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
;
89 struct request_queue
*q
= cmd
->request
->q
;
91 blk_mq_requeue_request(cmd
->request
);
92 blk_mq_kick_requeue_list(q
);
93 put_device(&sdev
->sdev_gendev
);
97 * __scsi_queue_insert - private queue insertion
98 * @cmd: The SCSI command being requeued
99 * @reason: The reason for the requeue
100 * @unbusy: Whether the queue should be unbusied
102 * This is a private queue insertion. The public interface
103 * scsi_queue_insert() always assumes the queue should be unbusied
104 * because it's always called before the completion. This function is
105 * for a requeue after completion, which should only occur in this
108 static void __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, int unbusy
)
110 struct scsi_device
*device
= cmd
->device
;
111 struct request_queue
*q
= device
->request_queue
;
114 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO
, cmd
,
115 "Inserting command %p into mlqueue\n", cmd
));
117 scsi_set_blocked(cmd
, reason
);
120 * Decrement the counters, since these commands are no longer
121 * active on the host/device.
124 scsi_device_unbusy(device
);
127 * Requeue this command. It will go before all other commands
128 * that are already in the queue. Schedule requeue work under
129 * lock such that the kblockd_schedule_work() call happens
130 * before blk_cleanup_queue() finishes.
134 scsi_mq_requeue_cmd(cmd
);
137 spin_lock_irqsave(q
->queue_lock
, flags
);
138 blk_requeue_request(q
, cmd
->request
);
139 kblockd_schedule_work(&device
->requeue_work
);
140 spin_unlock_irqrestore(q
->queue_lock
, flags
);
144 * Function: scsi_queue_insert()
146 * Purpose: Insert a command in the midlevel queue.
148 * Arguments: cmd - command that we are adding to queue.
149 * reason - why we are inserting command to queue.
151 * Lock status: Assumed that lock is not held upon entry.
155 * Notes: We do this for one of two cases. Either the host is busy
156 * and it cannot accept any more commands for the time being,
157 * or the device returned QUEUE_FULL and can accept no more
159 * Notes: This could be called either from an interrupt context or a
160 * normal process context.
162 void scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
164 __scsi_queue_insert(cmd
, reason
, 1);
167 * scsi_execute - insert request and wait for the result
170 * @data_direction: data direction
171 * @buffer: data buffer
172 * @bufflen: len of buffer
173 * @sense: optional sense buffer
174 * @timeout: request timeout in seconds
175 * @retries: number of times to retry request
176 * @flags: or into request flags;
177 * @resid: optional residual length
179 * returns the req->errors value which is the scsi_cmnd result
182 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
183 int data_direction
, void *buffer
, unsigned bufflen
,
184 unsigned char *sense
, int timeout
, int retries
, u64 flags
,
188 int write
= (data_direction
== DMA_TO_DEVICE
);
189 int ret
= DRIVER_ERROR
<< 24;
191 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_RECLAIM
);
194 blk_rq_set_block_pc(req
);
196 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
197 buffer
, bufflen
, __GFP_RECLAIM
))
200 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
201 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
204 req
->retries
= retries
;
205 req
->timeout
= timeout
;
206 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
209 * head injection *required* here otherwise quiesce won't work
211 blk_execute_rq(req
->q
, NULL
, req
, 1);
214 * Some devices (USB mass-storage in particular) may transfer
215 * garbage data together with a residue indicating that the data
216 * is invalid. Prevent the garbage from being misinterpreted
217 * and prevent security leaks by zeroing out the excess data.
219 if (unlikely(req
->resid_len
> 0 && req
->resid_len
<= bufflen
))
220 memset(buffer
+ (bufflen
- req
->resid_len
), 0, req
->resid_len
);
223 *resid
= req
->resid_len
;
226 blk_put_request(req
);
230 EXPORT_SYMBOL(scsi_execute
);
232 int scsi_execute_req_flags(struct scsi_device
*sdev
, const unsigned char *cmd
,
233 int data_direction
, void *buffer
, unsigned bufflen
,
234 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
,
235 int *resid
, u64 flags
)
241 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
243 return DRIVER_ERROR
<< 24;
245 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
246 sense
, timeout
, retries
, flags
, resid
);
248 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
253 EXPORT_SYMBOL(scsi_execute_req_flags
);
256 * Function: scsi_init_cmd_errh()
258 * Purpose: Initialize cmd fields related to error handling.
260 * Arguments: cmd - command that is ready to be queued.
262 * Notes: This function has the job of initializing a number of
263 * fields related to error handling. Typically this will
264 * be called once for each command, as required.
266 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
268 cmd
->serial_number
= 0;
269 scsi_set_resid(cmd
, 0);
270 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
271 if (cmd
->cmd_len
== 0)
272 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
275 void scsi_device_unbusy(struct scsi_device
*sdev
)
277 struct Scsi_Host
*shost
= sdev
->host
;
278 struct scsi_target
*starget
= scsi_target(sdev
);
281 atomic_dec(&shost
->host_busy
);
282 if (starget
->can_queue
> 0)
283 atomic_dec(&starget
->target_busy
);
285 if (unlikely(scsi_host_in_recovery(shost
) &&
286 (shost
->host_failed
|| shost
->host_eh_scheduled
))) {
287 spin_lock_irqsave(shost
->host_lock
, flags
);
288 scsi_eh_wakeup(shost
);
289 spin_unlock_irqrestore(shost
->host_lock
, flags
);
292 atomic_dec(&sdev
->device_busy
);
295 static void scsi_kick_queue(struct request_queue
*q
)
298 blk_mq_start_hw_queues(q
);
304 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
305 * and call blk_run_queue for all the scsi_devices on the target -
306 * including current_sdev first.
308 * Called with *no* scsi locks held.
310 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
312 struct Scsi_Host
*shost
= current_sdev
->host
;
313 struct scsi_device
*sdev
, *tmp
;
314 struct scsi_target
*starget
= scsi_target(current_sdev
);
317 spin_lock_irqsave(shost
->host_lock
, flags
);
318 starget
->starget_sdev_user
= NULL
;
319 spin_unlock_irqrestore(shost
->host_lock
, flags
);
322 * Call blk_run_queue for all LUNs on the target, starting with
323 * current_sdev. We race with others (to set starget_sdev_user),
324 * but in most cases, we will be first. Ideally, each LU on the
325 * target would get some limited time or requests on the target.
327 scsi_kick_queue(current_sdev
->request_queue
);
329 spin_lock_irqsave(shost
->host_lock
, flags
);
330 if (starget
->starget_sdev_user
)
332 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
333 same_target_siblings
) {
334 if (sdev
== current_sdev
)
336 if (scsi_device_get(sdev
))
339 spin_unlock_irqrestore(shost
->host_lock
, flags
);
340 scsi_kick_queue(sdev
->request_queue
);
341 spin_lock_irqsave(shost
->host_lock
, flags
);
343 scsi_device_put(sdev
);
346 spin_unlock_irqrestore(shost
->host_lock
, flags
);
349 static inline bool scsi_device_is_busy(struct scsi_device
*sdev
)
351 if (atomic_read(&sdev
->device_busy
) >= sdev
->queue_depth
)
353 if (atomic_read(&sdev
->device_blocked
) > 0)
358 static inline bool scsi_target_is_busy(struct scsi_target
*starget
)
360 if (starget
->can_queue
> 0) {
361 if (atomic_read(&starget
->target_busy
) >= starget
->can_queue
)
363 if (atomic_read(&starget
->target_blocked
) > 0)
369 static inline bool scsi_host_is_busy(struct Scsi_Host
*shost
)
371 if (shost
->can_queue
> 0 &&
372 atomic_read(&shost
->host_busy
) >= shost
->can_queue
)
374 if (atomic_read(&shost
->host_blocked
) > 0)
376 if (shost
->host_self_blocked
)
381 static void scsi_starved_list_run(struct Scsi_Host
*shost
)
383 LIST_HEAD(starved_list
);
384 struct scsi_device
*sdev
;
387 spin_lock_irqsave(shost
->host_lock
, flags
);
388 list_splice_init(&shost
->starved_list
, &starved_list
);
390 while (!list_empty(&starved_list
)) {
391 struct request_queue
*slq
;
394 * As long as shost is accepting commands and we have
395 * starved queues, call blk_run_queue. scsi_request_fn
396 * drops the queue_lock and can add us back to the
399 * host_lock protects the starved_list and starved_entry.
400 * scsi_request_fn must get the host_lock before checking
401 * or modifying starved_list or starved_entry.
403 if (scsi_host_is_busy(shost
))
406 sdev
= list_entry(starved_list
.next
,
407 struct scsi_device
, starved_entry
);
408 list_del_init(&sdev
->starved_entry
);
409 if (scsi_target_is_busy(scsi_target(sdev
))) {
410 list_move_tail(&sdev
->starved_entry
,
411 &shost
->starved_list
);
416 * Once we drop the host lock, a racing scsi_remove_device()
417 * call may remove the sdev from the starved list and destroy
418 * it and the queue. Mitigate by taking a reference to the
419 * queue and never touching the sdev again after we drop the
420 * host lock. Note: if __scsi_remove_device() invokes
421 * blk_cleanup_queue() before the queue is run from this
422 * function then blk_run_queue() will return immediately since
423 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
425 slq
= sdev
->request_queue
;
426 if (!blk_get_queue(slq
))
428 spin_unlock_irqrestore(shost
->host_lock
, flags
);
430 scsi_kick_queue(slq
);
433 spin_lock_irqsave(shost
->host_lock
, flags
);
435 /* put any unprocessed entries back */
436 list_splice(&starved_list
, &shost
->starved_list
);
437 spin_unlock_irqrestore(shost
->host_lock
, flags
);
441 * Function: scsi_run_queue()
443 * Purpose: Select a proper request queue to serve next
445 * Arguments: q - last request's queue
449 * Notes: The previous command was completely finished, start
450 * a new one if possible.
452 static void scsi_run_queue(struct request_queue
*q
)
454 struct scsi_device
*sdev
= q
->queuedata
;
456 if (scsi_target(sdev
)->single_lun
)
457 scsi_single_lun_run(sdev
);
458 if (!list_empty(&sdev
->host
->starved_list
))
459 scsi_starved_list_run(sdev
->host
);
462 blk_mq_start_stopped_hw_queues(q
, false);
467 void scsi_requeue_run_queue(struct work_struct
*work
)
469 struct scsi_device
*sdev
;
470 struct request_queue
*q
;
472 sdev
= container_of(work
, struct scsi_device
, requeue_work
);
473 q
= sdev
->request_queue
;
478 * Function: scsi_requeue_command()
480 * Purpose: Handle post-processing of completed commands.
482 * Arguments: q - queue to operate on
483 * cmd - command that may need to be requeued.
487 * Notes: After command completion, there may be blocks left
488 * over which weren't finished by the previous command
489 * this can be for a number of reasons - the main one is
490 * I/O errors in the middle of the request, in which case
491 * we need to request the blocks that come after the bad
493 * Notes: Upon return, cmd is a stale pointer.
495 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
497 struct scsi_device
*sdev
= cmd
->device
;
498 struct request
*req
= cmd
->request
;
501 spin_lock_irqsave(q
->queue_lock
, flags
);
502 blk_unprep_request(req
);
504 scsi_put_command(cmd
);
505 blk_requeue_request(q
, req
);
506 spin_unlock_irqrestore(q
->queue_lock
, flags
);
510 put_device(&sdev
->sdev_gendev
);
513 void scsi_run_host_queues(struct Scsi_Host
*shost
)
515 struct scsi_device
*sdev
;
517 shost_for_each_device(sdev
, shost
)
518 scsi_run_queue(sdev
->request_queue
);
521 static void scsi_uninit_cmd(struct scsi_cmnd
*cmd
)
523 if (cmd
->request
->cmd_type
== REQ_TYPE_FS
) {
524 struct scsi_driver
*drv
= scsi_cmd_to_driver(cmd
);
526 if (drv
->uninit_command
)
527 drv
->uninit_command(cmd
);
531 static void scsi_mq_free_sgtables(struct scsi_cmnd
*cmd
)
533 struct scsi_data_buffer
*sdb
;
535 if (cmd
->sdb
.table
.nents
)
536 sg_free_table_chained(&cmd
->sdb
.table
, true);
537 if (cmd
->request
->next_rq
) {
538 sdb
= cmd
->request
->next_rq
->special
;
540 sg_free_table_chained(&sdb
->table
, true);
542 if (scsi_prot_sg_count(cmd
))
543 sg_free_table_chained(&cmd
->prot_sdb
->table
, true);
546 static void scsi_mq_uninit_cmd(struct scsi_cmnd
*cmd
)
548 struct scsi_device
*sdev
= cmd
->device
;
549 struct Scsi_Host
*shost
= sdev
->host
;
552 scsi_mq_free_sgtables(cmd
);
553 scsi_uninit_cmd(cmd
);
555 if (shost
->use_cmd_list
) {
556 BUG_ON(list_empty(&cmd
->list
));
557 spin_lock_irqsave(&sdev
->list_lock
, flags
);
558 list_del_init(&cmd
->list
);
559 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
564 * Function: scsi_release_buffers()
566 * Purpose: Free resources allocate for a scsi_command.
568 * Arguments: cmd - command that we are bailing.
570 * Lock status: Assumed that no lock is held upon entry.
574 * Notes: In the event that an upper level driver rejects a
575 * command, we must release resources allocated during
576 * the __init_io() function. Primarily this would involve
577 * the scatter-gather table.
579 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
581 if (cmd
->sdb
.table
.nents
)
582 sg_free_table_chained(&cmd
->sdb
.table
, false);
584 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
586 if (scsi_prot_sg_count(cmd
))
587 sg_free_table_chained(&cmd
->prot_sdb
->table
, false);
590 static void scsi_release_bidi_buffers(struct scsi_cmnd
*cmd
)
592 struct scsi_data_buffer
*bidi_sdb
= cmd
->request
->next_rq
->special
;
594 sg_free_table_chained(&bidi_sdb
->table
, false);
595 kmem_cache_free(scsi_sdb_cache
, bidi_sdb
);
596 cmd
->request
->next_rq
->special
= NULL
;
599 static bool scsi_end_request(struct request
*req
, int error
,
600 unsigned int bytes
, unsigned int bidi_bytes
)
602 struct scsi_cmnd
*cmd
= req
->special
;
603 struct scsi_device
*sdev
= cmd
->device
;
604 struct request_queue
*q
= sdev
->request_queue
;
606 if (blk_update_request(req
, error
, bytes
))
609 /* Bidi request must be completed as a whole */
610 if (unlikely(bidi_bytes
) &&
611 blk_update_request(req
->next_rq
, error
, bidi_bytes
))
614 if (blk_queue_add_random(q
))
615 add_disk_randomness(req
->rq_disk
);
619 * In the MQ case the command gets freed by __blk_mq_end_request,
620 * so we have to do all cleanup that depends on it earlier.
622 * We also can't kick the queues from irq context, so we
623 * will have to defer it to a workqueue.
625 scsi_mq_uninit_cmd(cmd
);
627 __blk_mq_end_request(req
, error
);
629 if (scsi_target(sdev
)->single_lun
||
630 !list_empty(&sdev
->host
->starved_list
))
631 kblockd_schedule_work(&sdev
->requeue_work
);
633 blk_mq_start_stopped_hw_queues(q
, true);
638 scsi_release_bidi_buffers(cmd
);
640 spin_lock_irqsave(q
->queue_lock
, flags
);
641 blk_finish_request(req
, error
);
642 spin_unlock_irqrestore(q
->queue_lock
, flags
);
644 scsi_release_buffers(cmd
);
646 scsi_put_command(cmd
);
650 put_device(&sdev
->sdev_gendev
);
655 * __scsi_error_from_host_byte - translate SCSI error code into errno
656 * @cmd: SCSI command (unused)
657 * @result: scsi error code
659 * Translate SCSI error code into standard UNIX errno.
661 * -ENOLINK temporary transport failure
662 * -EREMOTEIO permanent target failure, do not retry
663 * -EBADE permanent nexus failure, retry on other path
664 * -ENOSPC No write space available
665 * -ENODATA Medium error
666 * -EIO unspecified I/O error
668 static int __scsi_error_from_host_byte(struct scsi_cmnd
*cmd
, int result
)
672 switch(host_byte(result
)) {
673 case DID_TRANSPORT_FAILFAST
:
676 case DID_TARGET_FAILURE
:
677 set_host_byte(cmd
, DID_OK
);
680 case DID_NEXUS_FAILURE
:
681 set_host_byte(cmd
, DID_OK
);
684 case DID_ALLOC_FAILURE
:
685 set_host_byte(cmd
, DID_OK
);
688 case DID_MEDIUM_ERROR
:
689 set_host_byte(cmd
, DID_OK
);
701 * Function: scsi_io_completion()
703 * Purpose: Completion processing for block device I/O requests.
705 * Arguments: cmd - command that is finished.
707 * Lock status: Assumed that no lock is held upon entry.
711 * Notes: We will finish off the specified number of sectors. If we
712 * are done, the command block will be released and the queue
713 * function will be goosed. If we are not done then we have to
714 * figure out what to do next:
716 * a) We can call scsi_requeue_command(). The request
717 * will be unprepared and put back on the queue. Then
718 * a new command will be created for it. This should
719 * be used if we made forward progress, or if we want
720 * to switch from READ(10) to READ(6) for example.
722 * b) We can call __scsi_queue_insert(). The request will
723 * be put back on the queue and retried using the same
724 * command as before, possibly after a delay.
726 * c) We can call scsi_end_request() with -EIO to fail
727 * the remainder of the request.
729 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
731 int result
= cmd
->result
;
732 struct request_queue
*q
= cmd
->device
->request_queue
;
733 struct request
*req
= cmd
->request
;
735 struct scsi_sense_hdr sshdr
;
736 bool sense_valid
= false;
737 int sense_deferred
= 0, level
= 0;
738 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
739 ACTION_DELAYED_RETRY
} action
;
740 unsigned long wait_for
= (cmd
->allowed
+ 1) * req
->timeout
;
743 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
745 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
748 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
750 if (sense_valid
&& req
->sense
) {
752 * SG_IO wants current and deferred errors
754 int len
= 8 + cmd
->sense_buffer
[7];
756 if (len
> SCSI_SENSE_BUFFERSIZE
)
757 len
= SCSI_SENSE_BUFFERSIZE
;
758 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
759 req
->sense_len
= len
;
762 error
= __scsi_error_from_host_byte(cmd
, result
);
765 * __scsi_error_from_host_byte may have reset the host_byte
767 req
->errors
= cmd
->result
;
769 req
->resid_len
= scsi_get_resid(cmd
);
771 if (scsi_bidi_cmnd(cmd
)) {
773 * Bidi commands Must be complete as a whole,
774 * both sides at once.
776 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
777 if (scsi_end_request(req
, 0, blk_rq_bytes(req
),
778 blk_rq_bytes(req
->next_rq
)))
782 } else if (blk_rq_bytes(req
) == 0 && result
&& !sense_deferred
) {
784 * Certain non BLOCK_PC requests are commands that don't
785 * actually transfer anything (FLUSH), so cannot use
786 * good_bytes != blk_rq_bytes(req) as the signal for an error.
787 * This sets the error explicitly for the problem case.
789 error
= __scsi_error_from_host_byte(cmd
, result
);
792 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
793 BUG_ON(blk_bidi_rq(req
));
796 * Next deal with any sectors which we were able to correctly
799 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO
, cmd
,
800 "%u sectors total, %d bytes done.\n",
801 blk_rq_sectors(req
), good_bytes
));
804 * Recovered errors need reporting, but they're always treated
805 * as success, so fiddle the result code here. For BLOCK_PC
806 * we already took a copy of the original into rq->errors which
807 * is what gets returned to the user
809 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
810 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
811 * print since caller wants ATA registers. Only occurs on
812 * SCSI ATA PASS_THROUGH commands when CK_COND=1
814 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
816 else if (!(req
->cmd_flags
& REQ_QUIET
))
817 scsi_print_sense(cmd
);
819 /* BLOCK_PC may have set error */
824 * If we finished all bytes in the request we are done now.
826 if (!scsi_end_request(req
, error
, good_bytes
, 0))
830 * Kill remainder if no retrys.
832 if (error
&& scsi_noretry_cmd(cmd
)) {
833 if (scsi_end_request(req
, error
, blk_rq_bytes(req
), 0))
839 * If there had been no error, but we have leftover bytes in the
840 * requeues just queue the command up again.
845 error
= __scsi_error_from_host_byte(cmd
, result
);
847 if (host_byte(result
) == DID_RESET
) {
848 /* Third party bus reset or reset for error recovery
849 * reasons. Just retry the command and see what
852 action
= ACTION_RETRY
;
853 } else if (sense_valid
&& !sense_deferred
) {
854 switch (sshdr
.sense_key
) {
856 if (cmd
->device
->removable
) {
857 /* Detected disc change. Set a bit
858 * and quietly refuse further access.
860 cmd
->device
->changed
= 1;
861 action
= ACTION_FAIL
;
863 /* Must have been a power glitch, or a
864 * bus reset. Could not have been a
865 * media change, so we just retry the
866 * command and see what happens.
868 action
= ACTION_RETRY
;
871 case ILLEGAL_REQUEST
:
872 /* If we had an ILLEGAL REQUEST returned, then
873 * we may have performed an unsupported
874 * command. The only thing this should be
875 * would be a ten byte read where only a six
876 * byte read was supported. Also, on a system
877 * where READ CAPACITY failed, we may have
878 * read past the end of the disk.
880 if ((cmd
->device
->use_10_for_rw
&&
881 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
882 (cmd
->cmnd
[0] == READ_10
||
883 cmd
->cmnd
[0] == WRITE_10
)) {
884 /* This will issue a new 6-byte command. */
885 cmd
->device
->use_10_for_rw
= 0;
886 action
= ACTION_REPREP
;
887 } else if (sshdr
.asc
== 0x10) /* DIX */ {
888 action
= ACTION_FAIL
;
890 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
891 } else if (sshdr
.asc
== 0x20 || sshdr
.asc
== 0x24) {
892 action
= ACTION_FAIL
;
895 action
= ACTION_FAIL
;
897 case ABORTED_COMMAND
:
898 action
= ACTION_FAIL
;
899 if (sshdr
.asc
== 0x10) /* DIF */
903 /* If the device is in the process of becoming
904 * ready, or has a temporary blockage, retry.
906 if (sshdr
.asc
== 0x04) {
907 switch (sshdr
.ascq
) {
908 case 0x01: /* becoming ready */
909 case 0x04: /* format in progress */
910 case 0x05: /* rebuild in progress */
911 case 0x06: /* recalculation in progress */
912 case 0x07: /* operation in progress */
913 case 0x08: /* Long write in progress */
914 case 0x09: /* self test in progress */
915 case 0x14: /* space allocation in progress */
916 action
= ACTION_DELAYED_RETRY
;
919 action
= ACTION_FAIL
;
923 action
= ACTION_FAIL
;
925 case VOLUME_OVERFLOW
:
926 /* See SSC3rXX or current. */
927 action
= ACTION_FAIL
;
930 action
= ACTION_FAIL
;
934 action
= ACTION_FAIL
;
936 if (action
!= ACTION_FAIL
&&
937 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
))
938 action
= ACTION_FAIL
;
942 /* Give up and fail the remainder of the request */
943 if (!(req
->cmd_flags
& REQ_QUIET
)) {
944 static DEFINE_RATELIMIT_STATE(_rs
,
945 DEFAULT_RATELIMIT_INTERVAL
,
946 DEFAULT_RATELIMIT_BURST
);
948 if (unlikely(scsi_logging_level
))
949 level
= SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT
,
950 SCSI_LOG_MLCOMPLETE_BITS
);
953 * if logging is enabled the failure will be printed
954 * in scsi_log_completion(), so avoid duplicate messages
956 if (!level
&& __ratelimit(&_rs
)) {
957 scsi_print_result(cmd
, NULL
, FAILED
);
958 if (driver_byte(result
) & DRIVER_SENSE
)
959 scsi_print_sense(cmd
);
960 scsi_print_command(cmd
);
963 if (!scsi_end_request(req
, error
, blk_rq_err_bytes(req
), 0))
968 /* Unprep the request and put it back at the head of the queue.
969 * A new command will be prepared and issued.
972 cmd
->request
->cmd_flags
&= ~REQ_DONTPREP
;
973 scsi_mq_uninit_cmd(cmd
);
974 scsi_mq_requeue_cmd(cmd
);
976 scsi_release_buffers(cmd
);
977 scsi_requeue_command(q
, cmd
);
981 /* Retry the same command immediately */
982 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
984 case ACTION_DELAYED_RETRY
:
985 /* Retry the same command after a delay */
986 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
991 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
)
996 * If sg table allocation fails, requeue request later.
998 if (unlikely(sg_alloc_table_chained(&sdb
->table
, req
->nr_phys_segments
,
1000 return BLKPREP_DEFER
;
1003 * Next, walk the list, and fill in the addresses and sizes of
1006 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1007 BUG_ON(count
> sdb
->table
.nents
);
1008 sdb
->table
.nents
= count
;
1009 sdb
->length
= blk_rq_bytes(req
);
1014 * Function: scsi_init_io()
1016 * Purpose: SCSI I/O initialize function.
1018 * Arguments: cmd - Command descriptor we wish to initialize
1020 * Returns: 0 on success
1021 * BLKPREP_DEFER if the failure is retryable
1022 * BLKPREP_KILL if the failure is fatal
1024 int scsi_init_io(struct scsi_cmnd
*cmd
)
1026 struct scsi_device
*sdev
= cmd
->device
;
1027 struct request
*rq
= cmd
->request
;
1028 bool is_mq
= (rq
->mq_ctx
!= NULL
);
1031 BUG_ON(!rq
->nr_phys_segments
);
1033 error
= scsi_init_sgtable(rq
, &cmd
->sdb
);
1037 if (blk_bidi_rq(rq
)) {
1038 if (!rq
->q
->mq_ops
) {
1039 struct scsi_data_buffer
*bidi_sdb
=
1040 kmem_cache_zalloc(scsi_sdb_cache
, GFP_ATOMIC
);
1042 error
= BLKPREP_DEFER
;
1046 rq
->next_rq
->special
= bidi_sdb
;
1049 error
= scsi_init_sgtable(rq
->next_rq
, rq
->next_rq
->special
);
1054 if (blk_integrity_rq(rq
)) {
1055 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
1058 if (prot_sdb
== NULL
) {
1060 * This can happen if someone (e.g. multipath)
1061 * queues a command to a device on an adapter
1062 * that does not support DIX.
1065 error
= BLKPREP_KILL
;
1069 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
1071 if (sg_alloc_table_chained(&prot_sdb
->table
, ivecs
,
1072 prot_sdb
->table
.sgl
)) {
1073 error
= BLKPREP_DEFER
;
1077 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1078 prot_sdb
->table
.sgl
);
1079 BUG_ON(unlikely(count
> ivecs
));
1080 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1082 cmd
->prot_sdb
= prot_sdb
;
1083 cmd
->prot_sdb
->table
.nents
= count
;
1089 scsi_mq_free_sgtables(cmd
);
1091 scsi_release_buffers(cmd
);
1092 cmd
->request
->special
= NULL
;
1093 scsi_put_command(cmd
);
1094 put_device(&sdev
->sdev_gendev
);
1098 EXPORT_SYMBOL(scsi_init_io
);
1100 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1101 struct request
*req
)
1103 struct scsi_cmnd
*cmd
;
1105 if (!req
->special
) {
1106 /* Bail if we can't get a reference to the device */
1107 if (!get_device(&sdev
->sdev_gendev
))
1110 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1111 if (unlikely(!cmd
)) {
1112 put_device(&sdev
->sdev_gendev
);
1120 /* pull a tag out of the request if we have one */
1121 cmd
->tag
= req
->tag
;
1124 cmd
->cmnd
= req
->cmd
;
1125 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1130 static int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1132 struct scsi_cmnd
*cmd
= req
->special
;
1135 * BLOCK_PC requests may transfer data, in which case they must
1136 * a bio attached to them. Or they might contain a SCSI command
1137 * that does not transfer data, in which case they may optionally
1138 * submit a request without an attached bio.
1141 int ret
= scsi_init_io(cmd
);
1145 BUG_ON(blk_rq_bytes(req
));
1147 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1150 cmd
->cmd_len
= req
->cmd_len
;
1151 cmd
->transfersize
= blk_rq_bytes(req
);
1152 cmd
->allowed
= req
->retries
;
1157 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1158 * that still need to be translated to SCSI CDBs from the ULD.
1160 static int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1162 struct scsi_cmnd
*cmd
= req
->special
;
1164 if (unlikely(sdev
->handler
&& sdev
->handler
->prep_fn
)) {
1165 int ret
= sdev
->handler
->prep_fn(sdev
, req
);
1166 if (ret
!= BLKPREP_OK
)
1170 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1171 return scsi_cmd_to_driver(cmd
)->init_command(cmd
);
1174 static int scsi_setup_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1176 struct scsi_cmnd
*cmd
= req
->special
;
1178 if (!blk_rq_bytes(req
))
1179 cmd
->sc_data_direction
= DMA_NONE
;
1180 else if (rq_data_dir(req
) == WRITE
)
1181 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1183 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1185 switch (req
->cmd_type
) {
1187 return scsi_setup_fs_cmnd(sdev
, req
);
1188 case REQ_TYPE_BLOCK_PC
:
1189 return scsi_setup_blk_pc_cmnd(sdev
, req
);
1191 return BLKPREP_KILL
;
1196 scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1198 int ret
= BLKPREP_OK
;
1201 * If the device is not in running state we will reject some
1204 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1205 switch (sdev
->sdev_state
) {
1207 case SDEV_TRANSPORT_OFFLINE
:
1209 * If the device is offline we refuse to process any
1210 * commands. The device must be brought online
1211 * before trying any recovery commands.
1213 sdev_printk(KERN_ERR
, sdev
,
1214 "rejecting I/O to offline device\n");
1219 * If the device is fully deleted, we refuse to
1220 * process any commands as well.
1222 sdev_printk(KERN_ERR
, sdev
,
1223 "rejecting I/O to dead device\n");
1227 case SDEV_CREATED_BLOCK
:
1228 ret
= BLKPREP_DEFER
;
1232 * If the devices is blocked we defer normal commands.
1234 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1235 ret
= BLKPREP_DEFER
;
1239 * For any other not fully online state we only allow
1240 * special commands. In particular any user initiated
1241 * command is not allowed.
1243 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1252 scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1254 struct scsi_device
*sdev
= q
->queuedata
;
1258 case BLKPREP_INVALID
:
1259 req
->errors
= DID_NO_CONNECT
<< 16;
1260 /* release the command and kill it */
1262 struct scsi_cmnd
*cmd
= req
->special
;
1263 scsi_release_buffers(cmd
);
1264 scsi_put_command(cmd
);
1265 put_device(&sdev
->sdev_gendev
);
1266 req
->special
= NULL
;
1271 * If we defer, the blk_peek_request() returns NULL, but the
1272 * queue must be restarted, so we schedule a callback to happen
1275 if (atomic_read(&sdev
->device_busy
) == 0)
1276 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1279 req
->cmd_flags
|= REQ_DONTPREP
;
1285 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1287 struct scsi_device
*sdev
= q
->queuedata
;
1288 struct scsi_cmnd
*cmd
;
1291 ret
= scsi_prep_state_check(sdev
, req
);
1292 if (ret
!= BLKPREP_OK
)
1295 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1296 if (unlikely(!cmd
)) {
1297 ret
= BLKPREP_DEFER
;
1301 ret
= scsi_setup_cmnd(sdev
, req
);
1303 return scsi_prep_return(q
, req
, ret
);
1306 static void scsi_unprep_fn(struct request_queue
*q
, struct request
*req
)
1308 scsi_uninit_cmd(req
->special
);
1312 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1315 * Called with the queue_lock held.
1317 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1318 struct scsi_device
*sdev
)
1322 busy
= atomic_inc_return(&sdev
->device_busy
) - 1;
1323 if (atomic_read(&sdev
->device_blocked
)) {
1328 * unblock after device_blocked iterates to zero
1330 if (atomic_dec_return(&sdev
->device_blocked
) > 0) {
1332 * For the MQ case we take care of this in the caller.
1335 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1338 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO
, sdev
,
1339 "unblocking device at zero depth\n"));
1342 if (busy
>= sdev
->queue_depth
)
1347 atomic_dec(&sdev
->device_busy
);
1352 * scsi_target_queue_ready: checks if there we can send commands to target
1353 * @sdev: scsi device on starget to check.
1355 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1356 struct scsi_device
*sdev
)
1358 struct scsi_target
*starget
= scsi_target(sdev
);
1361 if (starget
->single_lun
) {
1362 spin_lock_irq(shost
->host_lock
);
1363 if (starget
->starget_sdev_user
&&
1364 starget
->starget_sdev_user
!= sdev
) {
1365 spin_unlock_irq(shost
->host_lock
);
1368 starget
->starget_sdev_user
= sdev
;
1369 spin_unlock_irq(shost
->host_lock
);
1372 if (starget
->can_queue
<= 0)
1375 busy
= atomic_inc_return(&starget
->target_busy
) - 1;
1376 if (atomic_read(&starget
->target_blocked
) > 0) {
1381 * unblock after target_blocked iterates to zero
1383 if (atomic_dec_return(&starget
->target_blocked
) > 0)
1386 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1387 "unblocking target at zero depth\n"));
1390 if (busy
>= starget
->can_queue
)
1396 spin_lock_irq(shost
->host_lock
);
1397 list_move_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1398 spin_unlock_irq(shost
->host_lock
);
1400 if (starget
->can_queue
> 0)
1401 atomic_dec(&starget
->target_busy
);
1406 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1407 * return 0. We must end up running the queue again whenever 0 is
1408 * returned, else IO can hang.
1410 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1411 struct Scsi_Host
*shost
,
1412 struct scsi_device
*sdev
)
1416 if (scsi_host_in_recovery(shost
))
1419 busy
= atomic_inc_return(&shost
->host_busy
) - 1;
1420 if (atomic_read(&shost
->host_blocked
) > 0) {
1425 * unblock after host_blocked iterates to zero
1427 if (atomic_dec_return(&shost
->host_blocked
) > 0)
1431 shost_printk(KERN_INFO
, shost
,
1432 "unblocking host at zero depth\n"));
1435 if (shost
->can_queue
> 0 && busy
>= shost
->can_queue
)
1437 if (shost
->host_self_blocked
)
1440 /* We're OK to process the command, so we can't be starved */
1441 if (!list_empty(&sdev
->starved_entry
)) {
1442 spin_lock_irq(shost
->host_lock
);
1443 if (!list_empty(&sdev
->starved_entry
))
1444 list_del_init(&sdev
->starved_entry
);
1445 spin_unlock_irq(shost
->host_lock
);
1451 spin_lock_irq(shost
->host_lock
);
1452 if (list_empty(&sdev
->starved_entry
))
1453 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1454 spin_unlock_irq(shost
->host_lock
);
1456 atomic_dec(&shost
->host_busy
);
1461 * Busy state exporting function for request stacking drivers.
1463 * For efficiency, no lock is taken to check the busy state of
1464 * shost/starget/sdev, since the returned value is not guaranteed and
1465 * may be changed after request stacking drivers call the function,
1466 * regardless of taking lock or not.
1468 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1469 * needs to return 'not busy'. Otherwise, request stacking drivers
1470 * may hold requests forever.
1472 static int scsi_lld_busy(struct request_queue
*q
)
1474 struct scsi_device
*sdev
= q
->queuedata
;
1475 struct Scsi_Host
*shost
;
1477 if (blk_queue_dying(q
))
1483 * Ignore host/starget busy state.
1484 * Since block layer does not have a concept of fairness across
1485 * multiple queues, congestion of host/starget needs to be handled
1488 if (scsi_host_in_recovery(shost
) || scsi_device_is_busy(sdev
))
1495 * Kill a request for a dead device
1497 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1499 struct scsi_cmnd
*cmd
= req
->special
;
1500 struct scsi_device
*sdev
;
1501 struct scsi_target
*starget
;
1502 struct Scsi_Host
*shost
;
1504 blk_start_request(req
);
1506 scmd_printk(KERN_INFO
, cmd
, "killing request\n");
1509 starget
= scsi_target(sdev
);
1511 scsi_init_cmd_errh(cmd
);
1512 cmd
->result
= DID_NO_CONNECT
<< 16;
1513 atomic_inc(&cmd
->device
->iorequest_cnt
);
1516 * SCSI request completion path will do scsi_device_unbusy(),
1517 * bump busy counts. To bump the counters, we need to dance
1518 * with the locks as normal issue path does.
1520 atomic_inc(&sdev
->device_busy
);
1521 atomic_inc(&shost
->host_busy
);
1522 if (starget
->can_queue
> 0)
1523 atomic_inc(&starget
->target_busy
);
1525 blk_complete_request(req
);
1528 static void scsi_softirq_done(struct request
*rq
)
1530 struct scsi_cmnd
*cmd
= rq
->special
;
1531 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1534 INIT_LIST_HEAD(&cmd
->eh_entry
);
1536 atomic_inc(&cmd
->device
->iodone_cnt
);
1538 atomic_inc(&cmd
->device
->ioerr_cnt
);
1540 disposition
= scsi_decide_disposition(cmd
);
1541 if (disposition
!= SUCCESS
&&
1542 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1543 sdev_printk(KERN_ERR
, cmd
->device
,
1544 "timing out command, waited %lus\n",
1546 disposition
= SUCCESS
;
1549 scsi_log_completion(cmd
, disposition
);
1551 switch (disposition
) {
1553 scsi_finish_command(cmd
);
1556 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1558 case ADD_TO_MLQUEUE
:
1559 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1562 if (!scsi_eh_scmd_add(cmd
, 0))
1563 scsi_finish_command(cmd
);
1568 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1569 * @cmd: command block we are dispatching.
1571 * Return: nonzero return request was rejected and device's queue needs to be
1574 static int scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
1576 struct Scsi_Host
*host
= cmd
->device
->host
;
1579 atomic_inc(&cmd
->device
->iorequest_cnt
);
1581 /* check if the device is still usable */
1582 if (unlikely(cmd
->device
->sdev_state
== SDEV_DEL
)) {
1583 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1584 * returns an immediate error upwards, and signals
1585 * that the device is no longer present */
1586 cmd
->result
= DID_NO_CONNECT
<< 16;
1590 /* Check to see if the scsi lld made this device blocked. */
1591 if (unlikely(scsi_device_blocked(cmd
->device
))) {
1593 * in blocked state, the command is just put back on
1594 * the device queue. The suspend state has already
1595 * blocked the queue so future requests should not
1596 * occur until the device transitions out of the
1599 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1600 "queuecommand : device blocked\n"));
1601 return SCSI_MLQUEUE_DEVICE_BUSY
;
1604 /* Store the LUN value in cmnd, if needed. */
1605 if (cmd
->device
->lun_in_cdb
)
1606 cmd
->cmnd
[1] = (cmd
->cmnd
[1] & 0x1f) |
1607 (cmd
->device
->lun
<< 5 & 0xe0);
1612 * Before we queue this command, check if the command
1613 * length exceeds what the host adapter can handle.
1615 if (cmd
->cmd_len
> cmd
->device
->host
->max_cmd_len
) {
1616 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1617 "queuecommand : command too long. "
1618 "cdb_size=%d host->max_cmd_len=%d\n",
1619 cmd
->cmd_len
, cmd
->device
->host
->max_cmd_len
));
1620 cmd
->result
= (DID_ABORT
<< 16);
1624 if (unlikely(host
->shost_state
== SHOST_DEL
)) {
1625 cmd
->result
= (DID_NO_CONNECT
<< 16);
1630 trace_scsi_dispatch_cmd_start(cmd
);
1631 rtn
= host
->hostt
->queuecommand(host
, cmd
);
1633 trace_scsi_dispatch_cmd_error(cmd
, rtn
);
1634 if (rtn
!= SCSI_MLQUEUE_DEVICE_BUSY
&&
1635 rtn
!= SCSI_MLQUEUE_TARGET_BUSY
)
1636 rtn
= SCSI_MLQUEUE_HOST_BUSY
;
1638 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO
, cmd
,
1639 "queuecommand : request rejected\n"));
1644 cmd
->scsi_done(cmd
);
1649 * scsi_done - Invoke completion on finished SCSI command.
1650 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1651 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1653 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1654 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1655 * calls blk_complete_request() for further processing.
1657 * This function is interrupt context safe.
1659 static void scsi_done(struct scsi_cmnd
*cmd
)
1661 trace_scsi_dispatch_cmd_done(cmd
);
1662 blk_complete_request(cmd
->request
);
1666 * Function: scsi_request_fn()
1668 * Purpose: Main strategy routine for SCSI.
1670 * Arguments: q - Pointer to actual queue.
1674 * Lock status: IO request lock assumed to be held when called.
1676 static void scsi_request_fn(struct request_queue
*q
)
1677 __releases(q
->queue_lock
)
1678 __acquires(q
->queue_lock
)
1680 struct scsi_device
*sdev
= q
->queuedata
;
1681 struct Scsi_Host
*shost
;
1682 struct scsi_cmnd
*cmd
;
1683 struct request
*req
;
1686 * To start with, we keep looping until the queue is empty, or until
1687 * the host is no longer able to accept any more requests.
1693 * get next queueable request. We do this early to make sure
1694 * that the request is fully prepared even if we cannot
1697 req
= blk_peek_request(q
);
1701 if (unlikely(!scsi_device_online(sdev
))) {
1702 sdev_printk(KERN_ERR
, sdev
,
1703 "rejecting I/O to offline device\n");
1704 scsi_kill_request(req
, q
);
1708 if (!scsi_dev_queue_ready(q
, sdev
))
1712 * Remove the request from the request list.
1714 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1715 blk_start_request(req
);
1717 spin_unlock_irq(q
->queue_lock
);
1719 if (unlikely(cmd
== NULL
)) {
1720 printk(KERN_CRIT
"impossible request in %s.\n"
1721 "please mail a stack trace to "
1722 "linux-scsi@vger.kernel.org\n",
1724 blk_dump_rq_flags(req
, "foo");
1729 * We hit this when the driver is using a host wide
1730 * tag map. For device level tag maps the queue_depth check
1731 * in the device ready fn would prevent us from trying
1732 * to allocate a tag. Since the map is a shared host resource
1733 * we add the dev to the starved list so it eventually gets
1734 * a run when a tag is freed.
1736 if (blk_queue_tagged(q
) && !(req
->cmd_flags
& REQ_QUEUED
)) {
1737 spin_lock_irq(shost
->host_lock
);
1738 if (list_empty(&sdev
->starved_entry
))
1739 list_add_tail(&sdev
->starved_entry
,
1740 &shost
->starved_list
);
1741 spin_unlock_irq(shost
->host_lock
);
1745 if (!scsi_target_queue_ready(shost
, sdev
))
1748 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1749 goto host_not_ready
;
1751 if (sdev
->simple_tags
)
1752 cmd
->flags
|= SCMD_TAGGED
;
1754 cmd
->flags
&= ~SCMD_TAGGED
;
1757 * Finally, initialize any error handling parameters, and set up
1758 * the timers for timeouts.
1760 scsi_init_cmd_errh(cmd
);
1763 * Dispatch the command to the low-level driver.
1765 cmd
->scsi_done
= scsi_done
;
1766 rtn
= scsi_dispatch_cmd(cmd
);
1768 scsi_queue_insert(cmd
, rtn
);
1769 spin_lock_irq(q
->queue_lock
);
1772 spin_lock_irq(q
->queue_lock
);
1778 if (scsi_target(sdev
)->can_queue
> 0)
1779 atomic_dec(&scsi_target(sdev
)->target_busy
);
1782 * lock q, handle tag, requeue req, and decrement device_busy. We
1783 * must return with queue_lock held.
1785 * Decrementing device_busy without checking it is OK, as all such
1786 * cases (host limits or settings) should run the queue at some
1789 spin_lock_irq(q
->queue_lock
);
1790 blk_requeue_request(q
, req
);
1791 atomic_dec(&sdev
->device_busy
);
1793 if (!atomic_read(&sdev
->device_busy
) && !scsi_device_blocked(sdev
))
1794 blk_delay_queue(q
, SCSI_QUEUE_DELAY
);
1797 static inline int prep_to_mq(int ret
)
1803 return BLK_MQ_RQ_QUEUE_BUSY
;
1805 return BLK_MQ_RQ_QUEUE_ERROR
;
1809 static int scsi_mq_prep_fn(struct request
*req
)
1811 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1812 struct scsi_device
*sdev
= req
->q
->queuedata
;
1813 struct Scsi_Host
*shost
= sdev
->host
;
1814 unsigned char *sense_buf
= cmd
->sense_buffer
;
1815 struct scatterlist
*sg
;
1817 memset(cmd
, 0, sizeof(struct scsi_cmnd
));
1823 cmd
->sense_buffer
= sense_buf
;
1825 cmd
->tag
= req
->tag
;
1827 cmd
->cmnd
= req
->cmd
;
1828 cmd
->prot_op
= SCSI_PROT_NORMAL
;
1830 INIT_LIST_HEAD(&cmd
->list
);
1831 INIT_DELAYED_WORK(&cmd
->abort_work
, scmd_eh_abort_handler
);
1832 cmd
->jiffies_at_alloc
= jiffies
;
1834 if (shost
->use_cmd_list
) {
1835 spin_lock_irq(&sdev
->list_lock
);
1836 list_add_tail(&cmd
->list
, &sdev
->cmd_list
);
1837 spin_unlock_irq(&sdev
->list_lock
);
1840 sg
= (void *)cmd
+ sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
;
1841 cmd
->sdb
.table
.sgl
= sg
;
1843 if (scsi_host_get_prot(shost
)) {
1844 cmd
->prot_sdb
= (void *)sg
+
1846 shost
->sg_tablesize
, SG_CHUNK_SIZE
) *
1847 sizeof(struct scatterlist
);
1848 memset(cmd
->prot_sdb
, 0, sizeof(struct scsi_data_buffer
));
1850 cmd
->prot_sdb
->table
.sgl
=
1851 (struct scatterlist
*)(cmd
->prot_sdb
+ 1);
1854 if (blk_bidi_rq(req
)) {
1855 struct request
*next_rq
= req
->next_rq
;
1856 struct scsi_data_buffer
*bidi_sdb
= blk_mq_rq_to_pdu(next_rq
);
1858 memset(bidi_sdb
, 0, sizeof(struct scsi_data_buffer
));
1859 bidi_sdb
->table
.sgl
=
1860 (struct scatterlist
*)(bidi_sdb
+ 1);
1862 next_rq
->special
= bidi_sdb
;
1865 blk_mq_start_request(req
);
1867 return scsi_setup_cmnd(sdev
, req
);
1870 static void scsi_mq_done(struct scsi_cmnd
*cmd
)
1872 trace_scsi_dispatch_cmd_done(cmd
);
1873 blk_mq_complete_request(cmd
->request
, cmd
->request
->errors
);
1876 static int scsi_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1877 const struct blk_mq_queue_data
*bd
)
1879 struct request
*req
= bd
->rq
;
1880 struct request_queue
*q
= req
->q
;
1881 struct scsi_device
*sdev
= q
->queuedata
;
1882 struct Scsi_Host
*shost
= sdev
->host
;
1883 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(req
);
1887 ret
= prep_to_mq(scsi_prep_state_check(sdev
, req
));
1891 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1892 if (!get_device(&sdev
->sdev_gendev
))
1895 if (!scsi_dev_queue_ready(q
, sdev
))
1896 goto out_put_device
;
1897 if (!scsi_target_queue_ready(shost
, sdev
))
1898 goto out_dec_device_busy
;
1899 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1900 goto out_dec_target_busy
;
1903 if (!(req
->cmd_flags
& REQ_DONTPREP
)) {
1904 ret
= prep_to_mq(scsi_mq_prep_fn(req
));
1906 goto out_dec_host_busy
;
1907 req
->cmd_flags
|= REQ_DONTPREP
;
1909 blk_mq_start_request(req
);
1912 if (sdev
->simple_tags
)
1913 cmd
->flags
|= SCMD_TAGGED
;
1915 cmd
->flags
&= ~SCMD_TAGGED
;
1917 scsi_init_cmd_errh(cmd
);
1918 cmd
->scsi_done
= scsi_mq_done
;
1920 reason
= scsi_dispatch_cmd(cmd
);
1922 scsi_set_blocked(cmd
, reason
);
1923 ret
= BLK_MQ_RQ_QUEUE_BUSY
;
1924 goto out_dec_host_busy
;
1927 return BLK_MQ_RQ_QUEUE_OK
;
1930 atomic_dec(&shost
->host_busy
);
1931 out_dec_target_busy
:
1932 if (scsi_target(sdev
)->can_queue
> 0)
1933 atomic_dec(&scsi_target(sdev
)->target_busy
);
1934 out_dec_device_busy
:
1935 atomic_dec(&sdev
->device_busy
);
1937 put_device(&sdev
->sdev_gendev
);
1940 case BLK_MQ_RQ_QUEUE_BUSY
:
1941 blk_mq_stop_hw_queue(hctx
);
1942 if (atomic_read(&sdev
->device_busy
) == 0 &&
1943 !scsi_device_blocked(sdev
))
1944 blk_mq_delay_queue(hctx
, SCSI_QUEUE_DELAY
);
1946 case BLK_MQ_RQ_QUEUE_ERROR
:
1948 * Make sure to release all allocated ressources when
1949 * we hit an error, as we will never see this command
1952 if (req
->cmd_flags
& REQ_DONTPREP
)
1953 scsi_mq_uninit_cmd(cmd
);
1961 static enum blk_eh_timer_return
scsi_timeout(struct request
*req
,
1965 return BLK_EH_RESET_TIMER
;
1966 return scsi_times_out(req
);
1969 static int scsi_init_request(void *data
, struct request
*rq
,
1970 unsigned int hctx_idx
, unsigned int request_idx
,
1971 unsigned int numa_node
)
1973 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1975 cmd
->sense_buffer
= kzalloc_node(SCSI_SENSE_BUFFERSIZE
, GFP_KERNEL
,
1977 if (!cmd
->sense_buffer
)
1982 static void scsi_exit_request(void *data
, struct request
*rq
,
1983 unsigned int hctx_idx
, unsigned int request_idx
)
1985 struct scsi_cmnd
*cmd
= blk_mq_rq_to_pdu(rq
);
1987 kfree(cmd
->sense_buffer
);
1990 static u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1992 struct device
*host_dev
;
1993 u64 bounce_limit
= 0xffffffff;
1995 if (shost
->unchecked_isa_dma
)
1996 return BLK_BOUNCE_ISA
;
1998 * Platforms with virtual-DMA translation
1999 * hardware have no practical limit.
2001 if (!PCI_DMA_BUS_IS_PHYS
)
2002 return BLK_BOUNCE_ANY
;
2004 host_dev
= scsi_get_device(shost
);
2005 if (host_dev
&& host_dev
->dma_mask
)
2006 bounce_limit
= (u64
)dma_max_pfn(host_dev
) << PAGE_SHIFT
;
2008 return bounce_limit
;
2011 static void __scsi_init_queue(struct Scsi_Host
*shost
, struct request_queue
*q
)
2013 struct device
*dev
= shost
->dma_dev
;
2016 * this limit is imposed by hardware restrictions
2018 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
2021 if (scsi_host_prot_dma(shost
)) {
2022 shost
->sg_prot_tablesize
=
2023 min_not_zero(shost
->sg_prot_tablesize
,
2024 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
2025 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
2026 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
2029 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
2030 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
2031 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
2032 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
2034 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
2036 if (!shost
->use_clustering
)
2037 q
->limits
.cluster
= 0;
2040 * set a reasonable default alignment on word boundaries: the
2041 * host and device may alter it using
2042 * blk_queue_update_dma_alignment() later.
2044 blk_queue_dma_alignment(q
, 0x03);
2047 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
2048 request_fn_proc
*request_fn
)
2050 struct request_queue
*q
;
2052 q
= blk_init_queue(request_fn
, NULL
);
2055 __scsi_init_queue(shost
, q
);
2058 EXPORT_SYMBOL(__scsi_alloc_queue
);
2060 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
2062 struct request_queue
*q
;
2064 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
2068 blk_queue_prep_rq(q
, scsi_prep_fn
);
2069 blk_queue_unprep_rq(q
, scsi_unprep_fn
);
2070 blk_queue_softirq_done(q
, scsi_softirq_done
);
2071 blk_queue_rq_timed_out(q
, scsi_times_out
);
2072 blk_queue_lld_busy(q
, scsi_lld_busy
);
2076 static struct blk_mq_ops scsi_mq_ops
= {
2077 .map_queue
= blk_mq_map_queue
,
2078 .queue_rq
= scsi_queue_rq
,
2079 .complete
= scsi_softirq_done
,
2080 .timeout
= scsi_timeout
,
2081 .init_request
= scsi_init_request
,
2082 .exit_request
= scsi_exit_request
,
2085 struct request_queue
*scsi_mq_alloc_queue(struct scsi_device
*sdev
)
2087 sdev
->request_queue
= blk_mq_init_queue(&sdev
->host
->tag_set
);
2088 if (IS_ERR(sdev
->request_queue
))
2091 sdev
->request_queue
->queuedata
= sdev
;
2092 __scsi_init_queue(sdev
->host
, sdev
->request_queue
);
2093 return sdev
->request_queue
;
2096 int scsi_mq_setup_tags(struct Scsi_Host
*shost
)
2098 unsigned int cmd_size
, sgl_size
, tbl_size
;
2100 tbl_size
= shost
->sg_tablesize
;
2101 if (tbl_size
> SG_CHUNK_SIZE
)
2102 tbl_size
= SG_CHUNK_SIZE
;
2103 sgl_size
= tbl_size
* sizeof(struct scatterlist
);
2104 cmd_size
= sizeof(struct scsi_cmnd
) + shost
->hostt
->cmd_size
+ sgl_size
;
2105 if (scsi_host_get_prot(shost
))
2106 cmd_size
+= sizeof(struct scsi_data_buffer
) + sgl_size
;
2108 memset(&shost
->tag_set
, 0, sizeof(shost
->tag_set
));
2109 shost
->tag_set
.ops
= &scsi_mq_ops
;
2110 shost
->tag_set
.nr_hw_queues
= shost
->nr_hw_queues
? : 1;
2111 shost
->tag_set
.queue_depth
= shost
->can_queue
;
2112 shost
->tag_set
.cmd_size
= cmd_size
;
2113 shost
->tag_set
.numa_node
= NUMA_NO_NODE
;
2114 shost
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2115 shost
->tag_set
.flags
|=
2116 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost
->hostt
->tag_alloc_policy
);
2117 shost
->tag_set
.driver_data
= shost
;
2119 return blk_mq_alloc_tag_set(&shost
->tag_set
);
2122 void scsi_mq_destroy_tags(struct Scsi_Host
*shost
)
2124 blk_mq_free_tag_set(&shost
->tag_set
);
2128 * Function: scsi_block_requests()
2130 * Purpose: Utility function used by low-level drivers to prevent further
2131 * commands from being queued to the device.
2133 * Arguments: shost - Host in question
2137 * Lock status: No locks are assumed held.
2139 * Notes: There is no timer nor any other means by which the requests
2140 * get unblocked other than the low-level driver calling
2141 * scsi_unblock_requests().
2143 void scsi_block_requests(struct Scsi_Host
*shost
)
2145 shost
->host_self_blocked
= 1;
2147 EXPORT_SYMBOL(scsi_block_requests
);
2150 * Function: scsi_unblock_requests()
2152 * Purpose: Utility function used by low-level drivers to allow further
2153 * commands from being queued to the device.
2155 * Arguments: shost - Host in question
2159 * Lock status: No locks are assumed held.
2161 * Notes: There is no timer nor any other means by which the requests
2162 * get unblocked other than the low-level driver calling
2163 * scsi_unblock_requests().
2165 * This is done as an API function so that changes to the
2166 * internals of the scsi mid-layer won't require wholesale
2167 * changes to drivers that use this feature.
2169 void scsi_unblock_requests(struct Scsi_Host
*shost
)
2171 shost
->host_self_blocked
= 0;
2172 scsi_run_host_queues(shost
);
2174 EXPORT_SYMBOL(scsi_unblock_requests
);
2176 int __init
scsi_init_queue(void)
2178 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
2179 sizeof(struct scsi_data_buffer
),
2181 if (!scsi_sdb_cache
) {
2182 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
2189 void scsi_exit_queue(void)
2191 kmem_cache_destroy(scsi_sdb_cache
);
2195 * scsi_mode_select - issue a mode select
2196 * @sdev: SCSI device to be queried
2197 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2198 * @sp: Save page bit (0 == don't save, 1 == save)
2199 * @modepage: mode page being requested
2200 * @buffer: request buffer (may not be smaller than eight bytes)
2201 * @len: length of request buffer.
2202 * @timeout: command timeout
2203 * @retries: number of retries before failing
2204 * @data: returns a structure abstracting the mode header data
2205 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2206 * must be SCSI_SENSE_BUFFERSIZE big.
2208 * Returns zero if successful; negative error number or scsi
2213 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
2214 unsigned char *buffer
, int len
, int timeout
, int retries
,
2215 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2217 unsigned char cmd
[10];
2218 unsigned char *real_buffer
;
2221 memset(cmd
, 0, sizeof(cmd
));
2222 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
2224 if (sdev
->use_10_for_ms
) {
2227 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
2230 memcpy(real_buffer
+ 8, buffer
, len
);
2234 real_buffer
[2] = data
->medium_type
;
2235 real_buffer
[3] = data
->device_specific
;
2236 real_buffer
[4] = data
->longlba
? 0x01 : 0;
2238 real_buffer
[6] = data
->block_descriptor_length
>> 8;
2239 real_buffer
[7] = data
->block_descriptor_length
;
2241 cmd
[0] = MODE_SELECT_10
;
2245 if (len
> 255 || data
->block_descriptor_length
> 255 ||
2249 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
2252 memcpy(real_buffer
+ 4, buffer
, len
);
2255 real_buffer
[1] = data
->medium_type
;
2256 real_buffer
[2] = data
->device_specific
;
2257 real_buffer
[3] = data
->block_descriptor_length
;
2260 cmd
[0] = MODE_SELECT
;
2264 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
2265 sshdr
, timeout
, retries
, NULL
);
2269 EXPORT_SYMBOL_GPL(scsi_mode_select
);
2272 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2273 * @sdev: SCSI device to be queried
2274 * @dbd: set if mode sense will allow block descriptors to be returned
2275 * @modepage: mode page being requested
2276 * @buffer: request buffer (may not be smaller than eight bytes)
2277 * @len: length of request buffer.
2278 * @timeout: command timeout
2279 * @retries: number of retries before failing
2280 * @data: returns a structure abstracting the mode header data
2281 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2282 * must be SCSI_SENSE_BUFFERSIZE big.
2284 * Returns zero if unsuccessful, or the header offset (either 4
2285 * or 8 depending on whether a six or ten byte command was
2286 * issued) if successful.
2289 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
2290 unsigned char *buffer
, int len
, int timeout
, int retries
,
2291 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
2293 unsigned char cmd
[12];
2296 int result
, retry_count
= retries
;
2297 struct scsi_sense_hdr my_sshdr
;
2299 memset(data
, 0, sizeof(*data
));
2300 memset(&cmd
[0], 0, 12);
2301 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
2304 /* caller might not be interested in sense, but we need it */
2309 use_10_for_ms
= sdev
->use_10_for_ms
;
2311 if (use_10_for_ms
) {
2315 cmd
[0] = MODE_SENSE_10
;
2322 cmd
[0] = MODE_SENSE
;
2327 memset(buffer
, 0, len
);
2329 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
2330 sshdr
, timeout
, retries
, NULL
);
2332 /* This code looks awful: what it's doing is making sure an
2333 * ILLEGAL REQUEST sense return identifies the actual command
2334 * byte as the problem. MODE_SENSE commands can return
2335 * ILLEGAL REQUEST if the code page isn't supported */
2337 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
2338 (driver_byte(result
) & DRIVER_SENSE
)) {
2339 if (scsi_sense_valid(sshdr
)) {
2340 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
2341 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
2343 * Invalid command operation code
2345 sdev
->use_10_for_ms
= 0;
2351 if(scsi_status_is_good(result
)) {
2352 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
2353 (modepage
== 6 || modepage
== 8))) {
2354 /* Initio breakage? */
2357 data
->medium_type
= 0;
2358 data
->device_specific
= 0;
2360 data
->block_descriptor_length
= 0;
2361 } else if(use_10_for_ms
) {
2362 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
2363 data
->medium_type
= buffer
[2];
2364 data
->device_specific
= buffer
[3];
2365 data
->longlba
= buffer
[4] & 0x01;
2366 data
->block_descriptor_length
= buffer
[6]*256
2369 data
->length
= buffer
[0] + 1;
2370 data
->medium_type
= buffer
[1];
2371 data
->device_specific
= buffer
[2];
2372 data
->block_descriptor_length
= buffer
[3];
2374 data
->header_length
= header_length
;
2375 } else if ((status_byte(result
) == CHECK_CONDITION
) &&
2376 scsi_sense_valid(sshdr
) &&
2377 sshdr
->sense_key
== UNIT_ATTENTION
&& retry_count
) {
2384 EXPORT_SYMBOL(scsi_mode_sense
);
2387 * scsi_test_unit_ready - test if unit is ready
2388 * @sdev: scsi device to change the state of.
2389 * @timeout: command timeout
2390 * @retries: number of retries before failing
2391 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2392 * returning sense. Make sure that this is cleared before passing
2395 * Returns zero if unsuccessful or an error if TUR failed. For
2396 * removable media, UNIT_ATTENTION sets ->changed flag.
2399 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
2400 struct scsi_sense_hdr
*sshdr_external
)
2403 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
2405 struct scsi_sense_hdr
*sshdr
;
2408 if (!sshdr_external
)
2409 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
2411 sshdr
= sshdr_external
;
2413 /* try to eat the UNIT_ATTENTION if there are enough retries */
2415 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2416 timeout
, retries
, NULL
);
2417 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2418 sshdr
->sense_key
== UNIT_ATTENTION
)
2420 } while (scsi_sense_valid(sshdr
) &&
2421 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2423 if (!sshdr_external
)
2427 EXPORT_SYMBOL(scsi_test_unit_ready
);
2430 * scsi_device_set_state - Take the given device through the device state model.
2431 * @sdev: scsi device to change the state of.
2432 * @state: state to change to.
2434 * Returns zero if unsuccessful or an error if the requested
2435 * transition is illegal.
2438 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2440 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2442 if (state
== oldstate
)
2448 case SDEV_CREATED_BLOCK
:
2459 case SDEV_TRANSPORT_OFFLINE
:
2472 case SDEV_TRANSPORT_OFFLINE
:
2480 case SDEV_TRANSPORT_OFFLINE
:
2495 case SDEV_CREATED_BLOCK
:
2502 case SDEV_CREATED_BLOCK
:
2517 case SDEV_TRANSPORT_OFFLINE
:
2530 case SDEV_TRANSPORT_OFFLINE
:
2532 case SDEV_CREATED_BLOCK
:
2540 sdev
->sdev_state
= state
;
2544 SCSI_LOG_ERROR_RECOVERY(1,
2545 sdev_printk(KERN_ERR
, sdev
,
2546 "Illegal state transition %s->%s",
2547 scsi_device_state_name(oldstate
),
2548 scsi_device_state_name(state
))
2552 EXPORT_SYMBOL(scsi_device_set_state
);
2555 * sdev_evt_emit - emit a single SCSI device uevent
2556 * @sdev: associated SCSI device
2557 * @evt: event to emit
2559 * Send a single uevent (scsi_event) to the associated scsi_device.
2561 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2566 switch (evt
->evt_type
) {
2567 case SDEV_EVT_MEDIA_CHANGE
:
2568 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2570 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2571 scsi_rescan_device(&sdev
->sdev_gendev
);
2572 envp
[idx
++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2574 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2575 envp
[idx
++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2577 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2578 envp
[idx
++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2580 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2581 envp
[idx
++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2583 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2584 envp
[idx
++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2586 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2587 envp
[idx
++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2596 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2600 * sdev_evt_thread - send a uevent for each scsi event
2601 * @work: work struct for scsi_device
2603 * Dispatch queued events to their associated scsi_device kobjects
2606 void scsi_evt_thread(struct work_struct
*work
)
2608 struct scsi_device
*sdev
;
2609 enum scsi_device_event evt_type
;
2610 LIST_HEAD(event_list
);
2612 sdev
= container_of(work
, struct scsi_device
, event_work
);
2614 for (evt_type
= SDEV_EVT_FIRST
; evt_type
<= SDEV_EVT_LAST
; evt_type
++)
2615 if (test_and_clear_bit(evt_type
, sdev
->pending_events
))
2616 sdev_evt_send_simple(sdev
, evt_type
, GFP_KERNEL
);
2619 struct scsi_event
*evt
;
2620 struct list_head
*this, *tmp
;
2621 unsigned long flags
;
2623 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2624 list_splice_init(&sdev
->event_list
, &event_list
);
2625 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2627 if (list_empty(&event_list
))
2630 list_for_each_safe(this, tmp
, &event_list
) {
2631 evt
= list_entry(this, struct scsi_event
, node
);
2632 list_del(&evt
->node
);
2633 scsi_evt_emit(sdev
, evt
);
2640 * sdev_evt_send - send asserted event to uevent thread
2641 * @sdev: scsi_device event occurred on
2642 * @evt: event to send
2644 * Assert scsi device event asynchronously.
2646 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2648 unsigned long flags
;
2651 /* FIXME: currently this check eliminates all media change events
2652 * for polled devices. Need to update to discriminate between AN
2653 * and polled events */
2654 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2660 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2661 list_add_tail(&evt
->node
, &sdev
->event_list
);
2662 schedule_work(&sdev
->event_work
);
2663 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2665 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2668 * sdev_evt_alloc - allocate a new scsi event
2669 * @evt_type: type of event to allocate
2670 * @gfpflags: GFP flags for allocation
2672 * Allocates and returns a new scsi_event.
2674 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2677 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2681 evt
->evt_type
= evt_type
;
2682 INIT_LIST_HEAD(&evt
->node
);
2684 /* evt_type-specific initialization, if any */
2686 case SDEV_EVT_MEDIA_CHANGE
:
2687 case SDEV_EVT_INQUIRY_CHANGE_REPORTED
:
2688 case SDEV_EVT_CAPACITY_CHANGE_REPORTED
:
2689 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
:
2690 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
:
2691 case SDEV_EVT_LUN_CHANGE_REPORTED
:
2692 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
:
2700 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2703 * sdev_evt_send_simple - send asserted event to uevent thread
2704 * @sdev: scsi_device event occurred on
2705 * @evt_type: type of event to send
2706 * @gfpflags: GFP flags for allocation
2708 * Assert scsi device event asynchronously, given an event type.
2710 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2711 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2713 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2715 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2720 sdev_evt_send(sdev
, evt
);
2722 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2725 * scsi_device_quiesce - Block user issued commands.
2726 * @sdev: scsi device to quiesce.
2728 * This works by trying to transition to the SDEV_QUIESCE state
2729 * (which must be a legal transition). When the device is in this
2730 * state, only special requests will be accepted, all others will
2731 * be deferred. Since special requests may also be requeued requests,
2732 * a successful return doesn't guarantee the device will be
2733 * totally quiescent.
2735 * Must be called with user context, may sleep.
2737 * Returns zero if unsuccessful or an error if not.
2740 scsi_device_quiesce(struct scsi_device
*sdev
)
2742 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2746 scsi_run_queue(sdev
->request_queue
);
2747 while (atomic_read(&sdev
->device_busy
)) {
2748 msleep_interruptible(200);
2749 scsi_run_queue(sdev
->request_queue
);
2753 EXPORT_SYMBOL(scsi_device_quiesce
);
2756 * scsi_device_resume - Restart user issued commands to a quiesced device.
2757 * @sdev: scsi device to resume.
2759 * Moves the device from quiesced back to running and restarts the
2762 * Must be called with user context, may sleep.
2764 void scsi_device_resume(struct scsi_device
*sdev
)
2766 /* check if the device state was mutated prior to resume, and if
2767 * so assume the state is being managed elsewhere (for example
2768 * device deleted during suspend)
2770 if (sdev
->sdev_state
!= SDEV_QUIESCE
||
2771 scsi_device_set_state(sdev
, SDEV_RUNNING
))
2773 scsi_run_queue(sdev
->request_queue
);
2775 EXPORT_SYMBOL(scsi_device_resume
);
2778 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2780 scsi_device_quiesce(sdev
);
2784 scsi_target_quiesce(struct scsi_target
*starget
)
2786 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2788 EXPORT_SYMBOL(scsi_target_quiesce
);
2791 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2793 scsi_device_resume(sdev
);
2797 scsi_target_resume(struct scsi_target
*starget
)
2799 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2801 EXPORT_SYMBOL(scsi_target_resume
);
2804 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2805 * @sdev: device to block
2807 * Block request made by scsi lld's to temporarily stop all
2808 * scsi commands on the specified device. Called from interrupt
2809 * or normal process context.
2811 * Returns zero if successful or error if not
2814 * This routine transitions the device to the SDEV_BLOCK state
2815 * (which must be a legal transition). When the device is in this
2816 * state, all commands are deferred until the scsi lld reenables
2817 * the device with scsi_device_unblock or device_block_tmo fires.
2820 scsi_internal_device_block(struct scsi_device
*sdev
)
2822 struct request_queue
*q
= sdev
->request_queue
;
2823 unsigned long flags
;
2826 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2828 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2835 * The device has transitioned to SDEV_BLOCK. Stop the
2836 * block layer from calling the midlayer with this device's
2840 blk_mq_stop_hw_queues(q
);
2842 spin_lock_irqsave(q
->queue_lock
, flags
);
2844 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2849 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2852 * scsi_internal_device_unblock - resume a device after a block request
2853 * @sdev: device to resume
2854 * @new_state: state to set devices to after unblocking
2856 * Called by scsi lld's or the midlayer to restart the device queue
2857 * for the previously suspended scsi device. Called from interrupt or
2858 * normal process context.
2860 * Returns zero if successful or error if not.
2863 * This routine transitions the device to the SDEV_RUNNING state
2864 * or to one of the offline states (which must be a legal transition)
2865 * allowing the midlayer to goose the queue for this device.
2868 scsi_internal_device_unblock(struct scsi_device
*sdev
,
2869 enum scsi_device_state new_state
)
2871 struct request_queue
*q
= sdev
->request_queue
;
2872 unsigned long flags
;
2875 * Try to transition the scsi device to SDEV_RUNNING or one of the
2876 * offlined states and goose the device queue if successful.
2878 if ((sdev
->sdev_state
== SDEV_BLOCK
) ||
2879 (sdev
->sdev_state
== SDEV_TRANSPORT_OFFLINE
))
2880 sdev
->sdev_state
= new_state
;
2881 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
) {
2882 if (new_state
== SDEV_TRANSPORT_OFFLINE
||
2883 new_state
== SDEV_OFFLINE
)
2884 sdev
->sdev_state
= new_state
;
2886 sdev
->sdev_state
= SDEV_CREATED
;
2887 } else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2888 sdev
->sdev_state
!= SDEV_OFFLINE
)
2892 blk_mq_start_stopped_hw_queues(q
, false);
2894 spin_lock_irqsave(q
->queue_lock
, flags
);
2896 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2901 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2904 device_block(struct scsi_device
*sdev
, void *data
)
2906 scsi_internal_device_block(sdev
);
2910 target_block(struct device
*dev
, void *data
)
2912 if (scsi_is_target_device(dev
))
2913 starget_for_each_device(to_scsi_target(dev
), NULL
,
2919 scsi_target_block(struct device
*dev
)
2921 if (scsi_is_target_device(dev
))
2922 starget_for_each_device(to_scsi_target(dev
), NULL
,
2925 device_for_each_child(dev
, NULL
, target_block
);
2927 EXPORT_SYMBOL_GPL(scsi_target_block
);
2930 device_unblock(struct scsi_device
*sdev
, void *data
)
2932 scsi_internal_device_unblock(sdev
, *(enum scsi_device_state
*)data
);
2936 target_unblock(struct device
*dev
, void *data
)
2938 if (scsi_is_target_device(dev
))
2939 starget_for_each_device(to_scsi_target(dev
), data
,
2945 scsi_target_unblock(struct device
*dev
, enum scsi_device_state new_state
)
2947 if (scsi_is_target_device(dev
))
2948 starget_for_each_device(to_scsi_target(dev
), &new_state
,
2951 device_for_each_child(dev
, &new_state
, target_unblock
);
2953 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2956 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2957 * @sgl: scatter-gather list
2958 * @sg_count: number of segments in sg
2959 * @offset: offset in bytes into sg, on return offset into the mapped area
2960 * @len: bytes to map, on return number of bytes mapped
2962 * Returns virtual address of the start of the mapped page
2964 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2965 size_t *offset
, size_t *len
)
2968 size_t sg_len
= 0, len_complete
= 0;
2969 struct scatterlist
*sg
;
2972 WARN_ON(!irqs_disabled());
2974 for_each_sg(sgl
, sg
, sg_count
, i
) {
2975 len_complete
= sg_len
; /* Complete sg-entries */
2976 sg_len
+= sg
->length
;
2977 if (sg_len
> *offset
)
2981 if (unlikely(i
== sg_count
)) {
2982 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2984 __func__
, sg_len
, *offset
, sg_count
);
2989 /* Offset starting from the beginning of first page in this sg-entry */
2990 *offset
= *offset
- len_complete
+ sg
->offset
;
2992 /* Assumption: contiguous pages can be accessed as "page + i" */
2993 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2994 *offset
&= ~PAGE_MASK
;
2996 /* Bytes in this sg-entry from *offset to the end of the page */
2997 sg_len
= PAGE_SIZE
- *offset
;
3001 return kmap_atomic(page
);
3003 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
3006 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3007 * @virt: virtual address to be unmapped
3009 void scsi_kunmap_atomic_sg(void *virt
)
3011 kunmap_atomic(virt
);
3013 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);
3015 void sdev_disable_disk_events(struct scsi_device
*sdev
)
3017 atomic_inc(&sdev
->disk_events_disable_depth
);
3019 EXPORT_SYMBOL(sdev_disable_disk_events
);
3021 void sdev_enable_disk_events(struct scsi_device
*sdev
)
3023 if (WARN_ON_ONCE(atomic_read(&sdev
->disk_events_disable_depth
) <= 0))
3025 atomic_dec(&sdev
->disk_events_disable_depth
);
3027 EXPORT_SYMBOL(sdev_enable_disk_events
);
3030 * scsi_vpd_lun_id - return a unique device identification
3031 * @sdev: SCSI device
3032 * @id: buffer for the identification
3033 * @id_len: length of the buffer
3035 * Copies a unique device identification into @id based
3036 * on the information in the VPD page 0x83 of the device.
3037 * The string will be formatted as a SCSI name string.
3039 * Returns the length of the identification or error on failure.
3040 * If the identifier is longer than the supplied buffer the actual
3041 * identifier length is returned and the buffer is not zero-padded.
3043 int scsi_vpd_lun_id(struct scsi_device
*sdev
, char *id
, size_t id_len
)
3045 u8 cur_id_type
= 0xff;
3047 unsigned char *d
, *cur_id_str
;
3048 unsigned char __rcu
*vpd_pg83
;
3049 int id_size
= -EINVAL
;
3052 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3059 * Look for the correct descriptor.
3060 * Order of preference for lun descriptor:
3061 * - SCSI name string
3062 * - NAA IEEE Registered Extended
3063 * - EUI-64 based 16-byte
3064 * - EUI-64 based 12-byte
3065 * - NAA IEEE Registered
3066 * - NAA IEEE Extended
3068 * as longer descriptors reduce the likelyhood
3069 * of identification clashes.
3072 /* The id string must be at least 20 bytes + terminating NULL byte */
3078 memset(id
, 0, id_len
);
3080 while (d
< vpd_pg83
+ sdev
->vpd_pg83_len
) {
3081 /* Skip designators not referring to the LUN */
3082 if ((d
[1] & 0x30) != 0x00)
3085 switch (d
[1] & 0xf) {
3088 if (cur_id_size
> d
[3])
3090 /* Prefer anything */
3091 if (cur_id_type
> 0x01 && cur_id_type
!= 0xff)
3094 if (cur_id_size
+ 4 > id_len
)
3095 cur_id_size
= id_len
- 4;
3097 cur_id_type
= d
[1] & 0xf;
3098 id_size
= snprintf(id
, id_len
, "t10.%*pE",
3099 cur_id_size
, cur_id_str
);
3103 if (cur_id_size
> d
[3])
3105 /* Prefer NAA IEEE Registered Extended */
3106 if (cur_id_type
== 0x3 &&
3107 cur_id_size
== d
[3])
3111 cur_id_type
= d
[1] & 0xf;
3112 switch (cur_id_size
) {
3114 id_size
= snprintf(id
, id_len
,
3119 id_size
= snprintf(id
, id_len
,
3124 id_size
= snprintf(id
, id_len
,
3135 if (cur_id_size
> d
[3])
3139 cur_id_type
= d
[1] & 0xf;
3140 switch (cur_id_size
) {
3142 id_size
= snprintf(id
, id_len
,
3147 id_size
= snprintf(id
, id_len
,
3157 /* SCSI name string */
3158 if (cur_id_size
+ 4 > d
[3])
3160 /* Prefer others for truncated descriptor */
3161 if (cur_id_size
&& d
[3] > id_len
)
3163 cur_id_size
= id_size
= d
[3];
3165 cur_id_type
= d
[1] & 0xf;
3166 if (cur_id_size
>= id_len
)
3167 cur_id_size
= id_len
- 1;
3168 memcpy(id
, cur_id_str
, cur_id_size
);
3169 /* Decrease priority for truncated descriptor */
3170 if (cur_id_size
!= id_size
)
3183 EXPORT_SYMBOL(scsi_vpd_lun_id
);
3186 * scsi_vpd_tpg_id - return a target port group identifier
3187 * @sdev: SCSI device
3189 * Returns the Target Port Group identifier from the information
3190 * froom VPD page 0x83 of the device.
3192 * Returns the identifier or error on failure.
3194 int scsi_vpd_tpg_id(struct scsi_device
*sdev
, int *rel_id
)
3197 unsigned char __rcu
*vpd_pg83
;
3198 int group_id
= -EAGAIN
, rel_port
= -1;
3201 vpd_pg83
= rcu_dereference(sdev
->vpd_pg83
);
3207 d
= sdev
->vpd_pg83
+ 4;
3208 while (d
< sdev
->vpd_pg83
+ sdev
->vpd_pg83_len
) {
3209 switch (d
[1] & 0xf) {
3211 /* Relative target port */
3212 rel_port
= get_unaligned_be16(&d
[6]);
3215 /* Target port group */
3216 group_id
= get_unaligned_be16(&d
[6]);
3225 if (group_id
>= 0 && rel_id
&& rel_port
!= -1)
3230 EXPORT_SYMBOL(scsi_vpd_tpg_id
);