmd
->sense_buffer
)
1993 static void scsi_exit_request(void *data
, struct request
*rq
,
1994 unsigned int hctx_idx
, unsigned int request_idx
)
1996 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1998 kfree(cmd
->sense_buffer
);
2001 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
2003 struct device
*host_dev
;
2004 u64 bounce_limit
= 0xffffffff;
2006 if (shost
->unchecked_isa_dma
)
2007 return BLK_BOUNCE_ISA
;
2009 * Platforms with virtual-DMA translation
2010 * hardware have no practical limit.
2012 if (!PCI_DMA_BUS_IS_PHYS
)
2013 return BLK_BOUNCE_ANY
;
2015 host_dev
= scsi_get_device(shost
);
2016 if (host_dev
&& host_dev
->dma_mask
)
2017 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
2019 return bounce_limit
;
2022 static void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
2024 struct device
*dev
= shost
->dma_dev
;
2027 * this limit is imposed by hardware restrictions
2029 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
2032 if (scsi_host_prot_dma(shost
)) {
2033 shost
->sg_prot_tablesize
=
2034 min_not_zero(shost
->sg_prot_tablesize
,
2035 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
2036 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
2037 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
2040 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
2041 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
2042 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
2043 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
2045 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2047 if (!shost
->use_clustering
)
2048 q
->limits
.cluster
= 0;
2051 * set a reasonable default alignment on word boundaries: the
2052 * host and device may alter it using
2053 * blk_queue_update_dma_alignment() later.
2055 blk_queue_dma_alignment(q
, 0x03);
2058 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
2059 request_fn_proc
*request_fn
)
2061 struct request_queue
*q
;
2063 q
= blk_init_queue(request_fn
, NULL
);
2066 __scsi_init_queue(shost
, q
);
2069 EXPORT_SYMBOL(__scsi_alloc_queue
);
2071 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
2073 struct request_queue
*q
;
2075 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
2079 blk_queue_prep_rq(q
, scsi_prep_fn
);
2080 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2081 blk_queue_softirq_done(q
, scsi_softirq_done
);
2082 blk_queue_rq_timed_out(q
, scsi_times_out
);
2083 blk_queue_lld_busy(q
, scsi_lld_busy
);
2087 static struct blk_mq_ops scsi_mq_ops
= {
2088 .queue_rq
= scsi_queue_rq
,
2089 .complete
= scsi_softirq_done
,
2090 .timeout
= scsi_timeout
,
2091 .init_request
= scsi_init_request
,
2092 .exit_request
= scsi_exit_request
,
2095 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2097 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2098 if (IS_ERR(sdev
->request_queue
))
2101 sdev
->request_queue
->queuedata
= sdev
;
2102 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2103 return sdev
->request_queue
;
2106 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2108 unsigned int cmd_size
, sgl_size
, tbl_size
;
2110 tbl_size
= shost
->sg_tablesize
;
2111 if (tbl_size
> SG_CHUNK_SIZE
)
2112 tbl_size
= SG_CHUNK_SIZE
;
2113 sgl_size
= tbl_size
* sizeof(struct scatterlist
);
2114 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2115 if (scsi_host_get_prot(shost
))
2116 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2118 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2119 shost
->tag_set
.ops
= &scsi_mq_ops
;
2120 shost
->tag_set
.nr_hw_queues
= shost
->nr_hw_queues
? : 1;
2121 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2122 shost
->tag_set
.cmd_size
= cmd_size
;
2123 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2124 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2125 shost
->tag_set
.flags
|=
2126 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost
->hostt
->tag_alloc_policy
);
2127 shost
->tag_set
.driver_data
= shost
;
2129 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2132 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2134 blk_mq_free_tag_set(&shost
->tag_set
);
2138 * Function: scsi_block_requests()
2140 * Purpose: Utility function used by low-level drivers to prevent further
2141 * commands from being queued to the device.
2143 * Arguments: shost - Host in question
2147 * Lock status: No locks are assumed held.
2149 * Notes: There is no timer nor any other means by which the requests
2150 * get unblocked other than the low-level driver calling
2151 * scsi_unblock_requests().
2153 void scsi_block_requests(struct Scsi_Host
*shost
)
2155 shost
->host_self_blocked
= 1;
2157 EXPORT_SYMBOL(scsi_block_requests
);
2160 * Function: scsi_unblock_requests()
2162 * Purpose: Utility function used by low-level drivers to allow further
2163 * commands from being queued to the device.
2165 * Arguments: shost - Host in question
2169 * Lock status: No locks are assumed held.
2171 * Notes: There is no timer nor any other means by which the requests
2172 * get unblocked other than the low-level driver calling
2173 * scsi_unblock_requests().
2175 * This is done as an API function so that changes to the
2176 * internals of the scsi mid-layer won't require wholesale
2177 * changes to drivers that use this feature.
2179 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2181 shost
->host_self_blocked
= 0;
2182 scsi_run_host_queues(shost
);
2184 EXPORT_SYMBOL(scsi_unblock_requests
);
2186 int __init
scsi_init_queue(void)
2188 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2189 sizeof(struct scsi_data_buffer
),
2191 if (!scsi_sdb_cache
) {
2192 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2199 void scsi_exit_queue(void)
2201 kmem_cache_destroy(scsi_sdb_cache
);
2205 * scsi_mode_select - issue a mode select
2206 * @sdev: SCSI device to be queried
2207 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2208 * @sp: Save page bit (0 == don't save, 1 == save)
2209 * @modepage: mode page being requested
2210 * @buffer: request buffer (may not be smaller than eight bytes)
2211 * @len: length of request buffer.
2212 * @timeout: command timeout
2213 * @retries: number of retries before failing
2214 * @data: returns a structure abstracting the mode header data
2215 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2216 * must be SCSI_SENSE_BUFFERSIZE big.
2218 * Returns zero if successful; negative error number or scsi
2223 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2224 unsigned char *buffer
, int len
, int timeout
, int retries
,
2225 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2227 unsigned char cmd
[10];
2228 unsigned char *real_buffer
;
2231 memset(cmd
, 0, sizeof(cmd
));
2232 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2234 if (sdev
->use_10_for_ms
) {
2237 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2240 memcpy(real_buffer
+ 8, buffer
, len
);
2244 real_buffer
[2] = data
->medium_type
;
2245 real_buffer
[3] = data
->device_specific
;
2246 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2248 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2249 real_buffer
[7] = data
->block_descriptor_length
;
2251 cmd
[0] = MODE_SELECT_10
;
2255 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2259 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2262 memcpy(real_buffer
+ 4, buffer
, len
);
2265 real_buffer
[1] = data
->medium_type
;
2266 real_buffer
[2] = data
->device_specific
;
2267 real_buffer
[3] = data
->block_descriptor_length
;
2270 cmd
[0] = MODE_SELECT
;
2274 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2275 sshdr
, timeout
, retries
, NULL
);
2279 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2282 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2283 * @sdev: SCSI device to be queried
2284 * @dbd: set if mode sense will allow block descriptors to be returned
2285 * @modepage: mode page being requested
2286 * @buffer: request buffer (may not be smaller than eight bytes)
2287 * @len: length of request buffer.
2288 * @timeout: command timeout
2289 * @retries: number of retries before failing
2290 * @data: returns a structure abstracting the mode header data
2291 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2292 * must be SCSI_SENSE_BUFFERSIZE big.
2294 * Returns zero if unsuccessful, or the header offset (either 4
2295 * or 8 depending on whether a six or ten byte command was
2296 * issued) if successful.
2299 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2300 unsigned char *buffer
, int len
, int timeout
, int retries
,
2301 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2303 unsigned char cmd
[12];
2306 int result
, retry_count
= retries
;
2307 struct scsi_sense_hdr my_sshdr
;
2309 memset(data
, 0, sizeof(*data
));
2310 memset(&cmd
[0], 0, 12);
2311 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2314 /* caller might not be interested in sense, but we need it */
2319 use_10_for_ms
= sdev
->use_10_for_ms
;
2321 if (use_10_for_ms
) {
2325 cmd
[0] = MODE_SENSE_10
;
2332 cmd
[0] = MODE_SENSE
;
2337 memset(buffer
, 0, len
);
2339 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2340 sshdr
, timeout
, retries
, NULL
);
2342 /* This code looks awful: what it's doing is making sure an
2343 * ILLEGAL REQUEST sense return identifies the actual command
2344 * byte as the problem. MODE_SENSE commands can return
2345 * ILLEGAL REQUEST if the code page isn't supported */
2347 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2348 (driver_byte(result
) & DRIVER_SENSE
)) {
2349 if (scsi_sense_valid(sshdr
)) {
2350 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2351 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2353 * Invalid command operation code
2355 sdev
->use_10_for_ms
= 0;
2361 if(scsi_status_is_good(result
)) {
2362 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2363 (modepage
== 6 || modepage
== 8))) {
2364 /* Initio breakage? */
2367 data
->medium_type
= 0;
2368 data
->device_specific
= 0;
2370 data
->block_descriptor_length
= 0;
2371 } else if(use_10_for_ms
) {
2372 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2373 data
->medium_type
= buffer
[2];
2374 data
->device_specific
= buffer
[3];
2375 data
->longlba
= buffer
[4] & 0x01;
2376 data
->block_descriptor_length
= buffer
[6]*256
2379 data
->length
= buffer
[0] + 1;
2380 data
->medium_type
= buffer
[1];
2381 data
->device_specific
= buffer
[2];
2382 data
->block_descriptor_length
= buffer
[3];
2384 data
->header_length
= header_length
;
2385 } else if ((status_byte(result
) == CHECK_CONDITION
) &&
2386 scsi_sense_valid(sshdr
) &&
2387 sshdr
->sense_key
== UNIT_ATTENTION
&& retry_count
) {
2394 EXPORT_SYMBOL(scsi_mode_sense
);
2397 * scsi_test_unit_ready - test if unit is ready
2398 * @sdev: scsi device to change the state of.
2399 * @timeout: command timeout
2400 * @retries: number of retries before failing
2401 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2402 * returning sense. Make sure that this is cleared before passing
2405 * Returns zero if unsuccessful or an error if TUR failed. For
2406 * removable media, UNIT_ATTENTION sets ->changed flag.
2409 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2410 struct scsi_sense_hdr
*sshdr_external
)
2413 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2415 struct scsi_sense_hdr
*sshdr
;
2418 if (!sshdr_external
)
2419 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
2421 sshdr
= sshdr_external
;
2423 /* try to eat the UNIT_ATTENTION if there are enough retries */
2425 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2426 timeout
, retries
, NULL
);
2427 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2428 sshdr
->sense_key
== UNIT_ATTENTION
)
2430 } while (scsi_sense_valid(sshdr
) &&
2431 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2433 if (!sshdr_external
)
2437 EXPORT_SYMBOL(scsi_test_unit_ready
);
2440 * scsi_device_set_state - Take the given device through the device state model.
2441 * @sdev: scsi device to change the state of.
2442 * @state: state to change to.
2444 * Returns zero if unsuccessful or an error if the requested
2445 * transition is illegal.
2448 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2450 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2452 if (state
== oldstate
)
2458 case SDEV_CREATED_BLOCK
:
2469 case SDEV_TRANSPORT_OFFLINE
:
2482 case SDEV_TRANSPORT_OFFLINE
:
2490 case SDEV_TRANSPORT_OFFLINE
:
2505 case SDEV_CREATED_BLOCK
:
2512 case SDEV_CREATED_BLOCK
:
2527 case SDEV_TRANSPORT_OFFLINE
:
2540 case SDEV_TRANSPORT_OFFLINE
:
2542 case SDEV_CREATED_BLOCK
:
2550 sdev
->sdev_state
= state
;
2554 SCSI_LOG_ERROR_RECOVERY(1,
2555 sdev_printk(KERN_ERR
, sdev
,
2556 "Illegal state transition %s->%s",
2557 scsi_device_state_name(oldstate
),
2558 scsi_device_state_name(state
))
2562 EXPORT_SYMBOL(scsi_device_set_state
);
2565 * sdev_evt_emit - emit a single SCSI device uevent
2566 * @sdev: associated SCSI device
2567 * @evt: event to emit
2569 * Send a single uevent (scsi_event) to the associated scsi_device.
2571 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2576 switch (evt
->evt_type
) {
2577 case SDEV_EVT_MEDIA_CHANGE
:
2578 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2580 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2581 scsi_rescan_device(&sdev
->sdev_gendev
);
2582 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2584 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2585 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2587 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2588 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2590 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2591 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2593 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2594 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2596 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2597 envp
[idx
++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2606 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2610 * sdev_evt_thread - send a uevent for each scsi event
2611 * @work: work struct for scsi_device
2613 * Dispatch queued events to their associated scsi_device kobjects
2616 void scsi_evt_thread(struct work_struct
*work
)
2618 struct scsi_device
*sdev
;
2619 enum scsi_device_event evt_type
;
2620 LIST_HEAD(event_list
);
2622 sdev
= container_of(work
, struct scsi_device
, event_work
);
2624 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2625 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2626 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2629 struct scsi_event
*evt
;
2630 struct list_head
*this, *tmp
;
2631 unsigned long flags
;
2633 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2634 list_splice_init(&sdev
->event_list
, &event_list
);
2635 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2637 if (list_empty(&event_list
))
2640 list_for_each_safe(this, tmp
, &event_list
) {
2641 evt
= list_entry(this, struct scsi_event
, node
);
2642 list_del(&evt
->node
);
2643 scsi_evt_emit(sdev
, evt
);
2650 * sdev_evt_send - send asserted event to uevent thread
2651 * @sdev: scsi_device event occurred on
2652 * @evt: event to send
2654 * Assert scsi device event asynchronously.
2656 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2658 unsigned long flags
;
2661 /* FIXME: currently this check eliminates all media change events
2662 * for polled devices. Need to update to discriminate between AN
2663 * and polled events */
2664 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2670 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2671 list_add_tail(&evt
->node
, &sdev
->event_list
);
2672 schedule_work(&sdev
->event_work
);
2673 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2675 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2678 * sdev_evt_alloc - allocate a new scsi event
2679 * @evt_type: type of event to allocate
2680 * @gfpflags: GFP flags for allocation
2682 * Allocates and returns a new scsi_event.
2684 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2687 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2691 evt
->evt_type
= evt_type
;
2692 INIT_LIST_HEAD(&evt
->node
);
2694 /* evt_type-specific initialization, if any */
2696 case SDEV_EVT_MEDIA_CHANGE
:
2697 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2698 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2699 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2700 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2701 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2702 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2710 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2713 * sdev_evt_send_simple - send asserted event to uevent thread
2714 * @sdev: scsi_device event occurred on
2715 * @evt_type: type of event to send
2716 * @gfpflags: GFP flags for allocation
2718 * Assert scsi device event asynchronously, given an event type.
2720 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2721 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2723 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2725 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2730 sdev_evt_send(sdev
, evt
);
2732 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2735 * scsi_device_quiesce - Block user issued commands.
2736 * @sdev: scsi device to quiesce.
2738 * This works by trying to transition to the SDEV_QUIESCE state
2739 * (which must be a legal transition). When the device is in this
2740 * state, only special requests will be accepted, all others will
2741 * be deferred. Since special requests may also be requeued requests,
2742 * a successful return doesn't guarantee the device will be
2743 * totally quiescent.
2745 * Must be called with user context, may sleep.
2747 * Returns zero if unsuccessful or an error if not.
2750 scsi_device_quiesce(struct scsi_device
*sdev
)
2752 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2756 scsi_run_queue(sdev
->request_queue
);
2757 while (atomic_read(&sdev
->device_busy
)) {
2758 msleep_interruptible(200);
2759 scsi_run_queue(sdev
->request_queue
);
2763 EXPORT_SYMBOL(scsi_device_quiesce
);
2766 * scsi_device_resume - Restart user issued commands to a quiesced device.
2767 * @sdev: scsi device to resume.
2769 * Moves the device from quiesced back to running and restarts the
2772 * Must be called with user context, may sleep.
2774 void scsi_device_resume(struct scsi_device
*sdev
)
2776 /* check if the device state was mutated prior to resume, and if
2777 * so assume the state is being managed elsewhere (for example
2778 * device deleted during suspend)
2780 if (sdev
->sdev_state
!= SDEV_QUIESCE
||
2781 scsi_device_set_state(sdev
, SDEV_RUNNING
))
2783 scsi_run_queue(sdev
->request_queue
);
2785 EXPORT_SYMBOL(scsi_device_resume
);
2788 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2790 scsi_device_quiesce(sdev
);
2794 scsi_target_quiesce(struct scsi_target
*starget
)
2796 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2798 EXPORT_SYMBOL(scsi_target_quiesce
);
2801 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2803 scsi_device_resume(sdev
);
2807 scsi_target_resume(struct scsi_target
*starget
)
2809 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2811 EXPORT_SYMBOL(scsi_target_resume
);
2814 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2815 * @sdev: device to block
2817 * Block request made by scsi lld's to temporarily stop all
2818 * scsi commands on the specified device. Called from interrupt
2819 * or normal process context.
2821 * Returns zero if successful or error if not
2824 * This routine transitions the device to the SDEV_BLOCK state
2825 * (which must be a legal transition). When the device is in this
2826 * state, all commands are deferred until the scsi lld reenables
2827 * the device with scsi_device_unblock or device_block_tmo fires.
2830 scsi_internal_device_block(struct scsi_device
*sdev
)
2832 struct request_queue
*q
= sdev
->request_queue
;
2833 unsigned long flags
;
2836 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2838 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2845 * The device has transitioned to SDEV_BLOCK. Stop the
2846 * block layer from calling the midlayer with this device's
2850 blk_mq_stop_hw_queues(q
);
2852 spin_lock_irqsave(q
->queue_lock
, flags
);
2854 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2859 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2862 * scsi_internal_device_unblock - resume a device after a block request
2863 * @sdev: device to resume
2864 * @new_state: state to set devices to after unblocking
2866 * Called by scsi lld's or the midlayer to restart the device queue
2867 * for the previously suspended scsi device. Called from interrupt or
2868 * normal process context.
2870 * Returns zero if successful or error if not.
2873 * This routine transitions the device to the SDEV_RUNNING state
2874 * or to one of the offline states (which must be a legal transition)
2875 * allowing the midlayer to goose the queue for this device.
2878 scsi_internal_device_unblock(struct scsi_device
*sdev
,
2879 enum scsi_device_state new_state
)
2881 struct request_queue
*q
= sdev
->request_queue
;
2882 unsigned long flags
;
2885 * Try to transition the scsi device to SDEV_RUNNING or one of the
2886 * offlined states and goose the device queue if successful.
2888 if ((sdev
->sdev_state
== SDEV_BLOCK
) ||
2889 (sdev
->sdev_state
== SDEV_TRANSPORT_OFFLINE
))
2890 sdev
->sdev_state
= new_state
;
2891 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
) {
2892 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2893 new_state
== SDEV_OFFLINE
)
2894 sdev
->sdev_state
= new_state
;
2896 sdev
->sdev_state
= SDEV_CREATED
;
2897 } else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2898 sdev
->sdev_state
!= SDEV_OFFLINE
)
2902 blk_mq_start_stopped_hw_queues(q
, false);
2904 spin_lock_irqsave(q
->queue_lock
, flags
);
2906 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2911 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2914 device_block(struct scsi_device
*sdev
, void *data
)
2916 scsi_internal_device_block(sdev
);
2920 target_block(struct device
*dev
, void *data
)
2922 if (scsi_is_target_device(dev
))
2923 starget_for_each_device(to_scsi_target(dev
), NULL
,
2929 scsi_target_block(struct device
*dev
)
2931 if (scsi_is_target_device(dev
))
2932 starget_for_each_device(to_scsi_target(dev
), NULL
,
2935 device_for_each_child(dev
, NULL
, target_block
);
2937 EXPORT_SYMBOL_GPL(scsi_target_block
);
2940 device_unblock(struct scsi_device
*sdev
, void *data
)
2942 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2946 target_unblock(struct device
*dev
, void *data
)
2948 if (scsi_is_target_device(dev
))
2949 starget_for_each_device(to_scsi_target(dev
), data
,
2955 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2957 if (scsi_is_target_device(dev
))
2958 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2961 device_for_each_child(dev
, &new_state
, target_unblock
);
2963 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2966 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2967 * @sgl: scatter-gather list
2968 * @sg_count: number of segments in sg
2969 * @offset: offset in bytes into sg, on return offset into the mapped area
2970 * @len: bytes to map, on return number of bytes mapped
2972 * Returns virtual address of the start of the mapped page
2974 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2975 size_t *offset
, size_t *len
)
2978 size_t sg_len
= 0, len_complete
= 0;
2979 struct scatterlist
*sg
;
2982 WARN_ON(!irqs_disabled());
2984 for_each_sg(sgl
, sg
, sg_count
, i
) {
2985 len_complete
= sg_len
; /* Complete sg-entries */
2986 sg_len
+= sg
->length
;
2987 if (sg_len
> *offset
)
2991 if (unlikely(i
== sg_count
)) {
2992 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2994 __func__
, sg_len
, *offset
, sg_count
);
2999 /* Offset starting from the beginning of first page in this sg-entry */
3000 *offset
= *offset
- len_complete
+ sg
->offset
;
3002 /* Assumption: contiguous pages can be accessed as "page + i" */
3003 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
3004 *offset
&= ~PAGE_MASK
;
3006 /* Bytes in this sg-entry from *offset to the end of the page */
3007 sg_len
= PAGE_SIZE
- *offset
;
3011 return kmap_atomic(page
);
3013 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3016 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3017 * @virt: virtual address to be unmapped
3019 void scsi_kunmap_atomic_sg(void *virt
)
3021 kunmap_atomic(virt
);
3023 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3025 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3027 atomic_inc(&sdev
->disk_events_disable_depth
);
3029 EXPORT_SYMBOL(sdev_disable_disk_events
);
3031 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3033 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3035 atomic_dec(&sdev
->disk_events_disable_depth
);
3037 EXPORT_SYMBOL(sdev_enable_disk_events
);
3040 * scsi_vpd_lun_id - return a unique device identification
3041 * @sdev: SCSI device
3042 * @id: buffer for the identification
3043 * @id_len: length of the buffer
3045 * Copies a unique device identification into @id based
3046 * on the information in the VPD page 0x83 of the device.
3047 * The string will be formatted as a SCSI name string.
3049 * Returns the length of the identification or error on failure.
3050 * If the identifier is longer than the supplied buffer the actual
3051 * identifier length is returned and the buffer is not zero-padded.
3053 int scsi_vpd_lun_id(struct scsi_device
*sdev
, char *id
, size_t id_len
)
3055 u8 cur_id_type
= 0xff;
3057 unsigned char *d
, *cur_id_str
;
3058 unsigned char __rcu
*vpd_pg83
;
3059 int id_size
= -EINVAL
;
3062 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3069 * Look for the correct descriptor.
3070 * Order of preference for lun descriptor:
3071 * - SCSI name string
3072 * - NAA IEEE Registered Extended
3073 * - EUI-64 based 16-byte
3074 * - EUI-64 based 12-byte
3075 * - NAA IEEE Registered
3076 * - NAA IEEE Extended
3078 * as longer descriptors reduce the likelyhood
3079 * of identification clashes.
3082 /* The id string must be at least 20 bytes + terminating NULL byte */
3088 memset(id
, 0, id_len
);
3090 while (d
< vpd_pg83
+ sdev
->vpd_pg83_len
) {
3091 /* Skip designators not referring to the LUN */
3092 if ((d
[1] & 0x30) != 0x00)
3095 switch (d
[1] & 0xf) {
3098 if (cur_id_size
> d
[3])
3100 /* Prefer anything */
3101 if (cur_id_type
> 0x01 && cur_id_type
!= 0xff)
3104 if (cur_id_size
+ 4 > id_len
)
3105 cur_id_size
= id_len
- 4;
3107 cur_id_type
= d
[1] & 0xf;
3108 id_size
= snprintf(id
, id_len
, "t10.%*pE",
3109 cur_id_size
, cur_id_str
);
3113 if (cur_id_size
> d
[3])
3115 /* Prefer NAA IEEE Registered Extended */
3116 if (cur_id_type
== 0x3 &&
3117 cur_id_size
== d
[3])
3121 cur_id_type
= d
[1] & 0xf;
3122 switch (cur_id_size
) {
3124 id_size
= snprintf(id
, id_len
,
3129 id_size
= snprintf(id
, id_len
,
3134 id_size
= snprintf(id
, id_len
,
3145 if (cur_id_size
> d
[3])
3149 cur_id_type
= d
[1] & 0xf;
3150 switch (cur_id_size
) {
3152 id_size
= snprintf(id
, id_len
,
3157 id_size
= snprintf(id
, id_len
,
3167 /* SCSI name string */
3168 if (cur_id_size
+ 4 > d
[3])
3170 /* Prefer others for truncated descriptor */
3171 if (cur_id_size
&& d
[3] > id_len
)
3173 cur_id_size
= id_size
= d
[3];
3175 cur_id_type
= d
[1] & 0xf;
3176 if (cur_id_size
>= id_len
)
3177 cur_id_size
= id_len
- 1;
3178 memcpy(id
, cur_id_str
, cur_id_size
);
3179 /* Decrease priority for truncated descriptor */
3180 if (cur_id_size
!= id_size
)
3193 EXPORT_SYMBOL(scsi_vpd_lun_id
);
3196 * scsi_vpd_tpg_id - return a target port group identifier
3197 * @sdev: SCSI device
3199 * Returns the Target Port Group identifier from the information
3200 * froom VPD page 0x83 of the device.
3202 * Returns the identifier or error on failure.
3204 int scsi_vpd_tpg_id(struct scsi_device
*sdev
, int *rel_id
)
3207 unsigned char __rcu
*vpd_pg83
;
3208 int group_id
= -EAGAIN
, rel_port
= -1;
3211 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3217 d
= sdev
->vpd_pg83
+ 4;
3218 while (d
< sdev
->vpd_pg83
+ sdev
->vpd_pg83_len
) {
3219 switch (d
[1] & 0xf) {
3221 /* Relative target port */
3222 rel_port
= get_unaligned_be16(&d
[6]);
3225 /* Target port group */
3226 group_id
= get_unaligned_be16(&d
[6]);
3235 if (group_id
>= 0 && rel_id
&& rel_port
!= -1)
3240 EXPORT_SYMBOL(scsi_vpd_tpg_id
);