2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/init.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/hardirq.h>
23 #include <linux/scatterlist.h>
24 #include <linux/blk-mq.h>
25 #include <linux/ratelimit.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
42 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
43 #define SG_MEMPOOL_SIZE 2
45 struct scsi_host_sg_pool {
48 struct kmem_cache *slab;
52 #define SP(x) { .size = x, "sgpool-" __stringify(x) }
53 #if (SCSI_MAX_SG_SEGMENTS < 32)
54 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
56 static struct scsi_host_sg_pool scsi_sg_pools[] = {
59 #if (SCSI_MAX_SG_SEGMENTS > 32)
61 #if (SCSI_MAX_SG_SEGMENTS > 64)
63 #if (SCSI_MAX_SG_SEGMENTS > 128)
65 #if (SCSI_MAX_SG_SEGMENTS > 256)
66 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
71 SP(SCSI_MAX_SG_SEGMENTS)
75 struct kmem_cache *scsi_sdb_cache;
78 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
79 * not change behaviour from the previous unplug mechanism, experimentation
80 * may prove this needs changing.
82 #define SCSI_QUEUE_DELAY 3
85 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
87 struct Scsi_Host *host = cmd->device->host;
88 struct scsi_device *device = cmd->device;
89 struct scsi_target *starget = scsi_target(device);
92 * Set the appropriate busy bit for the device/host.
94 * If the host/device isn't busy, assume that something actually
95 * completed, and that we should be able to queue a command now.
97 * Note that the prior mid-layer assumption that any host could
98 * always queue at least one command is now broken. The mid-layer
99 * will implement a user specifiable stall (see
100 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
101 * if a command is requeued with no other commands outstanding
102 * either for the device or for the host.
105 case SCSI_MLQUEUE_HOST_BUSY:
106 atomic_set(&host->host_blocked, host->max_host_blocked);
108 case SCSI_MLQUEUE_DEVICE_BUSY:
109 case SCSI_MLQUEUE_EH_RETRY:
110 atomic_set(&device->device_blocked,
111 device->max_device_blocked);
113 case SCSI_MLQUEUE_TARGET_BUSY:
114 atomic_set(&starget->target_blocked,
115 starget->max_target_blocked);
120 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
122 struct scsi_device *sdev = cmd->device;
123 struct request_queue *q = cmd->request->q;
125 blk_mq_requeue_request(cmd->request);
126 blk_mq_kick_requeue_list(q);
127 put_device(&sdev->sdev_gendev);
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
142 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
144 struct scsi_device *device = cmd->device;
145 struct request_queue *q = device->request_queue;
148 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
149 "Inserting command %p into mlqueue\n", cmd));
151 scsi_set_blocked(cmd, reason);
154 * Decrement the counters, since these commands are no longer
155 * active on the host/device.
158 scsi_device_unbusy(device);
161 * Requeue this command. It will go before all other commands
162 * that are already in the queue. Schedule requeue work under
163 * lock such that the kblockd_schedule_work() call happens
164 * before blk_cleanup_queue() finishes.
168 scsi_mq_requeue_cmd(cmd);
171 spin_lock_irqsave(q->queue_lock, flags);
172 blk_requeue_request(q, cmd->request);
173 kblockd_schedule_work(&device->requeue_work);
174 spin_unlock_irqrestore(q->queue_lock, flags);
178 * Function: scsi_queue_insert()
180 * Purpose: Insert a command in the midlevel queue.
182 * Arguments: cmd - command that we are adding to queue.
183 * reason - why we are inserting command to queue.
185 * Lock status: Assumed that lock is not held upon entry.
189 * Notes: We do this for one of two cases. Either the host is busy
190 * and it cannot accept any more commands for the time being,
191 * or the device returned QUEUE_FULL and can accept no more
193 * Notes: This could be called either from an interrupt context or a
194 * normal process context.
196 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
198 __scsi_queue_insert(cmd, reason, 1);
201 * scsi_execute - insert request and wait for the result
204 * @data_direction: data direction
205 * @buffer: data buffer
206 * @bufflen: len of buffer
207 * @sense: optional sense buffer
208 * @timeout: request timeout in seconds
209 * @retries: number of times to retry request
210 * @flags: or into request flags;
211 * @resid: optional residual length
213 * returns the req->errors value which is the scsi_cmnd result
216 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
217 int data_direction, void *buffer, unsigned bufflen,
218 unsigned char *sense, int timeout, int retries, u64 flags,
222 int write = (data_direction == DMA_TO_DEVICE);
223 int ret = DRIVER_ERROR << 24;
225 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
228 blk_rq_set_block_pc(req);
230 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
231 buffer, bufflen, __GFP_RECLAIM))
234 req->cmd_len = COMMAND_SIZE(cmd[0]);
235 memcpy(req->cmd, cmd, req->cmd_len);
238 req->retries = retries;
239 req->timeout = timeout;
240 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
243 * head injection *required* here otherwise quiesce won't work
245 blk_execute_rq(req->q, NULL, req, 1);
248 * Some devices (USB mass-storage in particular) may transfer
249 * garbage data together with a residue indicating that the data
250 * is invalid. Prevent the garbage from being misinterpreted
251 * and prevent security leaks by zeroing out the excess data.
253 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
254 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
257 *resid = req->resid_len;
260 blk_put_request(req);
264 EXPORT_SYMBOL(scsi_execute);
266 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
267 int data_direction, void *buffer, unsigned bufflen,
268 struct scsi_sense_hdr *sshdr, int timeout, int retries,
269 int *resid, u64 flags)
275 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
277 return DRIVER_ERROR << 24;
279 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
280 sense, timeout, retries, flags, resid);
282 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
287 EXPORT_SYMBOL(scsi_execute_req_flags);
290 * Function: scsi_init_cmd_errh()
292 * Purpose: Initialize cmd fields related to error handling.
294 * Arguments: cmd - command that is ready to be queued.
296 * Notes: This function has the job of initializing a number of
297 * fields related to error handling. Typically this will
298 * be called once for each command, as required.
300 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
302 cmd->serial_number = 0;
303 scsi_set_resid(cmd, 0);
304 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
305 if (cmd->cmd_len == 0)
306 cmd->cmd_len = scsi_command_size(cmd->cmnd);
309 void scsi_device_unbusy(struct scsi_device *sdev)
311 struct Scsi_Host *shost = sdev->host;
312 struct scsi_target *starget = scsi_target(sdev);
315 atomic_dec(&shost->host_busy);
316 if (starget->can_queue > 0)
317 atomic_dec(&starget->target_busy);
319 if (unlikely(scsi_host_in_recovery(shost) &&
320 (shost->host_failed || shost->host_eh_scheduled))) {
321 spin_lock_irqsave(shost->host_lock, flags);
322 scsi_eh_wakeup(shost);
323 spin_unlock_irqrestore(shost->host_lock, flags);
326 atomic_dec(&sdev->device_busy);
329 static void scsi_kick_queue(struct request_queue *q)
332 blk_mq_start_hw_queues(q);
338 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
339 * and call blk_run_queue for all the scsi_devices on the target -
340 * including current_sdev first.
342 * Called with *no* scsi locks held.
344 static void scsi_single_lun_run(struct scsi_device *current_sdev)
346 struct Scsi_Host *shost = current_sdev->host;
347 struct scsi_device *sdev, *tmp;
348 struct scsi_target *starget = scsi_target(current_sdev);
351 spin_lock_irqsave(shost->host_lock, flags);
352 starget->starget_sdev_user = NULL;
353 spin_unlock_irqrestore(shost->host_lock, flags);
356 * Call blk_run_queue for all LUNs on the target, starting with
357 * current_sdev. We race with others (to set starget_sdev_user),
358 * but in most cases, we will be first. Ideally, each LU on the
359 * target would get some limited time or requests on the target.
361 scsi_kick_queue(current_sdev->request_queue);
363 spin_lock_irqsave(shost->host_lock, flags);
364 if (starget->starget_sdev_user)
366 list_for_each_entry_safe(sdev, tmp, &starget->devices,
367 same_target_siblings) {
368 if (sdev == current_sdev)
370 if (scsi_device_get(sdev))
373 spin_unlock_irqrestore(shost->host_lock, flags);
374 scsi_kick_queue(sdev->request_queue);
375 spin_lock_irqsave(shost->host_lock, flags);
377 scsi_device_put(sdev);
380 spin_unlock_irqrestore(shost->host_lock, flags);
383 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
385 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
387 if (atomic_read(&sdev->device_blocked) > 0)
392 static inline bool scsi_target_is_busy(struct scsi_target *starget)
394 if (starget->can_queue > 0) {
395 if (atomic_read(&starget->target_busy) >= starget->can_queue)
397 if (atomic_read(&starget->target_blocked) > 0)
403 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
405 if (shost->can_queue > 0 &&
406 atomic_read(&shost->host_busy) >= shost->can_queue)
408 if (atomic_read(&shost->host_blocked) > 0)
410 if (shost->host_self_blocked)
415 static void scsi_starved_list_run(struct Scsi_Host *shost)
417 LIST_HEAD(starved_list);
418 struct scsi_device *sdev;
421 spin_lock_irqsave(shost->host_lock, flags);
422 list_splice_init(&shost->starved_list, &starved_list);
424 while (!list_empty(&starved_list)) {
425 struct request_queue *slq;
428 * As long as shost is accepting commands and we have
429 * starved queues, call blk_run_queue. scsi_request_fn
430 * drops the queue_lock and can add us back to the
433 * host_lock protects the starved_list and starved_entry.
434 * scsi_request_fn must get the host_lock before checking
435 * or modifying starved_list or starved_entry.
437 if (scsi_host_is_busy(shost))
440 sdev = list_entry(starved_list.next,
441 struct scsi_device, starved_entry);
442 list_del_init(&sdev->starved_entry);
443 if (scsi_target_is_busy(scsi_target(sdev))) {
444 list_move_tail(&sdev->starved_entry,
445 &shost->starved_list);
450 * Once we drop the host lock, a racing scsi_remove_device()
451 * call may remove the sdev from the starved list and destroy
452 * it and the queue. Mitigate by taking a reference to the
453 * queue and never touching the sdev again after we drop the
454 * host lock. Note: if __scsi_remove_device() invokes
455 * blk_cleanup_queue() before the queue is run from this
456 * function then blk_run_queue() will return immediately since
457 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
459 slq = sdev->request_queue;
460 if (!blk_get_queue(slq))
462 spin_unlock_irqrestore(shost->host_lock, flags);
464 scsi_kick_queue(slq);
467 spin_lock_irqsave(shost->host_lock, flags);
469 /* put any unprocessed entries back */
470 list_splice(&starved_list, &shost->starved_list);
471 spin_unlock_irqrestore(shost->host_lock, flags);
475 * Function: scsi_run_queue()
477 * Purpose: Select a proper request queue to serve next
479 * Arguments: q - last request's queue
483 * Notes: The previous command was completely finished, start
484 * a new one if possible.
486 static void scsi_run_queue(struct request_queue *q)
488 struct scsi_device *sdev = q->queuedata;
490 if (scsi_target(sdev)->single_lun)
491 scsi_single_lun_run(sdev);
492 if (!list_empty(&sdev->host->starved_list))
493 scsi_starved_list_run(sdev->host);
496 blk_mq_start_stopped_hw_queues(q, false);
501 void scsi_requeue_run_queue(struct work_struct *work)
503 struct scsi_device *sdev;
504 struct request_queue *q;
506 sdev = container_of(work, struct scsi_device, requeue_work);
507 q = sdev->request_queue;
512 * Function: scsi_requeue_command()
514 * Purpose: Handle post-processing of completed commands.
516 * Arguments: q - queue to operate on
517 * cmd - command that may need to be requeued.
521 * Notes: After command completion, there may be blocks left
522 * over which weren't finished by the previous command
523 * this can be for a number of reasons - the main one is
524 * I/O errors in the middle of the request, in which case
525 * we need to request the blocks that come after the bad
527 * Notes: Upon return, cmd is a stale pointer.
529 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
531 struct scsi_device *sdev = cmd->device;
532 struct request *req = cmd->request;
535 spin_lock_irqsave(q->queue_lock, flags);
536 blk_unprep_request(req);
538 scsi_put_command(cmd);
539 blk_requeue_request(q, req);
540 spin_unlock_irqrestore(q->queue_lock, flags);
544 put_device(&sdev->sdev_gendev);
547 void scsi_run_host_queues(struct Scsi_Host *shost)
549 struct scsi_device *sdev;
551 shost_for_each_device(sdev, shost)
552 scsi_run_queue(sdev->request_queue);
555 static inline unsigned int scsi_sgtable_index(unsigned short nents)
559 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
564 index = get_count_order(nents) - 3;
569 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
571 struct scsi_host_sg_pool *sgp;
573 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
574 mempool_free(sgl, sgp->pool);
577 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
579 struct scsi_host_sg_pool *sgp;
581 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
582 return mempool_alloc(sgp->pool, gfp_mask);
585 static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
587 if (mq && sdb->table.orig_nents <= SCSI_MAX_SG_SEGMENTS)
589 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
592 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, bool mq)
594 struct scatterlist *first_chunk = NULL;
600 if (nents <= SCSI_MAX_SG_SEGMENTS) {
601 sdb->table.nents = sdb->table.orig_nents = nents;
602 sg_init_table(sdb->table.sgl, nents);
605 first_chunk = sdb->table.sgl;
608 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
609 first_chunk, GFP_ATOMIC, scsi_sg_alloc);
611 scsi_free_sgtable(sdb, mq);
615 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
617 if (cmd->request->cmd_type == REQ_TYPE_FS) {
618 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
620 if (drv->uninit_command)
621 drv->uninit_command(cmd);
625 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
627 if (cmd->sdb.table.nents)
628 scsi_free_sgtable(&cmd->sdb, true);
629 if (cmd->request->next_rq && cmd->request->next_rq->special)
630 scsi_free_sgtable(cmd->request->next_rq->special, true);
631 if (scsi_prot_sg_count(cmd))
632 scsi_free_sgtable(cmd->prot_sdb, true);
635 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
637 struct scsi_device *sdev = cmd->device;
638 struct Scsi_Host *shost = sdev->host;
641 scsi_mq_free_sgtables(cmd);
642 scsi_uninit_cmd(cmd);
644 if (shost->use_cmd_list) {
645 BUG_ON(list_empty(&cmd->list));
646 spin_lock_irqsave(&sdev->list_lock, flags);
647 list_del_init(&cmd->list);
648 spin_unlock_irqrestore(&sdev->list_lock, flags);
653 * Function: scsi_release_buffers()
655 * Purpose: Free resources allocate for a scsi_command.
657 * Arguments: cmd - command that we are bailing.
659 * Lock status: Assumed that no lock is held upon entry.
663 * Notes: In the event that an upper level driver rejects a
664 * command, we must release resources allocated during
665 * the __init_io() function. Primarily this would involve
666 * the scatter-gather table.
668 static void scsi_release_buffers(struct scsi_cmnd *cmd)
670 if (cmd->sdb.table.nents)
671 scsi_free_sgtable(&cmd->sdb, false);
673 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
675 if (scsi_prot_sg_count(cmd))
676 scsi_free_sgtable(cmd->prot_sdb, false);
679 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
681 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
683 scsi_free_sgtable(bidi_sdb, false);
684 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
685 cmd->request->next_rq->special = NULL;
688 static bool scsi_end_request(struct request *req, int error,
689 unsigned int bytes, unsigned int bidi_bytes)
691 struct scsi_cmnd *cmd = req->special;
692 struct scsi_device *sdev = cmd->device;
693 struct request_queue *q = sdev->request_queue;
695 if (blk_update_request(req, error, bytes))
698 /* Bidi request must be completed as a whole */
699 if (unlikely(bidi_bytes) &&
700 blk_update_request(req->next_rq, error, bidi_bytes))
703 if (blk_queue_add_random(q))
704 add_disk_randomness(req->rq_disk);
708 * In the MQ case the command gets freed by __blk_mq_end_request,
709 * so we have to do all cleanup that depends on it earlier.
711 * We also can't kick the queues from irq context, so we
712 * will have to defer it to a workqueue.
714 scsi_mq_uninit_cmd(cmd);
716 __blk_mq_end_request(req, error);
718 if (scsi_target(sdev)->single_lun ||
719 !list_empty(&sdev->host->starved_list))
720 kblockd_schedule_work(&sdev->requeue_work);
722 blk_mq_start_stopped_hw_queues(q, true);
727 scsi_release_bidi_buffers(cmd);
729 spin_lock_irqsave(q->queue_lock, flags);
730 blk_finish_request(req, error);
731 spin_unlock_irqrestore(q->queue_lock, flags);
733 scsi_release_buffers(cmd);
735 scsi_put_command(cmd);
739 put_device(&sdev->sdev_gendev);
744 * __scsi_error_from_host_byte - translate SCSI error code into errno
745 * @cmd: SCSI command (unused)
746 * @result: scsi error code
748 * Translate SCSI error code into standard UNIX errno.
750 * -ENOLINK temporary transport failure
751 * -EREMOTEIO permanent target failure, do not retry
752 * -EBADE permanent nexus failure, retry on other path
753 * -ENOSPC No write space available
754 * -ENODATA Medium error
755 * -EIO unspecified I/O error
757 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
761 switch(host_byte(result)) {
762 case DID_TRANSPORT_FAILFAST:
765 case DID_TARGET_FAILURE:
766 set_host_byte(cmd, DID_OK);
769 case DID_NEXUS_FAILURE:
770 set_host_byte(cmd, DID_OK);
773 case DID_ALLOC_FAILURE:
774 set_host_byte(cmd, DID_OK);
777 case DID_MEDIUM_ERROR:
778 set_host_byte(cmd, DID_OK);
790 * Function: scsi_io_completion()
792 * Purpose: Completion processing for block device I/O requests.
794 * Arguments: cmd - command that is finished.
796 * Lock status: Assumed that no lock is held upon entry.
800 * Notes: We will finish off the specified number of sectors. If we
801 * are done, the command block will be released and the queue
802 * function will be goosed. If we are not done then we have to
803 * figure out what to do next:
805 * a) We can call scsi_requeue_command(). The request
806 * will be unprepared and put back on the queue. Then
807 * a new command will be created for it. This should
808 * be used if we made forward progress, or if we want
809 * to switch from READ(10) to READ(6) for example.
811 * b) We can call __scsi_queue_insert(). The request will
812 * be put back on the queue and retried using the same
813 * command as before, possibly after a delay.
815 * c) We can call scsi_end_request() with -EIO to fail
816 * the remainder of the request.
818 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
820 int result = cmd->result;
821 struct request_queue *q = cmd->device->request_queue;
822 struct request *req = cmd->request;
824 struct scsi_sense_hdr sshdr;
825 bool sense_valid = false;
826 int sense_deferred = 0, level = 0;
827 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
828 ACTION_DELAYED_RETRY} action;
829 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
832 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
834 sense_deferred = scsi_sense_is_deferred(&sshdr);
837 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
839 if (sense_valid && req->sense) {
841 * SG_IO wants current and deferred errors
843 int len = 8 + cmd->sense_buffer[7];
845 if (len > SCSI_SENSE_BUFFERSIZE)
846 len = SCSI_SENSE_BUFFERSIZE;
847 memcpy(req->sense, cmd->sense_buffer, len);
848 req->sense_len = len;
851 error = __scsi_error_from_host_byte(cmd, result);
854 * __scsi_error_from_host_byte may have reset the host_byte
856 req->errors = cmd->result;
858 req->resid_len = scsi_get_resid(cmd);
860 if (scsi_bidi_cmnd(cmd)) {
862 * Bidi commands Must be complete as a whole,
863 * both sides at once.
865 req->next_rq->resid_len = scsi_in(cmd)->resid;
866 if (scsi_end_request(req, 0, blk_rq_bytes(req),
867 blk_rq_bytes(req->next_rq)))
871 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
873 * Certain non BLOCK_PC requests are commands that don't
874 * actually transfer anything (FLUSH), so cannot use
875 * good_bytes != blk_rq_bytes(req) as the signal for an error.
876 * This sets the error explicitly for the problem case.
878 error = __scsi_error_from_host_byte(cmd, result);
881 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
882 BUG_ON(blk_bidi_rq(req));
885 * Next deal with any sectors which we were able to correctly
888 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
889 "%u sectors total, %d bytes done.\n",
890 blk_rq_sectors(req), good_bytes));
893 * Recovered errors need reporting, but they're always treated
894 * as success, so fiddle the result code here. For BLOCK_PC
895 * we already took a copy of the original into rq->errors which
896 * is what gets returned to the user
898 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
899 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
900 * print since caller wants ATA registers. Only occurs on
901 * SCSI ATA PASS_THROUGH commands when CK_COND=1
903 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
905 else if (!(req->cmd_flags & REQ_QUIET))
906 scsi_print_sense(cmd);
908 /* BLOCK_PC may have set error */
913 * special case: failed zero length commands always need to
914 * drop down into the retry code. Otherwise, if we finished
915 * all bytes in the request we are done now.
917 if (!(blk_rq_bytes(req) == 0 && error) &&
918 !scsi_end_request(req, error, good_bytes, 0))
922 * Kill remainder if no retrys.
924 if (error && scsi_noretry_cmd(cmd)) {
925 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
931 * If there had been no error, but we have leftover bytes in the
932 * requeues just queue the command up again.
937 error = __scsi_error_from_host_byte(cmd, result);
939 if (host_byte(result) == DID_RESET) {
940 /* Third party bus reset or reset for error recovery
941 * reasons. Just retry the command and see what
944 action = ACTION_RETRY;
945 } else if (sense_valid && !sense_deferred) {
946 switch (sshdr.sense_key) {
948 if (cmd->device->removable) {
949 /* Detected disc change. Set a bit
950 * and quietly refuse further access.
952 cmd->device->changed = 1;
953 action = ACTION_FAIL;
955 /* Must have been a power glitch, or a
956 * bus reset. Could not have been a
957 * media change, so we just retry the
958 * command and see what happens.
960 action = ACTION_RETRY;
963 case ILLEGAL_REQUEST:
964 /* If we had an ILLEGAL REQUEST returned, then
965 * we may have performed an unsupported
966 * command. The only thing this should be
967 * would be a ten byte read where only a six
968 * byte read was supported. Also, on a system
969 * where READ CAPACITY failed, we may have
970 * read past the end of the disk.
972 if ((cmd->device->use_10_for_rw &&
973 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
974 (cmd->cmnd[0] == READ_10 ||
975 cmd->cmnd[0] == WRITE_10)) {
976 /* This will issue a new 6-byte command. */
977 cmd->device->use_10_for_rw = 0;
978 action = ACTION_REPREP;
979 } else if (sshdr.asc == 0x10) /* DIX */ {
980 action = ACTION_FAIL;
982 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
983 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
984 action = ACTION_FAIL;
987 action = ACTION_FAIL;
989 case ABORTED_COMMAND:
990 action = ACTION_FAIL;
991 if (sshdr.asc == 0x10) /* DIF */
995 /* If the device is in the process of becoming
996 * ready, or has a temporary blockage, retry.
998 if (sshdr.asc == 0x04) {
999 switch (sshdr.ascq) {
1000 case 0x01: /* becoming ready */
1001 case 0x04: /* format in progress */
1002 case 0x05: /* rebuild in progress */
1003 case 0x06: /* recalculation in progress */
1004 case 0x07: /* operation in progress */
1005 case 0x08: /* Long write in progress */
1006 case 0x09: /* self test in progress */
1007 case 0x14: /* space allocation in progress */
1008 action = ACTION_DELAYED_RETRY;
1011 action = ACTION_FAIL;
1015 action = ACTION_FAIL;
1017 case VOLUME_OVERFLOW:
1018 /* See SSC3rXX or current. */
1019 action = ACTION_FAIL;
1022 action = ACTION_FAIL;
1026 action = ACTION_FAIL;
1028 if (action != ACTION_FAIL &&
1029 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1030 action = ACTION_FAIL;
1034 /* Give up and fail the remainder of the request */
1035 if (!(req->cmd_flags & REQ_QUIET)) {
1036 static DEFINE_RATELIMIT_STATE(_rs,
1037 DEFAULT_RATELIMIT_INTERVAL,
1038 DEFAULT_RATELIMIT_BURST);
1040 if (unlikely(scsi_logging_level))
1041 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1042 SCSI_LOG_MLCOMPLETE_BITS);
1045 * if logging is enabled the failure will be printed
1046 * in scsi_log_completion(), so avoid duplicate messages
1048 if (!level && __ratelimit(&_rs)) {
1049 scsi_print_result(cmd, NULL, FAILED);
1050 if (driver_byte(result) & DRIVER_SENSE)
1051 scsi_print_sense(cmd);
1052 scsi_print_command(cmd);
1055 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1060 /* Unprep the request and put it back at the head of the queue.
1061 * A new command will be prepared and issued.
1064 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1065 scsi_mq_uninit_cmd(cmd);
1066 scsi_mq_requeue_cmd(cmd);
1068 scsi_release_buffers(cmd);
1069 scsi_requeue_command(q, cmd);
1073 /* Retry the same command immediately */
1074 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1076 case ACTION_DELAYED_RETRY:
1077 /* Retry the same command after a delay */
1078 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1083 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1088 * If sg table allocation fails, requeue request later.
1090 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1091 req->mq_ctx != NULL)))
1092 return BLKPREP_DEFER;
1095 * Next, walk the list, and fill in the addresses and sizes of
1098 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1099 BUG_ON(count > sdb->table.nents);
1100 sdb->table.nents = count;
1101 sdb->length = blk_rq_bytes(req);
1106 * Function: scsi_init_io()
1108 * Purpose: SCSI I/O initialize function.
1110 * Arguments: cmd - Command descriptor we wish to initialize
1112 * Returns: 0 on success
1113 * BLKPREP_DEFER if the failure is retryable
1114 * BLKPREP_KILL if the failure is fatal
1116 int scsi_init_io(struct scsi_cmnd *cmd)
1118 struct scsi_device *sdev = cmd->device;
1119 struct request *rq = cmd->request;
1120 bool is_mq = (rq->mq_ctx != NULL);
1123 BUG_ON(!rq->nr_phys_segments);
1125 error = scsi_init_sgtable(rq, &cmd->sdb);
1129 if (blk_bidi_rq(rq)) {
1130 if (!rq->q->mq_ops) {
1131 struct scsi_data_buffer *bidi_sdb =
1132 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1134 error = BLKPREP_DEFER;
1138 rq->next_rq->special = bidi_sdb;
1141 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1146 if (blk_integrity_rq(rq)) {
1147 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1150 if (prot_sdb == NULL) {
1152 * This can happen if someone (e.g. multipath)
1153 * queues a command to a device on an adapter
1154 * that does not support DIX.
1157 error = BLKPREP_KILL;
1161 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1163 if (scsi_alloc_sgtable(prot_sdb, ivecs, is_mq)) {
1164 error = BLKPREP_DEFER;
1168 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1169 prot_sdb->table.sgl);
1170 BUG_ON(unlikely(count > ivecs));
1171 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1173 cmd->prot_sdb = prot_sdb;
1174 cmd->prot_sdb->table.nents = count;
1180 scsi_mq_free_sgtables(cmd);
1182 scsi_release_buffers(cmd);
1183 cmd->request->special = NULL;
1184 scsi_put_command(cmd);
1185 put_device(&sdev->sdev_gendev);
1189 EXPORT_SYMBOL(scsi_init_io);
1191 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1192 struct request *req)
1194 struct scsi_cmnd *cmd;
1196 if (!req->special) {
1197 /* Bail if we can't get a reference to the device */
1198 if (!get_device(&sdev->sdev_gendev))
1201 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1202 if (unlikely(!cmd)) {
1203 put_device(&sdev->sdev_gendev);
1211 /* pull a tag out of the request if we have one */
1212 cmd->tag = req->tag;
1215 cmd->cmnd = req->cmd;
1216 cmd->prot_op = SCSI_PROT_NORMAL;
1221 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1223 struct scsi_cmnd *cmd = req->special;
1226 * BLOCK_PC requests may transfer data, in which case they must
1227 * a bio attached to them. Or they might contain a SCSI command
1228 * that does not transfer data, in which case they may optionally
1229 * submit a request without an attached bio.
1232 int ret = scsi_init_io(cmd);
1236 BUG_ON(blk_rq_bytes(req));
1238 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1241 cmd->cmd_len = req->cmd_len;
1242 cmd->transfersize = blk_rq_bytes(req);
1243 cmd->allowed = req->retries;
1248 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1249 * that still need to be translated to SCSI CDBs from the ULD.
1251 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1253 struct scsi_cmnd *cmd = req->special;
1255 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1256 int ret = sdev->handler->prep_fn(sdev, req);
1257 if (ret != BLKPREP_OK)
1261 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1262 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1265 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1267 struct scsi_cmnd *cmd = req->special;
1269 if (!blk_rq_bytes(req))
1270 cmd->sc_data_direction = DMA_NONE;
1271 else if (rq_data_dir(req) == WRITE)
1272 cmd->sc_data_direction = DMA_TO_DEVICE;
1274 cmd->sc_data_direction = DMA_FROM_DEVICE;
1276 switch (req->cmd_type) {
1278 return scsi_setup_fs_cmnd(sdev, req);
1279 case REQ_TYPE_BLOCK_PC:
1280 return scsi_setup_blk_pc_cmnd(sdev, req);
1282 return BLKPREP_KILL;
1287 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1289 int ret = BLKPREP_OK;
1292 * If the device is not in running state we will reject some
1295 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1296 switch (sdev->sdev_state) {
1298 case SDEV_TRANSPORT_OFFLINE:
1300 * If the device is offline we refuse to process any
1301 * commands. The device must be brought online
1302 * before trying any recovery commands.
1304 sdev_printk(KERN_ERR, sdev,
1305 "rejecting I/O to offline device\n");
1310 * If the device is fully deleted, we refuse to
1311 * process any commands as well.
1313 sdev_printk(KERN_ERR, sdev,
1314 "rejecting I/O to dead device\n");
1318 case SDEV_CREATED_BLOCK:
1319 ret = BLKPREP_DEFER;
1323 * If the devices is blocked we defer normal commands.
1325 if (!(req->cmd_flags & REQ_PREEMPT))
1326 ret = BLKPREP_DEFER;
1330 * For any other not fully online state we only allow
1331 * special commands. In particular any user initiated
1332 * command is not allowed.
1334 if (!(req->cmd_flags & REQ_PREEMPT))
1343 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1345 struct scsi_device *sdev = q->queuedata;
1349 req->errors = DID_NO_CONNECT << 16;
1350 /* release the command and kill it */
1352 struct scsi_cmnd *cmd = req->special;
1353 scsi_release_buffers(cmd);
1354 scsi_put_command(cmd);
1355 put_device(&sdev->sdev_gendev);
1356 req->special = NULL;
1361 * If we defer, the blk_peek_request() returns NULL, but the
1362 * queue must be restarted, so we schedule a callback to happen
1365 if (atomic_read(&sdev->device_busy) == 0)
1366 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1369 req->cmd_flags |= REQ_DONTPREP;
1375 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1377 struct scsi_device *sdev = q->queuedata;
1378 struct scsi_cmnd *cmd;
1381 ret = scsi_prep_state_check(sdev, req);
1382 if (ret != BLKPREP_OK)
1385 cmd = scsi_get_cmd_from_req(sdev, req);
1386 if (unlikely(!cmd)) {
1387 ret = BLKPREP_DEFER;
1391 ret = scsi_setup_cmnd(sdev, req);
1393 return scsi_prep_return(q, req, ret);
1396 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1398 scsi_uninit_cmd(req->special);
1402 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1405 * Called with the queue_lock held.
1407 static inline int scsi_dev_queue_ready(struct request_queue *q,
1408 struct scsi_device *sdev)
1412 busy = atomic_inc_return(&sdev->device_busy) - 1;
1413 if (atomic_read(&sdev->device_blocked)) {
1418 * unblock after device_blocked iterates to zero
1420 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1422 * For the MQ case we take care of this in the caller.
1425 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1428 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1429 "unblocking device at zero depth\n"));
1432 if (busy >= sdev->queue_depth)
1437 atomic_dec(&sdev->device_busy);
1442 * scsi_target_queue_ready: checks if there we can send commands to target
1443 * @sdev: scsi device on starget to check.
1445 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1446 struct scsi_device *sdev)
1448 struct scsi_target *starget = scsi_target(sdev);
1451 if (starget->single_lun) {
1452 spin_lock_irq(shost->host_lock);
1453 if (starget->starget_sdev_user &&
1454 starget->starget_sdev_user != sdev) {
1455 spin_unlock_irq(shost->host_lock);
1458 starget->starget_sdev_user = sdev;
1459 spin_unlock_irq(shost->host_lock);
1462 if (starget->can_queue <= 0)
1465 busy = atomic_inc_return(&starget->target_busy) - 1;
1466 if (atomic_read(&starget->target_blocked) > 0) {
1471 * unblock after target_blocked iterates to zero
1473 if (atomic_dec_return(&starget->target_blocked) > 0)
1476 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1477 "unblocking target at zero depth\n"));
1480 if (busy >= starget->can_queue)
1486 spin_lock_irq(shost->host_lock);
1487 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1488 spin_unlock_irq(shost->host_lock);
1490 if (starget->can_queue > 0)
1491 atomic_dec(&starget->target_busy);
1496 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1497 * return 0. We must end up running the queue again whenever 0 is
1498 * returned, else IO can hang.
1500 static inline int scsi_host_queue_ready(struct request_queue *q,
1501 struct Scsi_Host *shost,
1502 struct scsi_device *sdev)
1506 if (scsi_host_in_recovery(shost))
1509 busy = atomic_inc_return(&shost->host_busy) - 1;
1510 if (atomic_read(&shost->host_blocked) > 0) {
1515 * unblock after host_blocked iterates to zero
1517 if (atomic_dec_return(&shost->host_blocked) > 0)
1521 shost_printk(KERN_INFO, shost,
1522 "unblocking host at zero depth\n"));
1525 if (shost->can_queue > 0 && busy >= shost->can_queue)
1527 if (shost->host_self_blocked)
1530 /* We're OK to process the command, so we can't be starved */
1531 if (!list_empty(&sdev->starved_entry)) {
1532 spin_lock_irq(shost->host_lock);
1533 if (!list_empty(&sdev->starved_entry))
1534 list_del_init(&sdev->starved_entry);
1535 spin_unlock_irq(shost->host_lock);
1541 spin_lock_irq(shost->host_lock);
1542 if (list_empty(&sdev->starved_entry))
1543 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1544 spin_unlock_irq(shost->host_lock);
1546 atomic_dec(&shost->host_busy);
1551 * Busy state exporting function for request stacking drivers.
1553 * For efficiency, no lock is taken to check the busy state of
1554 * shost/starget/sdev, since the returned value is not guaranteed and
1555 * may be changed after request stacking drivers call the function,
1556 * regardless of taking lock or not.
1558 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1559 * needs to return 'not busy'. Otherwise, request stacking drivers
1560 * may hold requests forever.
1562 static int scsi_lld_busy(struct request_queue *q)
1564 struct scsi_device *sdev = q->queuedata;
1565 struct Scsi_Host *shost;
1567 if (blk_queue_dying(q))
1573 * Ignore host/starget busy state.
1574 * Since block layer does not have a concept of fairness across
1575 * multiple queues, congestion of host/starget needs to be handled
1578 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1585 * Kill a request for a dead device
1587 static void scsi_kill_request(struct request *req, struct request_queue *q)
1589 struct scsi_cmnd *cmd = req->special;
1590 struct scsi_device *sdev;
1591 struct scsi_target *starget;
1592 struct Scsi_Host *shost;
1594 blk_start_request(req);
1596 scmd_printk(KERN_INFO, cmd, "killing request\n");
1599 starget = scsi_target(sdev);
1601 scsi_init_cmd_errh(cmd);
1602 cmd->result = DID_NO_CONNECT << 16;
1603 atomic_inc(&cmd->device->iorequest_cnt);
1606 * SCSI request completion path will do scsi_device_unbusy(),
1607 * bump busy counts. To bump the counters, we need to dance
1608 * with the locks as normal issue path does.
1610 atomic_inc(&sdev->device_busy);
1611 atomic_inc(&shost->host_busy);
1612 if (starget->can_queue > 0)
1613 atomic_inc(&starget->target_busy);
1615 blk_complete_request(req);
1618 static void scsi_softirq_done(struct request *rq)
1620 struct scsi_cmnd *cmd = rq->special;
1621 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1624 INIT_LIST_HEAD(&cmd->eh_entry);
1626 atomic_inc(&cmd->device->iodone_cnt);
1628 atomic_inc(&cmd->device->ioerr_cnt);
1630 disposition = scsi_decide_disposition(cmd);
1631 if (disposition != SUCCESS &&
1632 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1633 sdev_printk(KERN_ERR, cmd->device,
1634 "timing out command, waited %lus\n",
1636 disposition = SUCCESS;
1639 scsi_log_completion(cmd, disposition);
1641 switch (disposition) {
1643 scsi_finish_command(cmd);
1646 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1648 case ADD_TO_MLQUEUE:
1649 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1652 if (!scsi_eh_scmd_add(cmd, 0))
1653 scsi_finish_command(cmd);
1658 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1659 * @cmd: command block we are dispatching.
1661 * Return: nonzero return request was rejected and device's queue needs to be
1664 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1666 struct Scsi_Host *host = cmd->device->host;
1669 atomic_inc(&cmd->device->iorequest_cnt);
1671 /* check if the device is still usable */
1672 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1673 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1674 * returns an immediate error upwards, and signals
1675 * that the device is no longer present */
1676 cmd->result = DID_NO_CONNECT << 16;
1680 /* Check to see if the scsi lld made this device blocked. */
1681 if (unlikely(scsi_device_blocked(cmd->device))) {
1683 * in blocked state, the command is just put back on
1684 * the device queue. The suspend state has already
1685 * blocked the queue so future requests should not
1686 * occur until the device transitions out of the
1689 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1690 "queuecommand : device blocked\n"));
1691 return SCSI_MLQUEUE_DEVICE_BUSY;
1694 /* Store the LUN value in cmnd, if needed. */
1695 if (cmd->device->lun_in_cdb)
1696 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1697 (cmd->device->lun << 5 & 0xe0);
1702 * Before we queue this command, check if the command
1703 * length exceeds what the host adapter can handle.
1705 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1706 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1707 "queuecommand : command too long. "
1708 "cdb_size=%d host->max_cmd_len=%d\n",
1709 cmd->cmd_len, cmd->device->host->max_cmd_len));
1710 cmd->result = (DID_ABORT << 16);
1714 if (unlikely(host->shost_state == SHOST_DEL)) {
1715 cmd->result = (DID_NO_CONNECT << 16);
1720 trace_scsi_dispatch_cmd_start(cmd);
1721 rtn = host->hostt->queuecommand(host, cmd);
1723 trace_scsi_dispatch_cmd_error(cmd, rtn);
1724 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1725 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1726 rtn = SCSI_MLQUEUE_HOST_BUSY;
1728 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1729 "queuecommand : request rejected\n"));
1734 cmd->scsi_done(cmd);
1739 * scsi_done - Invoke completion on finished SCSI command.
1740 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1741 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1743 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1744 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1745 * calls blk_complete_request() for further processing.
1747 * This function is interrupt context safe.
1749 static void scsi_done(struct scsi_cmnd *cmd)
1751 trace_scsi_dispatch_cmd_done(cmd);
1752 blk_complete_request(cmd->request);
1756 * Function: scsi_request_fn()
1758 * Purpose: Main strategy routine for SCSI.
1760 * Arguments: q - Pointer to actual queue.
1764 * Lock status: IO request lock assumed to be held when called.
1766 static void scsi_request_fn(struct request_queue *q)
1767 __releases(q->queue_lock)
1768 __acquires(q->queue_lock)
1770 struct scsi_device *sdev = q->queuedata;
1771 struct Scsi_Host *shost;
1772 struct scsi_cmnd *cmd;
1773 struct request *req;
1776 * To start with, we keep looping until the queue is empty, or until
1777 * the host is no longer able to accept any more requests.
1783 * get next queueable request. We do this early to make sure
1784 * that the request is fully prepared even if we cannot
1787 req = blk_peek_request(q);
1791 if (unlikely(!scsi_device_online(sdev))) {
1792 sdev_printk(KERN_ERR, sdev,
1793 "rejecting I/O to offline device\n");
1794 scsi_kill_request(req, q);
1798 if (!scsi_dev_queue_ready(q, sdev))
1802 * Remove the request from the request list.
1804 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1805 blk_start_request(req);
1807 spin_unlock_irq(q->queue_lock);
1809 if (unlikely(cmd == NULL)) {
1810 printk(KERN_CRIT "impossible request in %s.\n"
1811 "please mail a stack trace to "
1812 "linux-scsi@vger.kernel.org\n",
1814 blk_dump_rq_flags(req, "foo");
1819 * We hit this when the driver is using a host wide
1820 * tag map. For device level tag maps the queue_depth check
1821 * in the device ready fn would prevent us from trying
1822 * to allocate a tag. Since the map is a shared host resource
1823 * we add the dev to the starved list so it eventually gets
1824 * a run when a tag is freed.
1826 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1827 spin_lock_irq(shost->host_lock);
1828 if (list_empty(&sdev->starved_entry))
1829 list_add_tail(&sdev->starved_entry,
1830 &shost->starved_list);
1831 spin_unlock_irq(shost->host_lock);
1835 if (!scsi_target_queue_ready(shost, sdev))
1838 if (!scsi_host_queue_ready(q, shost, sdev))
1839 goto host_not_ready;
1841 if (sdev->simple_tags)
1842 cmd->flags |= SCMD_TAGGED;
1844 cmd->flags &= ~SCMD_TAGGED;
1847 * Finally, initialize any error handling parameters, and set up
1848 * the timers for timeouts.
1850 scsi_init_cmd_errh(cmd);
1853 * Dispatch the command to the low-level driver.
1855 cmd->scsi_done = scsi_done;
1856 rtn = scsi_dispatch_cmd(cmd);
1858 scsi_queue_insert(cmd, rtn);
1859 spin_lock_irq(q->queue_lock);
1862 spin_lock_irq(q->queue_lock);
1868 if (scsi_target(sdev)->can_queue > 0)
1869 atomic_dec(&scsi_target(sdev)->target_busy);
1872 * lock q, handle tag, requeue req, and decrement device_busy. We
1873 * must return with queue_lock held.
1875 * Decrementing device_busy without checking it is OK, as all such
1876 * cases (host limits or settings) should run the queue at some
1879 spin_lock_irq(q->queue_lock);
1880 blk_requeue_request(q, req);
1881 atomic_dec(&sdev->device_busy);
1883 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1884 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1887 static inline int prep_to_mq(int ret)
1893 return BLK_MQ_RQ_QUEUE_BUSY;
1895 return BLK_MQ_RQ_QUEUE_ERROR;
1899 static int scsi_mq_prep_fn(struct request *req)
1901 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1902 struct scsi_device *sdev = req->q->queuedata;
1903 struct Scsi_Host *shost = sdev->host;
1904 unsigned char *sense_buf = cmd->sense_buffer;
1905 struct scatterlist *sg;
1907 memset(cmd, 0, sizeof(struct scsi_cmnd));
1913 cmd->sense_buffer = sense_buf;
1915 cmd->tag = req->tag;
1917 cmd->cmnd = req->cmd;
1918 cmd->prot_op = SCSI_PROT_NORMAL;
1920 INIT_LIST_HEAD(&cmd->list);
1921 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1922 cmd->jiffies_at_alloc = jiffies;
1924 if (shost->use_cmd_list) {
1925 spin_lock_irq(&sdev->list_lock);
1926 list_add_tail(&cmd->list, &sdev->cmd_list);
1927 spin_unlock_irq(&sdev->list_lock);
1930 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1931 cmd->sdb.table.sgl = sg;
1933 if (scsi_host_get_prot(shost)) {
1934 cmd->prot_sdb = (void *)sg +
1936 shost->sg_tablesize, SCSI_MAX_SG_SEGMENTS) *
1937 sizeof(struct scatterlist);
1938 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1940 cmd->prot_sdb->table.sgl =
1941 (struct scatterlist *)(cmd->prot_sdb + 1);
1944 if (blk_bidi_rq(req)) {
1945 struct request *next_rq = req->next_rq;
1946 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1948 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1949 bidi_sdb->table.sgl =
1950 (struct scatterlist *)(bidi_sdb + 1);
1952 next_rq->special = bidi_sdb;
1955 blk_mq_start_request(req);
1957 return scsi_setup_cmnd(sdev, req);
1960 static void scsi_mq_done(struct scsi_cmnd *cmd)
1962 trace_scsi_dispatch_cmd_done(cmd);
1963 blk_mq_complete_request(cmd->request, cmd->request->errors);
1966 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1967 const struct blk_mq_queue_data *bd)
1969 struct request *req = bd->rq;
1970 struct request_queue *q = req->q;
1971 struct scsi_device *sdev = q->queuedata;
1972 struct Scsi_Host *shost = sdev->host;
1973 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1977 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1981 ret = BLK_MQ_RQ_QUEUE_BUSY;
1982 if (!get_device(&sdev->sdev_gendev))
1985 if (!scsi_dev_queue_ready(q, sdev))
1986 goto out_put_device;
1987 if (!scsi_target_queue_ready(shost, sdev))
1988 goto out_dec_device_busy;
1989 if (!scsi_host_queue_ready(q, shost, sdev))
1990 goto out_dec_target_busy;
1993 if (!(req->cmd_flags & REQ_DONTPREP)) {
1994 ret = prep_to_mq(scsi_mq_prep_fn(req));
1996 goto out_dec_host_busy;
1997 req->cmd_flags |= REQ_DONTPREP;
1999 blk_mq_start_request(req);
2002 if (sdev->simple_tags)
2003 cmd->flags |= SCMD_TAGGED;
2005 cmd->flags &= ~SCMD_TAGGED;
2007 scsi_init_cmd_errh(cmd);
2008 cmd->scsi_done = scsi_mq_done;
2010 reason = scsi_dispatch_cmd(cmd);
2012 scsi_set_blocked(cmd, reason);
2013 ret = BLK_MQ_RQ_QUEUE_BUSY;
2014 goto out_dec_host_busy;
2017 return BLK_MQ_RQ_QUEUE_OK;
2020 atomic_dec(&shost->host_busy);
2021 out_dec_target_busy:
2022 if (scsi_target(sdev)->can_queue > 0)
2023 atomic_dec(&scsi_target(sdev)->target_busy);
2024 out_dec_device_busy:
2025 atomic_dec(&sdev->device_busy);
2027 put_device(&sdev->sdev_gendev);
2030 case BLK_MQ_RQ_QUEUE_BUSY:
2031 blk_mq_stop_hw_queue(hctx);
2032 if (atomic_read(&sdev->device_busy) == 0 &&
2033 !scsi_device_blocked(sdev))
2034 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2036 case BLK_MQ_RQ_QUEUE_ERROR:
2038 * Make sure to release all allocated ressources when
2039 * we hit an error, as we will never see this command
2042 if (req->cmd_flags & REQ_DONTPREP)
2043 scsi_mq_uninit_cmd(cmd);
2051 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2055 return BLK_EH_RESET_TIMER;
2056 return scsi_times_out(req);
2059 static int scsi_init_request(void *data, struct request *rq,
2060 unsigned int hctx_idx, unsigned int request_idx,
2061 unsigned int numa_node)
2063 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2065 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
2067 if (!cmd->sense_buffer)
2072 static void scsi_exit_request(void *data, struct request *rq,
2073 unsigned int hctx_idx, unsigned int request_idx)
2075 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2077 kfree(cmd->sense_buffer);
2080 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2082 struct device *host_dev;
2083 u64 bounce_limit = 0xffffffff;
2085 if (shost->unchecked_isa_dma)
2086 return BLK_BOUNCE_ISA;
2088 * Platforms with virtual-DMA translation
2089 * hardware have no practical limit.
2091 if (!PCI_DMA_BUS_IS_PHYS)
2092 return BLK_BOUNCE_ANY;
2094 host_dev = scsi_get_device(shost);
2095 if (host_dev && host_dev->dma_mask)
2096 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2098 return bounce_limit;
2101 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2103 struct device *dev = shost->dma_dev;
2106 * this limit is imposed by hardware restrictions
2108 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2109 SCSI_MAX_SG_CHAIN_SEGMENTS));
2111 if (scsi_host_prot_dma(shost)) {
2112 shost->sg_prot_tablesize =
2113 min_not_zero(shost->sg_prot_tablesize,
2114 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2115 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2116 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2119 blk_queue_max_hw_sectors(q, shost->max_sectors);
2120 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2121 blk_queue_segment_boundary(q, shost->dma_boundary);
2122 dma_set_seg_boundary(dev, shost->dma_boundary);
2124 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2126 if (!shost->use_clustering)
2127 q->limits.cluster = 0;
2130 * set a reasonable default alignment on word boundaries: the
2131 * host and device may alter it using
2132 * blk_queue_update_dma_alignment() later.
2134 blk_queue_dma_alignment(q, 0x03);
2137 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2138 request_fn_proc *request_fn)
2140 struct request_queue *q;
2142 q = blk_init_queue(request_fn, NULL);
2145 __scsi_init_queue(shost, q);
2148 EXPORT_SYMBOL(__scsi_alloc_queue);
2150 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2152 struct request_queue *q;
2154 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2158 blk_queue_prep_rq(q, scsi_prep_fn);
2159 blk_queue_unprep_rq(q, scsi_unprep_fn);
2160 blk_queue_softirq_done(q, scsi_softirq_done);
2161 blk_queue_rq_timed_out(q, scsi_times_out);
2162 blk_queue_lld_busy(q, scsi_lld_busy);
2166 static struct blk_mq_ops scsi_mq_ops = {
2167 .map_queue = blk_mq_map_queue,
2168 .queue_rq = scsi_queue_rq,
2169 .complete = scsi_softirq_done,
2170 .timeout = scsi_timeout,
2171 .init_request = scsi_init_request,
2172 .exit_request = scsi_exit_request,
2175 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2177 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2178 if (IS_ERR(sdev->request_queue))
2181 sdev->request_queue->queuedata = sdev;
2182 __scsi_init_queue(sdev->host, sdev->request_queue);
2183 return sdev->request_queue;
2186 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2188 unsigned int cmd_size, sgl_size, tbl_size;
2190 tbl_size = shost->sg_tablesize;
2191 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2192 tbl_size = SCSI_MAX_SG_SEGMENTS;
2193 sgl_size = tbl_size * sizeof(struct scatterlist);
2194 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2195 if (scsi_host_get_prot(shost))
2196 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2198 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2199 shost->tag_set.ops = &scsi_mq_ops;
2200 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2201 shost->tag_set.queue_depth = shost->can_queue;
2202 shost->tag_set.cmd_size = cmd_size;
2203 shost->tag_set.numa_node = NUMA_NO_NODE;
2204 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2205 shost->tag_set.flags |=
2206 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2207 shost->tag_set.driver_data = shost;
2209 return blk_mq_alloc_tag_set(&shost->tag_set);
2212 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2214 blk_mq_free_tag_set(&shost->tag_set);
2218 * Function: scsi_block_requests()
2220 * Purpose: Utility function used by low-level drivers to prevent further
2221 * commands from being queued to the device.
2223 * Arguments: shost - Host in question
2227 * Lock status: No locks are assumed held.
2229 * Notes: There is no timer nor any other means by which the requests
2230 * get unblocked other than the low-level driver calling
2231 * scsi_unblock_requests().
2233 void scsi_block_requests(struct Scsi_Host *shost)
2235 shost->host_self_blocked = 1;
2237 EXPORT_SYMBOL(scsi_block_requests);
2240 * Function: scsi_unblock_requests()
2242 * Purpose: Utility function used by low-level drivers to allow further
2243 * commands from being queued to the device.
2245 * Arguments: shost - Host in question
2249 * Lock status: No locks are assumed held.
2251 * Notes: There is no timer nor any other means by which the requests
2252 * get unblocked other than the low-level driver calling
2253 * scsi_unblock_requests().
2255 * This is done as an API function so that changes to the
2256 * internals of the scsi mid-layer won't require wholesale
2257 * changes to drivers that use this feature.
2259 void scsi_unblock_requests(struct Scsi_Host *shost)
2261 shost->host_self_blocked = 0;
2262 scsi_run_host_queues(shost);
2264 EXPORT_SYMBOL(scsi_unblock_requests);
2266 int __init scsi_init_queue(void)
2270 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2271 sizeof(struct scsi_data_buffer),
2273 if (!scsi_sdb_cache) {
2274 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2278 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2279 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2280 int size = sgp->size * sizeof(struct scatterlist);
2282 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2283 SLAB_HWCACHE_ALIGN, NULL);
2285 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2290 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2293 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2302 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2303 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2305 mempool_destroy(sgp->pool);
2307 kmem_cache_destroy(sgp->slab);
2309 kmem_cache_destroy(scsi_sdb_cache);
2314 void scsi_exit_queue(void)
2318 kmem_cache_destroy(scsi_sdb_cache);
2320 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2321 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2322 mempool_destroy(sgp->pool);
2323 kmem_cache_destroy(sgp->slab);
2328 * scsi_mode_select - issue a mode select
2329 * @sdev: SCSI device to be queried
2330 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2331 * @sp: Save page bit (0 == don't save, 1 == save)
2332 * @modepage: mode page being requested
2333 * @buffer: request buffer (may not be smaller than eight bytes)
2334 * @len: length of request buffer.
2335 * @timeout: command timeout
2336 * @retries: number of retries before failing
2337 * @data: returns a structure abstracting the mode header data
2338 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2339 * must be SCSI_SENSE_BUFFERSIZE big.
2341 * Returns zero if successful; negative error number or scsi
2346 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2347 unsigned char *buffer, int len, int timeout, int retries,
2348 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2350 unsigned char cmd[10];
2351 unsigned char *real_buffer;
2354 memset(cmd, 0, sizeof(cmd));
2355 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2357 if (sdev->use_10_for_ms) {
2360 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2363 memcpy(real_buffer + 8, buffer, len);
2367 real_buffer[2] = data->medium_type;
2368 real_buffer[3] = data->device_specific;
2369 real_buffer[4] = data->longlba ? 0x01 : 0;
2371 real_buffer[6] = data->block_descriptor_length >> 8;
2372 real_buffer[7] = data->block_descriptor_length;
2374 cmd[0] = MODE_SELECT_10;
2378 if (len > 255 || data->block_descriptor_length > 255 ||
2382 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2385 memcpy(real_buffer + 4, buffer, len);
2388 real_buffer[1] = data->medium_type;
2389 real_buffer[2] = data->device_specific;
2390 real_buffer[3] = data->block_descriptor_length;
2393 cmd[0] = MODE_SELECT;
2397 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2398 sshdr, timeout, retries, NULL);
2402 EXPORT_SYMBOL_GPL(scsi_mode_select);
2405 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2406 * @sdev: SCSI device to be queried
2407 * @dbd: set if mode sense will allow block descriptors to be returned
2408 * @modepage: mode page being requested
2409 * @buffer: request buffer (may not be smaller than eight bytes)
2410 * @len: length of request buffer.
2411 * @timeout: command timeout
2412 * @retries: number of retries before failing
2413 * @data: returns a structure abstracting the mode header data
2414 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2415 * must be SCSI_SENSE_BUFFERSIZE big.
2417 * Returns zero if unsuccessful, or the header offset (either 4
2418 * or 8 depending on whether a six or ten byte command was
2419 * issued) if successful.
2422 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2423 unsigned char *buffer, int len, int timeout, int retries,
2424 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2426 unsigned char cmd[12];
2429 int result, retry_count = retries;
2430 struct scsi_sense_hdr my_sshdr;
2432 memset(data, 0, sizeof(*data));
2433 memset(&cmd[0], 0, 12);
2434 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2437 /* caller might not be interested in sense, but we need it */
2442 use_10_for_ms = sdev->use_10_for_ms;
2444 if (use_10_for_ms) {
2448 cmd[0] = MODE_SENSE_10;
2455 cmd[0] = MODE_SENSE;
2460 memset(buffer, 0, len);
2462 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2463 sshdr, timeout, retries, NULL);
2465 /* This code looks awful: what it's doing is making sure an
2466 * ILLEGAL REQUEST sense return identifies the actual command
2467 * byte as the problem. MODE_SENSE commands can return
2468 * ILLEGAL REQUEST if the code page isn't supported */
2470 if (use_10_for_ms && !scsi_status_is_good(result) &&
2471 (driver_byte(result) & DRIVER_SENSE)) {
2472 if (scsi_sense_valid(sshdr)) {
2473 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2474 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2476 * Invalid command operation code
2478 sdev->use_10_for_ms = 0;
2484 if(scsi_status_is_good(result)) {
2485 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2486 (modepage == 6 || modepage == 8))) {
2487 /* Initio breakage? */
2490 data->medium_type = 0;
2491 data->device_specific = 0;
2493 data->block_descriptor_length = 0;
2494 } else if(use_10_for_ms) {
2495 data->length = buffer[0]*256 + buffer[1] + 2;
2496 data->medium_type = buffer[2];
2497 data->device_specific = buffer[3];
2498 data->longlba = buffer[4] & 0x01;
2499 data->block_descriptor_length = buffer[6]*256
2502 data->length = buffer[0] + 1;
2503 data->medium_type = buffer[1];
2504 data->device_specific = buffer[2];
2505 data->block_descriptor_length = buffer[3];
2507 data->header_length = header_length;
2508 } else if ((status_byte(result) == CHECK_CONDITION) &&
2509 scsi_sense_valid(sshdr) &&
2510 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2517 EXPORT_SYMBOL(scsi_mode_sense);
2520 * scsi_test_unit_ready - test if unit is ready
2521 * @sdev: scsi device to change the state of.
2522 * @timeout: command timeout
2523 * @retries: number of retries before failing
2524 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2525 * returning sense. Make sure that this is cleared before passing
2528 * Returns zero if unsuccessful or an error if TUR failed. For
2529 * removable media, UNIT_ATTENTION sets ->changed flag.
2532 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2533 struct scsi_sense_hdr *sshdr_external)
2536 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2538 struct scsi_sense_hdr *sshdr;
2541 if (!sshdr_external)
2542 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2544 sshdr = sshdr_external;
2546 /* try to eat the UNIT_ATTENTION if there are enough retries */
2548 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2549 timeout, retries, NULL);
2550 if (sdev->removable && scsi_sense_valid(sshdr) &&
2551 sshdr->sense_key == UNIT_ATTENTION)
2553 } while (scsi_sense_valid(sshdr) &&
2554 sshdr->sense_key == UNIT_ATTENTION && --retries);
2556 if (!sshdr_external)
2560 EXPORT_SYMBOL(scsi_test_unit_ready);
2563 * scsi_device_set_state - Take the given device through the device state model.
2564 * @sdev: scsi device to change the state of.
2565 * @state: state to change to.
2567 * Returns zero if unsuccessful or an error if the requested
2568 * transition is illegal.
2571 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2573 enum scsi_device_state oldstate = sdev->sdev_state;
2575 if (state == oldstate)
2581 case SDEV_CREATED_BLOCK:
2592 case SDEV_TRANSPORT_OFFLINE:
2605 case SDEV_TRANSPORT_OFFLINE:
2613 case SDEV_TRANSPORT_OFFLINE:
2628 case SDEV_CREATED_BLOCK:
2635 case SDEV_CREATED_BLOCK:
2650 case SDEV_TRANSPORT_OFFLINE:
2663 case SDEV_TRANSPORT_OFFLINE:
2665 case SDEV_CREATED_BLOCK:
2673 sdev->sdev_state = state;
2677 SCSI_LOG_ERROR_RECOVERY(1,
2678 sdev_printk(KERN_ERR, sdev,
2679 "Illegal state transition %s->%s",
2680 scsi_device_state_name(oldstate),
2681 scsi_device_state_name(state))
2685 EXPORT_SYMBOL(scsi_device_set_state);
2688 * sdev_evt_emit - emit a single SCSI device uevent
2689 * @sdev: associated SCSI device
2690 * @evt: event to emit
2692 * Send a single uevent (scsi_event) to the associated scsi_device.
2694 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2699 switch (evt->evt_type) {
2700 case SDEV_EVT_MEDIA_CHANGE:
2701 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2703 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2704 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2706 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2707 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2709 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2710 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2712 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2713 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2715 case SDEV_EVT_LUN_CHANGE_REPORTED:
2716 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2718 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2719 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2728 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2732 * sdev_evt_thread - send a uevent for each scsi event
2733 * @work: work struct for scsi_device
2735 * Dispatch queued events to their associated scsi_device kobjects
2738 void scsi_evt_thread(struct work_struct *work)
2740 struct scsi_device *sdev;
2741 enum scsi_device_event evt_type;
2742 LIST_HEAD(event_list);
2744 sdev = container_of(work, struct scsi_device, event_work);
2746 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2747 if (test_and_clear_bit(evt_type, sdev->pending_events))
2748 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2751 struct scsi_event *evt;
2752 struct list_head *this, *tmp;
2753 unsigned long flags;
2755 spin_lock_irqsave(&sdev->list_lock, flags);
2756 list_splice_init(&sdev->event_list, &event_list);
2757 spin_unlock_irqrestore(&sdev->list_lock, flags);
2759 if (list_empty(&event_list))
2762 list_for_each_safe(this, tmp, &event_list) {
2763 evt = list_entry(this, struct scsi_event, node);
2764 list_del(&evt->node);
2765 scsi_evt_emit(sdev, evt);
2772 * sdev_evt_send - send asserted event to uevent thread
2773 * @sdev: scsi_device event occurred on
2774 * @evt: event to send
2776 * Assert scsi device event asynchronously.
2778 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2780 unsigned long flags;
2783 /* FIXME: currently this check eliminates all media change events
2784 * for polled devices. Need to update to discriminate between AN
2785 * and polled events */
2786 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2792 spin_lock_irqsave(&sdev->list_lock, flags);
2793 list_add_tail(&evt->node, &sdev->event_list);
2794 schedule_work(&sdev->event_work);
2795 spin_unlock_irqrestore(&sdev->list_lock, flags);
2797 EXPORT_SYMBOL_GPL(sdev_evt_send);
2800 * sdev_evt_alloc - allocate a new scsi event
2801 * @evt_type: type of event to allocate
2802 * @gfpflags: GFP flags for allocation
2804 * Allocates and returns a new scsi_event.
2806 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2809 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2813 evt->evt_type = evt_type;
2814 INIT_LIST_HEAD(&evt->node);
2816 /* evt_type-specific initialization, if any */
2818 case SDEV_EVT_MEDIA_CHANGE:
2819 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2820 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2821 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2822 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2823 case SDEV_EVT_LUN_CHANGE_REPORTED:
2824 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2832 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2835 * sdev_evt_send_simple - send asserted event to uevent thread
2836 * @sdev: scsi_device event occurred on
2837 * @evt_type: type of event to send
2838 * @gfpflags: GFP flags for allocation
2840 * Assert scsi device event asynchronously, given an event type.
2842 void sdev_evt_send_simple(struct scsi_device *sdev,
2843 enum scsi_device_event evt_type, gfp_t gfpflags)
2845 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2847 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2852 sdev_evt_send(sdev, evt);
2854 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2857 * scsi_device_quiesce - Block user issued commands.
2858 * @sdev: scsi device to quiesce.
2860 * This works by trying to transition to the SDEV_QUIESCE state
2861 * (which must be a legal transition). When the device is in this
2862 * state, only special requests will be accepted, all others will
2863 * be deferred. Since special requests may also be requeued requests,
2864 * a successful return doesn't guarantee the device will be
2865 * totally quiescent.
2867 * Must be called with user context, may sleep.
2869 * Returns zero if unsuccessful or an error if not.
2872 scsi_device_quiesce(struct scsi_device *sdev)
2874 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2878 scsi_run_queue(sdev->request_queue);
2879 while (atomic_read(&sdev->device_busy)) {
2880 msleep_interruptible(200);
2881 scsi_run_queue(sdev->request_queue);
2885 EXPORT_SYMBOL(scsi_device_quiesce);
2888 * scsi_device_resume - Restart user issued commands to a quiesced device.
2889 * @sdev: scsi device to resume.
2891 * Moves the device from quiesced back to running and restarts the
2894 * Must be called with user context, may sleep.
2896 void scsi_device_resume(struct scsi_device *sdev)
2898 /* check if the device state was mutated prior to resume, and if
2899 * so assume the state is being managed elsewhere (for example
2900 * device deleted during suspend)
2902 if (sdev->sdev_state != SDEV_QUIESCE ||
2903 scsi_device_set_state(sdev, SDEV_RUNNING))
2905 scsi_run_queue(sdev->request_queue);
2907 EXPORT_SYMBOL(scsi_device_resume);
2910 device_quiesce_fn(struct scsi_device *sdev, void *data)
2912 scsi_device_quiesce(sdev);
2916 scsi_target_quiesce(struct scsi_target *starget)
2918 starget_for_each_device(starget, NULL, device_quiesce_fn);
2920 EXPORT_SYMBOL(scsi_target_quiesce);
2923 device_resume_fn(struct scsi_device *sdev, void *data)
2925 scsi_device_resume(sdev);
2929 scsi_target_resume(struct scsi_target *starget)
2931 starget_for_each_device(starget, NULL, device_resume_fn);
2933 EXPORT_SYMBOL(scsi_target_resume);
2936 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2937 * @sdev: device to block
2939 * Block request made by scsi lld's to temporarily stop all
2940 * scsi commands on the specified device. Called from interrupt
2941 * or normal process context.
2943 * Returns zero if successful or error if not
2946 * This routine transitions the device to the SDEV_BLOCK state
2947 * (which must be a legal transition). When the device is in this
2948 * state, all commands are deferred until the scsi lld reenables
2949 * the device with scsi_device_unblock or device_block_tmo fires.
2952 scsi_internal_device_block(struct scsi_device *sdev)
2954 struct request_queue *q = sdev->request_queue;
2955 unsigned long flags;
2958 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2960 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2967 * The device has transitioned to SDEV_BLOCK. Stop the
2968 * block layer from calling the midlayer with this device's
2972 blk_mq_stop_hw_queues(q);
2974 spin_lock_irqsave(q->queue_lock, flags);
2976 spin_unlock_irqrestore(q->queue_lock, flags);
2981 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2984 * scsi_internal_device_unblock - resume a device after a block request
2985 * @sdev: device to resume
2986 * @new_state: state to set devices to after unblocking
2988 * Called by scsi lld's or the midlayer to restart the device queue
2989 * for the previously suspended scsi device. Called from interrupt or
2990 * normal process context.
2992 * Returns zero if successful or error if not.
2995 * This routine transitions the device to the SDEV_RUNNING state
2996 * or to one of the offline states (which must be a legal transition)
2997 * allowing the midlayer to goose the queue for this device.
3000 scsi_internal_device_unblock(struct scsi_device *sdev,
3001 enum scsi_device_state new_state)
3003 struct request_queue *q = sdev->request_queue;
3004 unsigned long flags;
3007 * Try to transition the scsi device to SDEV_RUNNING or one of the
3008 * offlined states and goose the device queue if successful.
3010 if ((sdev->sdev_state == SDEV_BLOCK) ||
3011 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3012 sdev->sdev_state = new_state;
3013 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3014 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3015 new_state == SDEV_OFFLINE)
3016 sdev->sdev_state = new_state;
3018 sdev->sdev_state = SDEV_CREATED;
3019 } else if (sdev->sdev_state != SDEV_CANCEL &&
3020 sdev->sdev_state != SDEV_OFFLINE)
3024 blk_mq_start_stopped_hw_queues(q, false);
3026 spin_lock_irqsave(q->queue_lock, flags);
3028 spin_unlock_irqrestore(q->queue_lock, flags);
3033 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3036 device_block(struct scsi_device *sdev, void *data)
3038 scsi_internal_device_block(sdev);
3042 target_block(struct device *dev, void *data)
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), NULL,
3051 scsi_target_block(struct device *dev)
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), NULL,
3057 device_for_each_child(dev, NULL, target_block);
3059 EXPORT_SYMBOL_GPL(scsi_target_block);
3062 device_unblock(struct scsi_device *sdev, void *data)
3064 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3068 target_unblock(struct device *dev, void *data)
3070 if (scsi_is_target_device(dev))
3071 starget_for_each_device(to_scsi_target(dev), data,
3077 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3079 if (scsi_is_target_device(dev))
3080 starget_for_each_device(to_scsi_target(dev), &new_state,
3083 device_for_each_child(dev, &new_state, target_unblock);
3085 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3088 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3089 * @sgl: scatter-gather list
3090 * @sg_count: number of segments in sg
3091 * @offset: offset in bytes into sg, on return offset into the mapped area
3092 * @len: bytes to map, on return number of bytes mapped
3094 * Returns virtual address of the start of the mapped page
3096 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3097 size_t *offset, size_t *len)
3100 size_t sg_len = 0, len_complete = 0;
3101 struct scatterlist *sg;
3104 WARN_ON(!irqs_disabled());
3106 for_each_sg(sgl, sg, sg_count, i) {
3107 len_complete = sg_len; /* Complete sg-entries */
3108 sg_len += sg->length;
3109 if (sg_len > *offset)
3113 if (unlikely(i == sg_count)) {
3114 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3116 __func__, sg_len, *offset, sg_count);
3121 /* Offset starting from the beginning of first page in this sg-entry */
3122 *offset = *offset - len_complete + sg->offset;
3124 /* Assumption: contiguous pages can be accessed as "page + i" */
3125 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3126 *offset &= ~PAGE_MASK;
3128 /* Bytes in this sg-entry from *offset to the end of the page */
3129 sg_len = PAGE_SIZE - *offset;
3133 return kmap_atomic(page);
3135 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3138 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3139 * @virt: virtual address to be unmapped
3141 void scsi_kunmap_atomic_sg(void *virt)
3143 kunmap_atomic(virt);
3145 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3147 void sdev_disable_disk_events(struct scsi_device *sdev)
3149 atomic_inc(&sdev->disk_events_disable_depth);
3151 EXPORT_SYMBOL(sdev_disable_disk_events);
3153 void sdev_enable_disk_events(struct scsi_device *sdev)
3155 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3157 atomic_dec(&sdev->disk_events_disable_depth);
3159 EXPORT_SYMBOL(sdev_enable_disk_events);