2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
49 static inline struct kmem_cache *
50 scsi_select_sense_cache(bool unchecked_isa_dma)
52 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56 unsigned char *sense_buffer)
58 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
63 gfp_t gfp_mask, int numa_node)
65 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
69 int scsi_init_sense_cache(struct Scsi_Host *shost)
71 struct kmem_cache *cache;
74 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
78 mutex_lock(&scsi_sense_cache_mutex);
79 if (shost->unchecked_isa_dma) {
80 scsi_sense_isadma_cache =
81 kmem_cache_create("scsi_sense_cache(DMA)",
82 SCSI_SENSE_BUFFERSIZE, 0,
83 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
84 if (!scsi_sense_isadma_cache)
88 kmem_cache_create_usercopy("scsi_sense_cache",
89 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
90 0, SCSI_SENSE_BUFFERSIZE, NULL);
91 if (!scsi_sense_cache)
95 mutex_unlock(&scsi_sense_cache_mutex);
100 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
101 * not change behaviour from the previous unplug mechanism, experimentation
102 * may prove this needs changing.
104 #define SCSI_QUEUE_DELAY 3
107 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
109 struct Scsi_Host *host = cmd->device->host;
110 struct scsi_device *device = cmd->device;
111 struct scsi_target *starget = scsi_target(device);
114 * Set the appropriate busy bit for the device/host.
116 * If the host/device isn't busy, assume that something actually
117 * completed, and that we should be able to queue a command now.
119 * Note that the prior mid-layer assumption that any host could
120 * always queue at least one command is now broken. The mid-layer
121 * will implement a user specifiable stall (see
122 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
123 * if a command is requeued with no other commands outstanding
124 * either for the device or for the host.
127 case SCSI_MLQUEUE_HOST_BUSY:
128 atomic_set(&host->host_blocked, host->max_host_blocked);
130 case SCSI_MLQUEUE_DEVICE_BUSY:
131 case SCSI_MLQUEUE_EH_RETRY:
132 atomic_set(&device->device_blocked,
133 device->max_device_blocked);
135 case SCSI_MLQUEUE_TARGET_BUSY:
136 atomic_set(&starget->target_blocked,
137 starget->max_target_blocked);
142 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
144 struct scsi_device *sdev = cmd->device;
146 if (cmd->request->rq_flags & RQF_DONTPREP) {
147 cmd->request->rq_flags &= ~RQF_DONTPREP;
148 scsi_mq_uninit_cmd(cmd);
152 blk_mq_requeue_request(cmd->request, true);
153 put_device(&sdev->sdev_gendev);
157 * __scsi_queue_insert - private queue insertion
158 * @cmd: The SCSI command being requeued
159 * @reason: The reason for the requeue
160 * @unbusy: Whether the queue should be unbusied
162 * This is a private queue insertion. The public interface
163 * scsi_queue_insert() always assumes the queue should be unbusied
164 * because it's always called before the completion. This function is
165 * for a requeue after completion, which should only occur in this
168 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
170 struct scsi_device *device = cmd->device;
172 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
173 "Inserting command %p into mlqueue\n", cmd));
175 scsi_set_blocked(cmd, reason);
178 * Decrement the counters, since these commands are no longer
179 * active on the host/device.
182 scsi_device_unbusy(device);
185 * Requeue this command. It will go before all other commands
186 * that are already in the queue. Schedule requeue work under
187 * lock such that the kblockd_schedule_work() call happens
188 * before blk_cleanup_queue() finishes.
193 * Before a SCSI command is dispatched,
194 * get_device(&sdev->sdev_gendev) is called and the host,
195 * target and device busy counters are increased. Since
196 * requeuing a request causes these actions to be repeated and
197 * since scsi_device_unbusy() has already been called,
198 * put_device(&device->sdev_gendev) must still be called. Call
199 * put_device() after blk_mq_requeue_request() to avoid that
200 * removal of the SCSI device can start before requeueing has
203 blk_mq_requeue_request(cmd->request, true);
204 put_device(&device->sdev_gendev);
208 * Function: scsi_queue_insert()
210 * Purpose: Insert a command in the midlevel queue.
212 * Arguments: cmd - command that we are adding to queue.
213 * reason - why we are inserting command to queue.
215 * Lock status: Assumed that lock is not held upon entry.
219 * Notes: We do this for one of two cases. Either the host is busy
220 * and it cannot accept any more commands for the time being,
221 * or the device returned QUEUE_FULL and can accept no more
223 * Notes: This could be called either from an interrupt context or a
224 * normal process context.
226 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
228 __scsi_queue_insert(cmd, reason, true);
233 * __scsi_execute - insert request and wait for the result
236 * @data_direction: data direction
237 * @buffer: data buffer
238 * @bufflen: len of buffer
239 * @sense: optional sense buffer
240 * @sshdr: optional decoded sense header
241 * @timeout: request timeout in seconds
242 * @retries: number of times to retry request
243 * @flags: flags for ->cmd_flags
244 * @rq_flags: flags for ->rq_flags
245 * @resid: optional residual length
247 * Returns the scsi_cmnd result field if a command was executed, or a negative
248 * Linux error code if we didn't get that far.
250 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
251 int data_direction, void *buffer, unsigned bufflen,
252 unsigned char *sense, struct scsi_sense_hdr *sshdr,
253 int timeout, int retries, u64 flags, req_flags_t rq_flags,
257 struct scsi_request *rq;
258 int ret = DRIVER_ERROR << 24;
260 req = blk_get_request(sdev->request_queue,
261 data_direction == DMA_TO_DEVICE ?
262 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
267 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
268 buffer, bufflen, GFP_NOIO))
271 rq->cmd_len = COMMAND_SIZE(cmd[0]);
272 memcpy(rq->cmd, cmd, rq->cmd_len);
273 rq->retries = retries;
274 req->timeout = timeout;
275 req->cmd_flags |= flags;
276 req->rq_flags |= rq_flags | RQF_QUIET;
279 * head injection *required* here otherwise quiesce won't work
281 blk_execute_rq(req->q, NULL, req, 1);
284 * Some devices (USB mass-storage in particular) may transfer
285 * garbage data together with a residue indicating that the data
286 * is invalid. Prevent the garbage from being misinterpreted
287 * and prevent security leaks by zeroing out the excess data.
289 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
290 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
293 *resid = rq->resid_len;
294 if (sense && rq->sense_len)
295 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
297 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
300 blk_put_request(req);
304 EXPORT_SYMBOL(__scsi_execute);
307 * Function: scsi_init_cmd_errh()
309 * Purpose: Initialize cmd fields related to error handling.
311 * Arguments: cmd - command that is ready to be queued.
313 * Notes: This function has the job of initializing a number of
314 * fields related to error handling. Typically this will
315 * be called once for each command, as required.
317 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
319 scsi_set_resid(cmd, 0);
320 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
321 if (cmd->cmd_len == 0)
322 cmd->cmd_len = scsi_command_size(cmd->cmnd);
326 * Decrement the host_busy counter and wake up the error handler if necessary.
327 * Avoid as follows that the error handler is not woken up if shost->host_busy
328 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
329 * with an RCU read lock in this function to ensure that this function in its
330 * entirety either finishes before scsi_eh_scmd_add() increases the
331 * host_failed counter or that it notices the shost state change made by
332 * scsi_eh_scmd_add().
334 static void scsi_dec_host_busy(struct Scsi_Host *shost)
339 atomic_dec(&shost->host_busy);
340 if (unlikely(scsi_host_in_recovery(shost))) {
341 spin_lock_irqsave(shost->host_lock, flags);
342 if (shost->host_failed || shost->host_eh_scheduled)
343 scsi_eh_wakeup(shost);
344 spin_unlock_irqrestore(shost->host_lock, flags);
349 void scsi_device_unbusy(struct scsi_device *sdev)
351 struct Scsi_Host *shost = sdev->host;
352 struct scsi_target *starget = scsi_target(sdev);
354 scsi_dec_host_busy(shost);
356 if (starget->can_queue > 0)
357 atomic_dec(&starget->target_busy);
359 atomic_dec(&sdev->device_busy);
362 static void scsi_kick_queue(struct request_queue *q)
364 blk_mq_run_hw_queues(q, false);
368 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
369 * and call blk_run_queue for all the scsi_devices on the target -
370 * including current_sdev first.
372 * Called with *no* scsi locks held.
374 static void scsi_single_lun_run(struct scsi_device *current_sdev)
376 struct Scsi_Host *shost = current_sdev->host;
377 struct scsi_device *sdev, *tmp;
378 struct scsi_target *starget = scsi_target(current_sdev);
381 spin_lock_irqsave(shost->host_lock, flags);
382 starget->starget_sdev_user = NULL;
383 spin_unlock_irqrestore(shost->host_lock, flags);
386 * Call blk_run_queue for all LUNs on the target, starting with
387 * current_sdev. We race with others (to set starget_sdev_user),
388 * but in most cases, we will be first. Ideally, each LU on the
389 * target would get some limited time or requests on the target.
391 scsi_kick_queue(current_sdev->request_queue);
393 spin_lock_irqsave(shost->host_lock, flags);
394 if (starget->starget_sdev_user)
396 list_for_each_entry_safe(sdev, tmp, &starget->devices,
397 same_target_siblings) {
398 if (sdev == current_sdev)
400 if (scsi_device_get(sdev))
403 spin_unlock_irqrestore(shost->host_lock, flags);
404 scsi_kick_queue(sdev->request_queue);
405 spin_lock_irqsave(shost->host_lock, flags);
407 scsi_device_put(sdev);
410 spin_unlock_irqrestore(shost->host_lock, flags);
413 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
415 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
417 if (atomic_read(&sdev->device_blocked) > 0)
422 static inline bool scsi_target_is_busy(struct scsi_target *starget)
424 if (starget->can_queue > 0) {
425 if (atomic_read(&starget->target_busy) >= starget->can_queue)
427 if (atomic_read(&starget->target_blocked) > 0)
433 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
435 if (shost->can_queue > 0 &&
436 atomic_read(&shost->host_busy) >= shost->can_queue)
438 if (atomic_read(&shost->host_blocked) > 0)
440 if (shost->host_self_blocked)
445 static void scsi_starved_list_run(struct Scsi_Host *shost)
447 LIST_HEAD(starved_list);
448 struct scsi_device *sdev;
451 spin_lock_irqsave(shost->host_lock, flags);
452 list_splice_init(&shost->starved_list, &starved_list);
454 while (!list_empty(&starved_list)) {
455 struct request_queue *slq;
458 * As long as shost is accepting commands and we have
459 * starved queues, call blk_run_queue. scsi_request_fn
460 * drops the queue_lock and can add us back to the
463 * host_lock protects the starved_list and starved_entry.
464 * scsi_request_fn must get the host_lock before checking
465 * or modifying starved_list or starved_entry.
467 if (scsi_host_is_busy(shost))
470 sdev = list_entry(starved_list.next,
471 struct scsi_device, starved_entry);
472 list_del_init(&sdev->starved_entry);
473 if (scsi_target_is_busy(scsi_target(sdev))) {
474 list_move_tail(&sdev->starved_entry,
475 &shost->starved_list);
480 * Once we drop the host lock, a racing scsi_remove_device()
481 * call may remove the sdev from the starved list and destroy
482 * it and the queue. Mitigate by taking a reference to the
483 * queue and never touching the sdev again after we drop the
484 * host lock. Note: if __scsi_remove_device() invokes
485 * blk_cleanup_queue() before the queue is run from this
486 * function then blk_run_queue() will return immediately since
487 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
489 slq = sdev->request_queue;
490 if (!blk_get_queue(slq))
492 spin_unlock_irqrestore(shost->host_lock, flags);
494 scsi_kick_queue(slq);
497 spin_lock_irqsave(shost->host_lock, flags);
499 /* put any unprocessed entries back */
500 list_splice(&starved_list, &shost->starved_list);
501 spin_unlock_irqrestore(shost->host_lock, flags);
505 * Function: scsi_run_queue()
507 * Purpose: Select a proper request queue to serve next
509 * Arguments: q - last request's queue
513 * Notes: The previous command was completely finished, start
514 * a new one if possible.
516 static void scsi_run_queue(struct request_queue *q)
518 struct scsi_device *sdev = q->queuedata;
520 if (scsi_target(sdev)->single_lun)
521 scsi_single_lun_run(sdev);
522 if (!list_empty(&sdev->host->starved_list))
523 scsi_starved_list_run(sdev->host);
525 blk_mq_run_hw_queues(q, false);
528 void scsi_requeue_run_queue(struct work_struct *work)
530 struct scsi_device *sdev;
531 struct request_queue *q;
533 sdev = container_of(work, struct scsi_device, requeue_work);
534 q = sdev->request_queue;
538 void scsi_run_host_queues(struct Scsi_Host *shost)
540 struct scsi_device *sdev;
542 shost_for_each_device(sdev, shost)
543 scsi_run_queue(sdev->request_queue);
546 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
548 if (!blk_rq_is_passthrough(cmd->request)) {
549 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
551 if (drv->uninit_command)
552 drv->uninit_command(cmd);
556 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
558 if (cmd->sdb.table.nents)
559 sg_free_table_chained(&cmd->sdb.table, true);
560 if (scsi_prot_sg_count(cmd))
561 sg_free_table_chained(&cmd->prot_sdb->table, true);
564 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
566 scsi_mq_free_sgtables(cmd);
567 scsi_uninit_cmd(cmd);
568 scsi_del_cmd_from_list(cmd);
571 /* Returns false when no more bytes to process, true if there are more */
572 static bool scsi_end_request(struct request *req, blk_status_t error,
575 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
576 struct scsi_device *sdev = cmd->device;
577 struct request_queue *q = sdev->request_queue;
579 if (blk_update_request(req, error, bytes))
582 if (blk_queue_add_random(q))
583 add_disk_randomness(req->rq_disk);
585 if (!blk_rq_is_scsi(req)) {
586 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
587 cmd->flags &= ~SCMD_INITIALIZED;
588 destroy_rcu_head(&cmd->rcu);
592 * In the MQ case the command gets freed by __blk_mq_end_request,
593 * so we have to do all cleanup that depends on it earlier.
595 * We also can't kick the queues from irq context, so we
596 * will have to defer it to a workqueue.
598 scsi_mq_uninit_cmd(cmd);
601 * queue is still alive, so grab the ref for preventing it
602 * from being cleaned up during running queue.
604 percpu_ref_get(&q->q_usage_counter);
606 __blk_mq_end_request(req, error);
608 if (scsi_target(sdev)->single_lun ||
609 !list_empty(&sdev->host->starved_list))
610 kblockd_schedule_work(&sdev->requeue_work);
612 blk_mq_run_hw_queues(q, true);
614 percpu_ref_put(&q->q_usage_counter);
615 put_device(&sdev->sdev_gendev);
620 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
622 * @result: scsi error code
624 * Translate a SCSI result code into a blk_status_t value. May reset the host
625 * byte of @cmd->result.
627 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
629 switch (host_byte(result)) {
632 * Also check the other bytes than the status byte in result
633 * to handle the case when a SCSI LLD sets result to
634 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
636 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
638 return BLK_STS_IOERR;
639 case DID_TRANSPORT_FAILFAST:
640 return BLK_STS_TRANSPORT;
641 case DID_TARGET_FAILURE:
642 set_host_byte(cmd, DID_OK);
643 return BLK_STS_TARGET;
644 case DID_NEXUS_FAILURE:
645 set_host_byte(cmd, DID_OK);
646 return BLK_STS_NEXUS;
647 case DID_ALLOC_FAILURE:
648 set_host_byte(cmd, DID_OK);
649 return BLK_STS_NOSPC;
650 case DID_MEDIUM_ERROR:
651 set_host_byte(cmd, DID_OK);
652 return BLK_STS_MEDIUM;
654 return BLK_STS_IOERR;
658 /* Helper for scsi_io_completion() when "reprep" action required. */
659 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
660 struct request_queue *q)
662 /* A new command will be prepared and issued. */
663 scsi_mq_requeue_cmd(cmd);
666 /* Helper for scsi_io_completion() when special action required. */
667 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
669 struct request_queue *q = cmd->device->request_queue;
670 struct request *req = cmd->request;
672 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
673 ACTION_DELAYED_RETRY} action;
674 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
675 struct scsi_sense_hdr sshdr;
677 bool sense_current = true; /* false implies "deferred sense" */
678 blk_status_t blk_stat;
680 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
682 sense_current = !scsi_sense_is_deferred(&sshdr);
684 blk_stat = scsi_result_to_blk_status(cmd, result);
686 if (host_byte(result) == DID_RESET) {
687 /* Third party bus reset or reset for error recovery
688 * reasons. Just retry the command and see what
691 action = ACTION_RETRY;
692 } else if (sense_valid && sense_current) {
693 switch (sshdr.sense_key) {
695 if (cmd->device->removable) {
696 /* Detected disc change. Set a bit
697 * and quietly refuse further access.
699 cmd->device->changed = 1;
700 action = ACTION_FAIL;
702 /* Must have been a power glitch, or a
703 * bus reset. Could not have been a
704 * media change, so we just retry the
705 * command and see what happens.
707 action = ACTION_RETRY;
710 case ILLEGAL_REQUEST:
711 /* If we had an ILLEGAL REQUEST returned, then
712 * we may have performed an unsupported
713 * command. The only thing this should be
714 * would be a ten byte read where only a six
715 * byte read was supported. Also, on a system
716 * where READ CAPACITY failed, we may have
717 * read past the end of the disk.
719 if ((cmd->device->use_10_for_rw &&
720 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
721 (cmd->cmnd[0] == READ_10 ||
722 cmd->cmnd[0] == WRITE_10)) {
723 /* This will issue a new 6-byte command. */
724 cmd->device->use_10_for_rw = 0;
725 action = ACTION_REPREP;
726 } else if (sshdr.asc == 0x10) /* DIX */ {
727 action = ACTION_FAIL;
728 blk_stat = BLK_STS_PROTECTION;
729 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
730 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
731 action = ACTION_FAIL;
732 blk_stat = BLK_STS_TARGET;
734 action = ACTION_FAIL;
736 case ABORTED_COMMAND:
737 action = ACTION_FAIL;
738 if (sshdr.asc == 0x10) /* DIF */
739 blk_stat = BLK_STS_PROTECTION;
742 /* If the device is in the process of becoming
743 * ready, or has a temporary blockage, retry.
745 if (sshdr.asc == 0x04) {
746 switch (sshdr.ascq) {
747 case 0x01: /* becoming ready */
748 case 0x04: /* format in progress */
749 case 0x05: /* rebuild in progress */
750 case 0x06: /* recalculation in progress */
751 case 0x07: /* operation in progress */
752 case 0x08: /* Long write in progress */
753 case 0x09: /* self test in progress */
754 case 0x14: /* space allocation in progress */
755 case 0x1a: /* start stop unit in progress */
756 case 0x1b: /* sanitize in progress */
757 case 0x1d: /* configuration in progress */
758 case 0x24: /* depopulation in progress */
759 action = ACTION_DELAYED_RETRY;
762 action = ACTION_FAIL;
766 action = ACTION_FAIL;
768 case VOLUME_OVERFLOW:
769 /* See SSC3rXX or current. */
770 action = ACTION_FAIL;
773 action = ACTION_FAIL;
777 action = ACTION_FAIL;
779 if (action != ACTION_FAIL &&
780 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
781 action = ACTION_FAIL;
785 /* Give up and fail the remainder of the request */
786 if (!(req->rq_flags & RQF_QUIET)) {
787 static DEFINE_RATELIMIT_STATE(_rs,
788 DEFAULT_RATELIMIT_INTERVAL,
789 DEFAULT_RATELIMIT_BURST);
791 if (unlikely(scsi_logging_level))
793 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
794 SCSI_LOG_MLCOMPLETE_BITS);
797 * if logging is enabled the failure will be printed
798 * in scsi_log_completion(), so avoid duplicate messages
800 if (!level && __ratelimit(&_rs)) {
801 scsi_print_result(cmd, NULL, FAILED);
802 if (driver_byte(result) == DRIVER_SENSE)
803 scsi_print_sense(cmd);
804 scsi_print_command(cmd);
807 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
811 scsi_io_completion_reprep(cmd, q);
814 /* Retry the same command immediately */
815 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
817 case ACTION_DELAYED_RETRY:
818 /* Retry the same command after a delay */
819 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
825 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
826 * new result that may suppress further error checking. Also modifies
827 * *blk_statp in some cases.
829 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
830 blk_status_t *blk_statp)
833 bool sense_current = true; /* false implies "deferred sense" */
834 struct request *req = cmd->request;
835 struct scsi_sense_hdr sshdr;
837 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
839 sense_current = !scsi_sense_is_deferred(&sshdr);
841 if (blk_rq_is_passthrough(req)) {
844 * SG_IO wants current and deferred errors
846 scsi_req(req)->sense_len =
847 min(8 + cmd->sense_buffer[7],
848 SCSI_SENSE_BUFFERSIZE);
851 *blk_statp = scsi_result_to_blk_status(cmd, result);
852 } else if (blk_rq_bytes(req) == 0 && sense_current) {
854 * Flush commands do not transfers any data, and thus cannot use
855 * good_bytes != blk_rq_bytes(req) as the signal for an error.
856 * This sets *blk_statp explicitly for the problem case.
858 *blk_statp = scsi_result_to_blk_status(cmd, result);
861 * Recovered errors need reporting, but they're always treated as
862 * success, so fiddle the result code here. For passthrough requests
863 * we already took a copy of the original into sreq->result which
864 * is what gets returned to the user
866 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
867 bool do_print = true;
869 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
870 * skip print since caller wants ATA registers. Only occurs
871 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
873 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
875 else if (req->rq_flags & RQF_QUIET)
878 scsi_print_sense(cmd);
880 /* for passthrough, *blk_statp may be set */
881 *blk_statp = BLK_STS_OK;
884 * Another corner case: the SCSI status byte is non-zero but 'good'.
885 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
886 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
887 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
888 * intermediate statuses (both obsolete in SAM-4) as good.
890 if (status_byte(result) && scsi_status_is_good(result)) {
892 *blk_statp = BLK_STS_OK;
898 * Function: scsi_io_completion()
900 * Purpose: Completion processing for block device I/O requests.
902 * Arguments: cmd - command that is finished.
904 * Lock status: Assumed that no lock is held upon entry.
908 * Notes: We will finish off the specified number of sectors. If we
909 * are done, the command block will be released and the queue
910 * function will be goosed. If we are not done then we have to
911 * figure out what to do next:
913 * a) We can call scsi_requeue_command(). The request
914 * will be unprepared and put back on the queue. Then
915 * a new command will be created for it. This should
916 * be used if we made forward progress, or if we want
917 * to switch from READ(10) to READ(6) for example.
919 * b) We can call __scsi_queue_insert(). The request will
920 * be put back on the queue and retried using the same
921 * command as before, possibly after a delay.
923 * c) We can call scsi_end_request() with blk_stat other than
924 * BLK_STS_OK, to fail the remainder of the request.
926 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
928 int result = cmd->result;
929 struct request_queue *q = cmd->device->request_queue;
930 struct request *req = cmd->request;
931 blk_status_t blk_stat = BLK_STS_OK;
933 if (unlikely(result)) /* a nz result may or may not be an error */
934 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
936 if (unlikely(blk_rq_is_passthrough(req))) {
938 * scsi_result_to_blk_status may have reset the host_byte
940 scsi_req(req)->result = cmd->result;
944 * Next deal with any sectors which we were able to correctly
947 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
948 "%u sectors total, %d bytes done.\n",
949 blk_rq_sectors(req), good_bytes));
952 * Next deal with any sectors which we were able to correctly
953 * handle. Failed, zero length commands always need to drop down
954 * to retry code. Fast path should return in this block.
956 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
957 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
958 return; /* no bytes remaining */
961 /* Kill remainder if no retries. */
962 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
963 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
965 "Bytes remaining after failed, no-retry command");
970 * If there had been no error, but we have leftover bytes in the
971 * requeues just queue the command up again.
973 if (likely(result == 0))
974 scsi_io_completion_reprep(cmd, q);
976 scsi_io_completion_action(cmd, result);
979 static blk_status_t scsi_init_sgtable(struct request *req,
980 struct scsi_data_buffer *sdb)
985 * If sg table allocation fails, requeue request later.
987 if (unlikely(sg_alloc_table_chained(&sdb->table,
988 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
989 return BLK_STS_RESOURCE;
992 * Next, walk the list, and fill in the addresses and sizes of
995 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
996 BUG_ON(count > sdb->table.nents);
997 sdb->table.nents = count;
998 sdb->length = blk_rq_payload_bytes(req);
1003 * Function: scsi_init_io()
1005 * Purpose: SCSI I/O initialize function.
1007 * Arguments: cmd - Command descriptor we wish to initialize
1009 * Returns: BLK_STS_OK on success
1010 * BLK_STS_RESOURCE if the failure is retryable
1011 * BLK_STS_IOERR if the failure is fatal
1013 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1015 struct request *rq = cmd->request;
1018 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1019 return BLK_STS_IOERR;
1021 ret = scsi_init_sgtable(rq, &cmd->sdb);
1025 if (blk_integrity_rq(rq)) {
1026 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1029 if (WARN_ON_ONCE(!prot_sdb)) {
1031 * This can happen if someone (e.g. multipath)
1032 * queues a command to a device on an adapter
1033 * that does not support DIX.
1035 ret = BLK_STS_IOERR;
1036 goto out_free_sgtables;
1039 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1041 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1042 prot_sdb->table.sgl)) {
1043 ret = BLK_STS_RESOURCE;
1044 goto out_free_sgtables;
1047 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1048 prot_sdb->table.sgl);
1049 BUG_ON(count > ivecs);
1050 BUG_ON(count > queue_max_integrity_segments(rq->q));
1052 cmd->prot_sdb = prot_sdb;
1053 cmd->prot_sdb->table.nents = count;
1058 scsi_mq_free_sgtables(cmd);
1061 EXPORT_SYMBOL(scsi_init_io);
1064 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1065 * @rq: Request associated with the SCSI command to be initialized.
1067 * This function initializes the members of struct scsi_cmnd that must be
1068 * initialized before request processing starts and that won't be
1069 * reinitialized if a SCSI command is requeued.
1071 * Called from inside blk_get_request() for pass-through requests and from
1072 * inside scsi_init_command() for filesystem requests.
1074 static void scsi_initialize_rq(struct request *rq)
1076 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1078 scsi_req_init(&cmd->req);
1079 init_rcu_head(&cmd->rcu);
1080 cmd->jiffies_at_alloc = jiffies;
1084 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1085 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1087 struct scsi_device *sdev = cmd->device;
1088 struct Scsi_Host *shost = sdev->host;
1089 unsigned long flags;
1091 if (shost->use_cmd_list) {
1092 spin_lock_irqsave(&sdev->list_lock, flags);
1093 list_add_tail(&cmd->list, &sdev->cmd_list);
1094 spin_unlock_irqrestore(&sdev->list_lock, flags);
1098 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1099 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1101 struct scsi_device *sdev = cmd->device;
1102 struct Scsi_Host *shost = sdev->host;
1103 unsigned long flags;
1105 if (shost->use_cmd_list) {
1106 spin_lock_irqsave(&sdev->list_lock, flags);
1107 BUG_ON(list_empty(&cmd->list));
1108 list_del_init(&cmd->list);
1109 spin_unlock_irqrestore(&sdev->list_lock, flags);
1113 /* Called after a request has been started. */
1114 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1116 void *buf = cmd->sense_buffer;
1117 void *prot = cmd->prot_sdb;
1118 struct request *rq = blk_mq_rq_from_pdu(cmd);
1119 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1120 unsigned long jiffies_at_alloc;
1123 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1124 flags |= SCMD_INITIALIZED;
1125 scsi_initialize_rq(rq);
1128 jiffies_at_alloc = cmd->jiffies_at_alloc;
1129 retries = cmd->retries;
1130 /* zero out the cmd, except for the embedded scsi_request */
1131 memset((char *)cmd + sizeof(cmd->req), 0,
1132 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1135 cmd->sense_buffer = buf;
1136 cmd->prot_sdb = prot;
1138 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1139 cmd->jiffies_at_alloc = jiffies_at_alloc;
1140 cmd->retries = retries;
1142 scsi_add_cmd_to_list(cmd);
1145 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1146 struct request *req)
1148 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1151 * Passthrough requests may transfer data, in which case they must
1152 * a bio attached to them. Or they might contain a SCSI command
1153 * that does not transfer data, in which case they may optionally
1154 * submit a request without an attached bio.
1157 blk_status_t ret = scsi_init_io(cmd);
1158 if (unlikely(ret != BLK_STS_OK))
1161 BUG_ON(blk_rq_bytes(req));
1163 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1166 cmd->cmd_len = scsi_req(req)->cmd_len;
1167 cmd->cmnd = scsi_req(req)->cmd;
1168 cmd->transfersize = blk_rq_bytes(req);
1169 cmd->allowed = scsi_req(req)->retries;
1174 * Setup a normal block command. These are simple request from filesystems
1175 * that still need to be translated to SCSI CDBs from the ULD.
1177 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1178 struct request *req)
1180 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1182 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1183 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1184 if (ret != BLK_STS_OK)
1188 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1189 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1190 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1193 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1194 struct request *req)
1196 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1198 if (!blk_rq_bytes(req))
1199 cmd->sc_data_direction = DMA_NONE;
1200 else if (rq_data_dir(req) == WRITE)
1201 cmd->sc_data_direction = DMA_TO_DEVICE;
1203 cmd->sc_data_direction = DMA_FROM_DEVICE;
1205 if (blk_rq_is_scsi(req))
1206 return scsi_setup_scsi_cmnd(sdev, req);
1208 return scsi_setup_fs_cmnd(sdev, req);
1212 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1214 switch (sdev->sdev_state) {
1216 case SDEV_TRANSPORT_OFFLINE:
1218 * If the device is offline we refuse to process any
1219 * commands. The device must be brought online
1220 * before trying any recovery commands.
1222 sdev_printk(KERN_ERR, sdev,
1223 "rejecting I/O to offline device\n");
1224 return BLK_STS_IOERR;
1227 * If the device is fully deleted, we refuse to
1228 * process any commands as well.
1230 sdev_printk(KERN_ERR, sdev,
1231 "rejecting I/O to dead device\n");
1232 return BLK_STS_IOERR;
1234 case SDEV_CREATED_BLOCK:
1235 return BLK_STS_RESOURCE;
1238 * If the devices is blocked we defer normal commands.
1240 if (req && !(req->rq_flags & RQF_PREEMPT))
1241 return BLK_STS_RESOURCE;
1245 * For any other not fully online state we only allow
1246 * special commands. In particular any user initiated
1247 * command is not allowed.
1249 if (req && !(req->rq_flags & RQF_PREEMPT))
1250 return BLK_STS_IOERR;
1256 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1259 * Called with the queue_lock held.
1261 static inline int scsi_dev_queue_ready(struct request_queue *q,
1262 struct scsi_device *sdev)
1266 busy = atomic_inc_return(&sdev->device_busy) - 1;
1267 if (atomic_read(&sdev->device_blocked)) {
1272 * unblock after device_blocked iterates to zero
1274 if (atomic_dec_return(&sdev->device_blocked) > 0)
1276 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1277 "unblocking device at zero depth\n"));
1280 if (busy >= sdev->queue_depth)
1285 atomic_dec(&sdev->device_busy);
1290 * scsi_target_queue_ready: checks if there we can send commands to target
1291 * @sdev: scsi device on starget to check.
1293 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1294 struct scsi_device *sdev)
1296 struct scsi_target *starget = scsi_target(sdev);
1299 if (starget->single_lun) {
1300 spin_lock_irq(shost->host_lock);
1301 if (starget->starget_sdev_user &&
1302 starget->starget_sdev_user != sdev) {
1303 spin_unlock_irq(shost->host_lock);
1306 starget->starget_sdev_user = sdev;
1307 spin_unlock_irq(shost->host_lock);
1310 if (starget->can_queue <= 0)
1313 busy = atomic_inc_return(&starget->target_busy) - 1;
1314 if (atomic_read(&starget->target_blocked) > 0) {
1319 * unblock after target_blocked iterates to zero
1321 if (atomic_dec_return(&starget->target_blocked) > 0)
1324 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1325 "unblocking target at zero depth\n"));
1328 if (busy >= starget->can_queue)
1334 spin_lock_irq(shost->host_lock);
1335 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1336 spin_unlock_irq(shost->host_lock);
1338 if (starget->can_queue > 0)
1339 atomic_dec(&starget->target_busy);
1344 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1345 * return 0. We must end up running the queue again whenever 0 is
1346 * returned, else IO can hang.
1348 static inline int scsi_host_queue_ready(struct request_queue *q,
1349 struct Scsi_Host *shost,
1350 struct scsi_device *sdev)
1354 if (scsi_host_in_recovery(shost))
1357 busy = atomic_inc_return(&shost->host_busy) - 1;
1358 if (atomic_read(&shost->host_blocked) > 0) {
1363 * unblock after host_blocked iterates to zero
1365 if (atomic_dec_return(&shost->host_blocked) > 0)
1369 shost_printk(KERN_INFO, shost,
1370 "unblocking host at zero depth\n"));
1373 if (shost->can_queue > 0 && busy >= shost->can_queue)
1375 if (shost->host_self_blocked)
1378 /* We're OK to process the command, so we can't be starved */
1379 if (!list_empty(&sdev->starved_entry)) {
1380 spin_lock_irq(shost->host_lock);
1381 if (!list_empty(&sdev->starved_entry))
1382 list_del_init(&sdev->starved_entry);
1383 spin_unlock_irq(shost->host_lock);
1389 spin_lock_irq(shost->host_lock);
1390 if (list_empty(&sdev->starved_entry))
1391 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1392 spin_unlock_irq(shost->host_lock);
1394 scsi_dec_host_busy(shost);
1399 * Busy state exporting function for request stacking drivers.
1401 * For efficiency, no lock is taken to check the busy state of
1402 * shost/starget/sdev, since the returned value is not guaranteed and
1403 * may be changed after request stacking drivers call the function,
1404 * regardless of taking lock or not.
1406 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1407 * needs to return 'not busy'. Otherwise, request stacking drivers
1408 * may hold requests forever.
1410 static bool scsi_mq_lld_busy(struct request_queue *q)
1412 struct scsi_device *sdev = q->queuedata;
1413 struct Scsi_Host *shost;
1415 if (blk_queue_dying(q))
1421 * Ignore host/starget busy state.
1422 * Since block layer does not have a concept of fairness across
1423 * multiple queues, congestion of host/starget needs to be handled
1426 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1432 static void scsi_softirq_done(struct request *rq)
1434 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1435 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1438 INIT_LIST_HEAD(&cmd->eh_entry);
1440 atomic_inc(&cmd->device->iodone_cnt);
1442 atomic_inc(&cmd->device->ioerr_cnt);
1444 disposition = scsi_decide_disposition(cmd);
1445 if (disposition != SUCCESS &&
1446 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1447 sdev_printk(KERN_ERR, cmd->device,
1448 "timing out command, waited %lus\n",
1450 disposition = SUCCESS;
1453 scsi_log_completion(cmd, disposition);
1455 switch (disposition) {
1457 scsi_finish_command(cmd);
1460 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1462 case ADD_TO_MLQUEUE:
1463 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1466 scsi_eh_scmd_add(cmd);
1472 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1473 * @cmd: command block we are dispatching.
1475 * Return: nonzero return request was rejected and device's queue needs to be
1478 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1480 struct Scsi_Host *host = cmd->device->host;
1483 atomic_inc(&cmd->device->iorequest_cnt);
1485 /* check if the device is still usable */
1486 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1487 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1488 * returns an immediate error upwards, and signals
1489 * that the device is no longer present */
1490 cmd->result = DID_NO_CONNECT << 16;
1494 /* Check to see if the scsi lld made this device blocked. */
1495 if (unlikely(scsi_device_blocked(cmd->device))) {
1497 * in blocked state, the command is just put back on
1498 * the device queue. The suspend state has already
1499 * blocked the queue so future requests should not
1500 * occur until the device transitions out of the
1503 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1504 "queuecommand : device blocked\n"));
1505 return SCSI_MLQUEUE_DEVICE_BUSY;
1508 /* Store the LUN value in cmnd, if needed. */
1509 if (cmd->device->lun_in_cdb)
1510 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1511 (cmd->device->lun << 5 & 0xe0);
1516 * Before we queue this command, check if the command
1517 * length exceeds what the host adapter can handle.
1519 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1520 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1521 "queuecommand : command too long. "
1522 "cdb_size=%d host->max_cmd_len=%d\n",
1523 cmd->cmd_len, cmd->device->host->max_cmd_len));
1524 cmd->result = (DID_ABORT << 16);
1528 if (unlikely(host->shost_state == SHOST_DEL)) {
1529 cmd->result = (DID_NO_CONNECT << 16);
1534 trace_scsi_dispatch_cmd_start(cmd);
1535 rtn = host->hostt->queuecommand(host, cmd);
1537 trace_scsi_dispatch_cmd_error(cmd, rtn);
1538 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1539 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1540 rtn = SCSI_MLQUEUE_HOST_BUSY;
1542 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1543 "queuecommand : request rejected\n"));
1548 cmd->scsi_done(cmd);
1552 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1553 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1555 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1556 sizeof(struct scatterlist);
1559 static blk_status_t scsi_mq_prep_fn(struct request *req)
1561 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1562 struct scsi_device *sdev = req->q->queuedata;
1563 struct Scsi_Host *shost = sdev->host;
1564 struct scatterlist *sg;
1566 scsi_init_command(sdev, cmd);
1569 cmd->tag = req->tag;
1570 cmd->prot_op = SCSI_PROT_NORMAL;
1572 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1573 cmd->sdb.table.sgl = sg;
1575 if (scsi_host_get_prot(shost)) {
1576 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1578 cmd->prot_sdb->table.sgl =
1579 (struct scatterlist *)(cmd->prot_sdb + 1);
1582 blk_mq_start_request(req);
1584 return scsi_setup_cmnd(sdev, req);
1587 static void scsi_mq_done(struct scsi_cmnd *cmd)
1589 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1591 trace_scsi_dispatch_cmd_done(cmd);
1594 * If the block layer didn't complete the request due to a timeout
1595 * injection, scsi must clear its internal completed state so that the
1596 * timeout handler will see it needs to escalate its own error
1599 if (unlikely(!blk_mq_complete_request(cmd->request)))
1600 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1603 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1605 struct request_queue *q = hctx->queue;
1606 struct scsi_device *sdev = q->queuedata;
1608 atomic_dec(&sdev->device_busy);
1609 put_device(&sdev->sdev_gendev);
1612 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1614 struct request_queue *q = hctx->queue;
1615 struct scsi_device *sdev = q->queuedata;
1617 if (!get_device(&sdev->sdev_gendev))
1619 if (!scsi_dev_queue_ready(q, sdev))
1620 goto out_put_device;
1625 put_device(&sdev->sdev_gendev);
1627 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1628 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1632 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1633 const struct blk_mq_queue_data *bd)
1635 struct request *req = bd->rq;
1636 struct request_queue *q = req->q;
1637 struct scsi_device *sdev = q->queuedata;
1638 struct Scsi_Host *shost = sdev->host;
1639 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1644 * If the device is not in running state we will reject some or all
1647 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1648 ret = scsi_prep_state_check(sdev, req);
1649 if (ret != BLK_STS_OK)
1650 goto out_put_budget;
1653 ret = BLK_STS_RESOURCE;
1654 if (!scsi_target_queue_ready(shost, sdev))
1655 goto out_put_budget;
1656 if (!scsi_host_queue_ready(q, shost, sdev))
1657 goto out_dec_target_busy;
1659 if (!(req->rq_flags & RQF_DONTPREP)) {
1660 ret = scsi_mq_prep_fn(req);
1661 if (ret != BLK_STS_OK)
1662 goto out_dec_host_busy;
1663 req->rq_flags |= RQF_DONTPREP;
1665 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1666 blk_mq_start_request(req);
1669 if (sdev->simple_tags)
1670 cmd->flags |= SCMD_TAGGED;
1672 cmd->flags &= ~SCMD_TAGGED;
1674 scsi_init_cmd_errh(cmd);
1675 cmd->scsi_done = scsi_mq_done;
1677 reason = scsi_dispatch_cmd(cmd);
1679 scsi_set_blocked(cmd, reason);
1680 ret = BLK_STS_RESOURCE;
1681 goto out_dec_host_busy;
1687 scsi_dec_host_busy(shost);
1688 out_dec_target_busy:
1689 if (scsi_target(sdev)->can_queue > 0)
1690 atomic_dec(&scsi_target(sdev)->target_busy);
1692 scsi_mq_put_budget(hctx);
1696 case BLK_STS_RESOURCE:
1697 if (atomic_read(&sdev->device_busy) ||
1698 scsi_device_blocked(sdev))
1699 ret = BLK_STS_DEV_RESOURCE;
1703 * Make sure to release all allocated ressources when
1704 * we hit an error, as we will never see this command
1707 if (req->rq_flags & RQF_DONTPREP)
1708 scsi_mq_uninit_cmd(cmd);
1714 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1718 return BLK_EH_RESET_TIMER;
1719 return scsi_times_out(req);
1722 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1723 unsigned int hctx_idx, unsigned int numa_node)
1725 struct Scsi_Host *shost = set->driver_data;
1726 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1727 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1728 struct scatterlist *sg;
1730 if (unchecked_isa_dma)
1731 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1732 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1733 GFP_KERNEL, numa_node);
1734 if (!cmd->sense_buffer)
1736 cmd->req.sense = cmd->sense_buffer;
1738 if (scsi_host_get_prot(shost)) {
1739 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1740 shost->hostt->cmd_size;
1741 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
1747 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1748 unsigned int hctx_idx)
1750 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1752 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1756 static int scsi_map_queues(struct blk_mq_tag_set *set)
1758 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1760 if (shost->hostt->map_queues)
1761 return shost->hostt->map_queues(shost);
1762 return blk_mq_map_queues(&set->map[0]);
1765 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1767 struct device *dev = shost->dma_dev;
1770 * this limit is imposed by hardware restrictions
1772 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1775 if (scsi_host_prot_dma(shost)) {
1776 shost->sg_prot_tablesize =
1777 min_not_zero(shost->sg_prot_tablesize,
1778 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1779 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1780 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1783 blk_queue_max_hw_sectors(q, shost->max_sectors);
1784 if (shost->unchecked_isa_dma)
1785 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1786 blk_queue_segment_boundary(q, shost->dma_boundary);
1787 dma_set_seg_boundary(dev, shost->dma_boundary);
1789 blk_queue_max_segment_size(q, shost->max_segment_size);
1790 dma_set_max_seg_size(dev, shost->max_segment_size);
1793 * Set a reasonable default alignment: The larger of 32-byte (dword),
1794 * which is a common minimum for HBAs, and the minimum DMA alignment,
1795 * which is set by the platform.
1797 * Devices that require a bigger alignment can increase it later.
1799 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1801 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1803 static const struct blk_mq_ops scsi_mq_ops = {
1804 .get_budget = scsi_mq_get_budget,
1805 .put_budget = scsi_mq_put_budget,
1806 .queue_rq = scsi_queue_rq,
1807 .complete = scsi_softirq_done,
1808 .timeout = scsi_timeout,
1809 #ifdef CONFIG_BLK_DEBUG_FS
1810 .show_rq = scsi_show_rq,
1812 .init_request = scsi_mq_init_request,
1813 .exit_request = scsi_mq_exit_request,
1814 .initialize_rq_fn = scsi_initialize_rq,
1815 .busy = scsi_mq_lld_busy,
1816 .map_queues = scsi_map_queues,
1819 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1821 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1822 if (IS_ERR(sdev->request_queue))
1825 sdev->request_queue->queuedata = sdev;
1826 __scsi_init_queue(sdev->host, sdev->request_queue);
1827 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1828 return sdev->request_queue;
1831 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1833 unsigned int cmd_size, sgl_size;
1835 sgl_size = scsi_mq_sgl_size(shost);
1836 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1837 if (scsi_host_get_prot(shost))
1838 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
1840 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1841 shost->tag_set.ops = &scsi_mq_ops;
1842 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1843 shost->tag_set.queue_depth = shost->can_queue;
1844 shost->tag_set.cmd_size = cmd_size;
1845 shost->tag_set.numa_node = NUMA_NO_NODE;
1846 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1847 shost->tag_set.flags |=
1848 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1849 shost->tag_set.driver_data = shost;
1851 return blk_mq_alloc_tag_set(&shost->tag_set);
1854 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1856 blk_mq_free_tag_set(&shost->tag_set);
1860 * scsi_device_from_queue - return sdev associated with a request_queue
1861 * @q: The request queue to return the sdev from
1863 * Return the sdev associated with a request queue or NULL if the
1864 * request_queue does not reference a SCSI device.
1866 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1868 struct scsi_device *sdev = NULL;
1870 if (q->mq_ops == &scsi_mq_ops)
1871 sdev = q->queuedata;
1872 if (!sdev || !get_device(&sdev->sdev_gendev))
1877 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1880 * Function: scsi_block_requests()
1882 * Purpose: Utility function used by low-level drivers to prevent further
1883 * commands from being queued to the device.
1885 * Arguments: shost - Host in question
1889 * Lock status: No locks are assumed held.
1891 * Notes: There is no timer nor any other means by which the requests
1892 * get unblocked other than the low-level driver calling
1893 * scsi_unblock_requests().
1895 void scsi_block_requests(struct Scsi_Host *shost)
1897 shost->host_self_blocked = 1;
1899 EXPORT_SYMBOL(scsi_block_requests);
1902 * Function: scsi_unblock_requests()
1904 * Purpose: Utility function used by low-level drivers to allow further
1905 * commands from being queued to the device.
1907 * Arguments: shost - Host in question
1911 * Lock status: No locks are assumed held.
1913 * Notes: There is no timer nor any other means by which the requests
1914 * get unblocked other than the low-level driver calling
1915 * scsi_unblock_requests().
1917 * This is done as an API function so that changes to the
1918 * internals of the scsi mid-layer won't require wholesale
1919 * changes to drivers that use this feature.
1921 void scsi_unblock_requests(struct Scsi_Host *shost)
1923 shost->host_self_blocked = 0;
1924 scsi_run_host_queues(shost);
1926 EXPORT_SYMBOL(scsi_unblock_requests);
1928 int __init scsi_init_queue(void)
1930 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1931 sizeof(struct scsi_data_buffer),
1933 if (!scsi_sdb_cache) {
1934 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1941 void scsi_exit_queue(void)
1943 kmem_cache_destroy(scsi_sense_cache);
1944 kmem_cache_destroy(scsi_sense_isadma_cache);
1945 kmem_cache_destroy(scsi_sdb_cache);
1949 * scsi_mode_select - issue a mode select
1950 * @sdev: SCSI device to be queried
1951 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1952 * @sp: Save page bit (0 == don't save, 1 == save)
1953 * @modepage: mode page being requested
1954 * @buffer: request buffer (may not be smaller than eight bytes)
1955 * @len: length of request buffer.
1956 * @timeout: command timeout
1957 * @retries: number of retries before failing
1958 * @data: returns a structure abstracting the mode header data
1959 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1960 * must be SCSI_SENSE_BUFFERSIZE big.
1962 * Returns zero if successful; negative error number or scsi
1967 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1968 unsigned char *buffer, int len, int timeout, int retries,
1969 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1971 unsigned char cmd[10];
1972 unsigned char *real_buffer;
1975 memset(cmd, 0, sizeof(cmd));
1976 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1978 if (sdev->use_10_for_ms) {
1981 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1984 memcpy(real_buffer + 8, buffer, len);
1988 real_buffer[2] = data->medium_type;
1989 real_buffer[3] = data->device_specific;
1990 real_buffer[4] = data->longlba ? 0x01 : 0;
1992 real_buffer[6] = data->block_descriptor_length >> 8;
1993 real_buffer[7] = data->block_descriptor_length;
1995 cmd[0] = MODE_SELECT_10;
1999 if (len > 255 || data->block_descriptor_length > 255 ||
2003 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2006 memcpy(real_buffer + 4, buffer, len);
2009 real_buffer[1] = data->medium_type;
2010 real_buffer[2] = data->device_specific;
2011 real_buffer[3] = data->block_descriptor_length;
2014 cmd[0] = MODE_SELECT;
2018 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2019 sshdr, timeout, retries, NULL);
2023 EXPORT_SYMBOL_GPL(scsi_mode_select);
2026 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2027 * @sdev: SCSI device to be queried
2028 * @dbd: set if mode sense will allow block descriptors to be returned
2029 * @modepage: mode page being requested
2030 * @buffer: request buffer (may not be smaller than eight bytes)
2031 * @len: length of request buffer.
2032 * @timeout: command timeout
2033 * @retries: number of retries before failing
2034 * @data: returns a structure abstracting the mode header data
2035 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2036 * must be SCSI_SENSE_BUFFERSIZE big.
2038 * Returns zero if unsuccessful, or the header offset (either 4
2039 * or 8 depending on whether a six or ten byte command was
2040 * issued) if successful.
2043 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2044 unsigned char *buffer, int len, int timeout, int retries,
2045 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2047 unsigned char cmd[12];
2050 int result, retry_count = retries;
2051 struct scsi_sense_hdr my_sshdr;
2053 memset(data, 0, sizeof(*data));
2054 memset(&cmd[0], 0, 12);
2055 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2058 /* caller might not be interested in sense, but we need it */
2063 use_10_for_ms = sdev->use_10_for_ms;
2065 if (use_10_for_ms) {
2069 cmd[0] = MODE_SENSE_10;
2076 cmd[0] = MODE_SENSE;
2081 memset(buffer, 0, len);
2083 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2084 sshdr, timeout, retries, NULL);
2086 /* This code looks awful: what it's doing is making sure an
2087 * ILLEGAL REQUEST sense return identifies the actual command
2088 * byte as the problem. MODE_SENSE commands can return
2089 * ILLEGAL REQUEST if the code page isn't supported */
2091 if (use_10_for_ms && !scsi_status_is_good(result) &&
2092 driver_byte(result) == DRIVER_SENSE) {
2093 if (scsi_sense_valid(sshdr)) {
2094 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2095 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2097 * Invalid command operation code
2099 sdev->use_10_for_ms = 0;
2105 if(scsi_status_is_good(result)) {
2106 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2107 (modepage == 6 || modepage == 8))) {
2108 /* Initio breakage? */
2111 data->medium_type = 0;
2112 data->device_specific = 0;
2114 data->block_descriptor_length = 0;
2115 } else if(use_10_for_ms) {
2116 data->length = buffer[0]*256 + buffer[1] + 2;
2117 data->medium_type = buffer[2];
2118 data->device_specific = buffer[3];
2119 data->longlba = buffer[4] & 0x01;
2120 data->block_descriptor_length = buffer[6]*256
2123 data->length = buffer[0] + 1;
2124 data->medium_type = buffer[1];
2125 data->device_specific = buffer[2];
2126 data->block_descriptor_length = buffer[3];
2128 data->header_length = header_length;
2129 } else if ((status_byte(result) == CHECK_CONDITION) &&
2130 scsi_sense_valid(sshdr) &&
2131 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2138 EXPORT_SYMBOL(scsi_mode_sense);
2141 * scsi_test_unit_ready - test if unit is ready
2142 * @sdev: scsi device to change the state of.
2143 * @timeout: command timeout
2144 * @retries: number of retries before failing
2145 * @sshdr: outpout pointer for decoded sense information.
2147 * Returns zero if unsuccessful or an error if TUR failed. For
2148 * removable media, UNIT_ATTENTION sets ->changed flag.
2151 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2152 struct scsi_sense_hdr *sshdr)
2155 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2159 /* try to eat the UNIT_ATTENTION if there are enough retries */
2161 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2163 if (sdev->removable && scsi_sense_valid(sshdr) &&
2164 sshdr->sense_key == UNIT_ATTENTION)
2166 } while (scsi_sense_valid(sshdr) &&
2167 sshdr->sense_key == UNIT_ATTENTION && --retries);
2171 EXPORT_SYMBOL(scsi_test_unit_ready);
2174 * scsi_device_set_state - Take the given device through the device state model.
2175 * @sdev: scsi device to change the state of.
2176 * @state: state to change to.
2178 * Returns zero if successful or an error if the requested
2179 * transition is illegal.
2182 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2184 enum scsi_device_state oldstate = sdev->sdev_state;
2186 if (state == oldstate)
2192 case SDEV_CREATED_BLOCK:
2203 case SDEV_TRANSPORT_OFFLINE:
2216 case SDEV_TRANSPORT_OFFLINE:
2224 case SDEV_TRANSPORT_OFFLINE:
2239 case SDEV_CREATED_BLOCK:
2247 case SDEV_CREATED_BLOCK:
2262 case SDEV_TRANSPORT_OFFLINE:
2274 case SDEV_TRANSPORT_OFFLINE:
2277 case SDEV_CREATED_BLOCK:
2285 sdev->sdev_state = state;
2289 SCSI_LOG_ERROR_RECOVERY(1,
2290 sdev_printk(KERN_ERR, sdev,
2291 "Illegal state transition %s->%s",
2292 scsi_device_state_name(oldstate),
2293 scsi_device_state_name(state))
2297 EXPORT_SYMBOL(scsi_device_set_state);
2300 * sdev_evt_emit - emit a single SCSI device uevent
2301 * @sdev: associated SCSI device
2302 * @evt: event to emit
2304 * Send a single uevent (scsi_event) to the associated scsi_device.
2306 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2311 switch (evt->evt_type) {
2312 case SDEV_EVT_MEDIA_CHANGE:
2313 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2315 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2316 scsi_rescan_device(&sdev->sdev_gendev);
2317 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2319 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2320 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2322 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2323 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2325 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2326 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2328 case SDEV_EVT_LUN_CHANGE_REPORTED:
2329 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2331 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2332 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2334 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2335 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2344 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2348 * sdev_evt_thread - send a uevent for each scsi event
2349 * @work: work struct for scsi_device
2351 * Dispatch queued events to their associated scsi_device kobjects
2354 void scsi_evt_thread(struct work_struct *work)
2356 struct scsi_device *sdev;
2357 enum scsi_device_event evt_type;
2358 LIST_HEAD(event_list);
2360 sdev = container_of(work, struct scsi_device, event_work);
2362 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2363 if (test_and_clear_bit(evt_type, sdev->pending_events))
2364 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2367 struct scsi_event *evt;
2368 struct list_head *this, *tmp;
2369 unsigned long flags;
2371 spin_lock_irqsave(&sdev->list_lock, flags);
2372 list_splice_init(&sdev->event_list, &event_list);
2373 spin_unlock_irqrestore(&sdev->list_lock, flags);
2375 if (list_empty(&event_list))
2378 list_for_each_safe(this, tmp, &event_list) {
2379 evt = list_entry(this, struct scsi_event, node);
2380 list_del(&evt->node);
2381 scsi_evt_emit(sdev, evt);
2388 * sdev_evt_send - send asserted event to uevent thread
2389 * @sdev: scsi_device event occurred on
2390 * @evt: event to send
2392 * Assert scsi device event asynchronously.
2394 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2396 unsigned long flags;
2399 /* FIXME: currently this check eliminates all media change events
2400 * for polled devices. Need to update to discriminate between AN
2401 * and polled events */
2402 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2408 spin_lock_irqsave(&sdev->list_lock, flags);
2409 list_add_tail(&evt->node, &sdev->event_list);
2410 schedule_work(&sdev->event_work);
2411 spin_unlock_irqrestore(&sdev->list_lock, flags);
2413 EXPORT_SYMBOL_GPL(sdev_evt_send);
2416 * sdev_evt_alloc - allocate a new scsi event
2417 * @evt_type: type of event to allocate
2418 * @gfpflags: GFP flags for allocation
2420 * Allocates and returns a new scsi_event.
2422 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2425 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2429 evt->evt_type = evt_type;
2430 INIT_LIST_HEAD(&evt->node);
2432 /* evt_type-specific initialization, if any */
2434 case SDEV_EVT_MEDIA_CHANGE:
2435 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2436 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2437 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2438 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2439 case SDEV_EVT_LUN_CHANGE_REPORTED:
2440 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2441 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2449 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2452 * sdev_evt_send_simple - send asserted event to uevent thread
2453 * @sdev: scsi_device event occurred on
2454 * @evt_type: type of event to send
2455 * @gfpflags: GFP flags for allocation
2457 * Assert scsi device event asynchronously, given an event type.
2459 void sdev_evt_send_simple(struct scsi_device *sdev,
2460 enum scsi_device_event evt_type, gfp_t gfpflags)
2462 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2464 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2469 sdev_evt_send(sdev, evt);
2471 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2474 * scsi_device_quiesce - Block user issued commands.
2475 * @sdev: scsi device to quiesce.
2477 * This works by trying to transition to the SDEV_QUIESCE state
2478 * (which must be a legal transition). When the device is in this
2479 * state, only special requests will be accepted, all others will
2480 * be deferred. Since special requests may also be requeued requests,
2481 * a successful return doesn't guarantee the device will be
2482 * totally quiescent.
2484 * Must be called with user context, may sleep.
2486 * Returns zero if unsuccessful or an error if not.
2489 scsi_device_quiesce(struct scsi_device *sdev)
2491 struct request_queue *q = sdev->request_queue;
2495 * It is allowed to call scsi_device_quiesce() multiple times from
2496 * the same context but concurrent scsi_device_quiesce() calls are
2499 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2501 if (sdev->quiesced_by == current)
2506 blk_mq_freeze_queue(q);
2508 * Ensure that the effect of blk_set_pm_only() will be visible
2509 * for percpu_ref_tryget() callers that occur after the queue
2510 * unfreeze even if the queue was already frozen before this function
2511 * was called. See also https://lwn.net/Articles/573497/.
2514 blk_mq_unfreeze_queue(q);
2516 mutex_lock(&sdev->state_mutex);
2517 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2519 sdev->quiesced_by = current;
2521 blk_clear_pm_only(q);
2522 mutex_unlock(&sdev->state_mutex);
2526 EXPORT_SYMBOL(scsi_device_quiesce);
2529 * scsi_device_resume - Restart user issued commands to a quiesced device.
2530 * @sdev: scsi device to resume.
2532 * Moves the device from quiesced back to running and restarts the
2535 * Must be called with user context, may sleep.
2537 void scsi_device_resume(struct scsi_device *sdev)
2539 /* check if the device state was mutated prior to resume, and if
2540 * so assume the state is being managed elsewhere (for example
2541 * device deleted during suspend)
2543 mutex_lock(&sdev->state_mutex);
2544 sdev->quiesced_by = NULL;
2545 blk_clear_pm_only(sdev->request_queue);
2546 if (sdev->sdev_state == SDEV_QUIESCE)
2547 scsi_device_set_state(sdev, SDEV_RUNNING);
2548 mutex_unlock(&sdev->state_mutex);
2550 EXPORT_SYMBOL(scsi_device_resume);
2553 device_quiesce_fn(struct scsi_device *sdev, void *data)
2555 scsi_device_quiesce(sdev);
2559 scsi_target_quiesce(struct scsi_target *starget)
2561 starget_for_each_device(starget, NULL, device_quiesce_fn);
2563 EXPORT_SYMBOL(scsi_target_quiesce);
2566 device_resume_fn(struct scsi_device *sdev, void *data)
2568 scsi_device_resume(sdev);
2572 scsi_target_resume(struct scsi_target *starget)
2574 starget_for_each_device(starget, NULL, device_resume_fn);
2576 EXPORT_SYMBOL(scsi_target_resume);
2579 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2580 * @sdev: device to block
2582 * Pause SCSI command processing on the specified device. Does not sleep.
2584 * Returns zero if successful or a negative error code upon failure.
2587 * This routine transitions the device to the SDEV_BLOCK state (which must be
2588 * a legal transition). When the device is in this state, command processing
2589 * is paused until the device leaves the SDEV_BLOCK state. See also
2590 * scsi_internal_device_unblock_nowait().
2592 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2594 struct request_queue *q = sdev->request_queue;
2597 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2599 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2606 * The device has transitioned to SDEV_BLOCK. Stop the
2607 * block layer from calling the midlayer with this device's
2610 blk_mq_quiesce_queue_nowait(q);
2613 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2616 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2617 * @sdev: device to block
2619 * Pause SCSI command processing on the specified device and wait until all
2620 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2622 * Returns zero if successful or a negative error code upon failure.
2625 * This routine transitions the device to the SDEV_BLOCK state (which must be
2626 * a legal transition). When the device is in this state, command processing
2627 * is paused until the device leaves the SDEV_BLOCK state. See also
2628 * scsi_internal_device_unblock().
2630 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2631 * scsi_internal_device_block() has blocked a SCSI device and also
2632 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2634 static int scsi_internal_device_block(struct scsi_device *sdev)
2636 struct request_queue *q = sdev->request_queue;
2639 mutex_lock(&sdev->state_mutex);
2640 err = scsi_internal_device_block_nowait(sdev);
2642 blk_mq_quiesce_queue(q);
2643 mutex_unlock(&sdev->state_mutex);
2648 void scsi_start_queue(struct scsi_device *sdev)
2650 struct request_queue *q = sdev->request_queue;
2652 blk_mq_unquiesce_queue(q);
2656 * scsi_internal_device_unblock_nowait - resume a device after a block request
2657 * @sdev: device to resume
2658 * @new_state: state to set the device to after unblocking
2660 * Restart the device queue for a previously suspended SCSI device. Does not
2663 * Returns zero if successful or a negative error code upon failure.
2666 * This routine transitions the device to the SDEV_RUNNING state or to one of
2667 * the offline states (which must be a legal transition) allowing the midlayer
2668 * to goose the queue for this device.
2670 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2671 enum scsi_device_state new_state)
2674 * Try to transition the scsi device to SDEV_RUNNING or one of the
2675 * offlined states and goose the device queue if successful.
2677 switch (sdev->sdev_state) {
2679 case SDEV_TRANSPORT_OFFLINE:
2680 sdev->sdev_state = new_state;
2682 case SDEV_CREATED_BLOCK:
2683 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2684 new_state == SDEV_OFFLINE)
2685 sdev->sdev_state = new_state;
2687 sdev->sdev_state = SDEV_CREATED;
2695 scsi_start_queue(sdev);
2699 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2702 * scsi_internal_device_unblock - resume a device after a block request
2703 * @sdev: device to resume
2704 * @new_state: state to set the device to after unblocking
2706 * Restart the device queue for a previously suspended SCSI device. May sleep.
2708 * Returns zero if successful or a negative error code upon failure.
2711 * This routine transitions the device to the SDEV_RUNNING state or to one of
2712 * the offline states (which must be a legal transition) allowing the midlayer
2713 * to goose the queue for this device.
2715 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2716 enum scsi_device_state new_state)
2720 mutex_lock(&sdev->state_mutex);
2721 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2722 mutex_unlock(&sdev->state_mutex);
2728 device_block(struct scsi_device *sdev, void *data)
2730 scsi_internal_device_block(sdev);
2734 target_block(struct device *dev, void *data)
2736 if (scsi_is_target_device(dev))
2737 starget_for_each_device(to_scsi_target(dev), NULL,
2743 scsi_target_block(struct device *dev)
2745 if (scsi_is_target_device(dev))
2746 starget_for_each_device(to_scsi_target(dev), NULL,
2749 device_for_each_child(dev, NULL, target_block);
2751 EXPORT_SYMBOL_GPL(scsi_target_block);
2754 device_unblock(struct scsi_device *sdev, void *data)
2756 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2760 target_unblock(struct device *dev, void *data)
2762 if (scsi_is_target_device(dev))
2763 starget_for_each_device(to_scsi_target(dev), data,
2769 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2771 if (scsi_is_target_device(dev))
2772 starget_for_each_device(to_scsi_target(dev), &new_state,
2775 device_for_each_child(dev, &new_state, target_unblock);
2777 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2780 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2781 * @sgl: scatter-gather list
2782 * @sg_count: number of segments in sg
2783 * @offset: offset in bytes into sg, on return offset into the mapped area
2784 * @len: bytes to map, on return number of bytes mapped
2786 * Returns virtual address of the start of the mapped page
2788 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2789 size_t *offset, size_t *len)
2792 size_t sg_len = 0, len_complete = 0;
2793 struct scatterlist *sg;
2796 WARN_ON(!irqs_disabled());
2798 for_each_sg(sgl, sg, sg_count, i) {
2799 len_complete = sg_len; /* Complete sg-entries */
2800 sg_len += sg->length;
2801 if (sg_len > *offset)
2805 if (unlikely(i == sg_count)) {
2806 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2808 __func__, sg_len, *offset, sg_count);
2813 /* Offset starting from the beginning of first page in this sg-entry */
2814 *offset = *offset - len_complete + sg->offset;
2816 /* Assumption: contiguous pages can be accessed as "page + i" */
2817 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2818 *offset &= ~PAGE_MASK;
2820 /* Bytes in this sg-entry from *offset to the end of the page */
2821 sg_len = PAGE_SIZE - *offset;
2825 return kmap_atomic(page);
2827 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2830 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2831 * @virt: virtual address to be unmapped
2833 void scsi_kunmap_atomic_sg(void *virt)
2835 kunmap_atomic(virt);
2837 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2839 void sdev_disable_disk_events(struct scsi_device *sdev)
2841 atomic_inc(&sdev->disk_events_disable_depth);
2843 EXPORT_SYMBOL(sdev_disable_disk_events);
2845 void sdev_enable_disk_events(struct scsi_device *sdev)
2847 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2849 atomic_dec(&sdev->disk_events_disable_depth);
2851 EXPORT_SYMBOL(sdev_enable_disk_events);
2854 * scsi_vpd_lun_id - return a unique device identification
2855 * @sdev: SCSI device
2856 * @id: buffer for the identification
2857 * @id_len: length of the buffer
2859 * Copies a unique device identification into @id based
2860 * on the information in the VPD page 0x83 of the device.
2861 * The string will be formatted as a SCSI name string.
2863 * Returns the length of the identification or error on failure.
2864 * If the identifier is longer than the supplied buffer the actual
2865 * identifier length is returned and the buffer is not zero-padded.
2867 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2869 u8 cur_id_type = 0xff;
2871 const unsigned char *d, *cur_id_str;
2872 const struct scsi_vpd *vpd_pg83;
2873 int id_size = -EINVAL;
2876 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2883 * Look for the correct descriptor.
2884 * Order of preference for lun descriptor:
2885 * - SCSI name string
2886 * - NAA IEEE Registered Extended
2887 * - EUI-64 based 16-byte
2888 * - EUI-64 based 12-byte
2889 * - NAA IEEE Registered
2890 * - NAA IEEE Extended
2892 * as longer descriptors reduce the likelyhood
2893 * of identification clashes.
2896 /* The id string must be at least 20 bytes + terminating NULL byte */
2902 memset(id, 0, id_len);
2903 d = vpd_pg83->data + 4;
2904 while (d < vpd_pg83->data + vpd_pg83->len) {
2905 /* Skip designators not referring to the LUN */
2906 if ((d[1] & 0x30) != 0x00)
2909 switch (d[1] & 0xf) {
2912 if (cur_id_size > d[3])
2914 /* Prefer anything */
2915 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2918 if (cur_id_size + 4 > id_len)
2919 cur_id_size = id_len - 4;
2921 cur_id_type = d[1] & 0xf;
2922 id_size = snprintf(id, id_len, "t10.%*pE",
2923 cur_id_size, cur_id_str);
2927 if (cur_id_size > d[3])
2929 /* Prefer NAA IEEE Registered Extended */
2930 if (cur_id_type == 0x3 &&
2931 cur_id_size == d[3])
2935 cur_id_type = d[1] & 0xf;
2936 switch (cur_id_size) {
2938 id_size = snprintf(id, id_len,
2943 id_size = snprintf(id, id_len,
2948 id_size = snprintf(id, id_len,
2959 if (cur_id_size > d[3])
2963 cur_id_type = d[1] & 0xf;
2964 switch (cur_id_size) {
2966 id_size = snprintf(id, id_len,
2971 id_size = snprintf(id, id_len,
2981 /* SCSI name string */
2982 if (cur_id_size + 4 > d[3])
2984 /* Prefer others for truncated descriptor */
2985 if (cur_id_size && d[3] > id_len)
2987 cur_id_size = id_size = d[3];
2989 cur_id_type = d[1] & 0xf;
2990 if (cur_id_size >= id_len)
2991 cur_id_size = id_len - 1;
2992 memcpy(id, cur_id_str, cur_id_size);
2993 /* Decrease priority for truncated descriptor */
2994 if (cur_id_size != id_size)
3007 EXPORT_SYMBOL(scsi_vpd_lun_id);
3010 * scsi_vpd_tpg_id - return a target port group identifier
3011 * @sdev: SCSI device
3013 * Returns the Target Port Group identifier from the information
3014 * froom VPD page 0x83 of the device.
3016 * Returns the identifier or error on failure.
3018 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3020 const unsigned char *d;
3021 const struct scsi_vpd *vpd_pg83;
3022 int group_id = -EAGAIN, rel_port = -1;
3025 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3031 d = vpd_pg83->data + 4;
3032 while (d < vpd_pg83->data + vpd_pg83->len) {
3033 switch (d[1] & 0xf) {
3035 /* Relative target port */
3036 rel_port = get_unaligned_be16(&d[6]);
3039 /* Target port group */
3040 group_id = get_unaligned_be16(&d[6]);
3049 if (group_id >= 0 && rel_id && rel_port != -1)
3054 EXPORT_SYMBOL(scsi_vpd_tpg_id);