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>
35 #include <trace/events/scsi.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
41 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
42 #define SG_MEMPOOL_SIZE 2
44 struct scsi_host_sg_pool {
47 struct kmem_cache *slab;
51 #define SP(x) { .size = x, "sgpool-" __stringify(x) }
52 #if (SCSI_MAX_SG_SEGMENTS < 32)
53 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
55 static struct scsi_host_sg_pool scsi_sg_pools[] = {
58 #if (SCSI_MAX_SG_SEGMENTS > 32)
60 #if (SCSI_MAX_SG_SEGMENTS > 64)
62 #if (SCSI_MAX_SG_SEGMENTS > 128)
64 #if (SCSI_MAX_SG_SEGMENTS > 256)
65 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
70 SP(SCSI_MAX_SG_SEGMENTS)
74 struct kmem_cache *scsi_sdb_cache;
77 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
78 * not change behaviour from the previous unplug mechanism, experimentation
79 * may prove this needs changing.
81 #define SCSI_QUEUE_DELAY 3
84 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
86 struct Scsi_Host *host = cmd->device->host;
87 struct scsi_device *device = cmd->device;
88 struct scsi_target *starget = scsi_target(device);
91 * Set the appropriate busy bit for the device/host.
93 * If the host/device isn't busy, assume that something actually
94 * completed, and that we should be able to queue a command now.
96 * Note that the prior mid-layer assumption that any host could
97 * always queue at least one command is now broken. The mid-layer
98 * will implement a user specifiable stall (see
99 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
100 * if a command is requeued with no other commands outstanding
101 * either for the device or for the host.
104 case SCSI_MLQUEUE_HOST_BUSY:
105 atomic_set(&host->host_blocked, host->max_host_blocked);
107 case SCSI_MLQUEUE_DEVICE_BUSY:
108 case SCSI_MLQUEUE_EH_RETRY:
109 atomic_set(&device->device_blocked,
110 device->max_device_blocked);
112 case SCSI_MLQUEUE_TARGET_BUSY:
113 atomic_set(&starget->target_blocked,
114 starget->max_target_blocked);
119 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
121 struct scsi_device *sdev = cmd->device;
122 struct request_queue *q = cmd->request->q;
124 blk_mq_requeue_request(cmd->request);
125 blk_mq_kick_requeue_list(q);
126 put_device(&sdev->sdev_gendev);
130 * __scsi_queue_insert - private queue insertion
131 * @cmd: The SCSI command being requeued
132 * @reason: The reason for the requeue
133 * @unbusy: Whether the queue should be unbusied
135 * This is a private queue insertion. The public interface
136 * scsi_queue_insert() always assumes the queue should be unbusied
137 * because it's always called before the completion. This function is
138 * for a requeue after completion, which should only occur in this
141 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
143 struct scsi_device *device = cmd->device;
144 struct request_queue *q = device->request_queue;
147 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
148 "Inserting command %p into mlqueue\n", cmd));
150 scsi_set_blocked(cmd, reason);
153 * Decrement the counters, since these commands are no longer
154 * active on the host/device.
157 scsi_device_unbusy(device);
160 * Requeue this command. It will go before all other commands
161 * that are already in the queue. Schedule requeue work under
162 * lock such that the kblockd_schedule_work() call happens
163 * before blk_cleanup_queue() finishes.
167 scsi_mq_requeue_cmd(cmd);
170 spin_lock_irqsave(q->queue_lock, flags);
171 blk_requeue_request(q, cmd->request);
172 kblockd_schedule_work(&device->requeue_work);
173 spin_unlock_irqrestore(q->queue_lock, flags);
177 * Function: scsi_queue_insert()
179 * Purpose: Insert a command in the midlevel queue.
181 * Arguments: cmd - command that we are adding to queue.
182 * reason - why we are inserting command to queue.
184 * Lock status: Assumed that lock is not held upon entry.
188 * Notes: We do this for one of two cases. Either the host is busy
189 * and it cannot accept any more commands for the time being,
190 * or the device returned QUEUE_FULL and can accept no more
192 * Notes: This could be called either from an interrupt context or a
193 * normal process context.
195 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
197 __scsi_queue_insert(cmd, reason, 1);
200 * scsi_execute - insert request and wait for the result
203 * @data_direction: data direction
204 * @buffer: data buffer
205 * @bufflen: len of buffer
206 * @sense: optional sense buffer
207 * @timeout: request timeout in seconds
208 * @retries: number of times to retry request
209 * @flags: or into request flags;
210 * @resid: optional residual length
212 * returns the req->errors value which is the scsi_cmnd result
215 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
216 int data_direction, void *buffer, unsigned bufflen,
217 unsigned char *sense, int timeout, int retries, u64 flags,
221 int write = (data_direction == DMA_TO_DEVICE);
222 int ret = DRIVER_ERROR << 24;
224 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
227 blk_rq_set_block_pc(req);
229 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
230 buffer, bufflen, __GFP_WAIT))
233 req->cmd_len = COMMAND_SIZE(cmd[0]);
234 memcpy(req->cmd, cmd, req->cmd_len);
237 req->retries = retries;
238 req->timeout = timeout;
239 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
242 * head injection *required* here otherwise quiesce won't work
244 blk_execute_rq(req->q, NULL, req, 1);
247 * Some devices (USB mass-storage in particular) may transfer
248 * garbage data together with a residue indicating that the data
249 * is invalid. Prevent the garbage from being misinterpreted
250 * and prevent security leaks by zeroing out the excess data.
252 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
253 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
256 *resid = req->resid_len;
259 blk_put_request(req);
263 EXPORT_SYMBOL(scsi_execute);
265 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
266 int data_direction, void *buffer, unsigned bufflen,
267 struct scsi_sense_hdr *sshdr, int timeout, int retries,
268 int *resid, u64 flags)
274 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
276 return DRIVER_ERROR << 24;
278 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
279 sense, timeout, retries, flags, resid);
281 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
286 EXPORT_SYMBOL(scsi_execute_req_flags);
289 * Function: scsi_init_cmd_errh()
291 * Purpose: Initialize cmd fields related to error handling.
293 * Arguments: cmd - command that is ready to be queued.
295 * Notes: This function has the job of initializing a number of
296 * fields related to error handling. Typically this will
297 * be called once for each command, as required.
299 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
301 cmd->serial_number = 0;
302 scsi_set_resid(cmd, 0);
303 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
304 if (cmd->cmd_len == 0)
305 cmd->cmd_len = scsi_command_size(cmd->cmnd);
308 void scsi_device_unbusy(struct scsi_device *sdev)
310 struct Scsi_Host *shost = sdev->host;
311 struct scsi_target *starget = scsi_target(sdev);
314 atomic_dec(&shost->host_busy);
315 if (starget->can_queue > 0)
316 atomic_dec(&starget->target_busy);
318 if (unlikely(scsi_host_in_recovery(shost) &&
319 (shost->host_failed || shost->host_eh_scheduled))) {
320 spin_lock_irqsave(shost->host_lock, flags);
321 scsi_eh_wakeup(shost);
322 spin_unlock_irqrestore(shost->host_lock, flags);
325 atomic_dec(&sdev->device_busy);
328 static void scsi_kick_queue(struct request_queue *q)
331 blk_mq_start_hw_queues(q);
337 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
338 * and call blk_run_queue for all the scsi_devices on the target -
339 * including current_sdev first.
341 * Called with *no* scsi locks held.
343 static void scsi_single_lun_run(struct scsi_device *current_sdev)
345 struct Scsi_Host *shost = current_sdev->host;
346 struct scsi_device *sdev, *tmp;
347 struct scsi_target *starget = scsi_target(current_sdev);
350 spin_lock_irqsave(shost->host_lock, flags);
351 starget->starget_sdev_user = NULL;
352 spin_unlock_irqrestore(shost->host_lock, flags);
355 * Call blk_run_queue for all LUNs on the target, starting with
356 * current_sdev. We race with others (to set starget_sdev_user),
357 * but in most cases, we will be first. Ideally, each LU on the
358 * target would get some limited time or requests on the target.
360 scsi_kick_queue(current_sdev->request_queue);
362 spin_lock_irqsave(shost->host_lock, flags);
363 if (starget->starget_sdev_user)
365 list_for_each_entry_safe(sdev, tmp, &starget->devices,
366 same_target_siblings) {
367 if (sdev == current_sdev)
369 if (scsi_device_get(sdev))
372 spin_unlock_irqrestore(shost->host_lock, flags);
373 scsi_kick_queue(sdev->request_queue);
374 spin_lock_irqsave(shost->host_lock, flags);
376 scsi_device_put(sdev);
379 spin_unlock_irqrestore(shost->host_lock, flags);
382 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
384 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
386 if (atomic_read(&sdev->device_blocked) > 0)
391 static inline bool scsi_target_is_busy(struct scsi_target *starget)
393 if (starget->can_queue > 0) {
394 if (atomic_read(&starget->target_busy) >= starget->can_queue)
396 if (atomic_read(&starget->target_blocked) > 0)
402 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
404 if (shost->can_queue > 0 &&
405 atomic_read(&shost->host_busy) >= shost->can_queue)
407 if (atomic_read(&shost->host_blocked) > 0)
409 if (shost->host_self_blocked)
414 static void scsi_starved_list_run(struct Scsi_Host *shost)
416 LIST_HEAD(starved_list);
417 struct scsi_device *sdev;
420 spin_lock_irqsave(shost->host_lock, flags);
421 list_splice_init(&shost->starved_list, &starved_list);
423 while (!list_empty(&starved_list)) {
424 struct request_queue *slq;
427 * As long as shost is accepting commands and we have
428 * starved queues, call blk_run_queue. scsi_request_fn
429 * drops the queue_lock and can add us back to the
432 * host_lock protects the starved_list and starved_entry.
433 * scsi_request_fn must get the host_lock before checking
434 * or modifying starved_list or starved_entry.
436 if (scsi_host_is_busy(shost))
439 sdev = list_entry(starved_list.next,
440 struct scsi_device, starved_entry);
441 list_del_init(&sdev->starved_entry);
442 if (scsi_target_is_busy(scsi_target(sdev))) {
443 list_move_tail(&sdev->starved_entry,
444 &shost->starved_list);
449 * Once we drop the host lock, a racing scsi_remove_device()
450 * call may remove the sdev from the starved list and destroy
451 * it and the queue. Mitigate by taking a reference to the
452 * queue and never touching the sdev again after we drop the
453 * host lock. Note: if __scsi_remove_device() invokes
454 * blk_cleanup_queue() before the queue is run from this
455 * function then blk_run_queue() will return immediately since
456 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
458 slq = sdev->request_queue;
459 if (!blk_get_queue(slq))
461 spin_unlock_irqrestore(shost->host_lock, flags);
463 scsi_kick_queue(slq);
466 spin_lock_irqsave(shost->host_lock, flags);
468 /* put any unprocessed entries back */
469 list_splice(&starved_list, &shost->starved_list);
470 spin_unlock_irqrestore(shost->host_lock, flags);
474 * Function: scsi_run_queue()
476 * Purpose: Select a proper request queue to serve next
478 * Arguments: q - last request's queue
482 * Notes: The previous command was completely finished, start
483 * a new one if possible.
485 static void scsi_run_queue(struct request_queue *q)
487 struct scsi_device *sdev = q->queuedata;
489 if (scsi_target(sdev)->single_lun)
490 scsi_single_lun_run(sdev);
491 if (!list_empty(&sdev->host->starved_list))
492 scsi_starved_list_run(sdev->host);
495 blk_mq_start_stopped_hw_queues(q, false);
500 void scsi_requeue_run_queue(struct work_struct *work)
502 struct scsi_device *sdev;
503 struct request_queue *q;
505 sdev = container_of(work, struct scsi_device, requeue_work);
506 q = sdev->request_queue;
511 * Function: scsi_requeue_command()
513 * Purpose: Handle post-processing of completed commands.
515 * Arguments: q - queue to operate on
516 * cmd - command that may need to be requeued.
520 * Notes: After command completion, there may be blocks left
521 * over which weren't finished by the previous command
522 * this can be for a number of reasons - the main one is
523 * I/O errors in the middle of the request, in which case
524 * we need to request the blocks that come after the bad
526 * Notes: Upon return, cmd is a stale pointer.
528 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
530 struct scsi_device *sdev = cmd->device;
531 struct request *req = cmd->request;
534 spin_lock_irqsave(q->queue_lock, flags);
535 blk_unprep_request(req);
537 scsi_put_command(cmd);
538 blk_requeue_request(q, req);
539 spin_unlock_irqrestore(q->queue_lock, flags);
543 put_device(&sdev->sdev_gendev);
546 void scsi_run_host_queues(struct Scsi_Host *shost)
548 struct scsi_device *sdev;
550 shost_for_each_device(sdev, shost)
551 scsi_run_queue(sdev->request_queue);
554 static inline unsigned int scsi_sgtable_index(unsigned short nents)
558 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
563 index = get_count_order(nents) - 3;
568 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
570 struct scsi_host_sg_pool *sgp;
572 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
573 mempool_free(sgl, sgp->pool);
576 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
578 struct scsi_host_sg_pool *sgp;
580 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
581 return mempool_alloc(sgp->pool, gfp_mask);
584 static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
586 if (mq && sdb->table.nents <= SCSI_MAX_SG_SEGMENTS)
588 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
591 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, bool mq)
593 struct scatterlist *first_chunk = NULL;
599 if (nents <= SCSI_MAX_SG_SEGMENTS) {
600 sdb->table.nents = nents;
601 sg_init_table(sdb->table.sgl, sdb->table.nents);
604 first_chunk = sdb->table.sgl;
607 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
608 first_chunk, GFP_ATOMIC, scsi_sg_alloc);
610 scsi_free_sgtable(sdb, mq);
614 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
616 if (cmd->request->cmd_type == REQ_TYPE_FS) {
617 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
619 if (drv->uninit_command)
620 drv->uninit_command(cmd);
624 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
626 if (cmd->sdb.table.nents)
627 scsi_free_sgtable(&cmd->sdb, true);
628 if (cmd->request->next_rq && cmd->request->next_rq->special)
629 scsi_free_sgtable(cmd->request->next_rq->special, true);
630 if (scsi_prot_sg_count(cmd))
631 scsi_free_sgtable(cmd->prot_sdb, true);
634 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
636 struct scsi_device *sdev = cmd->device;
637 struct Scsi_Host *shost = sdev->host;
640 scsi_mq_free_sgtables(cmd);
641 scsi_uninit_cmd(cmd);
643 if (shost->use_cmd_list) {
644 BUG_ON(list_empty(&cmd->list));
645 spin_lock_irqsave(&sdev->list_lock, flags);
646 list_del_init(&cmd->list);
647 spin_unlock_irqrestore(&sdev->list_lock, flags);
652 * Function: scsi_release_buffers()
654 * Purpose: Free resources allocate for a scsi_command.
656 * Arguments: cmd - command that we are bailing.
658 * Lock status: Assumed that no lock is held upon entry.
662 * Notes: In the event that an upper level driver rejects a
663 * command, we must release resources allocated during
664 * the __init_io() function. Primarily this would involve
665 * the scatter-gather table.
667 static void scsi_release_buffers(struct scsi_cmnd *cmd)
669 if (cmd->sdb.table.nents)
670 scsi_free_sgtable(&cmd->sdb, false);
672 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
674 if (scsi_prot_sg_count(cmd))
675 scsi_free_sgtable(cmd->prot_sdb, false);
678 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
680 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
682 scsi_free_sgtable(bidi_sdb, false);
683 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
684 cmd->request->next_rq->special = NULL;
687 static bool scsi_end_request(struct request *req, int error,
688 unsigned int bytes, unsigned int bidi_bytes)
690 struct scsi_cmnd *cmd = req->special;
691 struct scsi_device *sdev = cmd->device;
692 struct request_queue *q = sdev->request_queue;
694 if (blk_update_request(req, error, bytes))
697 /* Bidi request must be completed as a whole */
698 if (unlikely(bidi_bytes) &&
699 blk_update_request(req->next_rq, error, bidi_bytes))
702 if (blk_queue_add_random(q))
703 add_disk_randomness(req->rq_disk);
707 * In the MQ case the command gets freed by __blk_mq_end_request,
708 * so we have to do all cleanup that depends on it earlier.
710 * We also can't kick the queues from irq context, so we
711 * will have to defer it to a workqueue.
713 scsi_mq_uninit_cmd(cmd);
715 __blk_mq_end_request(req, error);
717 if (scsi_target(sdev)->single_lun ||
718 !list_empty(&sdev->host->starved_list))
719 kblockd_schedule_work(&sdev->requeue_work);
721 blk_mq_start_stopped_hw_queues(q, true);
726 scsi_release_bidi_buffers(cmd);
728 spin_lock_irqsave(q->queue_lock, flags);
729 blk_finish_request(req, error);
730 spin_unlock_irqrestore(q->queue_lock, flags);
732 scsi_release_buffers(cmd);
734 scsi_put_command(cmd);
738 put_device(&sdev->sdev_gendev);
743 * __scsi_error_from_host_byte - translate SCSI error code into errno
744 * @cmd: SCSI command (unused)
745 * @result: scsi error code
747 * Translate SCSI error code into standard UNIX errno.
749 * -ENOLINK temporary transport failure
750 * -EREMOTEIO permanent target failure, do not retry
751 * -EBADE permanent nexus failure, retry on other path
752 * -ENOSPC No write space available
753 * -ENODATA Medium error
754 * -EIO unspecified I/O error
756 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
760 switch(host_byte(result)) {
761 case DID_TRANSPORT_FAILFAST:
764 case DID_TARGET_FAILURE:
765 set_host_byte(cmd, DID_OK);
768 case DID_NEXUS_FAILURE:
769 set_host_byte(cmd, DID_OK);
772 case DID_ALLOC_FAILURE:
773 set_host_byte(cmd, DID_OK);
776 case DID_MEDIUM_ERROR:
777 set_host_byte(cmd, DID_OK);
789 * Function: scsi_io_completion()
791 * Purpose: Completion processing for block device I/O requests.
793 * Arguments: cmd - command that is finished.
795 * Lock status: Assumed that no lock is held upon entry.
799 * Notes: We will finish off the specified number of sectors. If we
800 * are done, the command block will be released and the queue
801 * function will be goosed. If we are not done then we have to
802 * figure out what to do next:
804 * a) We can call scsi_requeue_command(). The request
805 * will be unprepared and put back on the queue. Then
806 * a new command will be created for it. This should
807 * be used if we made forward progress, or if we want
808 * to switch from READ(10) to READ(6) for example.
810 * b) We can call __scsi_queue_insert(). The request will
811 * be put back on the queue and retried using the same
812 * command as before, possibly after a delay.
814 * c) We can call scsi_end_request() with -EIO to fail
815 * the remainder of the request.
817 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
819 int result = cmd->result;
820 struct request_queue *q = cmd->device->request_queue;
821 struct request *req = cmd->request;
823 struct scsi_sense_hdr sshdr;
824 bool sense_valid = false;
825 int sense_deferred = 0, level = 0;
826 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
827 ACTION_DELAYED_RETRY} action;
828 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
831 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
833 sense_deferred = scsi_sense_is_deferred(&sshdr);
836 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
838 if (sense_valid && req->sense) {
840 * SG_IO wants current and deferred errors
842 int len = 8 + cmd->sense_buffer[7];
844 if (len > SCSI_SENSE_BUFFERSIZE)
845 len = SCSI_SENSE_BUFFERSIZE;
846 memcpy(req->sense, cmd->sense_buffer, len);
847 req->sense_len = len;
850 error = __scsi_error_from_host_byte(cmd, result);
853 * __scsi_error_from_host_byte may have reset the host_byte
855 req->errors = cmd->result;
857 req->resid_len = scsi_get_resid(cmd);
859 if (scsi_bidi_cmnd(cmd)) {
861 * Bidi commands Must be complete as a whole,
862 * both sides at once.
864 req->next_rq->resid_len = scsi_in(cmd)->resid;
865 if (scsi_end_request(req, 0, blk_rq_bytes(req),
866 blk_rq_bytes(req->next_rq)))
870 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
872 * Certain non BLOCK_PC requests are commands that don't
873 * actually transfer anything (FLUSH), so cannot use
874 * good_bytes != blk_rq_bytes(req) as the signal for an error.
875 * This sets the error explicitly for the problem case.
877 error = __scsi_error_from_host_byte(cmd, result);
880 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
881 BUG_ON(blk_bidi_rq(req));
884 * Next deal with any sectors which we were able to correctly
887 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
888 "%u sectors total, %d bytes done.\n",
889 blk_rq_sectors(req), good_bytes));
892 * Recovered errors need reporting, but they're always treated
893 * as success, so fiddle the result code here. For BLOCK_PC
894 * we already took a copy of the original into rq->errors which
895 * is what gets returned to the user
897 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
898 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
899 * print since caller wants ATA registers. Only occurs on
900 * SCSI ATA PASS_THROUGH commands when CK_COND=1
902 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
904 else if (!(req->cmd_flags & REQ_QUIET))
905 scsi_print_sense(cmd);
907 /* BLOCK_PC may have set error */
912 * If we finished all bytes in the request we are done now.
914 if (!scsi_end_request(req, error, good_bytes, 0))
918 * Kill remainder if no retrys.
920 if (error && scsi_noretry_cmd(cmd)) {
921 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
927 * If there had been no error, but we have leftover bytes in the
928 * requeues just queue the command up again.
933 error = __scsi_error_from_host_byte(cmd, result);
935 if (host_byte(result) == DID_RESET) {
936 /* Third party bus reset or reset for error recovery
937 * reasons. Just retry the command and see what
940 action = ACTION_RETRY;
941 } else if (sense_valid && !sense_deferred) {
942 switch (sshdr.sense_key) {
944 if (cmd->device->removable) {
945 /* Detected disc change. Set a bit
946 * and quietly refuse further access.
948 cmd->device->changed = 1;
949 action = ACTION_FAIL;
951 /* Must have been a power glitch, or a
952 * bus reset. Could not have been a
953 * media change, so we just retry the
954 * command and see what happens.
956 action = ACTION_RETRY;
959 case ILLEGAL_REQUEST:
960 /* If we had an ILLEGAL REQUEST returned, then
961 * we may have performed an unsupported
962 * command. The only thing this should be
963 * would be a ten byte read where only a six
964 * byte read was supported. Also, on a system
965 * where READ CAPACITY failed, we may have
966 * read past the end of the disk.
968 if ((cmd->device->use_10_for_rw &&
969 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
970 (cmd->cmnd[0] == READ_10 ||
971 cmd->cmnd[0] == WRITE_10)) {
972 /* This will issue a new 6-byte command. */
973 cmd->device->use_10_for_rw = 0;
974 action = ACTION_REPREP;
975 } else if (sshdr.asc == 0x10) /* DIX */ {
976 action = ACTION_FAIL;
978 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
979 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
980 action = ACTION_FAIL;
983 action = ACTION_FAIL;
985 case ABORTED_COMMAND:
986 action = ACTION_FAIL;
987 if (sshdr.asc == 0x10) /* DIF */
991 /* If the device is in the process of becoming
992 * ready, or has a temporary blockage, retry.
994 if (sshdr.asc == 0x04) {
995 switch (sshdr.ascq) {
996 case 0x01: /* becoming ready */
997 case 0x04: /* format in progress */
998 case 0x05: /* rebuild in progress */
999 case 0x06: /* recalculation in progress */
1000 case 0x07: /* operation in progress */
1001 case 0x08: /* Long write in progress */
1002 case 0x09: /* self test in progress */
1003 case 0x14: /* space allocation in progress */
1004 action = ACTION_DELAYED_RETRY;
1007 action = ACTION_FAIL;
1011 action = ACTION_FAIL;
1013 case VOLUME_OVERFLOW:
1014 /* See SSC3rXX or current. */
1015 action = ACTION_FAIL;
1018 action = ACTION_FAIL;
1022 action = ACTION_FAIL;
1024 if (action != ACTION_FAIL &&
1025 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1026 action = ACTION_FAIL;
1030 /* Give up and fail the remainder of the request */
1031 if (!(req->cmd_flags & REQ_QUIET)) {
1032 static DEFINE_RATELIMIT_STATE(_rs,
1033 DEFAULT_RATELIMIT_INTERVAL,
1034 DEFAULT_RATELIMIT_BURST);
1036 if (unlikely(scsi_logging_level))
1037 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1038 SCSI_LOG_MLCOMPLETE_BITS);
1041 * if logging is enabled the failure will be printed
1042 * in scsi_log_completion(), so avoid duplicate messages
1044 if (!level && __ratelimit(&_rs)) {
1045 scsi_print_result(cmd, NULL, FAILED);
1046 if (driver_byte(result) & DRIVER_SENSE)
1047 scsi_print_sense(cmd);
1048 scsi_print_command(cmd);
1051 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1056 /* Unprep the request and put it back at the head of the queue.
1057 * A new command will be prepared and issued.
1060 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1061 scsi_mq_uninit_cmd(cmd);
1062 scsi_mq_requeue_cmd(cmd);
1064 scsi_release_buffers(cmd);
1065 scsi_requeue_command(q, cmd);
1069 /* Retry the same command immediately */
1070 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1072 case ACTION_DELAYED_RETRY:
1073 /* Retry the same command after a delay */
1074 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1079 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1084 * If sg table allocation fails, requeue request later.
1086 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1087 req->mq_ctx != NULL)))
1088 return BLKPREP_DEFER;
1091 * Next, walk the list, and fill in the addresses and sizes of
1094 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1095 BUG_ON(count > sdb->table.nents);
1096 sdb->table.nents = count;
1097 sdb->length = blk_rq_bytes(req);
1102 * Function: scsi_init_io()
1104 * Purpose: SCSI I/O initialize function.
1106 * Arguments: cmd - Command descriptor we wish to initialize
1108 * Returns: 0 on success
1109 * BLKPREP_DEFER if the failure is retryable
1110 * BLKPREP_KILL if the failure is fatal
1112 int scsi_init_io(struct scsi_cmnd *cmd)
1114 struct scsi_device *sdev = cmd->device;
1115 struct request *rq = cmd->request;
1116 bool is_mq = (rq->mq_ctx != NULL);
1119 BUG_ON(!rq->nr_phys_segments);
1121 error = scsi_init_sgtable(rq, &cmd->sdb);
1125 if (blk_bidi_rq(rq)) {
1126 if (!rq->q->mq_ops) {
1127 struct scsi_data_buffer *bidi_sdb =
1128 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1130 error = BLKPREP_DEFER;
1134 rq->next_rq->special = bidi_sdb;
1137 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1142 if (blk_integrity_rq(rq)) {
1143 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1146 BUG_ON(prot_sdb == NULL);
1147 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1149 if (scsi_alloc_sgtable(prot_sdb, ivecs, is_mq)) {
1150 error = BLKPREP_DEFER;
1154 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1155 prot_sdb->table.sgl);
1156 BUG_ON(unlikely(count > ivecs));
1157 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1159 cmd->prot_sdb = prot_sdb;
1160 cmd->prot_sdb->table.nents = count;
1166 scsi_mq_free_sgtables(cmd);
1168 scsi_release_buffers(cmd);
1169 cmd->request->special = NULL;
1170 scsi_put_command(cmd);
1171 put_device(&sdev->sdev_gendev);
1175 EXPORT_SYMBOL(scsi_init_io);
1177 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1178 struct request *req)
1180 struct scsi_cmnd *cmd;
1182 if (!req->special) {
1183 /* Bail if we can't get a reference to the device */
1184 if (!get_device(&sdev->sdev_gendev))
1187 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1188 if (unlikely(!cmd)) {
1189 put_device(&sdev->sdev_gendev);
1197 /* pull a tag out of the request if we have one */
1198 cmd->tag = req->tag;
1201 cmd->cmnd = req->cmd;
1202 cmd->prot_op = SCSI_PROT_NORMAL;
1207 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1209 struct scsi_cmnd *cmd = req->special;
1212 * BLOCK_PC requests may transfer data, in which case they must
1213 * a bio attached to them. Or they might contain a SCSI command
1214 * that does not transfer data, in which case they may optionally
1215 * submit a request without an attached bio.
1218 int ret = scsi_init_io(cmd);
1222 BUG_ON(blk_rq_bytes(req));
1224 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1227 cmd->cmd_len = req->cmd_len;
1228 cmd->transfersize = blk_rq_bytes(req);
1229 cmd->allowed = req->retries;
1234 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1235 * that still need to be translated to SCSI CDBs from the ULD.
1237 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1239 struct scsi_cmnd *cmd = req->special;
1241 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1242 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1243 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1244 if (ret != BLKPREP_OK)
1248 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1249 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1252 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1254 struct scsi_cmnd *cmd = req->special;
1256 if (!blk_rq_bytes(req))
1257 cmd->sc_data_direction = DMA_NONE;
1258 else if (rq_data_dir(req) == WRITE)
1259 cmd->sc_data_direction = DMA_TO_DEVICE;
1261 cmd->sc_data_direction = DMA_FROM_DEVICE;
1263 switch (req->cmd_type) {
1265 return scsi_setup_fs_cmnd(sdev, req);
1266 case REQ_TYPE_BLOCK_PC:
1267 return scsi_setup_blk_pc_cmnd(sdev, req);
1269 return BLKPREP_KILL;
1274 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1276 int ret = BLKPREP_OK;
1279 * If the device is not in running state we will reject some
1282 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1283 switch (sdev->sdev_state) {
1285 case SDEV_TRANSPORT_OFFLINE:
1287 * If the device is offline we refuse to process any
1288 * commands. The device must be brought online
1289 * before trying any recovery commands.
1291 sdev_printk(KERN_ERR, sdev,
1292 "rejecting I/O to offline device\n");
1297 * If the device is fully deleted, we refuse to
1298 * process any commands as well.
1300 sdev_printk(KERN_ERR, sdev,
1301 "rejecting I/O to dead device\n");
1306 case SDEV_CREATED_BLOCK:
1308 * If the devices is blocked we defer normal commands.
1310 if (!(req->cmd_flags & REQ_PREEMPT))
1311 ret = BLKPREP_DEFER;
1315 * For any other not fully online state we only allow
1316 * special commands. In particular any user initiated
1317 * command is not allowed.
1319 if (!(req->cmd_flags & REQ_PREEMPT))
1328 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1330 struct scsi_device *sdev = q->queuedata;
1334 req->errors = DID_NO_CONNECT << 16;
1335 /* release the command and kill it */
1337 struct scsi_cmnd *cmd = req->special;
1338 scsi_release_buffers(cmd);
1339 scsi_put_command(cmd);
1340 put_device(&sdev->sdev_gendev);
1341 req->special = NULL;
1346 * If we defer, the blk_peek_request() returns NULL, but the
1347 * queue must be restarted, so we schedule a callback to happen
1350 if (atomic_read(&sdev->device_busy) == 0)
1351 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1354 req->cmd_flags |= REQ_DONTPREP;
1360 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1362 struct scsi_device *sdev = q->queuedata;
1363 struct scsi_cmnd *cmd;
1366 ret = scsi_prep_state_check(sdev, req);
1367 if (ret != BLKPREP_OK)
1370 cmd = scsi_get_cmd_from_req(sdev, req);
1371 if (unlikely(!cmd)) {
1372 ret = BLKPREP_DEFER;
1376 ret = scsi_setup_cmnd(sdev, req);
1378 return scsi_prep_return(q, req, ret);
1381 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1383 scsi_uninit_cmd(req->special);
1387 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1390 * Called with the queue_lock held.
1392 static inline int scsi_dev_queue_ready(struct request_queue *q,
1393 struct scsi_device *sdev)
1397 busy = atomic_inc_return(&sdev->device_busy) - 1;
1398 if (atomic_read(&sdev->device_blocked)) {
1403 * unblock after device_blocked iterates to zero
1405 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1407 * For the MQ case we take care of this in the caller.
1410 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1413 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1414 "unblocking device at zero depth\n"));
1417 if (busy >= sdev->queue_depth)
1422 atomic_dec(&sdev->device_busy);
1427 * scsi_target_queue_ready: checks if there we can send commands to target
1428 * @sdev: scsi device on starget to check.
1430 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1431 struct scsi_device *sdev)
1433 struct scsi_target *starget = scsi_target(sdev);
1436 if (starget->single_lun) {
1437 spin_lock_irq(shost->host_lock);
1438 if (starget->starget_sdev_user &&
1439 starget->starget_sdev_user != sdev) {
1440 spin_unlock_irq(shost->host_lock);
1443 starget->starget_sdev_user = sdev;
1444 spin_unlock_irq(shost->host_lock);
1447 if (starget->can_queue <= 0)
1450 busy = atomic_inc_return(&starget->target_busy) - 1;
1451 if (atomic_read(&starget->target_blocked) > 0) {
1456 * unblock after target_blocked iterates to zero
1458 if (atomic_dec_return(&starget->target_blocked) > 0)
1461 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1462 "unblocking target at zero depth\n"));
1465 if (busy >= starget->can_queue)
1471 spin_lock_irq(shost->host_lock);
1472 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1473 spin_unlock_irq(shost->host_lock);
1475 if (starget->can_queue > 0)
1476 atomic_dec(&starget->target_busy);
1481 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1482 * return 0. We must end up running the queue again whenever 0 is
1483 * returned, else IO can hang.
1485 static inline int scsi_host_queue_ready(struct request_queue *q,
1486 struct Scsi_Host *shost,
1487 struct scsi_device *sdev)
1491 if (scsi_host_in_recovery(shost))
1494 busy = atomic_inc_return(&shost->host_busy) - 1;
1495 if (atomic_read(&shost->host_blocked) > 0) {
1500 * unblock after host_blocked iterates to zero
1502 if (atomic_dec_return(&shost->host_blocked) > 0)
1506 shost_printk(KERN_INFO, shost,
1507 "unblocking host at zero depth\n"));
1510 if (shost->can_queue > 0 && busy >= shost->can_queue)
1512 if (shost->host_self_blocked)
1515 /* We're OK to process the command, so we can't be starved */
1516 if (!list_empty(&sdev->starved_entry)) {
1517 spin_lock_irq(shost->host_lock);
1518 if (!list_empty(&sdev->starved_entry))
1519 list_del_init(&sdev->starved_entry);
1520 spin_unlock_irq(shost->host_lock);
1526 spin_lock_irq(shost->host_lock);
1527 if (list_empty(&sdev->starved_entry))
1528 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1529 spin_unlock_irq(shost->host_lock);
1531 atomic_dec(&shost->host_busy);
1536 * Busy state exporting function for request stacking drivers.
1538 * For efficiency, no lock is taken to check the busy state of
1539 * shost/starget/sdev, since the returned value is not guaranteed and
1540 * may be changed after request stacking drivers call the function,
1541 * regardless of taking lock or not.
1543 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1544 * needs to return 'not busy'. Otherwise, request stacking drivers
1545 * may hold requests forever.
1547 static int scsi_lld_busy(struct request_queue *q)
1549 struct scsi_device *sdev = q->queuedata;
1550 struct Scsi_Host *shost;
1552 if (blk_queue_dying(q))
1558 * Ignore host/starget busy state.
1559 * Since block layer does not have a concept of fairness across
1560 * multiple queues, congestion of host/starget needs to be handled
1563 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1570 * Kill a request for a dead device
1572 static void scsi_kill_request(struct request *req, struct request_queue *q)
1574 struct scsi_cmnd *cmd = req->special;
1575 struct scsi_device *sdev;
1576 struct scsi_target *starget;
1577 struct Scsi_Host *shost;
1579 blk_start_request(req);
1581 scmd_printk(KERN_INFO, cmd, "killing request\n");
1584 starget = scsi_target(sdev);
1586 scsi_init_cmd_errh(cmd);
1587 cmd->result = DID_NO_CONNECT << 16;
1588 atomic_inc(&cmd->device->iorequest_cnt);
1591 * SCSI request completion path will do scsi_device_unbusy(),
1592 * bump busy counts. To bump the counters, we need to dance
1593 * with the locks as normal issue path does.
1595 atomic_inc(&sdev->device_busy);
1596 atomic_inc(&shost->host_busy);
1597 if (starget->can_queue > 0)
1598 atomic_inc(&starget->target_busy);
1600 blk_complete_request(req);
1603 static void scsi_softirq_done(struct request *rq)
1605 struct scsi_cmnd *cmd = rq->special;
1606 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1609 INIT_LIST_HEAD(&cmd->eh_entry);
1611 atomic_inc(&cmd->device->iodone_cnt);
1613 atomic_inc(&cmd->device->ioerr_cnt);
1615 disposition = scsi_decide_disposition(cmd);
1616 if (disposition != SUCCESS &&
1617 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1618 sdev_printk(KERN_ERR, cmd->device,
1619 "timing out command, waited %lus\n",
1621 disposition = SUCCESS;
1624 scsi_log_completion(cmd, disposition);
1626 switch (disposition) {
1628 scsi_finish_command(cmd);
1631 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1633 case ADD_TO_MLQUEUE:
1634 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1637 if (!scsi_eh_scmd_add(cmd, 0))
1638 scsi_finish_command(cmd);
1643 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1644 * @cmd: command block we are dispatching.
1646 * Return: nonzero return request was rejected and device's queue needs to be
1649 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1651 struct Scsi_Host *host = cmd->device->host;
1654 atomic_inc(&cmd->device->iorequest_cnt);
1656 /* check if the device is still usable */
1657 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1658 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1659 * returns an immediate error upwards, and signals
1660 * that the device is no longer present */
1661 cmd->result = DID_NO_CONNECT << 16;
1665 /* Check to see if the scsi lld made this device blocked. */
1666 if (unlikely(scsi_device_blocked(cmd->device))) {
1668 * in blocked state, the command is just put back on
1669 * the device queue. The suspend state has already
1670 * blocked the queue so future requests should not
1671 * occur until the device transitions out of the
1674 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1675 "queuecommand : device blocked\n"));
1676 return SCSI_MLQUEUE_DEVICE_BUSY;
1679 /* Store the LUN value in cmnd, if needed. */
1680 if (cmd->device->lun_in_cdb)
1681 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1682 (cmd->device->lun << 5 & 0xe0);
1687 * Before we queue this command, check if the command
1688 * length exceeds what the host adapter can handle.
1690 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1691 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1692 "queuecommand : command too long. "
1693 "cdb_size=%d host->max_cmd_len=%d\n",
1694 cmd->cmd_len, cmd->device->host->max_cmd_len));
1695 cmd->result = (DID_ABORT << 16);
1699 if (unlikely(host->shost_state == SHOST_DEL)) {
1700 cmd->result = (DID_NO_CONNECT << 16);
1705 trace_scsi_dispatch_cmd_start(cmd);
1706 rtn = host->hostt->queuecommand(host, cmd);
1708 trace_scsi_dispatch_cmd_error(cmd, rtn);
1709 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1710 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1711 rtn = SCSI_MLQUEUE_HOST_BUSY;
1713 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1714 "queuecommand : request rejected\n"));
1719 cmd->scsi_done(cmd);
1724 * scsi_done - Invoke completion on finished SCSI command.
1725 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1726 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1728 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1729 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1730 * calls blk_complete_request() for further processing.
1732 * This function is interrupt context safe.
1734 static void scsi_done(struct scsi_cmnd *cmd)
1736 trace_scsi_dispatch_cmd_done(cmd);
1737 blk_complete_request(cmd->request);
1741 * Function: scsi_request_fn()
1743 * Purpose: Main strategy routine for SCSI.
1745 * Arguments: q - Pointer to actual queue.
1749 * Lock status: IO request lock assumed to be held when called.
1751 static void scsi_request_fn(struct request_queue *q)
1752 __releases(q->queue_lock)
1753 __acquires(q->queue_lock)
1755 struct scsi_device *sdev = q->queuedata;
1756 struct Scsi_Host *shost;
1757 struct scsi_cmnd *cmd;
1758 struct request *req;
1761 * To start with, we keep looping until the queue is empty, or until
1762 * the host is no longer able to accept any more requests.
1768 * get next queueable request. We do this early to make sure
1769 * that the request is fully prepared even if we cannot
1772 req = blk_peek_request(q);
1776 if (unlikely(!scsi_device_online(sdev))) {
1777 sdev_printk(KERN_ERR, sdev,
1778 "rejecting I/O to offline device\n");
1779 scsi_kill_request(req, q);
1783 if (!scsi_dev_queue_ready(q, sdev))
1787 * Remove the request from the request list.
1789 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1790 blk_start_request(req);
1792 spin_unlock_irq(q->queue_lock);
1794 if (unlikely(cmd == NULL)) {
1795 printk(KERN_CRIT "impossible request in %s.\n"
1796 "please mail a stack trace to "
1797 "linux-scsi@vger.kernel.org\n",
1799 blk_dump_rq_flags(req, "foo");
1804 * We hit this when the driver is using a host wide
1805 * tag map. For device level tag maps the queue_depth check
1806 * in the device ready fn would prevent us from trying
1807 * to allocate a tag. Since the map is a shared host resource
1808 * we add the dev to the starved list so it eventually gets
1809 * a run when a tag is freed.
1811 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1812 spin_lock_irq(shost->host_lock);
1813 if (list_empty(&sdev->starved_entry))
1814 list_add_tail(&sdev->starved_entry,
1815 &shost->starved_list);
1816 spin_unlock_irq(shost->host_lock);
1820 if (!scsi_target_queue_ready(shost, sdev))
1823 if (!scsi_host_queue_ready(q, shost, sdev))
1824 goto host_not_ready;
1826 if (sdev->simple_tags)
1827 cmd->flags |= SCMD_TAGGED;
1829 cmd->flags &= ~SCMD_TAGGED;
1832 * Finally, initialize any error handling parameters, and set up
1833 * the timers for timeouts.
1835 scsi_init_cmd_errh(cmd);
1838 * Dispatch the command to the low-level driver.
1840 cmd->scsi_done = scsi_done;
1841 rtn = scsi_dispatch_cmd(cmd);
1843 scsi_queue_insert(cmd, rtn);
1844 spin_lock_irq(q->queue_lock);
1847 spin_lock_irq(q->queue_lock);
1853 if (scsi_target(sdev)->can_queue > 0)
1854 atomic_dec(&scsi_target(sdev)->target_busy);
1857 * lock q, handle tag, requeue req, and decrement device_busy. We
1858 * must return with queue_lock held.
1860 * Decrementing device_busy without checking it is OK, as all such
1861 * cases (host limits or settings) should run the queue at some
1864 spin_lock_irq(q->queue_lock);
1865 blk_requeue_request(q, req);
1866 atomic_dec(&sdev->device_busy);
1868 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1869 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1872 static inline int prep_to_mq(int ret)
1878 return BLK_MQ_RQ_QUEUE_BUSY;
1880 return BLK_MQ_RQ_QUEUE_ERROR;
1884 static int scsi_mq_prep_fn(struct request *req)
1886 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1887 struct scsi_device *sdev = req->q->queuedata;
1888 struct Scsi_Host *shost = sdev->host;
1889 unsigned char *sense_buf = cmd->sense_buffer;
1890 struct scatterlist *sg;
1892 memset(cmd, 0, sizeof(struct scsi_cmnd));
1898 cmd->sense_buffer = sense_buf;
1900 cmd->tag = req->tag;
1902 cmd->cmnd = req->cmd;
1903 cmd->prot_op = SCSI_PROT_NORMAL;
1905 INIT_LIST_HEAD(&cmd->list);
1906 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1907 cmd->jiffies_at_alloc = jiffies;
1909 if (shost->use_cmd_list) {
1910 spin_lock_irq(&sdev->list_lock);
1911 list_add_tail(&cmd->list, &sdev->cmd_list);
1912 spin_unlock_irq(&sdev->list_lock);
1915 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1916 cmd->sdb.table.sgl = sg;
1918 if (scsi_host_get_prot(shost)) {
1919 cmd->prot_sdb = (void *)sg +
1921 shost->sg_tablesize, SCSI_MAX_SG_SEGMENTS) *
1922 sizeof(struct scatterlist);
1923 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1925 cmd->prot_sdb->table.sgl =
1926 (struct scatterlist *)(cmd->prot_sdb + 1);
1929 if (blk_bidi_rq(req)) {
1930 struct request *next_rq = req->next_rq;
1931 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1933 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1934 bidi_sdb->table.sgl =
1935 (struct scatterlist *)(bidi_sdb + 1);
1937 next_rq->special = bidi_sdb;
1940 blk_mq_start_request(req);
1942 return scsi_setup_cmnd(sdev, req);
1945 static void scsi_mq_done(struct scsi_cmnd *cmd)
1947 trace_scsi_dispatch_cmd_done(cmd);
1948 blk_mq_complete_request(cmd->request);
1951 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1952 const struct blk_mq_queue_data *bd)
1954 struct request *req = bd->rq;
1955 struct request_queue *q = req->q;
1956 struct scsi_device *sdev = q->queuedata;
1957 struct Scsi_Host *shost = sdev->host;
1958 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1962 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1966 ret = BLK_MQ_RQ_QUEUE_BUSY;
1967 if (!get_device(&sdev->sdev_gendev))
1970 if (!scsi_dev_queue_ready(q, sdev))
1971 goto out_put_device;
1972 if (!scsi_target_queue_ready(shost, sdev))
1973 goto out_dec_device_busy;
1974 if (!scsi_host_queue_ready(q, shost, sdev))
1975 goto out_dec_target_busy;
1978 if (!(req->cmd_flags & REQ_DONTPREP)) {
1979 ret = prep_to_mq(scsi_mq_prep_fn(req));
1981 goto out_dec_host_busy;
1982 req->cmd_flags |= REQ_DONTPREP;
1984 blk_mq_start_request(req);
1987 if (sdev->simple_tags)
1988 cmd->flags |= SCMD_TAGGED;
1990 cmd->flags &= ~SCMD_TAGGED;
1992 scsi_init_cmd_errh(cmd);
1993 cmd->scsi_done = scsi_mq_done;
1995 reason = scsi_dispatch_cmd(cmd);
1997 scsi_set_blocked(cmd, reason);
1998 ret = BLK_MQ_RQ_QUEUE_BUSY;
1999 goto out_dec_host_busy;
2002 return BLK_MQ_RQ_QUEUE_OK;
2005 atomic_dec(&shost->host_busy);
2006 out_dec_target_busy:
2007 if (scsi_target(sdev)->can_queue > 0)
2008 atomic_dec(&scsi_target(sdev)->target_busy);
2009 out_dec_device_busy:
2010 atomic_dec(&sdev->device_busy);
2012 put_device(&sdev->sdev_gendev);
2015 case BLK_MQ_RQ_QUEUE_BUSY:
2016 blk_mq_stop_hw_queue(hctx);
2017 if (atomic_read(&sdev->device_busy) == 0 &&
2018 !scsi_device_blocked(sdev))
2019 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2021 case BLK_MQ_RQ_QUEUE_ERROR:
2023 * Make sure to release all allocated ressources when
2024 * we hit an error, as we will never see this command
2027 if (req->cmd_flags & REQ_DONTPREP)
2028 scsi_mq_uninit_cmd(cmd);
2036 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2040 return BLK_EH_RESET_TIMER;
2041 return scsi_times_out(req);
2044 static int scsi_init_request(void *data, struct request *rq,
2045 unsigned int hctx_idx, unsigned int request_idx,
2046 unsigned int numa_node)
2048 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2050 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
2052 if (!cmd->sense_buffer)
2057 static void scsi_exit_request(void *data, struct request *rq,
2058 unsigned int hctx_idx, unsigned int request_idx)
2060 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2062 kfree(cmd->sense_buffer);
2065 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2067 struct device *host_dev;
2068 u64 bounce_limit = 0xffffffff;
2070 if (shost->unchecked_isa_dma)
2071 return BLK_BOUNCE_ISA;
2073 * Platforms with virtual-DMA translation
2074 * hardware have no practical limit.
2076 if (!PCI_DMA_BUS_IS_PHYS)
2077 return BLK_BOUNCE_ANY;
2079 host_dev = scsi_get_device(shost);
2080 if (host_dev && host_dev->dma_mask)
2081 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2083 return bounce_limit;
2086 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2088 struct device *dev = shost->dma_dev;
2091 * this limit is imposed by hardware restrictions
2093 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2094 SCSI_MAX_SG_CHAIN_SEGMENTS));
2096 if (scsi_host_prot_dma(shost)) {
2097 shost->sg_prot_tablesize =
2098 min_not_zero(shost->sg_prot_tablesize,
2099 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2100 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2101 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2104 blk_queue_max_hw_sectors(q, shost->max_sectors);
2105 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2106 blk_queue_segment_boundary(q, shost->dma_boundary);
2107 dma_set_seg_boundary(dev, shost->dma_boundary);
2109 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2111 if (!shost->use_clustering)
2112 q->limits.cluster = 0;
2115 * set a reasonable default alignment on word boundaries: the
2116 * host and device may alter it using
2117 * blk_queue_update_dma_alignment() later.
2119 blk_queue_dma_alignment(q, 0x03);
2122 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2123 request_fn_proc *request_fn)
2125 struct request_queue *q;
2127 q = blk_init_queue(request_fn, NULL);
2130 __scsi_init_queue(shost, q);
2133 EXPORT_SYMBOL(__scsi_alloc_queue);
2135 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2137 struct request_queue *q;
2139 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2143 blk_queue_prep_rq(q, scsi_prep_fn);
2144 blk_queue_unprep_rq(q, scsi_unprep_fn);
2145 blk_queue_softirq_done(q, scsi_softirq_done);
2146 blk_queue_rq_timed_out(q, scsi_times_out);
2147 blk_queue_lld_busy(q, scsi_lld_busy);
2151 static struct blk_mq_ops scsi_mq_ops = {
2152 .map_queue = blk_mq_map_queue,
2153 .queue_rq = scsi_queue_rq,
2154 .complete = scsi_softirq_done,
2155 .timeout = scsi_timeout,
2156 .init_request = scsi_init_request,
2157 .exit_request = scsi_exit_request,
2160 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2162 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2163 if (IS_ERR(sdev->request_queue))
2166 sdev->request_queue->queuedata = sdev;
2167 __scsi_init_queue(sdev->host, sdev->request_queue);
2168 return sdev->request_queue;
2171 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2173 unsigned int cmd_size, sgl_size, tbl_size;
2175 tbl_size = shost->sg_tablesize;
2176 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2177 tbl_size = SCSI_MAX_SG_SEGMENTS;
2178 sgl_size = tbl_size * sizeof(struct scatterlist);
2179 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2180 if (scsi_host_get_prot(shost))
2181 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2183 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2184 shost->tag_set.ops = &scsi_mq_ops;
2185 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2186 shost->tag_set.queue_depth = shost->can_queue;
2187 shost->tag_set.cmd_size = cmd_size;
2188 shost->tag_set.numa_node = NUMA_NO_NODE;
2189 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2190 shost->tag_set.driver_data = shost;
2192 return blk_mq_alloc_tag_set(&shost->tag_set);
2195 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2197 blk_mq_free_tag_set(&shost->tag_set);
2201 * Function: scsi_block_requests()
2203 * Purpose: Utility function used by low-level drivers to prevent further
2204 * commands from being queued to the device.
2206 * Arguments: shost - Host in question
2210 * Lock status: No locks are assumed held.
2212 * Notes: There is no timer nor any other means by which the requests
2213 * get unblocked other than the low-level driver calling
2214 * scsi_unblock_requests().
2216 void scsi_block_requests(struct Scsi_Host *shost)
2218 shost->host_self_blocked = 1;
2220 EXPORT_SYMBOL(scsi_block_requests);
2223 * Function: scsi_unblock_requests()
2225 * Purpose: Utility function used by low-level drivers to allow further
2226 * commands from being queued to the device.
2228 * Arguments: shost - Host in question
2232 * Lock status: No locks are assumed held.
2234 * Notes: There is no timer nor any other means by which the requests
2235 * get unblocked other than the low-level driver calling
2236 * scsi_unblock_requests().
2238 * This is done as an API function so that changes to the
2239 * internals of the scsi mid-layer won't require wholesale
2240 * changes to drivers that use this feature.
2242 void scsi_unblock_requests(struct Scsi_Host *shost)
2244 shost->host_self_blocked = 0;
2245 scsi_run_host_queues(shost);
2247 EXPORT_SYMBOL(scsi_unblock_requests);
2249 int __init scsi_init_queue(void)
2253 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2254 sizeof(struct scsi_data_buffer),
2256 if (!scsi_sdb_cache) {
2257 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2261 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2262 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2263 int size = sgp->size * sizeof(struct scatterlist);
2265 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2266 SLAB_HWCACHE_ALIGN, NULL);
2268 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2273 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2276 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2285 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2286 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2288 mempool_destroy(sgp->pool);
2290 kmem_cache_destroy(sgp->slab);
2292 kmem_cache_destroy(scsi_sdb_cache);
2297 void scsi_exit_queue(void)
2301 kmem_cache_destroy(scsi_sdb_cache);
2303 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2304 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2305 mempool_destroy(sgp->pool);
2306 kmem_cache_destroy(sgp->slab);
2311 * scsi_mode_select - issue a mode select
2312 * @sdev: SCSI device to be queried
2313 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2314 * @sp: Save page bit (0 == don't save, 1 == save)
2315 * @modepage: mode page being requested
2316 * @buffer: request buffer (may not be smaller than eight bytes)
2317 * @len: length of request buffer.
2318 * @timeout: command timeout
2319 * @retries: number of retries before failing
2320 * @data: returns a structure abstracting the mode header data
2321 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2322 * must be SCSI_SENSE_BUFFERSIZE big.
2324 * Returns zero if successful; negative error number or scsi
2329 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2330 unsigned char *buffer, int len, int timeout, int retries,
2331 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2333 unsigned char cmd[10];
2334 unsigned char *real_buffer;
2337 memset(cmd, 0, sizeof(cmd));
2338 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2340 if (sdev->use_10_for_ms) {
2343 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2346 memcpy(real_buffer + 8, buffer, len);
2350 real_buffer[2] = data->medium_type;
2351 real_buffer[3] = data->device_specific;
2352 real_buffer[4] = data->longlba ? 0x01 : 0;
2354 real_buffer[6] = data->block_descriptor_length >> 8;
2355 real_buffer[7] = data->block_descriptor_length;
2357 cmd[0] = MODE_SELECT_10;
2361 if (len > 255 || data->block_descriptor_length > 255 ||
2365 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2368 memcpy(real_buffer + 4, buffer, len);
2371 real_buffer[1] = data->medium_type;
2372 real_buffer[2] = data->device_specific;
2373 real_buffer[3] = data->block_descriptor_length;
2376 cmd[0] = MODE_SELECT;
2380 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2381 sshdr, timeout, retries, NULL);
2385 EXPORT_SYMBOL_GPL(scsi_mode_select);
2388 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2389 * @sdev: SCSI device to be queried
2390 * @dbd: set if mode sense will allow block descriptors to be returned
2391 * @modepage: mode page being requested
2392 * @buffer: request buffer (may not be smaller than eight bytes)
2393 * @len: length of request buffer.
2394 * @timeout: command timeout
2395 * @retries: number of retries before failing
2396 * @data: returns a structure abstracting the mode header data
2397 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2398 * must be SCSI_SENSE_BUFFERSIZE big.
2400 * Returns zero if unsuccessful, or the header offset (either 4
2401 * or 8 depending on whether a six or ten byte command was
2402 * issued) if successful.
2405 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2406 unsigned char *buffer, int len, int timeout, int retries,
2407 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2409 unsigned char cmd[12];
2413 struct scsi_sense_hdr my_sshdr;
2415 memset(data, 0, sizeof(*data));
2416 memset(&cmd[0], 0, 12);
2417 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2420 /* caller might not be interested in sense, but we need it */
2425 use_10_for_ms = sdev->use_10_for_ms;
2427 if (use_10_for_ms) {
2431 cmd[0] = MODE_SENSE_10;
2438 cmd[0] = MODE_SENSE;
2443 memset(buffer, 0, len);
2445 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2446 sshdr, timeout, retries, NULL);
2448 /* This code looks awful: what it's doing is making sure an
2449 * ILLEGAL REQUEST sense return identifies the actual command
2450 * byte as the problem. MODE_SENSE commands can return
2451 * ILLEGAL REQUEST if the code page isn't supported */
2453 if (use_10_for_ms && !scsi_status_is_good(result) &&
2454 (driver_byte(result) & DRIVER_SENSE)) {
2455 if (scsi_sense_valid(sshdr)) {
2456 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2457 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2459 * Invalid command operation code
2461 sdev->use_10_for_ms = 0;
2467 if(scsi_status_is_good(result)) {
2468 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2469 (modepage == 6 || modepage == 8))) {
2470 /* Initio breakage? */
2473 data->medium_type = 0;
2474 data->device_specific = 0;
2476 data->block_descriptor_length = 0;
2477 } else if(use_10_for_ms) {
2478 data->length = buffer[0]*256 + buffer[1] + 2;
2479 data->medium_type = buffer[2];
2480 data->device_specific = buffer[3];
2481 data->longlba = buffer[4] & 0x01;
2482 data->block_descriptor_length = buffer[6]*256
2485 data->length = buffer[0] + 1;
2486 data->medium_type = buffer[1];
2487 data->device_specific = buffer[2];
2488 data->block_descriptor_length = buffer[3];
2490 data->header_length = header_length;
2495 EXPORT_SYMBOL(scsi_mode_sense);
2498 * scsi_test_unit_ready - test if unit is ready
2499 * @sdev: scsi device to change the state of.
2500 * @timeout: command timeout
2501 * @retries: number of retries before failing
2502 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2503 * returning sense. Make sure that this is cleared before passing
2506 * Returns zero if unsuccessful or an error if TUR failed. For
2507 * removable media, UNIT_ATTENTION sets ->changed flag.
2510 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2511 struct scsi_sense_hdr *sshdr_external)
2514 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2516 struct scsi_sense_hdr *sshdr;
2519 if (!sshdr_external)
2520 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2522 sshdr = sshdr_external;
2524 /* try to eat the UNIT_ATTENTION if there are enough retries */
2526 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2527 timeout, retries, NULL);
2528 if (sdev->removable && scsi_sense_valid(sshdr) &&
2529 sshdr->sense_key == UNIT_ATTENTION)
2531 } while (scsi_sense_valid(sshdr) &&
2532 sshdr->sense_key == UNIT_ATTENTION && --retries);
2534 if (!sshdr_external)
2538 EXPORT_SYMBOL(scsi_test_unit_ready);
2541 * scsi_device_set_state - Take the given device through the device state model.
2542 * @sdev: scsi device to change the state of.
2543 * @state: state to change to.
2545 * Returns zero if unsuccessful or an error if the requested
2546 * transition is illegal.
2549 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2551 enum scsi_device_state oldstate = sdev->sdev_state;
2553 if (state == oldstate)
2559 case SDEV_CREATED_BLOCK:
2570 case SDEV_TRANSPORT_OFFLINE:
2583 case SDEV_TRANSPORT_OFFLINE:
2591 case SDEV_TRANSPORT_OFFLINE:
2606 case SDEV_CREATED_BLOCK:
2613 case SDEV_CREATED_BLOCK:
2628 case SDEV_TRANSPORT_OFFLINE:
2641 case SDEV_TRANSPORT_OFFLINE:
2643 case SDEV_CREATED_BLOCK:
2651 sdev->sdev_state = state;
2655 SCSI_LOG_ERROR_RECOVERY(1,
2656 sdev_printk(KERN_ERR, sdev,
2657 "Illegal state transition %s->%s",
2658 scsi_device_state_name(oldstate),
2659 scsi_device_state_name(state))
2663 EXPORT_SYMBOL(scsi_device_set_state);
2666 * sdev_evt_emit - emit a single SCSI device uevent
2667 * @sdev: associated SCSI device
2668 * @evt: event to emit
2670 * Send a single uevent (scsi_event) to the associated scsi_device.
2672 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2677 switch (evt->evt_type) {
2678 case SDEV_EVT_MEDIA_CHANGE:
2679 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2681 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2682 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2684 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2685 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2687 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2688 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2690 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2691 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2693 case SDEV_EVT_LUN_CHANGE_REPORTED:
2694 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2703 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2707 * sdev_evt_thread - send a uevent for each scsi event
2708 * @work: work struct for scsi_device
2710 * Dispatch queued events to their associated scsi_device kobjects
2713 void scsi_evt_thread(struct work_struct *work)
2715 struct scsi_device *sdev;
2716 enum scsi_device_event evt_type;
2717 LIST_HEAD(event_list);
2719 sdev = container_of(work, struct scsi_device, event_work);
2721 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2722 if (test_and_clear_bit(evt_type, sdev->pending_events))
2723 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2726 struct scsi_event *evt;
2727 struct list_head *this, *tmp;
2728 unsigned long flags;
2730 spin_lock_irqsave(&sdev->list_lock, flags);
2731 list_splice_init(&sdev->event_list, &event_list);
2732 spin_unlock_irqrestore(&sdev->list_lock, flags);
2734 if (list_empty(&event_list))
2737 list_for_each_safe(this, tmp, &event_list) {
2738 evt = list_entry(this, struct scsi_event, node);
2739 list_del(&evt->node);
2740 scsi_evt_emit(sdev, evt);
2747 * sdev_evt_send - send asserted event to uevent thread
2748 * @sdev: scsi_device event occurred on
2749 * @evt: event to send
2751 * Assert scsi device event asynchronously.
2753 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2755 unsigned long flags;
2758 /* FIXME: currently this check eliminates all media change events
2759 * for polled devices. Need to update to discriminate between AN
2760 * and polled events */
2761 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2767 spin_lock_irqsave(&sdev->list_lock, flags);
2768 list_add_tail(&evt->node, &sdev->event_list);
2769 schedule_work(&sdev->event_work);
2770 spin_unlock_irqrestore(&sdev->list_lock, flags);
2772 EXPORT_SYMBOL_GPL(sdev_evt_send);
2775 * sdev_evt_alloc - allocate a new scsi event
2776 * @evt_type: type of event to allocate
2777 * @gfpflags: GFP flags for allocation
2779 * Allocates and returns a new scsi_event.
2781 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2784 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2788 evt->evt_type = evt_type;
2789 INIT_LIST_HEAD(&evt->node);
2791 /* evt_type-specific initialization, if any */
2793 case SDEV_EVT_MEDIA_CHANGE:
2794 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2795 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2796 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2797 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2798 case SDEV_EVT_LUN_CHANGE_REPORTED:
2806 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2809 * sdev_evt_send_simple - send asserted event to uevent thread
2810 * @sdev: scsi_device event occurred on
2811 * @evt_type: type of event to send
2812 * @gfpflags: GFP flags for allocation
2814 * Assert scsi device event asynchronously, given an event type.
2816 void sdev_evt_send_simple(struct scsi_device *sdev,
2817 enum scsi_device_event evt_type, gfp_t gfpflags)
2819 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2821 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2826 sdev_evt_send(sdev, evt);
2828 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2831 * scsi_device_quiesce - Block user issued commands.
2832 * @sdev: scsi device to quiesce.
2834 * This works by trying to transition to the SDEV_QUIESCE state
2835 * (which must be a legal transition). When the device is in this
2836 * state, only special requests will be accepted, all others will
2837 * be deferred. Since special requests may also be requeued requests,
2838 * a successful return doesn't guarantee the device will be
2839 * totally quiescent.
2841 * Must be called with user context, may sleep.
2843 * Returns zero if unsuccessful or an error if not.
2846 scsi_device_quiesce(struct scsi_device *sdev)
2848 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2852 scsi_run_queue(sdev->request_queue);
2853 while (atomic_read(&sdev->device_busy)) {
2854 msleep_interruptible(200);
2855 scsi_run_queue(sdev->request_queue);
2859 EXPORT_SYMBOL(scsi_device_quiesce);
2862 * scsi_device_resume - Restart user issued commands to a quiesced device.
2863 * @sdev: scsi device to resume.
2865 * Moves the device from quiesced back to running and restarts the
2868 * Must be called with user context, may sleep.
2870 void scsi_device_resume(struct scsi_device *sdev)
2872 /* check if the device state was mutated prior to resume, and if
2873 * so assume the state is being managed elsewhere (for example
2874 * device deleted during suspend)
2876 if (sdev->sdev_state != SDEV_QUIESCE ||
2877 scsi_device_set_state(sdev, SDEV_RUNNING))
2879 scsi_run_queue(sdev->request_queue);
2881 EXPORT_SYMBOL(scsi_device_resume);
2884 device_quiesce_fn(struct scsi_device *sdev, void *data)
2886 scsi_device_quiesce(sdev);
2890 scsi_target_quiesce(struct scsi_target *starget)
2892 starget_for_each_device(starget, NULL, device_quiesce_fn);
2894 EXPORT_SYMBOL(scsi_target_quiesce);
2897 device_resume_fn(struct scsi_device *sdev, void *data)
2899 scsi_device_resume(sdev);
2903 scsi_target_resume(struct scsi_target *starget)
2905 starget_for_each_device(starget, NULL, device_resume_fn);
2907 EXPORT_SYMBOL(scsi_target_resume);
2910 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2911 * @sdev: device to block
2913 * Block request made by scsi lld's to temporarily stop all
2914 * scsi commands on the specified device. Called from interrupt
2915 * or normal process context.
2917 * Returns zero if successful or error if not
2920 * This routine transitions the device to the SDEV_BLOCK state
2921 * (which must be a legal transition). When the device is in this
2922 * state, all commands are deferred until the scsi lld reenables
2923 * the device with scsi_device_unblock or device_block_tmo fires.
2926 scsi_internal_device_block(struct scsi_device *sdev)
2928 struct request_queue *q = sdev->request_queue;
2929 unsigned long flags;
2932 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2934 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2941 * The device has transitioned to SDEV_BLOCK. Stop the
2942 * block layer from calling the midlayer with this device's
2946 blk_mq_stop_hw_queues(q);
2948 spin_lock_irqsave(q->queue_lock, flags);
2950 spin_unlock_irqrestore(q->queue_lock, flags);
2955 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2958 * scsi_internal_device_unblock - resume a device after a block request
2959 * @sdev: device to resume
2960 * @new_state: state to set devices to after unblocking
2962 * Called by scsi lld's or the midlayer to restart the device queue
2963 * for the previously suspended scsi device. Called from interrupt or
2964 * normal process context.
2966 * Returns zero if successful or error if not.
2969 * This routine transitions the device to the SDEV_RUNNING state
2970 * or to one of the offline states (which must be a legal transition)
2971 * allowing the midlayer to goose the queue for this device.
2974 scsi_internal_device_unblock(struct scsi_device *sdev,
2975 enum scsi_device_state new_state)
2977 struct request_queue *q = sdev->request_queue;
2978 unsigned long flags;
2981 * Try to transition the scsi device to SDEV_RUNNING or one of the
2982 * offlined states and goose the device queue if successful.
2984 if ((sdev->sdev_state == SDEV_BLOCK) ||
2985 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2986 sdev->sdev_state = new_state;
2987 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2988 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2989 new_state == SDEV_OFFLINE)
2990 sdev->sdev_state = new_state;
2992 sdev->sdev_state = SDEV_CREATED;
2993 } else if (sdev->sdev_state != SDEV_CANCEL &&
2994 sdev->sdev_state != SDEV_OFFLINE)
2998 blk_mq_start_stopped_hw_queues(q, false);
3000 spin_lock_irqsave(q->queue_lock, flags);
3002 spin_unlock_irqrestore(q->queue_lock, flags);
3007 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3010 device_block(struct scsi_device *sdev, void *data)
3012 scsi_internal_device_block(sdev);
3016 target_block(struct device *dev, void *data)
3018 if (scsi_is_target_device(dev))
3019 starget_for_each_device(to_scsi_target(dev), NULL,
3025 scsi_target_block(struct device *dev)
3027 if (scsi_is_target_device(dev))
3028 starget_for_each_device(to_scsi_target(dev), NULL,
3031 device_for_each_child(dev, NULL, target_block);
3033 EXPORT_SYMBOL_GPL(scsi_target_block);
3036 device_unblock(struct scsi_device *sdev, void *data)
3038 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3042 target_unblock(struct device *dev, void *data)
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), data,
3051 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), &new_state,
3057 device_for_each_child(dev, &new_state, target_unblock);
3059 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3062 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3063 * @sgl: scatter-gather list
3064 * @sg_count: number of segments in sg
3065 * @offset: offset in bytes into sg, on return offset into the mapped area
3066 * @len: bytes to map, on return number of bytes mapped
3068 * Returns virtual address of the start of the mapped page
3070 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3071 size_t *offset, size_t *len)
3074 size_t sg_len = 0, len_complete = 0;
3075 struct scatterlist *sg;
3078 WARN_ON(!irqs_disabled());
3080 for_each_sg(sgl, sg, sg_count, i) {
3081 len_complete = sg_len; /* Complete sg-entries */
3082 sg_len += sg->length;
3083 if (sg_len > *offset)
3087 if (unlikely(i == sg_count)) {
3088 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3090 __func__, sg_len, *offset, sg_count);
3095 /* Offset starting from the beginning of first page in this sg-entry */
3096 *offset = *offset - len_complete + sg->offset;
3098 /* Assumption: contiguous pages can be accessed as "page + i" */
3099 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3100 *offset &= ~PAGE_MASK;
3102 /* Bytes in this sg-entry from *offset to the end of the page */
3103 sg_len = PAGE_SIZE - *offset;
3107 return kmap_atomic(page);
3109 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3112 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3113 * @virt: virtual address to be unmapped
3115 void scsi_kunmap_atomic_sg(void *virt)
3117 kunmap_atomic(virt);
3119 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3121 void sdev_disable_disk_events(struct scsi_device *sdev)
3123 atomic_inc(&sdev->disk_events_disable_depth);
3125 EXPORT_SYMBOL(sdev_disable_disk_events);
3127 void sdev_enable_disk_events(struct scsi_device *sdev)
3129 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3131 atomic_dec(&sdev->disk_events_disable_depth);
3133 EXPORT_SYMBOL(sdev_enable_disk_events);