2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/init.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/hardirq.h>
23 #include <linux/scatterlist.h>
24 #include <linux/blk-mq.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
34 #include <trace/events/scsi.h>
36 #include "scsi_priv.h"
37 #include "scsi_logging.h"
40 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
41 #define SG_MEMPOOL_SIZE 2
43 struct scsi_host_sg_pool {
46 struct kmem_cache *slab;
50 #define SP(x) { .size = x, "sgpool-" __stringify(x) }
51 #if (SCSI_MAX_SG_SEGMENTS < 32)
52 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
54 static struct scsi_host_sg_pool scsi_sg_pools[] = {
57 #if (SCSI_MAX_SG_SEGMENTS > 32)
59 #if (SCSI_MAX_SG_SEGMENTS > 64)
61 #if (SCSI_MAX_SG_SEGMENTS > 128)
63 #if (SCSI_MAX_SG_SEGMENTS > 256)
64 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
69 SP(SCSI_MAX_SG_SEGMENTS)
73 struct kmem_cache *scsi_sdb_cache;
76 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
77 * not change behaviour from the previous unplug mechanism, experimentation
78 * may prove this needs changing.
80 #define SCSI_QUEUE_DELAY 3
83 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
85 struct Scsi_Host *host = cmd->device->host;
86 struct scsi_device *device = cmd->device;
87 struct scsi_target *starget = scsi_target(device);
90 * Set the appropriate busy bit for the device/host.
92 * If the host/device isn't busy, assume that something actually
93 * completed, and that we should be able to queue a command now.
95 * Note that the prior mid-layer assumption that any host could
96 * always queue at least one command is now broken. The mid-layer
97 * will implement a user specifiable stall (see
98 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
99 * if a command is requeued with no other commands outstanding
100 * either for the device or for the host.
103 case SCSI_MLQUEUE_HOST_BUSY:
104 atomic_set(&host->host_blocked, host->max_host_blocked);
106 case SCSI_MLQUEUE_DEVICE_BUSY:
107 case SCSI_MLQUEUE_EH_RETRY:
108 atomic_set(&device->device_blocked,
109 device->max_device_blocked);
111 case SCSI_MLQUEUE_TARGET_BUSY:
112 atomic_set(&starget->target_blocked,
113 starget->max_target_blocked);
118 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
120 struct scsi_device *sdev = cmd->device;
121 struct request_queue *q = cmd->request->q;
123 blk_mq_requeue_request(cmd->request);
124 blk_mq_kick_requeue_list(q);
125 put_device(&sdev->sdev_gendev);
129 * __scsi_queue_insert - private queue insertion
130 * @cmd: The SCSI command being requeued
131 * @reason: The reason for the requeue
132 * @unbusy: Whether the queue should be unbusied
134 * This is a private queue insertion. The public interface
135 * scsi_queue_insert() always assumes the queue should be unbusied
136 * because it's always called before the completion. This function is
137 * for a requeue after completion, which should only occur in this
140 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
142 struct scsi_device *device = cmd->device;
143 struct request_queue *q = device->request_queue;
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 "Inserting command %p into mlqueue\n", cmd));
149 scsi_set_blocked(cmd, reason);
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
156 scsi_device_unbusy(device);
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_cleanup_queue() finishes.
166 scsi_mq_requeue_cmd(cmd);
169 spin_lock_irqsave(q->queue_lock, flags);
170 blk_requeue_request(q, cmd->request);
171 kblockd_schedule_work(&device->requeue_work);
172 spin_unlock_irqrestore(q->queue_lock, flags);
176 * Function: scsi_queue_insert()
178 * Purpose: Insert a command in the midlevel queue.
180 * Arguments: cmd - command that we are adding to queue.
181 * reason - why we are inserting command to queue.
183 * Lock status: Assumed that lock is not held upon entry.
187 * Notes: We do this for one of two cases. Either the host is busy
188 * and it cannot accept any more commands for the time being,
189 * or the device returned QUEUE_FULL and can accept no more
191 * Notes: This could be called either from an interrupt context or a
192 * normal process context.
194 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
196 __scsi_queue_insert(cmd, reason, 1);
199 * scsi_execute - insert request and wait for the result
202 * @data_direction: data direction
203 * @buffer: data buffer
204 * @bufflen: len of buffer
205 * @sense: optional sense buffer
206 * @timeout: request timeout in seconds
207 * @retries: number of times to retry request
208 * @flags: or into request flags;
209 * @resid: optional residual length
211 * returns the req->errors value which is the scsi_cmnd result
214 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
215 int data_direction, void *buffer, unsigned bufflen,
216 unsigned char *sense, int timeout, int retries, u64 flags,
220 int write = (data_direction == DMA_TO_DEVICE);
221 int ret = DRIVER_ERROR << 24;
223 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
226 blk_rq_set_block_pc(req);
228 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
229 buffer, bufflen, __GFP_WAIT))
232 req->cmd_len = COMMAND_SIZE(cmd[0]);
233 memcpy(req->cmd, cmd, req->cmd_len);
236 req->retries = retries;
237 req->timeout = timeout;
238 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
241 * head injection *required* here otherwise quiesce won't work
243 blk_execute_rq(req->q, NULL, req, 1);
246 * Some devices (USB mass-storage in particular) may transfer
247 * garbage data together with a residue indicating that the data
248 * is invalid. Prevent the garbage from being misinterpreted
249 * and prevent security leaks by zeroing out the excess data.
251 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
252 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
255 *resid = req->resid_len;
258 blk_put_request(req);
262 EXPORT_SYMBOL(scsi_execute);
264 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
265 int data_direction, void *buffer, unsigned bufflen,
266 struct scsi_sense_hdr *sshdr, int timeout, int retries,
267 int *resid, u64 flags)
273 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
275 return DRIVER_ERROR << 24;
277 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
278 sense, timeout, retries, flags, resid);
280 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
285 EXPORT_SYMBOL(scsi_execute_req_flags);
288 * Function: scsi_init_cmd_errh()
290 * Purpose: Initialize cmd fields related to error handling.
292 * Arguments: cmd - command that is ready to be queued.
294 * Notes: This function has the job of initializing a number of
295 * fields related to error handling. Typically this will
296 * be called once for each command, as required.
298 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
300 cmd->serial_number = 0;
301 scsi_set_resid(cmd, 0);
302 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
303 if (cmd->cmd_len == 0)
304 cmd->cmd_len = scsi_command_size(cmd->cmnd);
307 void scsi_device_unbusy(struct scsi_device *sdev)
309 struct Scsi_Host *shost = sdev->host;
310 struct scsi_target *starget = scsi_target(sdev);
313 atomic_dec(&shost->host_busy);
314 if (starget->can_queue > 0)
315 atomic_dec(&starget->target_busy);
317 if (unlikely(scsi_host_in_recovery(shost) &&
318 (shost->host_failed || shost->host_eh_scheduled))) {
319 spin_lock_irqsave(shost->host_lock, flags);
320 scsi_eh_wakeup(shost);
321 spin_unlock_irqrestore(shost->host_lock, flags);
324 atomic_dec(&sdev->device_busy);
327 static void scsi_kick_queue(struct request_queue *q)
330 blk_mq_start_hw_queues(q);
336 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
337 * and call blk_run_queue for all the scsi_devices on the target -
338 * including current_sdev first.
340 * Called with *no* scsi locks held.
342 static void scsi_single_lun_run(struct scsi_device *current_sdev)
344 struct Scsi_Host *shost = current_sdev->host;
345 struct scsi_device *sdev, *tmp;
346 struct scsi_target *starget = scsi_target(current_sdev);
349 spin_lock_irqsave(shost->host_lock, flags);
350 starget->starget_sdev_user = NULL;
351 spin_unlock_irqrestore(shost->host_lock, flags);
354 * Call blk_run_queue for all LUNs on the target, starting with
355 * current_sdev. We race with others (to set starget_sdev_user),
356 * but in most cases, we will be first. Ideally, each LU on the
357 * target would get some limited time or requests on the target.
359 scsi_kick_queue(current_sdev->request_queue);
361 spin_lock_irqsave(shost->host_lock, flags);
362 if (starget->starget_sdev_user)
364 list_for_each_entry_safe(sdev, tmp, &starget->devices,
365 same_target_siblings) {
366 if (sdev == current_sdev)
368 if (scsi_device_get(sdev))
371 spin_unlock_irqrestore(shost->host_lock, flags);
372 scsi_kick_queue(sdev->request_queue);
373 spin_lock_irqsave(shost->host_lock, flags);
375 scsi_device_put(sdev);
378 spin_unlock_irqrestore(shost->host_lock, flags);
381 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
383 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
385 if (atomic_read(&sdev->device_blocked) > 0)
390 static inline bool scsi_target_is_busy(struct scsi_target *starget)
392 if (starget->can_queue > 0) {
393 if (atomic_read(&starget->target_busy) >= starget->can_queue)
395 if (atomic_read(&starget->target_blocked) > 0)
401 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
403 if (shost->can_queue > 0 &&
404 atomic_read(&shost->host_busy) >= shost->can_queue)
406 if (atomic_read(&shost->host_blocked) > 0)
408 if (shost->host_self_blocked)
413 static void scsi_starved_list_run(struct Scsi_Host *shost)
415 LIST_HEAD(starved_list);
416 struct scsi_device *sdev;
419 spin_lock_irqsave(shost->host_lock, flags);
420 list_splice_init(&shost->starved_list, &starved_list);
422 while (!list_empty(&starved_list)) {
423 struct request_queue *slq;
426 * As long as shost is accepting commands and we have
427 * starved queues, call blk_run_queue. scsi_request_fn
428 * drops the queue_lock and can add us back to the
431 * host_lock protects the starved_list and starved_entry.
432 * scsi_request_fn must get the host_lock before checking
433 * or modifying starved_list or starved_entry.
435 if (scsi_host_is_busy(shost))
438 sdev = list_entry(starved_list.next,
439 struct scsi_device, starved_entry);
440 list_del_init(&sdev->starved_entry);
441 if (scsi_target_is_busy(scsi_target(sdev))) {
442 list_move_tail(&sdev->starved_entry,
443 &shost->starved_list);
448 * Once we drop the host lock, a racing scsi_remove_device()
449 * call may remove the sdev from the starved list and destroy
450 * it and the queue. Mitigate by taking a reference to the
451 * queue and never touching the sdev again after we drop the
452 * host lock. Note: if __scsi_remove_device() invokes
453 * blk_cleanup_queue() before the queue is run from this
454 * function then blk_run_queue() will return immediately since
455 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
457 slq = sdev->request_queue;
458 if (!blk_get_queue(slq))
460 spin_unlock_irqrestore(shost->host_lock, flags);
462 scsi_kick_queue(slq);
465 spin_lock_irqsave(shost->host_lock, flags);
467 /* put any unprocessed entries back */
468 list_splice(&starved_list, &shost->starved_list);
469 spin_unlock_irqrestore(shost->host_lock, flags);
473 * Function: scsi_run_queue()
475 * Purpose: Select a proper request queue to serve next
477 * Arguments: q - last request's queue
481 * Notes: The previous command was completely finished, start
482 * a new one if possible.
484 static void scsi_run_queue(struct request_queue *q)
486 struct scsi_device *sdev = q->queuedata;
488 if (scsi_target(sdev)->single_lun)
489 scsi_single_lun_run(sdev);
490 if (!list_empty(&sdev->host->starved_list))
491 scsi_starved_list_run(sdev->host);
494 blk_mq_start_stopped_hw_queues(q, false);
499 void scsi_requeue_run_queue(struct work_struct *work)
501 struct scsi_device *sdev;
502 struct request_queue *q;
504 sdev = container_of(work, struct scsi_device, requeue_work);
505 q = sdev->request_queue;
510 * Function: scsi_requeue_command()
512 * Purpose: Handle post-processing of completed commands.
514 * Arguments: q - queue to operate on
515 * cmd - command that may need to be requeued.
519 * Notes: After command completion, there may be blocks left
520 * over which weren't finished by the previous command
521 * this can be for a number of reasons - the main one is
522 * I/O errors in the middle of the request, in which case
523 * we need to request the blocks that come after the bad
525 * Notes: Upon return, cmd is a stale pointer.
527 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
529 struct scsi_device *sdev = cmd->device;
530 struct request *req = cmd->request;
533 spin_lock_irqsave(q->queue_lock, flags);
534 blk_unprep_request(req);
536 scsi_put_command(cmd);
537 blk_requeue_request(q, req);
538 spin_unlock_irqrestore(q->queue_lock, flags);
542 put_device(&sdev->sdev_gendev);
545 void scsi_next_command(struct scsi_cmnd *cmd)
547 struct scsi_device *sdev = cmd->device;
548 struct request_queue *q = sdev->request_queue;
550 scsi_put_command(cmd);
553 put_device(&sdev->sdev_gendev);
556 void scsi_run_host_queues(struct Scsi_Host *shost)
558 struct scsi_device *sdev;
560 shost_for_each_device(sdev, shost)
561 scsi_run_queue(sdev->request_queue);
564 static inline unsigned int scsi_sgtable_index(unsigned short nents)
568 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
573 index = get_count_order(nents) - 3;
578 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
580 struct scsi_host_sg_pool *sgp;
582 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
583 mempool_free(sgl, sgp->pool);
586 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
588 struct scsi_host_sg_pool *sgp;
590 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
591 return mempool_alloc(sgp->pool, gfp_mask);
594 static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
596 if (mq && sdb->table.nents <= SCSI_MAX_SG_SEGMENTS)
598 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
601 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
602 gfp_t gfp_mask, bool mq)
604 struct scatterlist *first_chunk = NULL;
610 if (nents <= SCSI_MAX_SG_SEGMENTS) {
611 sdb->table.nents = nents;
612 sg_init_table(sdb->table.sgl, sdb->table.nents);
615 first_chunk = sdb->table.sgl;
618 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
619 first_chunk, gfp_mask, scsi_sg_alloc);
621 scsi_free_sgtable(sdb, mq);
625 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
627 if (cmd->request->cmd_type == REQ_TYPE_FS) {
628 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
630 if (drv->uninit_command)
631 drv->uninit_command(cmd);
635 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
637 if (cmd->sdb.table.nents)
638 scsi_free_sgtable(&cmd->sdb, true);
639 if (cmd->request->next_rq && cmd->request->next_rq->special)
640 scsi_free_sgtable(cmd->request->next_rq->special, true);
641 if (scsi_prot_sg_count(cmd))
642 scsi_free_sgtable(cmd->prot_sdb, true);
645 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
647 struct scsi_device *sdev = cmd->device;
648 struct Scsi_Host *shost = sdev->host;
651 scsi_mq_free_sgtables(cmd);
652 scsi_uninit_cmd(cmd);
654 if (shost->use_cmd_list) {
655 BUG_ON(list_empty(&cmd->list));
656 spin_lock_irqsave(&sdev->list_lock, flags);
657 list_del_init(&cmd->list);
658 spin_unlock_irqrestore(&sdev->list_lock, flags);
663 * Function: scsi_release_buffers()
665 * Purpose: Free resources allocate for a scsi_command.
667 * Arguments: cmd - command that we are bailing.
669 * Lock status: Assumed that no lock is held upon entry.
673 * Notes: In the event that an upper level driver rejects a
674 * command, we must release resources allocated during
675 * the __init_io() function. Primarily this would involve
676 * the scatter-gather table.
678 static void scsi_release_buffers(struct scsi_cmnd *cmd)
680 if (cmd->sdb.table.nents)
681 scsi_free_sgtable(&cmd->sdb, false);
683 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
685 if (scsi_prot_sg_count(cmd))
686 scsi_free_sgtable(cmd->prot_sdb, false);
689 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
691 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
693 scsi_free_sgtable(bidi_sdb, false);
694 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
695 cmd->request->next_rq->special = NULL;
698 static bool scsi_end_request(struct request *req, int error,
699 unsigned int bytes, unsigned int bidi_bytes)
701 struct scsi_cmnd *cmd = req->special;
702 struct scsi_device *sdev = cmd->device;
703 struct request_queue *q = sdev->request_queue;
705 if (blk_update_request(req, error, bytes))
708 /* Bidi request must be completed as a whole */
709 if (unlikely(bidi_bytes) &&
710 blk_update_request(req->next_rq, error, bidi_bytes))
713 if (blk_queue_add_random(q))
714 add_disk_randomness(req->rq_disk);
718 * In the MQ case the command gets freed by __blk_mq_end_request,
719 * so we have to do all cleanup that depends on it earlier.
721 * We also can't kick the queues from irq context, so we
722 * will have to defer it to a workqueue.
724 scsi_mq_uninit_cmd(cmd);
726 __blk_mq_end_request(req, error);
728 if (scsi_target(sdev)->single_lun ||
729 !list_empty(&sdev->host->starved_list))
730 kblockd_schedule_work(&sdev->requeue_work);
732 blk_mq_start_stopped_hw_queues(q, true);
734 put_device(&sdev->sdev_gendev);
739 scsi_release_bidi_buffers(cmd);
741 spin_lock_irqsave(q->queue_lock, flags);
742 blk_finish_request(req, error);
743 spin_unlock_irqrestore(q->queue_lock, flags);
745 scsi_release_buffers(cmd);
746 scsi_next_command(cmd);
753 * __scsi_error_from_host_byte - translate SCSI error code into errno
754 * @cmd: SCSI command (unused)
755 * @result: scsi error code
757 * Translate SCSI error code into standard UNIX errno.
759 * -ENOLINK temporary transport failure
760 * -EREMOTEIO permanent target failure, do not retry
761 * -EBADE permanent nexus failure, retry on other path
762 * -ENOSPC No write space available
763 * -ENODATA Medium error
764 * -EIO unspecified I/O error
766 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
770 switch(host_byte(result)) {
771 case DID_TRANSPORT_FAILFAST:
774 case DID_TARGET_FAILURE:
775 set_host_byte(cmd, DID_OK);
778 case DID_NEXUS_FAILURE:
779 set_host_byte(cmd, DID_OK);
782 case DID_ALLOC_FAILURE:
783 set_host_byte(cmd, DID_OK);
786 case DID_MEDIUM_ERROR:
787 set_host_byte(cmd, DID_OK);
799 * Function: scsi_io_completion()
801 * Purpose: Completion processing for block device I/O requests.
803 * Arguments: cmd - command that is finished.
805 * Lock status: Assumed that no lock is held upon entry.
809 * Notes: We will finish off the specified number of sectors. If we
810 * are done, the command block will be released and the queue
811 * function will be goosed. If we are not done then we have to
812 * figure out what to do next:
814 * a) We can call scsi_requeue_command(). The request
815 * will be unprepared and put back on the queue. Then
816 * a new command will be created for it. This should
817 * be used if we made forward progress, or if we want
818 * to switch from READ(10) to READ(6) for example.
820 * b) We can call __scsi_queue_insert(). The request will
821 * be put back on the queue and retried using the same
822 * command as before, possibly after a delay.
824 * c) We can call scsi_end_request() with -EIO to fail
825 * the remainder of the request.
827 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
829 int result = cmd->result;
830 struct request_queue *q = cmd->device->request_queue;
831 struct request *req = cmd->request;
833 struct scsi_sense_hdr sshdr;
834 bool sense_valid = false;
835 int sense_deferred = 0, level = 0;
836 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
837 ACTION_DELAYED_RETRY} action;
838 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
841 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
843 sense_deferred = scsi_sense_is_deferred(&sshdr);
846 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
848 if (sense_valid && req->sense) {
850 * SG_IO wants current and deferred errors
852 int len = 8 + cmd->sense_buffer[7];
854 if (len > SCSI_SENSE_BUFFERSIZE)
855 len = SCSI_SENSE_BUFFERSIZE;
856 memcpy(req->sense, cmd->sense_buffer, len);
857 req->sense_len = len;
860 error = __scsi_error_from_host_byte(cmd, result);
863 * __scsi_error_from_host_byte may have reset the host_byte
865 req->errors = cmd->result;
867 req->resid_len = scsi_get_resid(cmd);
869 if (scsi_bidi_cmnd(cmd)) {
871 * Bidi commands Must be complete as a whole,
872 * both sides at once.
874 req->next_rq->resid_len = scsi_in(cmd)->resid;
875 if (scsi_end_request(req, 0, blk_rq_bytes(req),
876 blk_rq_bytes(req->next_rq)))
880 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
882 * Certain non BLOCK_PC requests are commands that don't
883 * actually transfer anything (FLUSH), so cannot use
884 * good_bytes != blk_rq_bytes(req) as the signal for an error.
885 * This sets the error explicitly for the problem case.
887 error = __scsi_error_from_host_byte(cmd, result);
890 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
891 BUG_ON(blk_bidi_rq(req));
894 * Next deal with any sectors which we were able to correctly
897 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
898 "%u sectors total, %d bytes done.\n",
899 blk_rq_sectors(req), good_bytes));
902 * Recovered errors need reporting, but they're always treated
903 * as success, so fiddle the result code here. For BLOCK_PC
904 * we already took a copy of the original into rq->errors which
905 * is what gets returned to the user
907 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
908 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
909 * print since caller wants ATA registers. Only occurs on
910 * SCSI ATA PASS_THROUGH commands when CK_COND=1
912 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
914 else if (!(req->cmd_flags & REQ_QUIET))
915 scsi_print_sense(cmd);
917 /* BLOCK_PC may have set error */
922 * If we finished all bytes in the request we are done now.
924 if (!scsi_end_request(req, error, good_bytes, 0))
928 * Kill remainder if no retrys.
930 if (error && scsi_noretry_cmd(cmd)) {
931 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
937 * If there had been no error, but we have leftover bytes in the
938 * requeues just queue the command up again.
943 error = __scsi_error_from_host_byte(cmd, result);
945 if (host_byte(result) == DID_RESET) {
946 /* Third party bus reset or reset for error recovery
947 * reasons. Just retry the command and see what
950 action = ACTION_RETRY;
951 } else if (sense_valid && !sense_deferred) {
952 switch (sshdr.sense_key) {
954 if (cmd->device->removable) {
955 /* Detected disc change. Set a bit
956 * and quietly refuse further access.
958 cmd->device->changed = 1;
959 action = ACTION_FAIL;
961 /* Must have been a power glitch, or a
962 * bus reset. Could not have been a
963 * media change, so we just retry the
964 * command and see what happens.
966 action = ACTION_RETRY;
969 case ILLEGAL_REQUEST:
970 /* If we had an ILLEGAL REQUEST returned, then
971 * we may have performed an unsupported
972 * command. The only thing this should be
973 * would be a ten byte read where only a six
974 * byte read was supported. Also, on a system
975 * where READ CAPACITY failed, we may have
976 * read past the end of the disk.
978 if ((cmd->device->use_10_for_rw &&
979 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
980 (cmd->cmnd[0] == READ_10 ||
981 cmd->cmnd[0] == WRITE_10)) {
982 /* This will issue a new 6-byte command. */
983 cmd->device->use_10_for_rw = 0;
984 action = ACTION_REPREP;
985 } else if (sshdr.asc == 0x10) /* DIX */ {
986 action = ACTION_FAIL;
988 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
989 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
990 action = ACTION_FAIL;
993 action = ACTION_FAIL;
995 case ABORTED_COMMAND:
996 action = ACTION_FAIL;
997 if (sshdr.asc == 0x10) /* DIF */
1001 /* If the device is in the process of becoming
1002 * ready, or has a temporary blockage, retry.
1004 if (sshdr.asc == 0x04) {
1005 switch (sshdr.ascq) {
1006 case 0x01: /* becoming ready */
1007 case 0x04: /* format in progress */
1008 case 0x05: /* rebuild in progress */
1009 case 0x06: /* recalculation in progress */
1010 case 0x07: /* operation in progress */
1011 case 0x08: /* Long write in progress */
1012 case 0x09: /* self test in progress */
1013 case 0x14: /* space allocation in progress */
1014 action = ACTION_DELAYED_RETRY;
1017 action = ACTION_FAIL;
1021 action = ACTION_FAIL;
1023 case VOLUME_OVERFLOW:
1024 /* See SSC3rXX or current. */
1025 action = ACTION_FAIL;
1028 action = ACTION_FAIL;
1032 action = ACTION_FAIL;
1034 if (action != ACTION_FAIL &&
1035 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1036 action = ACTION_FAIL;
1040 /* Give up and fail the remainder of the request */
1041 if (unlikely(scsi_logging_level))
1042 level = SCSI_LOG_LEVEL(SCSI_LOG_MLQUEUE_SHIFT,
1043 SCSI_LOG_MLQUEUE_BITS);
1045 * if logging is enabled the failure will be printed
1046 * in scsi_log_completion(), so avoid duplicate messages
1048 if (!level && !(req->cmd_flags & REQ_QUIET)) {
1049 scsi_print_result(cmd, NULL, FAILED);
1050 if (driver_byte(result) & DRIVER_SENSE)
1051 scsi_print_sense(cmd);
1052 scsi_print_command(cmd);
1054 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1059 /* Unprep the request and put it back at the head of the queue.
1060 * A new command will be prepared and issued.
1063 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1064 scsi_mq_uninit_cmd(cmd);
1065 scsi_mq_requeue_cmd(cmd);
1067 scsi_release_buffers(cmd);
1068 scsi_requeue_command(q, cmd);
1072 /* Retry the same command immediately */
1073 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1075 case ACTION_DELAYED_RETRY:
1076 /* Retry the same command after a delay */
1077 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1082 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 gfp_mask, 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, gfp_t gfp_mask)
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, gfp_mask);
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,
1147 if (blk_integrity_rq(rq)) {
1148 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1151 BUG_ON(prot_sdb == NULL);
1152 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1154 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask, is_mq)) {
1155 error = BLKPREP_DEFER;
1159 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1160 prot_sdb->table.sgl);
1161 BUG_ON(unlikely(count > ivecs));
1162 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1164 cmd->prot_sdb = prot_sdb;
1165 cmd->prot_sdb->table.nents = count;
1171 scsi_mq_free_sgtables(cmd);
1173 scsi_release_buffers(cmd);
1174 cmd->request->special = NULL;
1175 scsi_put_command(cmd);
1176 put_device(&sdev->sdev_gendev);
1180 EXPORT_SYMBOL(scsi_init_io);
1182 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1183 struct request *req)
1185 struct scsi_cmnd *cmd;
1187 if (!req->special) {
1188 /* Bail if we can't get a reference to the device */
1189 if (!get_device(&sdev->sdev_gendev))
1192 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1193 if (unlikely(!cmd)) {
1194 put_device(&sdev->sdev_gendev);
1202 /* pull a tag out of the request if we have one */
1203 cmd->tag = req->tag;
1206 cmd->cmnd = req->cmd;
1207 cmd->prot_op = SCSI_PROT_NORMAL;
1212 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1214 struct scsi_cmnd *cmd = req->special;
1217 * BLOCK_PC requests may transfer data, in which case they must
1218 * a bio attached to them. Or they might contain a SCSI command
1219 * that does not transfer data, in which case they may optionally
1220 * submit a request without an attached bio.
1223 int ret = scsi_init_io(cmd, GFP_ATOMIC);
1227 BUG_ON(blk_rq_bytes(req));
1229 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1232 cmd->cmd_len = req->cmd_len;
1233 cmd->transfersize = blk_rq_bytes(req);
1234 cmd->allowed = req->retries;
1239 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1240 * that still need to be translated to SCSI CDBs from the ULD.
1242 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1244 struct scsi_cmnd *cmd = req->special;
1246 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1247 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1248 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1249 if (ret != BLKPREP_OK)
1253 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1254 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1257 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1259 struct scsi_cmnd *cmd = req->special;
1261 if (!blk_rq_bytes(req))
1262 cmd->sc_data_direction = DMA_NONE;
1263 else if (rq_data_dir(req) == WRITE)
1264 cmd->sc_data_direction = DMA_TO_DEVICE;
1266 cmd->sc_data_direction = DMA_FROM_DEVICE;
1268 switch (req->cmd_type) {
1270 return scsi_setup_fs_cmnd(sdev, req);
1271 case REQ_TYPE_BLOCK_PC:
1272 return scsi_setup_blk_pc_cmnd(sdev, req);
1274 return BLKPREP_KILL;
1279 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1281 int ret = BLKPREP_OK;
1284 * If the device is not in running state we will reject some
1287 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1288 switch (sdev->sdev_state) {
1290 case SDEV_TRANSPORT_OFFLINE:
1292 * If the device is offline we refuse to process any
1293 * commands. The device must be brought online
1294 * before trying any recovery commands.
1296 sdev_printk(KERN_ERR, sdev,
1297 "rejecting I/O to offline device\n");
1302 * If the device is fully deleted, we refuse to
1303 * process any commands as well.
1305 sdev_printk(KERN_ERR, sdev,
1306 "rejecting I/O to dead device\n");
1311 case SDEV_CREATED_BLOCK:
1313 * If the devices is blocked we defer normal commands.
1315 if (!(req->cmd_flags & REQ_PREEMPT))
1316 ret = BLKPREP_DEFER;
1320 * For any other not fully online state we only allow
1321 * special commands. In particular any user initiated
1322 * command is not allowed.
1324 if (!(req->cmd_flags & REQ_PREEMPT))
1333 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1335 struct scsi_device *sdev = q->queuedata;
1339 req->errors = DID_NO_CONNECT << 16;
1340 /* release the command and kill it */
1342 struct scsi_cmnd *cmd = req->special;
1343 scsi_release_buffers(cmd);
1344 scsi_put_command(cmd);
1345 put_device(&sdev->sdev_gendev);
1346 req->special = NULL;
1351 * If we defer, the blk_peek_request() returns NULL, but the
1352 * queue must be restarted, so we schedule a callback to happen
1355 if (atomic_read(&sdev->device_busy) == 0)
1356 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1359 req->cmd_flags |= REQ_DONTPREP;
1365 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1367 struct scsi_device *sdev = q->queuedata;
1368 struct scsi_cmnd *cmd;
1371 ret = scsi_prep_state_check(sdev, req);
1372 if (ret != BLKPREP_OK)
1375 cmd = scsi_get_cmd_from_req(sdev, req);
1376 if (unlikely(!cmd)) {
1377 ret = BLKPREP_DEFER;
1381 ret = scsi_setup_cmnd(sdev, req);
1383 return scsi_prep_return(q, req, ret);
1386 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1388 scsi_uninit_cmd(req->special);
1392 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1395 * Called with the queue_lock held.
1397 static inline int scsi_dev_queue_ready(struct request_queue *q,
1398 struct scsi_device *sdev)
1402 busy = atomic_inc_return(&sdev->device_busy) - 1;
1403 if (atomic_read(&sdev->device_blocked)) {
1408 * unblock after device_blocked iterates to zero
1410 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1412 * For the MQ case we take care of this in the caller.
1415 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1418 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1419 "unblocking device at zero depth\n"));
1422 if (busy >= sdev->queue_depth)
1427 atomic_dec(&sdev->device_busy);
1432 * scsi_target_queue_ready: checks if there we can send commands to target
1433 * @sdev: scsi device on starget to check.
1435 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1436 struct scsi_device *sdev)
1438 struct scsi_target *starget = scsi_target(sdev);
1441 if (starget->single_lun) {
1442 spin_lock_irq(shost->host_lock);
1443 if (starget->starget_sdev_user &&
1444 starget->starget_sdev_user != sdev) {
1445 spin_unlock_irq(shost->host_lock);
1448 starget->starget_sdev_user = sdev;
1449 spin_unlock_irq(shost->host_lock);
1452 if (starget->can_queue <= 0)
1455 busy = atomic_inc_return(&starget->target_busy) - 1;
1456 if (atomic_read(&starget->target_blocked) > 0) {
1461 * unblock after target_blocked iterates to zero
1463 if (atomic_dec_return(&starget->target_blocked) > 0)
1466 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1467 "unblocking target at zero depth\n"));
1470 if (busy >= starget->can_queue)
1476 spin_lock_irq(shost->host_lock);
1477 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1478 spin_unlock_irq(shost->host_lock);
1480 if (starget->can_queue > 0)
1481 atomic_dec(&starget->target_busy);
1486 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1487 * return 0. We must end up running the queue again whenever 0 is
1488 * returned, else IO can hang.
1490 static inline int scsi_host_queue_ready(struct request_queue *q,
1491 struct Scsi_Host *shost,
1492 struct scsi_device *sdev)
1496 if (scsi_host_in_recovery(shost))
1499 busy = atomic_inc_return(&shost->host_busy) - 1;
1500 if (atomic_read(&shost->host_blocked) > 0) {
1505 * unblock after host_blocked iterates to zero
1507 if (atomic_dec_return(&shost->host_blocked) > 0)
1511 shost_printk(KERN_INFO, shost,
1512 "unblocking host at zero depth\n"));
1515 if (shost->can_queue > 0 && busy >= shost->can_queue)
1517 if (shost->host_self_blocked)
1520 /* We're OK to process the command, so we can't be starved */
1521 if (!list_empty(&sdev->starved_entry)) {
1522 spin_lock_irq(shost->host_lock);
1523 if (!list_empty(&sdev->starved_entry))
1524 list_del_init(&sdev->starved_entry);
1525 spin_unlock_irq(shost->host_lock);
1531 spin_lock_irq(shost->host_lock);
1532 if (list_empty(&sdev->starved_entry))
1533 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1534 spin_unlock_irq(shost->host_lock);
1536 atomic_dec(&shost->host_busy);
1541 * Busy state exporting function for request stacking drivers.
1543 * For efficiency, no lock is taken to check the busy state of
1544 * shost/starget/sdev, since the returned value is not guaranteed and
1545 * may be changed after request stacking drivers call the function,
1546 * regardless of taking lock or not.
1548 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1549 * needs to return 'not busy'. Otherwise, request stacking drivers
1550 * may hold requests forever.
1552 static int scsi_lld_busy(struct request_queue *q)
1554 struct scsi_device *sdev = q->queuedata;
1555 struct Scsi_Host *shost;
1557 if (blk_queue_dying(q))
1563 * Ignore host/starget busy state.
1564 * Since block layer does not have a concept of fairness across
1565 * multiple queues, congestion of host/starget needs to be handled
1568 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1575 * Kill a request for a dead device
1577 static void scsi_kill_request(struct request *req, struct request_queue *q)
1579 struct scsi_cmnd *cmd = req->special;
1580 struct scsi_device *sdev;
1581 struct scsi_target *starget;
1582 struct Scsi_Host *shost;
1584 blk_start_request(req);
1586 scmd_printk(KERN_INFO, cmd, "killing request\n");
1589 starget = scsi_target(sdev);
1591 scsi_init_cmd_errh(cmd);
1592 cmd->result = DID_NO_CONNECT << 16;
1593 atomic_inc(&cmd->device->iorequest_cnt);
1596 * SCSI request completion path will do scsi_device_unbusy(),
1597 * bump busy counts. To bump the counters, we need to dance
1598 * with the locks as normal issue path does.
1600 atomic_inc(&sdev->device_busy);
1601 atomic_inc(&shost->host_busy);
1602 if (starget->can_queue > 0)
1603 atomic_inc(&starget->target_busy);
1605 blk_complete_request(req);
1608 static void scsi_softirq_done(struct request *rq)
1610 struct scsi_cmnd *cmd = rq->special;
1611 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1614 INIT_LIST_HEAD(&cmd->eh_entry);
1616 atomic_inc(&cmd->device->iodone_cnt);
1618 atomic_inc(&cmd->device->ioerr_cnt);
1620 disposition = scsi_decide_disposition(cmd);
1621 if (disposition != SUCCESS &&
1622 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1623 sdev_printk(KERN_ERR, cmd->device,
1624 "timing out command, waited %lus\n",
1626 disposition = SUCCESS;
1629 scsi_log_completion(cmd, disposition);
1631 switch (disposition) {
1633 scsi_finish_command(cmd);
1636 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1638 case ADD_TO_MLQUEUE:
1639 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1642 if (!scsi_eh_scmd_add(cmd, 0))
1643 scsi_finish_command(cmd);
1648 * scsi_done - Invoke completion on finished SCSI command.
1649 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1650 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1652 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1653 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1654 * calls blk_complete_request() for further processing.
1656 * This function is interrupt context safe.
1658 static void scsi_done(struct scsi_cmnd *cmd)
1660 trace_scsi_dispatch_cmd_done(cmd);
1661 blk_complete_request(cmd->request);
1665 * Function: scsi_request_fn()
1667 * Purpose: Main strategy routine for SCSI.
1669 * Arguments: q - Pointer to actual queue.
1673 * Lock status: IO request lock assumed to be held when called.
1675 static void scsi_request_fn(struct request_queue *q)
1676 __releases(q->queue_lock)
1677 __acquires(q->queue_lock)
1679 struct scsi_device *sdev = q->queuedata;
1680 struct Scsi_Host *shost;
1681 struct scsi_cmnd *cmd;
1682 struct request *req;
1685 * To start with, we keep looping until the queue is empty, or until
1686 * the host is no longer able to accept any more requests.
1692 * get next queueable request. We do this early to make sure
1693 * that the request is fully prepared even if we cannot
1696 req = blk_peek_request(q);
1700 if (unlikely(!scsi_device_online(sdev))) {
1701 sdev_printk(KERN_ERR, sdev,
1702 "rejecting I/O to offline device\n");
1703 scsi_kill_request(req, q);
1707 if (!scsi_dev_queue_ready(q, sdev))
1711 * Remove the request from the request list.
1713 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1714 blk_start_request(req);
1716 spin_unlock_irq(q->queue_lock);
1718 if (unlikely(cmd == NULL)) {
1719 printk(KERN_CRIT "impossible request in %s.\n"
1720 "please mail a stack trace to "
1721 "linux-scsi@vger.kernel.org\n",
1723 blk_dump_rq_flags(req, "foo");
1728 * We hit this when the driver is using a host wide
1729 * tag map. For device level tag maps the queue_depth check
1730 * in the device ready fn would prevent us from trying
1731 * to allocate a tag. Since the map is a shared host resource
1732 * we add the dev to the starved list so it eventually gets
1733 * a run when a tag is freed.
1735 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1736 spin_lock_irq(shost->host_lock);
1737 if (list_empty(&sdev->starved_entry))
1738 list_add_tail(&sdev->starved_entry,
1739 &shost->starved_list);
1740 spin_unlock_irq(shost->host_lock);
1744 if (!scsi_target_queue_ready(shost, sdev))
1747 if (!scsi_host_queue_ready(q, shost, sdev))
1748 goto host_not_ready;
1751 * Finally, initialize any error handling parameters, and set up
1752 * the timers for timeouts.
1754 scsi_init_cmd_errh(cmd);
1757 * Dispatch the command to the low-level driver.
1759 cmd->scsi_done = scsi_done;
1760 rtn = scsi_dispatch_cmd(cmd);
1762 scsi_queue_insert(cmd, rtn);
1763 spin_lock_irq(q->queue_lock);
1766 spin_lock_irq(q->queue_lock);
1772 if (scsi_target(sdev)->can_queue > 0)
1773 atomic_dec(&scsi_target(sdev)->target_busy);
1776 * lock q, handle tag, requeue req, and decrement device_busy. We
1777 * must return with queue_lock held.
1779 * Decrementing device_busy without checking it is OK, as all such
1780 * cases (host limits or settings) should run the queue at some
1783 spin_lock_irq(q->queue_lock);
1784 blk_requeue_request(q, req);
1785 atomic_dec(&sdev->device_busy);
1787 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1788 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1791 static inline int prep_to_mq(int ret)
1797 return BLK_MQ_RQ_QUEUE_BUSY;
1799 return BLK_MQ_RQ_QUEUE_ERROR;
1803 static int scsi_mq_prep_fn(struct request *req)
1805 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1806 struct scsi_device *sdev = req->q->queuedata;
1807 struct Scsi_Host *shost = sdev->host;
1808 unsigned char *sense_buf = cmd->sense_buffer;
1809 struct scatterlist *sg;
1811 memset(cmd, 0, sizeof(struct scsi_cmnd));
1817 cmd->sense_buffer = sense_buf;
1819 cmd->tag = req->tag;
1821 cmd->cmnd = req->cmd;
1822 cmd->prot_op = SCSI_PROT_NORMAL;
1824 INIT_LIST_HEAD(&cmd->list);
1825 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1826 cmd->jiffies_at_alloc = jiffies;
1828 if (shost->use_cmd_list) {
1829 spin_lock_irq(&sdev->list_lock);
1830 list_add_tail(&cmd->list, &sdev->cmd_list);
1831 spin_unlock_irq(&sdev->list_lock);
1834 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1835 cmd->sdb.table.sgl = sg;
1837 if (scsi_host_get_prot(shost)) {
1838 cmd->prot_sdb = (void *)sg +
1839 shost->sg_tablesize * sizeof(struct scatterlist);
1840 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1842 cmd->prot_sdb->table.sgl =
1843 (struct scatterlist *)(cmd->prot_sdb + 1);
1846 if (blk_bidi_rq(req)) {
1847 struct request *next_rq = req->next_rq;
1848 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1850 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1851 bidi_sdb->table.sgl =
1852 (struct scatterlist *)(bidi_sdb + 1);
1854 next_rq->special = bidi_sdb;
1857 blk_mq_start_request(req);
1859 return scsi_setup_cmnd(sdev, req);
1862 static void scsi_mq_done(struct scsi_cmnd *cmd)
1864 trace_scsi_dispatch_cmd_done(cmd);
1865 blk_mq_complete_request(cmd->request);
1868 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req,
1871 struct request_queue *q = req->q;
1872 struct scsi_device *sdev = q->queuedata;
1873 struct Scsi_Host *shost = sdev->host;
1874 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1878 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1882 ret = BLK_MQ_RQ_QUEUE_BUSY;
1883 if (!get_device(&sdev->sdev_gendev))
1886 if (!scsi_dev_queue_ready(q, sdev))
1887 goto out_put_device;
1888 if (!scsi_target_queue_ready(shost, sdev))
1889 goto out_dec_device_busy;
1890 if (!scsi_host_queue_ready(q, shost, sdev))
1891 goto out_dec_target_busy;
1894 if (!(req->cmd_flags & REQ_DONTPREP)) {
1895 ret = prep_to_mq(scsi_mq_prep_fn(req));
1897 goto out_dec_host_busy;
1898 req->cmd_flags |= REQ_DONTPREP;
1900 blk_mq_start_request(req);
1903 if (blk_queue_tagged(q))
1904 req->cmd_flags |= REQ_QUEUED;
1906 req->cmd_flags &= ~REQ_QUEUED;
1908 scsi_init_cmd_errh(cmd);
1909 cmd->scsi_done = scsi_mq_done;
1911 reason = scsi_dispatch_cmd(cmd);
1913 scsi_set_blocked(cmd, reason);
1914 ret = BLK_MQ_RQ_QUEUE_BUSY;
1915 goto out_dec_host_busy;
1918 return BLK_MQ_RQ_QUEUE_OK;
1921 atomic_dec(&shost->host_busy);
1922 out_dec_target_busy:
1923 if (scsi_target(sdev)->can_queue > 0)
1924 atomic_dec(&scsi_target(sdev)->target_busy);
1925 out_dec_device_busy:
1926 atomic_dec(&sdev->device_busy);
1928 put_device(&sdev->sdev_gendev);
1931 case BLK_MQ_RQ_QUEUE_BUSY:
1932 blk_mq_stop_hw_queue(hctx);
1933 if (atomic_read(&sdev->device_busy) == 0 &&
1934 !scsi_device_blocked(sdev))
1935 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
1937 case BLK_MQ_RQ_QUEUE_ERROR:
1939 * Make sure to release all allocated ressources when
1940 * we hit an error, as we will never see this command
1943 if (req->cmd_flags & REQ_DONTPREP)
1944 scsi_mq_uninit_cmd(cmd);
1952 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1956 return BLK_EH_RESET_TIMER;
1957 return scsi_times_out(req);
1960 static int scsi_init_request(void *data, struct request *rq,
1961 unsigned int hctx_idx, unsigned int request_idx,
1962 unsigned int numa_node)
1964 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1966 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
1968 if (!cmd->sense_buffer)
1973 static void scsi_exit_request(void *data, struct request *rq,
1974 unsigned int hctx_idx, unsigned int request_idx)
1976 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1978 kfree(cmd->sense_buffer);
1981 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1983 struct device *host_dev;
1984 u64 bounce_limit = 0xffffffff;
1986 if (shost->unchecked_isa_dma)
1987 return BLK_BOUNCE_ISA;
1989 * Platforms with virtual-DMA translation
1990 * hardware have no practical limit.
1992 if (!PCI_DMA_BUS_IS_PHYS)
1993 return BLK_BOUNCE_ANY;
1995 host_dev = scsi_get_device(shost);
1996 if (host_dev && host_dev->dma_mask)
1997 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1999 return bounce_limit;
2002 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2004 struct device *dev = shost->dma_dev;
2007 * this limit is imposed by hardware restrictions
2009 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2010 SCSI_MAX_SG_CHAIN_SEGMENTS));
2012 if (scsi_host_prot_dma(shost)) {
2013 shost->sg_prot_tablesize =
2014 min_not_zero(shost->sg_prot_tablesize,
2015 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2016 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2017 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2020 blk_queue_max_hw_sectors(q, shost->max_sectors);
2021 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2022 blk_queue_segment_boundary(q, shost->dma_boundary);
2023 dma_set_seg_boundary(dev, shost->dma_boundary);
2025 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2027 if (!shost->use_clustering)
2028 q->limits.cluster = 0;
2031 * set a reasonable default alignment on word boundaries: the
2032 * host and device may alter it using
2033 * blk_queue_update_dma_alignment() later.
2035 blk_queue_dma_alignment(q, 0x03);
2038 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2039 request_fn_proc *request_fn)
2041 struct request_queue *q;
2043 q = blk_init_queue(request_fn, NULL);
2046 __scsi_init_queue(shost, q);
2049 EXPORT_SYMBOL(__scsi_alloc_queue);
2051 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2053 struct request_queue *q;
2055 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2059 blk_queue_prep_rq(q, scsi_prep_fn);
2060 blk_queue_unprep_rq(q, scsi_unprep_fn);
2061 blk_queue_softirq_done(q, scsi_softirq_done);
2062 blk_queue_rq_timed_out(q, scsi_times_out);
2063 blk_queue_lld_busy(q, scsi_lld_busy);
2067 static struct blk_mq_ops scsi_mq_ops = {
2068 .map_queue = blk_mq_map_queue,
2069 .queue_rq = scsi_queue_rq,
2070 .complete = scsi_softirq_done,
2071 .timeout = scsi_timeout,
2072 .init_request = scsi_init_request,
2073 .exit_request = scsi_exit_request,
2076 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2078 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2079 if (IS_ERR(sdev->request_queue))
2082 sdev->request_queue->queuedata = sdev;
2083 __scsi_init_queue(sdev->host, sdev->request_queue);
2084 return sdev->request_queue;
2087 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2089 unsigned int cmd_size, sgl_size, tbl_size;
2091 tbl_size = shost->sg_tablesize;
2092 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2093 tbl_size = SCSI_MAX_SG_SEGMENTS;
2094 sgl_size = tbl_size * sizeof(struct scatterlist);
2095 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2096 if (scsi_host_get_prot(shost))
2097 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2099 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2100 shost->tag_set.ops = &scsi_mq_ops;
2101 shost->tag_set.nr_hw_queues = 1;
2102 shost->tag_set.queue_depth = shost->can_queue;
2103 shost->tag_set.cmd_size = cmd_size;
2104 shost->tag_set.numa_node = NUMA_NO_NODE;
2105 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2106 shost->tag_set.driver_data = shost;
2108 return blk_mq_alloc_tag_set(&shost->tag_set);
2111 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2113 blk_mq_free_tag_set(&shost->tag_set);
2117 * Function: scsi_block_requests()
2119 * Purpose: Utility function used by low-level drivers to prevent further
2120 * commands from being queued to the device.
2122 * Arguments: shost - Host in question
2126 * Lock status: No locks are assumed held.
2128 * Notes: There is no timer nor any other means by which the requests
2129 * get unblocked other than the low-level driver calling
2130 * scsi_unblock_requests().
2132 void scsi_block_requests(struct Scsi_Host *shost)
2134 shost->host_self_blocked = 1;
2136 EXPORT_SYMBOL(scsi_block_requests);
2139 * Function: scsi_unblock_requests()
2141 * Purpose: Utility function used by low-level drivers to allow further
2142 * commands from being queued to the device.
2144 * Arguments: shost - Host in question
2148 * Lock status: No locks are assumed held.
2150 * Notes: There is no timer nor any other means by which the requests
2151 * get unblocked other than the low-level driver calling
2152 * scsi_unblock_requests().
2154 * This is done as an API function so that changes to the
2155 * internals of the scsi mid-layer won't require wholesale
2156 * changes to drivers that use this feature.
2158 void scsi_unblock_requests(struct Scsi_Host *shost)
2160 shost->host_self_blocked = 0;
2161 scsi_run_host_queues(shost);
2163 EXPORT_SYMBOL(scsi_unblock_requests);
2165 int __init scsi_init_queue(void)
2169 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2170 sizeof(struct scsi_data_buffer),
2172 if (!scsi_sdb_cache) {
2173 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2177 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2178 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2179 int size = sgp->size * sizeof(struct scatterlist);
2181 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2182 SLAB_HWCACHE_ALIGN, NULL);
2184 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2189 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2192 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2201 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2202 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2204 mempool_destroy(sgp->pool);
2206 kmem_cache_destroy(sgp->slab);
2208 kmem_cache_destroy(scsi_sdb_cache);
2213 void scsi_exit_queue(void)
2217 kmem_cache_destroy(scsi_sdb_cache);
2219 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2220 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2221 mempool_destroy(sgp->pool);
2222 kmem_cache_destroy(sgp->slab);
2227 * scsi_mode_select - issue a mode select
2228 * @sdev: SCSI device to be queried
2229 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2230 * @sp: Save page bit (0 == don't save, 1 == save)
2231 * @modepage: mode page being requested
2232 * @buffer: request buffer (may not be smaller than eight bytes)
2233 * @len: length of request buffer.
2234 * @timeout: command timeout
2235 * @retries: number of retries before failing
2236 * @data: returns a structure abstracting the mode header data
2237 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2238 * must be SCSI_SENSE_BUFFERSIZE big.
2240 * Returns zero if successful; negative error number or scsi
2245 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2246 unsigned char *buffer, int len, int timeout, int retries,
2247 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2249 unsigned char cmd[10];
2250 unsigned char *real_buffer;
2253 memset(cmd, 0, sizeof(cmd));
2254 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2256 if (sdev->use_10_for_ms) {
2259 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2262 memcpy(real_buffer + 8, buffer, len);
2266 real_buffer[2] = data->medium_type;
2267 real_buffer[3] = data->device_specific;
2268 real_buffer[4] = data->longlba ? 0x01 : 0;
2270 real_buffer[6] = data->block_descriptor_length >> 8;
2271 real_buffer[7] = data->block_descriptor_length;
2273 cmd[0] = MODE_SELECT_10;
2277 if (len > 255 || data->block_descriptor_length > 255 ||
2281 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2284 memcpy(real_buffer + 4, buffer, len);
2287 real_buffer[1] = data->medium_type;
2288 real_buffer[2] = data->device_specific;
2289 real_buffer[3] = data->block_descriptor_length;
2292 cmd[0] = MODE_SELECT;
2296 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2297 sshdr, timeout, retries, NULL);
2301 EXPORT_SYMBOL_GPL(scsi_mode_select);
2304 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2305 * @sdev: SCSI device to be queried
2306 * @dbd: set if mode sense will allow block descriptors to be returned
2307 * @modepage: mode page being requested
2308 * @buffer: request buffer (may not be smaller than eight bytes)
2309 * @len: length of request buffer.
2310 * @timeout: command timeout
2311 * @retries: number of retries before failing
2312 * @data: returns a structure abstracting the mode header data
2313 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2314 * must be SCSI_SENSE_BUFFERSIZE big.
2316 * Returns zero if unsuccessful, or the header offset (either 4
2317 * or 8 depending on whether a six or ten byte command was
2318 * issued) if successful.
2321 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2322 unsigned char *buffer, int len, int timeout, int retries,
2323 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2325 unsigned char cmd[12];
2329 struct scsi_sense_hdr my_sshdr;
2331 memset(data, 0, sizeof(*data));
2332 memset(&cmd[0], 0, 12);
2333 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2336 /* caller might not be interested in sense, but we need it */
2341 use_10_for_ms = sdev->use_10_for_ms;
2343 if (use_10_for_ms) {
2347 cmd[0] = MODE_SENSE_10;
2354 cmd[0] = MODE_SENSE;
2359 memset(buffer, 0, len);
2361 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2362 sshdr, timeout, retries, NULL);
2364 /* This code looks awful: what it's doing is making sure an
2365 * ILLEGAL REQUEST sense return identifies the actual command
2366 * byte as the problem. MODE_SENSE commands can return
2367 * ILLEGAL REQUEST if the code page isn't supported */
2369 if (use_10_for_ms && !scsi_status_is_good(result) &&
2370 (driver_byte(result) & DRIVER_SENSE)) {
2371 if (scsi_sense_valid(sshdr)) {
2372 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2373 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2375 * Invalid command operation code
2377 sdev->use_10_for_ms = 0;
2383 if(scsi_status_is_good(result)) {
2384 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2385 (modepage == 6 || modepage == 8))) {
2386 /* Initio breakage? */
2389 data->medium_type = 0;
2390 data->device_specific = 0;
2392 data->block_descriptor_length = 0;
2393 } else if(use_10_for_ms) {
2394 data->length = buffer[0]*256 + buffer[1] + 2;
2395 data->medium_type = buffer[2];
2396 data->device_specific = buffer[3];
2397 data->longlba = buffer[4] & 0x01;
2398 data->block_descriptor_length = buffer[6]*256
2401 data->length = buffer[0] + 1;
2402 data->medium_type = buffer[1];
2403 data->device_specific = buffer[2];
2404 data->block_descriptor_length = buffer[3];
2406 data->header_length = header_length;
2411 EXPORT_SYMBOL(scsi_mode_sense);
2414 * scsi_test_unit_ready - test if unit is ready
2415 * @sdev: scsi device to change the state of.
2416 * @timeout: command timeout
2417 * @retries: number of retries before failing
2418 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2419 * returning sense. Make sure that this is cleared before passing
2422 * Returns zero if unsuccessful or an error if TUR failed. For
2423 * removable media, UNIT_ATTENTION sets ->changed flag.
2426 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2427 struct scsi_sense_hdr *sshdr_external)
2430 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2432 struct scsi_sense_hdr *sshdr;
2435 if (!sshdr_external)
2436 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2438 sshdr = sshdr_external;
2440 /* try to eat the UNIT_ATTENTION if there are enough retries */
2442 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2443 timeout, retries, NULL);
2444 if (sdev->removable && scsi_sense_valid(sshdr) &&
2445 sshdr->sense_key == UNIT_ATTENTION)
2447 } while (scsi_sense_valid(sshdr) &&
2448 sshdr->sense_key == UNIT_ATTENTION && --retries);
2450 if (!sshdr_external)
2454 EXPORT_SYMBOL(scsi_test_unit_ready);
2457 * scsi_device_set_state - Take the given device through the device state model.
2458 * @sdev: scsi device to change the state of.
2459 * @state: state to change to.
2461 * Returns zero if unsuccessful or an error if the requested
2462 * transition is illegal.
2465 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2467 enum scsi_device_state oldstate = sdev->sdev_state;
2469 if (state == oldstate)
2475 case SDEV_CREATED_BLOCK:
2486 case SDEV_TRANSPORT_OFFLINE:
2499 case SDEV_TRANSPORT_OFFLINE:
2507 case SDEV_TRANSPORT_OFFLINE:
2522 case SDEV_CREATED_BLOCK:
2529 case SDEV_CREATED_BLOCK:
2544 case SDEV_TRANSPORT_OFFLINE:
2557 case SDEV_TRANSPORT_OFFLINE:
2559 case SDEV_CREATED_BLOCK:
2567 sdev->sdev_state = state;
2571 SCSI_LOG_ERROR_RECOVERY(1,
2572 sdev_printk(KERN_ERR, sdev,
2573 "Illegal state transition %s->%s",
2574 scsi_device_state_name(oldstate),
2575 scsi_device_state_name(state))
2579 EXPORT_SYMBOL(scsi_device_set_state);
2582 * sdev_evt_emit - emit a single SCSI device uevent
2583 * @sdev: associated SCSI device
2584 * @evt: event to emit
2586 * Send a single uevent (scsi_event) to the associated scsi_device.
2588 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2593 switch (evt->evt_type) {
2594 case SDEV_EVT_MEDIA_CHANGE:
2595 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2597 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2598 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2600 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2601 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2603 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2604 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2606 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2607 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2609 case SDEV_EVT_LUN_CHANGE_REPORTED:
2610 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2619 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2623 * sdev_evt_thread - send a uevent for each scsi event
2624 * @work: work struct for scsi_device
2626 * Dispatch queued events to their associated scsi_device kobjects
2629 void scsi_evt_thread(struct work_struct *work)
2631 struct scsi_device *sdev;
2632 enum scsi_device_event evt_type;
2633 LIST_HEAD(event_list);
2635 sdev = container_of(work, struct scsi_device, event_work);
2637 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2638 if (test_and_clear_bit(evt_type, sdev->pending_events))
2639 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2642 struct scsi_event *evt;
2643 struct list_head *this, *tmp;
2644 unsigned long flags;
2646 spin_lock_irqsave(&sdev->list_lock, flags);
2647 list_splice_init(&sdev->event_list, &event_list);
2648 spin_unlock_irqrestore(&sdev->list_lock, flags);
2650 if (list_empty(&event_list))
2653 list_for_each_safe(this, tmp, &event_list) {
2654 evt = list_entry(this, struct scsi_event, node);
2655 list_del(&evt->node);
2656 scsi_evt_emit(sdev, evt);
2663 * sdev_evt_send - send asserted event to uevent thread
2664 * @sdev: scsi_device event occurred on
2665 * @evt: event to send
2667 * Assert scsi device event asynchronously.
2669 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2671 unsigned long flags;
2674 /* FIXME: currently this check eliminates all media change events
2675 * for polled devices. Need to update to discriminate between AN
2676 * and polled events */
2677 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2683 spin_lock_irqsave(&sdev->list_lock, flags);
2684 list_add_tail(&evt->node, &sdev->event_list);
2685 schedule_work(&sdev->event_work);
2686 spin_unlock_irqrestore(&sdev->list_lock, flags);
2688 EXPORT_SYMBOL_GPL(sdev_evt_send);
2691 * sdev_evt_alloc - allocate a new scsi event
2692 * @evt_type: type of event to allocate
2693 * @gfpflags: GFP flags for allocation
2695 * Allocates and returns a new scsi_event.
2697 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2700 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2704 evt->evt_type = evt_type;
2705 INIT_LIST_HEAD(&evt->node);
2707 /* evt_type-specific initialization, if any */
2709 case SDEV_EVT_MEDIA_CHANGE:
2710 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2711 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2712 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2713 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2714 case SDEV_EVT_LUN_CHANGE_REPORTED:
2722 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2725 * sdev_evt_send_simple - send asserted event to uevent thread
2726 * @sdev: scsi_device event occurred on
2727 * @evt_type: type of event to send
2728 * @gfpflags: GFP flags for allocation
2730 * Assert scsi device event asynchronously, given an event type.
2732 void sdev_evt_send_simple(struct scsi_device *sdev,
2733 enum scsi_device_event evt_type, gfp_t gfpflags)
2735 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2737 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2742 sdev_evt_send(sdev, evt);
2744 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2747 * scsi_device_quiesce - Block user issued commands.
2748 * @sdev: scsi device to quiesce.
2750 * This works by trying to transition to the SDEV_QUIESCE state
2751 * (which must be a legal transition). When the device is in this
2752 * state, only special requests will be accepted, all others will
2753 * be deferred. Since special requests may also be requeued requests,
2754 * a successful return doesn't guarantee the device will be
2755 * totally quiescent.
2757 * Must be called with user context, may sleep.
2759 * Returns zero if unsuccessful or an error if not.
2762 scsi_device_quiesce(struct scsi_device *sdev)
2764 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2768 scsi_run_queue(sdev->request_queue);
2769 while (atomic_read(&sdev->device_busy)) {
2770 msleep_interruptible(200);
2771 scsi_run_queue(sdev->request_queue);
2775 EXPORT_SYMBOL(scsi_device_quiesce);
2778 * scsi_device_resume - Restart user issued commands to a quiesced device.
2779 * @sdev: scsi device to resume.
2781 * Moves the device from quiesced back to running and restarts the
2784 * Must be called with user context, may sleep.
2786 void scsi_device_resume(struct scsi_device *sdev)
2788 /* check if the device state was mutated prior to resume, and if
2789 * so assume the state is being managed elsewhere (for example
2790 * device deleted during suspend)
2792 if (sdev->sdev_state != SDEV_QUIESCE ||
2793 scsi_device_set_state(sdev, SDEV_RUNNING))
2795 scsi_run_queue(sdev->request_queue);
2797 EXPORT_SYMBOL(scsi_device_resume);
2800 device_quiesce_fn(struct scsi_device *sdev, void *data)
2802 scsi_device_quiesce(sdev);
2806 scsi_target_quiesce(struct scsi_target *starget)
2808 starget_for_each_device(starget, NULL, device_quiesce_fn);
2810 EXPORT_SYMBOL(scsi_target_quiesce);
2813 device_resume_fn(struct scsi_device *sdev, void *data)
2815 scsi_device_resume(sdev);
2819 scsi_target_resume(struct scsi_target *starget)
2821 starget_for_each_device(starget, NULL, device_resume_fn);
2823 EXPORT_SYMBOL(scsi_target_resume);
2826 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2827 * @sdev: device to block
2829 * Block request made by scsi lld's to temporarily stop all
2830 * scsi commands on the specified device. Called from interrupt
2831 * or normal process context.
2833 * Returns zero if successful or error if not
2836 * This routine transitions the device to the SDEV_BLOCK state
2837 * (which must be a legal transition). When the device is in this
2838 * state, all commands are deferred until the scsi lld reenables
2839 * the device with scsi_device_unblock or device_block_tmo fires.
2842 scsi_internal_device_block(struct scsi_device *sdev)
2844 struct request_queue *q = sdev->request_queue;
2845 unsigned long flags;
2848 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2850 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2857 * The device has transitioned to SDEV_BLOCK. Stop the
2858 * block layer from calling the midlayer with this device's
2862 blk_mq_stop_hw_queues(q);
2864 spin_lock_irqsave(q->queue_lock, flags);
2866 spin_unlock_irqrestore(q->queue_lock, flags);
2871 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2874 * scsi_internal_device_unblock - resume a device after a block request
2875 * @sdev: device to resume
2876 * @new_state: state to set devices to after unblocking
2878 * Called by scsi lld's or the midlayer to restart the device queue
2879 * for the previously suspended scsi device. Called from interrupt or
2880 * normal process context.
2882 * Returns zero if successful or error if not.
2885 * This routine transitions the device to the SDEV_RUNNING state
2886 * or to one of the offline states (which must be a legal transition)
2887 * allowing the midlayer to goose the queue for this device.
2890 scsi_internal_device_unblock(struct scsi_device *sdev,
2891 enum scsi_device_state new_state)
2893 struct request_queue *q = sdev->request_queue;
2894 unsigned long flags;
2897 * Try to transition the scsi device to SDEV_RUNNING or one of the
2898 * offlined states and goose the device queue if successful.
2900 if ((sdev->sdev_state == SDEV_BLOCK) ||
2901 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2902 sdev->sdev_state = new_state;
2903 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2904 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2905 new_state == SDEV_OFFLINE)
2906 sdev->sdev_state = new_state;
2908 sdev->sdev_state = SDEV_CREATED;
2909 } else if (sdev->sdev_state != SDEV_CANCEL &&
2910 sdev->sdev_state != SDEV_OFFLINE)
2914 blk_mq_start_stopped_hw_queues(q, false);
2916 spin_lock_irqsave(q->queue_lock, flags);
2918 spin_unlock_irqrestore(q->queue_lock, flags);
2923 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2926 device_block(struct scsi_device *sdev, void *data)
2928 scsi_internal_device_block(sdev);
2932 target_block(struct device *dev, void *data)
2934 if (scsi_is_target_device(dev))
2935 starget_for_each_device(to_scsi_target(dev), NULL,
2941 scsi_target_block(struct device *dev)
2943 if (scsi_is_target_device(dev))
2944 starget_for_each_device(to_scsi_target(dev), NULL,
2947 device_for_each_child(dev, NULL, target_block);
2949 EXPORT_SYMBOL_GPL(scsi_target_block);
2952 device_unblock(struct scsi_device *sdev, void *data)
2954 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2958 target_unblock(struct device *dev, void *data)
2960 if (scsi_is_target_device(dev))
2961 starget_for_each_device(to_scsi_target(dev), data,
2967 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2969 if (scsi_is_target_device(dev))
2970 starget_for_each_device(to_scsi_target(dev), &new_state,
2973 device_for_each_child(dev, &new_state, target_unblock);
2975 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2978 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2979 * @sgl: scatter-gather list
2980 * @sg_count: number of segments in sg
2981 * @offset: offset in bytes into sg, on return offset into the mapped area
2982 * @len: bytes to map, on return number of bytes mapped
2984 * Returns virtual address of the start of the mapped page
2986 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2987 size_t *offset, size_t *len)
2990 size_t sg_len = 0, len_complete = 0;
2991 struct scatterlist *sg;
2994 WARN_ON(!irqs_disabled());
2996 for_each_sg(sgl, sg, sg_count, i) {
2997 len_complete = sg_len; /* Complete sg-entries */
2998 sg_len += sg->length;
2999 if (sg_len > *offset)
3003 if (unlikely(i == sg_count)) {
3004 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3006 __func__, sg_len, *offset, sg_count);
3011 /* Offset starting from the beginning of first page in this sg-entry */
3012 *offset = *offset - len_complete + sg->offset;
3014 /* Assumption: contiguous pages can be accessed as "page + i" */
3015 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3016 *offset &= ~PAGE_MASK;
3018 /* Bytes in this sg-entry from *offset to the end of the page */
3019 sg_len = PAGE_SIZE - *offset;
3023 return kmap_atomic(page);
3025 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3028 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3029 * @virt: virtual address to be unmapped
3031 void scsi_kunmap_atomic_sg(void *virt)
3033 kunmap_atomic(virt);
3035 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3037 void sdev_disable_disk_events(struct scsi_device *sdev)
3039 atomic_inc(&sdev->disk_events_disable_depth);
3041 EXPORT_SYMBOL(sdev_disable_disk_events);
3043 void sdev_enable_disk_events(struct scsi_device *sdev)
3045 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3047 atomic_dec(&sdev->disk_events_disable_depth);
3049 EXPORT_SYMBOL(sdev_enable_disk_events);