1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/ratelimit.h>
25 #include <asm/unaligned.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
37 #include <trace/events/scsi.h>
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
44 * Size of integrity metadata is usually small, 1 inline sg should
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define SCSI_INLINE_PROT_SG_CNT 0
49 #define SCSI_INLINE_SG_CNT 0
51 #define SCSI_INLINE_PROT_SG_CNT 1
52 #define SCSI_INLINE_SG_CNT 2
55 static struct kmem_cache *scsi_sdb_cache;
56 static struct kmem_cache *scsi_sense_cache;
57 static struct kmem_cache *scsi_sense_isadma_cache;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex);
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
62 static inline struct kmem_cache *
63 scsi_select_sense_cache(bool unchecked_isa_dma)
65 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69 unsigned char *sense_buffer)
71 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
76 gfp_t gfp_mask, int numa_node)
78 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
82 int scsi_init_sense_cache(struct Scsi_Host *shost)
84 struct kmem_cache *cache;
87 mutex_lock(&scsi_sense_cache_mutex);
88 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
92 if (shost->unchecked_isa_dma) {
93 scsi_sense_isadma_cache =
94 kmem_cache_create("scsi_sense_cache(DMA)",
95 SCSI_SENSE_BUFFERSIZE, 0,
96 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
97 if (!scsi_sense_isadma_cache)
101 kmem_cache_create_usercopy("scsi_sense_cache",
102 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103 0, SCSI_SENSE_BUFFERSIZE, NULL);
104 if (!scsi_sense_cache)
108 mutex_unlock(&scsi_sense_cache_mutex);
113 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114 * not change behaviour from the previous unplug mechanism, experimentation
115 * may prove this needs changing.
117 #define SCSI_QUEUE_DELAY 3
120 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
122 struct Scsi_Host *host = cmd->device->host;
123 struct scsi_device *device = cmd->device;
124 struct scsi_target *starget = scsi_target(device);
127 * Set the appropriate busy bit for the device/host.
129 * If the host/device isn't busy, assume that something actually
130 * completed, and that we should be able to queue a command now.
132 * Note that the prior mid-layer assumption that any host could
133 * always queue at least one command is now broken. The mid-layer
134 * will implement a user specifiable stall (see
135 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136 * if a command is requeued with no other commands outstanding
137 * either for the device or for the host.
140 case SCSI_MLQUEUE_HOST_BUSY:
141 atomic_set(&host->host_blocked, host->max_host_blocked);
143 case SCSI_MLQUEUE_DEVICE_BUSY:
144 case SCSI_MLQUEUE_EH_RETRY:
145 atomic_set(&device->device_blocked,
146 device->max_device_blocked);
148 case SCSI_MLQUEUE_TARGET_BUSY:
149 atomic_set(&starget->target_blocked,
150 starget->max_target_blocked);
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
157 if (cmd->request->rq_flags & RQF_DONTPREP) {
158 cmd->request->rq_flags &= ~RQF_DONTPREP;
159 scsi_mq_uninit_cmd(cmd);
163 blk_mq_requeue_request(cmd->request, true);
167 * __scsi_queue_insert - private queue insertion
168 * @cmd: The SCSI command being requeued
169 * @reason: The reason for the requeue
170 * @unbusy: Whether the queue should be unbusied
172 * This is a private queue insertion. The public interface
173 * scsi_queue_insert() always assumes the queue should be unbusied
174 * because it's always called before the completion. This function is
175 * for a requeue after completion, which should only occur in this
178 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
180 struct scsi_device *device = cmd->device;
182 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183 "Inserting command %p into mlqueue\n", cmd));
185 scsi_set_blocked(cmd, reason);
188 * Decrement the counters, since these commands are no longer
189 * active on the host/device.
192 scsi_device_unbusy(device);
195 * Requeue this command. It will go before all other commands
196 * that are already in the queue. Schedule requeue work under
197 * lock such that the kblockd_schedule_work() call happens
198 * before blk_cleanup_queue() finishes.
202 blk_mq_requeue_request(cmd->request, true);
206 * Function: scsi_queue_insert()
208 * Purpose: Insert a command in the midlevel queue.
210 * Arguments: cmd - command that we are adding to queue.
211 * reason - why we are inserting command to queue.
213 * Lock status: Assumed that lock is not held upon entry.
217 * Notes: We do this for one of two cases. Either the host is busy
218 * and it cannot accept any more commands for the time being,
219 * or the device returned QUEUE_FULL and can accept no more
221 * Notes: This could be called either from an interrupt context or a
222 * normal process context.
224 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
226 __scsi_queue_insert(cmd, reason, true);
231 * __scsi_execute - insert request and wait for the result
234 * @data_direction: data direction
235 * @buffer: data buffer
236 * @bufflen: len of buffer
237 * @sense: optional sense buffer
238 * @sshdr: optional decoded sense header
239 * @timeout: request timeout in seconds
240 * @retries: number of times to retry request
241 * @flags: flags for ->cmd_flags
242 * @rq_flags: flags for ->rq_flags
243 * @resid: optional residual length
245 * Returns the scsi_cmnd result field if a command was executed, or a negative
246 * Linux error code if we didn't get that far.
248 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
249 int data_direction, void *buffer, unsigned bufflen,
250 unsigned char *sense, struct scsi_sense_hdr *sshdr,
251 int timeout, int retries, u64 flags, req_flags_t rq_flags,
255 struct scsi_request *rq;
256 int ret = DRIVER_ERROR << 24;
258 req = blk_get_request(sdev->request_queue,
259 data_direction == DMA_TO_DEVICE ?
260 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
265 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
266 buffer, bufflen, GFP_NOIO))
269 rq->cmd_len = COMMAND_SIZE(cmd[0]);
270 memcpy(rq->cmd, cmd, rq->cmd_len);
271 rq->retries = retries;
272 req->timeout = timeout;
273 req->cmd_flags |= flags;
274 req->rq_flags |= rq_flags | RQF_QUIET;
277 * head injection *required* here otherwise quiesce won't work
279 blk_execute_rq(req->q, NULL, req, 1);
282 * Some devices (USB mass-storage in particular) may transfer
283 * garbage data together with a residue indicating that the data
284 * is invalid. Prevent the garbage from being misinterpreted
285 * and prevent security leaks by zeroing out the excess data.
287 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
291 *resid = rq->resid_len;
292 if (sense && rq->sense_len)
293 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
295 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
298 blk_put_request(req);
302 EXPORT_SYMBOL(__scsi_execute);
305 * Function: scsi_init_cmd_errh()
307 * Purpose: Initialize cmd fields related to error handling.
309 * Arguments: cmd - command that is ready to be queued.
311 * Notes: This function has the job of initializing a number of
312 * fields related to error handling. Typically this will
313 * be called once for each command, as required.
315 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
317 scsi_set_resid(cmd, 0);
318 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
319 if (cmd->cmd_len == 0)
320 cmd->cmd_len = scsi_command_size(cmd->cmnd);
324 * Decrement the host_busy counter and wake up the error handler if necessary.
325 * Avoid as follows that the error handler is not woken up if shost->host_busy
326 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327 * with an RCU read lock in this function to ensure that this function in its
328 * entirety either finishes before scsi_eh_scmd_add() increases the
329 * host_failed counter or that it notices the shost state change made by
330 * scsi_eh_scmd_add().
332 static void scsi_dec_host_busy(struct Scsi_Host *shost)
337 atomic_dec(&shost->host_busy);
338 if (unlikely(scsi_host_in_recovery(shost))) {
339 spin_lock_irqsave(shost->host_lock, flags);
340 if (shost->host_failed || shost->host_eh_scheduled)
341 scsi_eh_wakeup(shost);
342 spin_unlock_irqrestore(shost->host_lock, flags);
347 void scsi_device_unbusy(struct scsi_device *sdev)
349 struct Scsi_Host *shost = sdev->host;
350 struct scsi_target *starget = scsi_target(sdev);
352 scsi_dec_host_busy(shost);
354 if (starget->can_queue > 0)
355 atomic_dec(&starget->target_busy);
357 atomic_dec(&sdev->device_busy);
360 static void scsi_kick_queue(struct request_queue *q)
362 blk_mq_run_hw_queues(q, false);
366 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367 * and call blk_run_queue for all the scsi_devices on the target -
368 * including current_sdev first.
370 * Called with *no* scsi locks held.
372 static void scsi_single_lun_run(struct scsi_device *current_sdev)
374 struct Scsi_Host *shost = current_sdev->host;
375 struct scsi_device *sdev, *tmp;
376 struct scsi_target *starget = scsi_target(current_sdev);
379 spin_lock_irqsave(shost->host_lock, flags);
380 starget->starget_sdev_user = NULL;
381 spin_unlock_irqrestore(shost->host_lock, flags);
384 * Call blk_run_queue for all LUNs on the target, starting with
385 * current_sdev. We race with others (to set starget_sdev_user),
386 * but in most cases, we will be first. Ideally, each LU on the
387 * target would get some limited time or requests on the target.
389 scsi_kick_queue(current_sdev->request_queue);
391 spin_lock_irqsave(shost->host_lock, flags);
392 if (starget->starget_sdev_user)
394 list_for_each_entry_safe(sdev, tmp, &starget->devices,
395 same_target_siblings) {
396 if (sdev == current_sdev)
398 if (scsi_device_get(sdev))
401 spin_unlock_irqrestore(shost->host_lock, flags);
402 scsi_kick_queue(sdev->request_queue);
403 spin_lock_irqsave(shost->host_lock, flags);
405 scsi_device_put(sdev);
408 spin_unlock_irqrestore(shost->host_lock, flags);
411 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
413 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
415 if (atomic_read(&sdev->device_blocked) > 0)
420 static inline bool scsi_target_is_busy(struct scsi_target *starget)
422 if (starget->can_queue > 0) {
423 if (atomic_read(&starget->target_busy) >= starget->can_queue)
425 if (atomic_read(&starget->target_blocked) > 0)
431 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
433 if (shost->can_queue > 0 &&
434 atomic_read(&shost->host_busy) >= shost->can_queue)
436 if (atomic_read(&shost->host_blocked) > 0)
438 if (shost->host_self_blocked)
443 static void scsi_starved_list_run(struct Scsi_Host *shost)
445 LIST_HEAD(starved_list);
446 struct scsi_device *sdev;
449 spin_lock_irqsave(shost->host_lock, flags);
450 list_splice_init(&shost->starved_list, &starved_list);
452 while (!list_empty(&starved_list)) {
453 struct request_queue *slq;
456 * As long as shost is accepting commands and we have
457 * starved queues, call blk_run_queue. scsi_request_fn
458 * drops the queue_lock and can add us back to the
461 * host_lock protects the starved_list and starved_entry.
462 * scsi_request_fn must get the host_lock before checking
463 * or modifying starved_list or starved_entry.
465 if (scsi_host_is_busy(shost))
468 sdev = list_entry(starved_list.next,
469 struct scsi_device, starved_entry);
470 list_del_init(&sdev->starved_entry);
471 if (scsi_target_is_busy(scsi_target(sdev))) {
472 list_move_tail(&sdev->starved_entry,
473 &shost->starved_list);
478 * Once we drop the host lock, a racing scsi_remove_device()
479 * call may remove the sdev from the starved list and destroy
480 * it and the queue. Mitigate by taking a reference to the
481 * queue and never touching the sdev again after we drop the
482 * host lock. Note: if __scsi_remove_device() invokes
483 * blk_cleanup_queue() before the queue is run from this
484 * function then blk_run_queue() will return immediately since
485 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
487 slq = sdev->request_queue;
488 if (!blk_get_queue(slq))
490 spin_unlock_irqrestore(shost->host_lock, flags);
492 scsi_kick_queue(slq);
495 spin_lock_irqsave(shost->host_lock, flags);
497 /* put any unprocessed entries back */
498 list_splice(&starved_list, &shost->starved_list);
499 spin_unlock_irqrestore(shost->host_lock, flags);
503 * Function: scsi_run_queue()
505 * Purpose: Select a proper request queue to serve next
507 * Arguments: q - last request's queue
511 * Notes: The previous command was completely finished, start
512 * a new one if possible.
514 static void scsi_run_queue(struct request_queue *q)
516 struct scsi_device *sdev = q->queuedata;
518 if (scsi_target(sdev)->single_lun)
519 scsi_single_lun_run(sdev);
520 if (!list_empty(&sdev->host->starved_list))
521 scsi_starved_list_run(sdev->host);
523 blk_mq_run_hw_queues(q, false);
526 void scsi_requeue_run_queue(struct work_struct *work)
528 struct scsi_device *sdev;
529 struct request_queue *q;
531 sdev = container_of(work, struct scsi_device, requeue_work);
532 q = sdev->request_queue;
536 void scsi_run_host_queues(struct Scsi_Host *shost)
538 struct scsi_device *sdev;
540 shost_for_each_device(sdev, shost)
541 scsi_run_queue(sdev->request_queue);
544 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
546 if (!blk_rq_is_passthrough(cmd->request)) {
547 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
549 if (drv->uninit_command)
550 drv->uninit_command(cmd);
554 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
556 if (cmd->sdb.table.nents)
557 sg_free_table_chained(&cmd->sdb.table,
559 if (scsi_prot_sg_count(cmd))
560 sg_free_table_chained(&cmd->prot_sdb->table,
561 SCSI_INLINE_PROT_SG_CNT);
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;
591 * Calling rcu_barrier() is not necessary here because the
592 * SCSI error handler guarantees that the function called by
593 * call_rcu() has been called before scsi_end_request() is
596 destroy_rcu_head(&cmd->rcu);
599 * In the MQ case the command gets freed by __blk_mq_end_request,
600 * so we have to do all cleanup that depends on it earlier.
602 * We also can't kick the queues from irq context, so we
603 * will have to defer it to a workqueue.
605 scsi_mq_uninit_cmd(cmd);
608 * queue is still alive, so grab the ref for preventing it
609 * from being cleaned up during running queue.
611 percpu_ref_get(&q->q_usage_counter);
613 __blk_mq_end_request(req, error);
615 if (scsi_target(sdev)->single_lun ||
616 !list_empty(&sdev->host->starved_list))
617 kblockd_schedule_work(&sdev->requeue_work);
619 blk_mq_run_hw_queues(q, true);
621 percpu_ref_put(&q->q_usage_counter);
626 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
628 * @result: scsi error code
630 * Translate a SCSI result code into a blk_status_t value. May reset the host
631 * byte of @cmd->result.
633 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
635 switch (host_byte(result)) {
638 * Also check the other bytes than the status byte in result
639 * to handle the case when a SCSI LLD sets result to
640 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
642 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
644 return BLK_STS_IOERR;
645 case DID_TRANSPORT_FAILFAST:
646 return BLK_STS_TRANSPORT;
647 case DID_TARGET_FAILURE:
648 set_host_byte(cmd, DID_OK);
649 return BLK_STS_TARGET;
650 case DID_NEXUS_FAILURE:
651 set_host_byte(cmd, DID_OK);
652 return BLK_STS_NEXUS;
653 case DID_ALLOC_FAILURE:
654 set_host_byte(cmd, DID_OK);
655 return BLK_STS_NOSPC;
656 case DID_MEDIUM_ERROR:
657 set_host_byte(cmd, DID_OK);
658 return BLK_STS_MEDIUM;
660 return BLK_STS_IOERR;
664 /* Helper for scsi_io_completion() when "reprep" action required. */
665 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
666 struct request_queue *q)
668 /* A new command will be prepared and issued. */
669 scsi_mq_requeue_cmd(cmd);
672 /* Helper for scsi_io_completion() when special action required. */
673 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
675 struct request_queue *q = cmd->device->request_queue;
676 struct request *req = cmd->request;
678 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
679 ACTION_DELAYED_RETRY} action;
680 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
681 struct scsi_sense_hdr sshdr;
683 bool sense_current = true; /* false implies "deferred sense" */
684 blk_status_t blk_stat;
686 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
688 sense_current = !scsi_sense_is_deferred(&sshdr);
690 blk_stat = scsi_result_to_blk_status(cmd, result);
692 if (host_byte(result) == DID_RESET) {
693 /* Third party bus reset or reset for error recovery
694 * reasons. Just retry the command and see what
697 action = ACTION_RETRY;
698 } else if (sense_valid && sense_current) {
699 switch (sshdr.sense_key) {
701 if (cmd->device->removable) {
702 /* Detected disc change. Set a bit
703 * and quietly refuse further access.
705 cmd->device->changed = 1;
706 action = ACTION_FAIL;
708 /* Must have been a power glitch, or a
709 * bus reset. Could not have been a
710 * media change, so we just retry the
711 * command and see what happens.
713 action = ACTION_RETRY;
716 case ILLEGAL_REQUEST:
717 /* If we had an ILLEGAL REQUEST returned, then
718 * we may have performed an unsupported
719 * command. The only thing this should be
720 * would be a ten byte read where only a six
721 * byte read was supported. Also, on a system
722 * where READ CAPACITY failed, we may have
723 * read past the end of the disk.
725 if ((cmd->device->use_10_for_rw &&
726 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
727 (cmd->cmnd[0] == READ_10 ||
728 cmd->cmnd[0] == WRITE_10)) {
729 /* This will issue a new 6-byte command. */
730 cmd->device->use_10_for_rw = 0;
731 action = ACTION_REPREP;
732 } else if (sshdr.asc == 0x10) /* DIX */ {
733 action = ACTION_FAIL;
734 blk_stat = BLK_STS_PROTECTION;
735 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
736 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
737 action = ACTION_FAIL;
738 blk_stat = BLK_STS_TARGET;
740 action = ACTION_FAIL;
742 case ABORTED_COMMAND:
743 action = ACTION_FAIL;
744 if (sshdr.asc == 0x10) /* DIF */
745 blk_stat = BLK_STS_PROTECTION;
748 /* If the device is in the process of becoming
749 * ready, or has a temporary blockage, retry.
751 if (sshdr.asc == 0x04) {
752 switch (sshdr.ascq) {
753 case 0x01: /* becoming ready */
754 case 0x04: /* format in progress */
755 case 0x05: /* rebuild in progress */
756 case 0x06: /* recalculation in progress */
757 case 0x07: /* operation in progress */
758 case 0x08: /* Long write in progress */
759 case 0x09: /* self test in progress */
760 case 0x14: /* space allocation in progress */
761 case 0x1a: /* start stop unit in progress */
762 case 0x1b: /* sanitize in progress */
763 case 0x1d: /* configuration in progress */
764 case 0x24: /* depopulation in progress */
765 action = ACTION_DELAYED_RETRY;
768 action = ACTION_FAIL;
772 action = ACTION_FAIL;
774 case VOLUME_OVERFLOW:
775 /* See SSC3rXX or current. */
776 action = ACTION_FAIL;
779 action = ACTION_FAIL;
783 action = ACTION_FAIL;
785 if (action != ACTION_FAIL &&
786 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
787 action = ACTION_FAIL;
791 /* Give up and fail the remainder of the request */
792 if (!(req->rq_flags & RQF_QUIET)) {
793 static DEFINE_RATELIMIT_STATE(_rs,
794 DEFAULT_RATELIMIT_INTERVAL,
795 DEFAULT_RATELIMIT_BURST);
797 if (unlikely(scsi_logging_level))
799 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
800 SCSI_LOG_MLCOMPLETE_BITS);
803 * if logging is enabled the failure will be printed
804 * in scsi_log_completion(), so avoid duplicate messages
806 if (!level && __ratelimit(&_rs)) {
807 scsi_print_result(cmd, NULL, FAILED);
808 if (driver_byte(result) == DRIVER_SENSE)
809 scsi_print_sense(cmd);
810 scsi_print_command(cmd);
813 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
817 scsi_io_completion_reprep(cmd, q);
820 /* Retry the same command immediately */
821 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
823 case ACTION_DELAYED_RETRY:
824 /* Retry the same command after a delay */
825 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
831 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
832 * new result that may suppress further error checking. Also modifies
833 * *blk_statp in some cases.
835 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
836 blk_status_t *blk_statp)
839 bool sense_current = true; /* false implies "deferred sense" */
840 struct request *req = cmd->request;
841 struct scsi_sense_hdr sshdr;
843 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
845 sense_current = !scsi_sense_is_deferred(&sshdr);
847 if (blk_rq_is_passthrough(req)) {
850 * SG_IO wants current and deferred errors
852 scsi_req(req)->sense_len =
853 min(8 + cmd->sense_buffer[7],
854 SCSI_SENSE_BUFFERSIZE);
857 *blk_statp = scsi_result_to_blk_status(cmd, result);
858 } else if (blk_rq_bytes(req) == 0 && sense_current) {
860 * Flush commands do not transfers any data, and thus cannot use
861 * good_bytes != blk_rq_bytes(req) as the signal for an error.
862 * This sets *blk_statp explicitly for the problem case.
864 *blk_statp = scsi_result_to_blk_status(cmd, result);
867 * Recovered errors need reporting, but they're always treated as
868 * success, so fiddle the result code here. For passthrough requests
869 * we already took a copy of the original into sreq->result which
870 * is what gets returned to the user
872 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
873 bool do_print = true;
875 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
876 * skip print since caller wants ATA registers. Only occurs
877 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
879 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
881 else if (req->rq_flags & RQF_QUIET)
884 scsi_print_sense(cmd);
886 /* for passthrough, *blk_statp may be set */
887 *blk_statp = BLK_STS_OK;
890 * Another corner case: the SCSI status byte is non-zero but 'good'.
891 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
892 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
893 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
894 * intermediate statuses (both obsolete in SAM-4) as good.
896 if (status_byte(result) && scsi_status_is_good(result)) {
898 *blk_statp = BLK_STS_OK;
904 * Function: scsi_io_completion()
906 * Purpose: Completion processing for block device I/O requests.
908 * Arguments: cmd - command that is finished.
910 * Lock status: Assumed that no lock is held upon entry.
914 * Notes: We will finish off the specified number of sectors. If we
915 * are done, the command block will be released and the queue
916 * function will be goosed. If we are not done then we have to
917 * figure out what to do next:
919 * a) We can call scsi_requeue_command(). The request
920 * will be unprepared and put back on the queue. Then
921 * a new command will be created for it. This should
922 * be used if we made forward progress, or if we want
923 * to switch from READ(10) to READ(6) for example.
925 * b) We can call __scsi_queue_insert(). The request will
926 * be put back on the queue and retried using the same
927 * command as before, possibly after a delay.
929 * c) We can call scsi_end_request() with blk_stat other than
930 * BLK_STS_OK, to fail the remainder of the request.
932 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
934 int result = cmd->result;
935 struct request_queue *q = cmd->device->request_queue;
936 struct request *req = cmd->request;
937 blk_status_t blk_stat = BLK_STS_OK;
939 if (unlikely(result)) /* a nz result may or may not be an error */
940 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
942 if (unlikely(blk_rq_is_passthrough(req))) {
944 * scsi_result_to_blk_status may have reset the host_byte
946 scsi_req(req)->result = cmd->result;
950 * Next deal with any sectors which we were able to correctly
953 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
954 "%u sectors total, %d bytes done.\n",
955 blk_rq_sectors(req), good_bytes));
958 * Next deal with any sectors which we were able to correctly
959 * handle. Failed, zero length commands always need to drop down
960 * to retry code. Fast path should return in this block.
962 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
963 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
964 return; /* no bytes remaining */
967 /* Kill remainder if no retries. */
968 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
969 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
971 "Bytes remaining after failed, no-retry command");
976 * If there had been no error, but we have leftover bytes in the
977 * requeues just queue the command up again.
979 if (likely(result == 0))
980 scsi_io_completion_reprep(cmd, q);
982 scsi_io_completion_action(cmd, result);
985 static blk_status_t scsi_init_sgtable(struct request *req,
986 struct scsi_data_buffer *sdb)
991 * If sg table allocation fails, requeue request later.
993 if (unlikely(sg_alloc_table_chained(&sdb->table,
994 blk_rq_nr_phys_segments(req), sdb->table.sgl,
995 SCSI_INLINE_SG_CNT)))
996 return BLK_STS_RESOURCE;
999 * Next, walk the list, and fill in the addresses and sizes of
1002 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1003 BUG_ON(count > sdb->table.nents);
1004 sdb->table.nents = count;
1005 sdb->length = blk_rq_payload_bytes(req);
1010 * Function: scsi_init_io()
1012 * Purpose: SCSI I/O initialize function.
1014 * Arguments: cmd - Command descriptor we wish to initialize
1016 * Returns: BLK_STS_OK on success
1017 * BLK_STS_RESOURCE if the failure is retryable
1018 * BLK_STS_IOERR if the failure is fatal
1020 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1022 struct request *rq = cmd->request;
1025 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1026 return BLK_STS_IOERR;
1028 ret = scsi_init_sgtable(rq, &cmd->sdb);
1032 if (blk_integrity_rq(rq)) {
1033 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1036 if (WARN_ON_ONCE(!prot_sdb)) {
1038 * This can happen if someone (e.g. multipath)
1039 * queues a command to a device on an adapter
1040 * that does not support DIX.
1042 ret = BLK_STS_IOERR;
1043 goto out_free_sgtables;
1046 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1048 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049 prot_sdb->table.sgl,
1050 SCSI_INLINE_PROT_SG_CNT)) {
1051 ret = BLK_STS_RESOURCE;
1052 goto out_free_sgtables;
1055 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056 prot_sdb->table.sgl);
1057 BUG_ON(count > ivecs);
1058 BUG_ON(count > queue_max_integrity_segments(rq->q));
1060 cmd->prot_sdb = prot_sdb;
1061 cmd->prot_sdb->table.nents = count;
1066 scsi_mq_free_sgtables(cmd);
1069 EXPORT_SYMBOL(scsi_init_io);
1072 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073 * @rq: Request associated with the SCSI command to be initialized.
1075 * This function initializes the members of struct scsi_cmnd that must be
1076 * initialized before request processing starts and that won't be
1077 * reinitialized if a SCSI command is requeued.
1079 * Called from inside blk_get_request() for pass-through requests and from
1080 * inside scsi_init_command() for filesystem requests.
1082 static void scsi_initialize_rq(struct request *rq)
1084 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1086 scsi_req_init(&cmd->req);
1087 init_rcu_head(&cmd->rcu);
1088 cmd->jiffies_at_alloc = jiffies;
1093 * Only called when the request isn't completed by SCSI, and not freed by
1096 static void scsi_cleanup_rq(struct request *rq)
1098 if (rq->rq_flags & RQF_DONTPREP) {
1099 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1100 rq->rq_flags &= ~RQF_DONTPREP;
1104 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1105 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1107 struct scsi_device *sdev = cmd->device;
1108 struct Scsi_Host *shost = sdev->host;
1109 unsigned long flags;
1111 if (shost->use_cmd_list) {
1112 spin_lock_irqsave(&sdev->list_lock, flags);
1113 list_add_tail(&cmd->list, &sdev->cmd_list);
1114 spin_unlock_irqrestore(&sdev->list_lock, flags);
1118 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1119 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1121 struct scsi_device *sdev = cmd->device;
1122 struct Scsi_Host *shost = sdev->host;
1123 unsigned long flags;
1125 if (shost->use_cmd_list) {
1126 spin_lock_irqsave(&sdev->list_lock, flags);
1127 BUG_ON(list_empty(&cmd->list));
1128 list_del_init(&cmd->list);
1129 spin_unlock_irqrestore(&sdev->list_lock, flags);
1133 /* Called after a request has been started. */
1134 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1136 void *buf = cmd->sense_buffer;
1137 void *prot = cmd->prot_sdb;
1138 struct request *rq = blk_mq_rq_from_pdu(cmd);
1139 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1140 unsigned long jiffies_at_alloc;
1143 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1144 flags |= SCMD_INITIALIZED;
1145 scsi_initialize_rq(rq);
1148 jiffies_at_alloc = cmd->jiffies_at_alloc;
1149 retries = cmd->retries;
1150 /* zero out the cmd, except for the embedded scsi_request */
1151 memset((char *)cmd + sizeof(cmd->req), 0,
1152 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1155 cmd->sense_buffer = buf;
1156 cmd->prot_sdb = prot;
1158 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1159 cmd->jiffies_at_alloc = jiffies_at_alloc;
1160 cmd->retries = retries;
1162 scsi_add_cmd_to_list(cmd);
1165 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1166 struct request *req)
1168 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1171 * Passthrough requests may transfer data, in which case they must
1172 * a bio attached to them. Or they might contain a SCSI command
1173 * that does not transfer data, in which case they may optionally
1174 * submit a request without an attached bio.
1177 blk_status_t ret = scsi_init_io(cmd);
1178 if (unlikely(ret != BLK_STS_OK))
1181 BUG_ON(blk_rq_bytes(req));
1183 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1186 cmd->cmd_len = scsi_req(req)->cmd_len;
1187 cmd->cmnd = scsi_req(req)->cmd;
1188 cmd->transfersize = blk_rq_bytes(req);
1189 cmd->allowed = scsi_req(req)->retries;
1194 * Setup a normal block command. These are simple request from filesystems
1195 * that still need to be translated to SCSI CDBs from the ULD.
1197 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1198 struct request *req)
1200 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1202 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1203 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1204 if (ret != BLK_STS_OK)
1208 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1209 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1210 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1213 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1214 struct request *req)
1216 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1218 if (!blk_rq_bytes(req))
1219 cmd->sc_data_direction = DMA_NONE;
1220 else if (rq_data_dir(req) == WRITE)
1221 cmd->sc_data_direction = DMA_TO_DEVICE;
1223 cmd->sc_data_direction = DMA_FROM_DEVICE;
1225 if (blk_rq_is_scsi(req))
1226 return scsi_setup_scsi_cmnd(sdev, req);
1228 return scsi_setup_fs_cmnd(sdev, req);
1232 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1234 switch (sdev->sdev_state) {
1236 case SDEV_TRANSPORT_OFFLINE:
1238 * If the device is offline we refuse to process any
1239 * commands. The device must be brought online
1240 * before trying any recovery commands.
1242 sdev_printk(KERN_ERR, sdev,
1243 "rejecting I/O to offline device\n");
1244 return BLK_STS_IOERR;
1247 * If the device is fully deleted, we refuse to
1248 * process any commands as well.
1250 sdev_printk(KERN_ERR, sdev,
1251 "rejecting I/O to dead device\n");
1252 return BLK_STS_IOERR;
1254 case SDEV_CREATED_BLOCK:
1255 return BLK_STS_RESOURCE;
1258 * If the devices is blocked we defer normal commands.
1260 if (req && !(req->rq_flags & RQF_PREEMPT))
1261 return BLK_STS_RESOURCE;
1265 * For any other not fully online state we only allow
1266 * special commands. In particular any user initiated
1267 * command is not allowed.
1269 if (req && !(req->rq_flags & RQF_PREEMPT))
1270 return BLK_STS_IOERR;
1276 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1279 * Called with the queue_lock held.
1281 static inline int scsi_dev_queue_ready(struct request_queue *q,
1282 struct scsi_device *sdev)
1286 busy = atomic_inc_return(&sdev->device_busy) - 1;
1287 if (atomic_read(&sdev->device_blocked)) {
1292 * unblock after device_blocked iterates to zero
1294 if (atomic_dec_return(&sdev->device_blocked) > 0)
1296 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1297 "unblocking device at zero depth\n"));
1300 if (busy >= sdev->queue_depth)
1305 atomic_dec(&sdev->device_busy);
1310 * scsi_target_queue_ready: checks if there we can send commands to target
1311 * @sdev: scsi device on starget to check.
1313 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1314 struct scsi_device *sdev)
1316 struct scsi_target *starget = scsi_target(sdev);
1319 if (starget->single_lun) {
1320 spin_lock_irq(shost->host_lock);
1321 if (starget->starget_sdev_user &&
1322 starget->starget_sdev_user != sdev) {
1323 spin_unlock_irq(shost->host_lock);
1326 starget->starget_sdev_user = sdev;
1327 spin_unlock_irq(shost->host_lock);
1330 if (starget->can_queue <= 0)
1333 busy = atomic_inc_return(&starget->target_busy) - 1;
1334 if (atomic_read(&starget->target_blocked) > 0) {
1339 * unblock after target_blocked iterates to zero
1341 if (atomic_dec_return(&starget->target_blocked) > 0)
1344 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1345 "unblocking target at zero depth\n"));
1348 if (busy >= starget->can_queue)
1354 spin_lock_irq(shost->host_lock);
1355 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1356 spin_unlock_irq(shost->host_lock);
1358 if (starget->can_queue > 0)
1359 atomic_dec(&starget->target_busy);
1364 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1365 * return 0. We must end up running the queue again whenever 0 is
1366 * returned, else IO can hang.
1368 static inline int scsi_host_queue_ready(struct request_queue *q,
1369 struct Scsi_Host *shost,
1370 struct scsi_device *sdev)
1374 if (scsi_host_in_recovery(shost))
1377 busy = atomic_inc_return(&shost->host_busy) - 1;
1378 if (atomic_read(&shost->host_blocked) > 0) {
1383 * unblock after host_blocked iterates to zero
1385 if (atomic_dec_return(&shost->host_blocked) > 0)
1389 shost_printk(KERN_INFO, shost,
1390 "unblocking host at zero depth\n"));
1393 if (shost->can_queue > 0 && busy >= shost->can_queue)
1395 if (shost->host_self_blocked)
1398 /* We're OK to process the command, so we can't be starved */
1399 if (!list_empty(&sdev->starved_entry)) {
1400 spin_lock_irq(shost->host_lock);
1401 if (!list_empty(&sdev->starved_entry))
1402 list_del_init(&sdev->starved_entry);
1403 spin_unlock_irq(shost->host_lock);
1409 spin_lock_irq(shost->host_lock);
1410 if (list_empty(&sdev->starved_entry))
1411 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1412 spin_unlock_irq(shost->host_lock);
1414 scsi_dec_host_busy(shost);
1419 * Busy state exporting function for request stacking drivers.
1421 * For efficiency, no lock is taken to check the busy state of
1422 * shost/starget/sdev, since the returned value is not guaranteed and
1423 * may be changed after request stacking drivers call the function,
1424 * regardless of taking lock or not.
1426 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1427 * needs to return 'not busy'. Otherwise, request stacking drivers
1428 * may hold requests forever.
1430 static bool scsi_mq_lld_busy(struct request_queue *q)
1432 struct scsi_device *sdev = q->queuedata;
1433 struct Scsi_Host *shost;
1435 if (blk_queue_dying(q))
1441 * Ignore host/starget busy state.
1442 * Since block layer does not have a concept of fairness across
1443 * multiple queues, congestion of host/starget needs to be handled
1446 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1452 static void scsi_softirq_done(struct request *rq)
1454 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1455 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1458 INIT_LIST_HEAD(&cmd->eh_entry);
1460 atomic_inc(&cmd->device->iodone_cnt);
1462 atomic_inc(&cmd->device->ioerr_cnt);
1464 disposition = scsi_decide_disposition(cmd);
1465 if (disposition != SUCCESS &&
1466 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1467 scmd_printk(KERN_ERR, cmd,
1468 "timing out command, waited %lus\n",
1470 disposition = SUCCESS;
1473 scsi_log_completion(cmd, disposition);
1475 switch (disposition) {
1477 scsi_finish_command(cmd);
1480 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1482 case ADD_TO_MLQUEUE:
1483 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1486 scsi_eh_scmd_add(cmd);
1492 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1493 * @cmd: command block we are dispatching.
1495 * Return: nonzero return request was rejected and device's queue needs to be
1498 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1500 struct Scsi_Host *host = cmd->device->host;
1503 atomic_inc(&cmd->device->iorequest_cnt);
1505 /* check if the device is still usable */
1506 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1507 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1508 * returns an immediate error upwards, and signals
1509 * that the device is no longer present */
1510 cmd->result = DID_NO_CONNECT << 16;
1514 /* Check to see if the scsi lld made this device blocked. */
1515 if (unlikely(scsi_device_blocked(cmd->device))) {
1517 * in blocked state, the command is just put back on
1518 * the device queue. The suspend state has already
1519 * blocked the queue so future requests should not
1520 * occur until the device transitions out of the
1523 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1524 "queuecommand : device blocked\n"));
1525 return SCSI_MLQUEUE_DEVICE_BUSY;
1528 /* Store the LUN value in cmnd, if needed. */
1529 if (cmd->device->lun_in_cdb)
1530 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1531 (cmd->device->lun << 5 & 0xe0);
1536 * Before we queue this command, check if the command
1537 * length exceeds what the host adapter can handle.
1539 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1540 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1541 "queuecommand : command too long. "
1542 "cdb_size=%d host->max_cmd_len=%d\n",
1543 cmd->cmd_len, cmd->device->host->max_cmd_len));
1544 cmd->result = (DID_ABORT << 16);
1548 if (unlikely(host->shost_state == SHOST_DEL)) {
1549 cmd->result = (DID_NO_CONNECT << 16);
1554 trace_scsi_dispatch_cmd_start(cmd);
1555 rtn = host->hostt->queuecommand(host, cmd);
1557 trace_scsi_dispatch_cmd_error(cmd, rtn);
1558 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1559 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1560 rtn = SCSI_MLQUEUE_HOST_BUSY;
1562 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1563 "queuecommand : request rejected\n"));
1568 cmd->scsi_done(cmd);
1572 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1573 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1575 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1576 sizeof(struct scatterlist);
1579 static blk_status_t scsi_mq_prep_fn(struct request *req)
1581 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1582 struct scsi_device *sdev = req->q->queuedata;
1583 struct Scsi_Host *shost = sdev->host;
1584 struct scatterlist *sg;
1586 scsi_init_command(sdev, cmd);
1589 cmd->tag = req->tag;
1590 cmd->prot_op = SCSI_PROT_NORMAL;
1592 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1593 cmd->sdb.table.sgl = sg;
1595 if (scsi_host_get_prot(shost)) {
1596 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1598 cmd->prot_sdb->table.sgl =
1599 (struct scatterlist *)(cmd->prot_sdb + 1);
1602 blk_mq_start_request(req);
1604 return scsi_setup_cmnd(sdev, req);
1607 static void scsi_mq_done(struct scsi_cmnd *cmd)
1609 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1611 trace_scsi_dispatch_cmd_done(cmd);
1614 * If the block layer didn't complete the request due to a timeout
1615 * injection, scsi must clear its internal completed state so that the
1616 * timeout handler will see it needs to escalate its own error
1619 if (unlikely(!blk_mq_complete_request(cmd->request)))
1620 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1623 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1625 struct request_queue *q = hctx->queue;
1626 struct scsi_device *sdev = q->queuedata;
1628 atomic_dec(&sdev->device_busy);
1631 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1633 struct request_queue *q = hctx->queue;
1634 struct scsi_device *sdev = q->queuedata;
1636 if (scsi_dev_queue_ready(q, sdev))
1639 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1640 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1644 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1645 const struct blk_mq_queue_data *bd)
1647 struct request *req = bd->rq;
1648 struct request_queue *q = req->q;
1649 struct scsi_device *sdev = q->queuedata;
1650 struct Scsi_Host *shost = sdev->host;
1651 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1656 * If the device is not in running state we will reject some or all
1659 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1660 ret = scsi_prep_state_check(sdev, req);
1661 if (ret != BLK_STS_OK)
1662 goto out_put_budget;
1665 ret = BLK_STS_RESOURCE;
1666 if (!scsi_target_queue_ready(shost, sdev))
1667 goto out_put_budget;
1668 if (!scsi_host_queue_ready(q, shost, sdev))
1669 goto out_dec_target_busy;
1671 if (!(req->rq_flags & RQF_DONTPREP)) {
1672 ret = scsi_mq_prep_fn(req);
1673 if (ret != BLK_STS_OK)
1674 goto out_dec_host_busy;
1675 req->rq_flags |= RQF_DONTPREP;
1677 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1678 blk_mq_start_request(req);
1681 cmd->flags &= SCMD_PRESERVED_FLAGS;
1682 if (sdev->simple_tags)
1683 cmd->flags |= SCMD_TAGGED;
1685 cmd->flags |= SCMD_LAST;
1687 scsi_init_cmd_errh(cmd);
1688 cmd->scsi_done = scsi_mq_done;
1690 reason = scsi_dispatch_cmd(cmd);
1692 scsi_set_blocked(cmd, reason);
1693 ret = BLK_STS_RESOURCE;
1694 goto out_dec_host_busy;
1700 scsi_dec_host_busy(shost);
1701 out_dec_target_busy:
1702 if (scsi_target(sdev)->can_queue > 0)
1703 atomic_dec(&scsi_target(sdev)->target_busy);
1705 scsi_mq_put_budget(hctx);
1709 case BLK_STS_RESOURCE:
1710 if (atomic_read(&sdev->device_busy) ||
1711 scsi_device_blocked(sdev))
1712 ret = BLK_STS_DEV_RESOURCE;
1715 if (unlikely(!scsi_device_online(sdev)))
1716 scsi_req(req)->result = DID_NO_CONNECT << 16;
1718 scsi_req(req)->result = DID_ERROR << 16;
1720 * Make sure to release all allocated resources when
1721 * we hit an error, as we will never see this command
1724 if (req->rq_flags & RQF_DONTPREP)
1725 scsi_mq_uninit_cmd(cmd);
1731 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1735 return BLK_EH_RESET_TIMER;
1736 return scsi_times_out(req);
1739 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1740 unsigned int hctx_idx, unsigned int numa_node)
1742 struct Scsi_Host *shost = set->driver_data;
1743 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1744 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1745 struct scatterlist *sg;
1747 if (unchecked_isa_dma)
1748 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1749 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1750 GFP_KERNEL, numa_node);
1751 if (!cmd->sense_buffer)
1753 cmd->req.sense = cmd->sense_buffer;
1755 if (scsi_host_get_prot(shost)) {
1756 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1757 shost->hostt->cmd_size;
1758 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1764 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1765 unsigned int hctx_idx)
1767 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1769 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1773 static int scsi_map_queues(struct blk_mq_tag_set *set)
1775 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1777 if (shost->hostt->map_queues)
1778 return shost->hostt->map_queues(shost);
1779 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1782 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1784 struct device *dev = shost->dma_dev;
1787 * this limit is imposed by hardware restrictions
1789 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1792 if (scsi_host_prot_dma(shost)) {
1793 shost->sg_prot_tablesize =
1794 min_not_zero(shost->sg_prot_tablesize,
1795 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1796 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1797 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1800 if (dev->dma_mask) {
1801 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1802 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1804 blk_queue_max_hw_sectors(q, shost->max_sectors);
1805 if (shost->unchecked_isa_dma)
1806 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1807 blk_queue_segment_boundary(q, shost->dma_boundary);
1808 dma_set_seg_boundary(dev, shost->dma_boundary);
1810 blk_queue_max_segment_size(q, shost->max_segment_size);
1811 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1812 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1815 * Set a reasonable default alignment: The larger of 32-byte (dword),
1816 * which is a common minimum for HBAs, and the minimum DMA alignment,
1817 * which is set by the platform.
1819 * Devices that require a bigger alignment can increase it later.
1821 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1823 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1825 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1826 .get_budget = scsi_mq_get_budget,
1827 .put_budget = scsi_mq_put_budget,
1828 .queue_rq = scsi_queue_rq,
1829 .complete = scsi_softirq_done,
1830 .timeout = scsi_timeout,
1831 #ifdef CONFIG_BLK_DEBUG_FS
1832 .show_rq = scsi_show_rq,
1834 .init_request = scsi_mq_init_request,
1835 .exit_request = scsi_mq_exit_request,
1836 .initialize_rq_fn = scsi_initialize_rq,
1837 .cleanup_rq = scsi_cleanup_rq,
1838 .busy = scsi_mq_lld_busy,
1839 .map_queues = scsi_map_queues,
1843 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1845 struct request_queue *q = hctx->queue;
1846 struct scsi_device *sdev = q->queuedata;
1847 struct Scsi_Host *shost = sdev->host;
1849 shost->hostt->commit_rqs(shost, hctx->queue_num);
1852 static const struct blk_mq_ops scsi_mq_ops = {
1853 .get_budget = scsi_mq_get_budget,
1854 .put_budget = scsi_mq_put_budget,
1855 .queue_rq = scsi_queue_rq,
1856 .commit_rqs = scsi_commit_rqs,
1857 .complete = scsi_softirq_done,
1858 .timeout = scsi_timeout,
1859 #ifdef CONFIG_BLK_DEBUG_FS
1860 .show_rq = scsi_show_rq,
1862 .init_request = scsi_mq_init_request,
1863 .exit_request = scsi_mq_exit_request,
1864 .initialize_rq_fn = scsi_initialize_rq,
1865 .cleanup_rq = scsi_cleanup_rq,
1866 .busy = scsi_mq_lld_busy,
1867 .map_queues = scsi_map_queues,
1870 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1872 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1873 if (IS_ERR(sdev->request_queue))
1876 sdev->request_queue->queuedata = sdev;
1877 __scsi_init_queue(sdev->host, sdev->request_queue);
1878 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1879 return sdev->request_queue;
1882 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1884 unsigned int cmd_size, sgl_size;
1886 sgl_size = scsi_mq_inline_sgl_size(shost);
1887 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1888 if (scsi_host_get_prot(shost))
1889 cmd_size += sizeof(struct scsi_data_buffer) +
1890 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1892 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1893 if (shost->hostt->commit_rqs)
1894 shost->tag_set.ops = &scsi_mq_ops;
1896 shost->tag_set.ops = &scsi_mq_ops_no_commit;
1897 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1898 shost->tag_set.queue_depth = shost->can_queue;
1899 shost->tag_set.cmd_size = cmd_size;
1900 shost->tag_set.numa_node = NUMA_NO_NODE;
1901 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1902 shost->tag_set.flags |=
1903 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1904 shost->tag_set.driver_data = shost;
1906 return blk_mq_alloc_tag_set(&shost->tag_set);
1909 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1911 blk_mq_free_tag_set(&shost->tag_set);
1915 * scsi_device_from_queue - return sdev associated with a request_queue
1916 * @q: The request queue to return the sdev from
1918 * Return the sdev associated with a request queue or NULL if the
1919 * request_queue does not reference a SCSI device.
1921 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1923 struct scsi_device *sdev = NULL;
1925 if (q->mq_ops == &scsi_mq_ops)
1926 sdev = q->queuedata;
1927 if (!sdev || !get_device(&sdev->sdev_gendev))
1932 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1935 * Function: scsi_block_requests()
1937 * Purpose: Utility function used by low-level drivers to prevent further
1938 * commands from being queued to the device.
1940 * Arguments: shost - Host in question
1944 * Lock status: No locks are assumed held.
1946 * Notes: There is no timer nor any other means by which the requests
1947 * get unblocked other than the low-level driver calling
1948 * scsi_unblock_requests().
1950 void scsi_block_requests(struct Scsi_Host *shost)
1952 shost->host_self_blocked = 1;
1954 EXPORT_SYMBOL(scsi_block_requests);
1957 * Function: scsi_unblock_requests()
1959 * Purpose: Utility function used by low-level drivers to allow further
1960 * commands from being queued to the device.
1962 * Arguments: shost - Host in question
1966 * Lock status: No locks are assumed held.
1968 * Notes: There is no timer nor any other means by which the requests
1969 * get unblocked other than the low-level driver calling
1970 * scsi_unblock_requests().
1972 * This is done as an API function so that changes to the
1973 * internals of the scsi mid-layer won't require wholesale
1974 * changes to drivers that use this feature.
1976 void scsi_unblock_requests(struct Scsi_Host *shost)
1978 shost->host_self_blocked = 0;
1979 scsi_run_host_queues(shost);
1981 EXPORT_SYMBOL(scsi_unblock_requests);
1983 int __init scsi_init_queue(void)
1985 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1986 sizeof(struct scsi_data_buffer),
1988 if (!scsi_sdb_cache) {
1989 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1996 void scsi_exit_queue(void)
1998 kmem_cache_destroy(scsi_sense_cache);
1999 kmem_cache_destroy(scsi_sense_isadma_cache);
2000 kmem_cache_destroy(scsi_sdb_cache);
2004 * scsi_mode_select - issue a mode select
2005 * @sdev: SCSI device to be queried
2006 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2007 * @sp: Save page bit (0 == don't save, 1 == save)
2008 * @modepage: mode page being requested
2009 * @buffer: request buffer (may not be smaller than eight bytes)
2010 * @len: length of request buffer.
2011 * @timeout: command timeout
2012 * @retries: number of retries before failing
2013 * @data: returns a structure abstracting the mode header data
2014 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2015 * must be SCSI_SENSE_BUFFERSIZE big.
2017 * Returns zero if successful; negative error number or scsi
2022 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2023 unsigned char *buffer, int len, int timeout, int retries,
2024 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2026 unsigned char cmd[10];
2027 unsigned char *real_buffer;
2030 memset(cmd, 0, sizeof(cmd));
2031 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2033 if (sdev->use_10_for_ms) {
2036 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2039 memcpy(real_buffer + 8, buffer, len);
2043 real_buffer[2] = data->medium_type;
2044 real_buffer[3] = data->device_specific;
2045 real_buffer[4] = data->longlba ? 0x01 : 0;
2047 real_buffer[6] = data->block_descriptor_length >> 8;
2048 real_buffer[7] = data->block_descriptor_length;
2050 cmd[0] = MODE_SELECT_10;
2054 if (len > 255 || data->block_descriptor_length > 255 ||
2058 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2061 memcpy(real_buffer + 4, buffer, len);
2064 real_buffer[1] = data->medium_type;
2065 real_buffer[2] = data->device_specific;
2066 real_buffer[3] = data->block_descriptor_length;
2069 cmd[0] = MODE_SELECT;
2073 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2074 sshdr, timeout, retries, NULL);
2078 EXPORT_SYMBOL_GPL(scsi_mode_select);
2081 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2082 * @sdev: SCSI device to be queried
2083 * @dbd: set if mode sense will allow block descriptors to be returned
2084 * @modepage: mode page being requested
2085 * @buffer: request buffer (may not be smaller than eight bytes)
2086 * @len: length of request buffer.
2087 * @timeout: command timeout
2088 * @retries: number of retries before failing
2089 * @data: returns a structure abstracting the mode header data
2090 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2091 * must be SCSI_SENSE_BUFFERSIZE big.
2093 * Returns zero if unsuccessful, or the header offset (either 4
2094 * or 8 depending on whether a six or ten byte command was
2095 * issued) if successful.
2098 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2099 unsigned char *buffer, int len, int timeout, int retries,
2100 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2102 unsigned char cmd[12];
2105 int result, retry_count = retries;
2106 struct scsi_sense_hdr my_sshdr;
2108 memset(data, 0, sizeof(*data));
2109 memset(&cmd[0], 0, 12);
2110 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2113 /* caller might not be interested in sense, but we need it */
2118 use_10_for_ms = sdev->use_10_for_ms;
2120 if (use_10_for_ms) {
2124 cmd[0] = MODE_SENSE_10;
2131 cmd[0] = MODE_SENSE;
2136 memset(buffer, 0, len);
2138 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2139 sshdr, timeout, retries, NULL);
2141 /* This code looks awful: what it's doing is making sure an
2142 * ILLEGAL REQUEST sense return identifies the actual command
2143 * byte as the problem. MODE_SENSE commands can return
2144 * ILLEGAL REQUEST if the code page isn't supported */
2146 if (use_10_for_ms && !scsi_status_is_good(result) &&
2147 driver_byte(result) == DRIVER_SENSE) {
2148 if (scsi_sense_valid(sshdr)) {
2149 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2150 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2152 * Invalid command operation code
2154 sdev->use_10_for_ms = 0;
2160 if(scsi_status_is_good(result)) {
2161 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2162 (modepage == 6 || modepage == 8))) {
2163 /* Initio breakage? */
2166 data->medium_type = 0;
2167 data->device_specific = 0;
2169 data->block_descriptor_length = 0;
2170 } else if(use_10_for_ms) {
2171 data->length = buffer[0]*256 + buffer[1] + 2;
2172 data->medium_type = buffer[2];
2173 data->device_specific = buffer[3];
2174 data->longlba = buffer[4] & 0x01;
2175 data->block_descriptor_length = buffer[6]*256
2178 data->length = buffer[0] + 1;
2179 data->medium_type = buffer[1];
2180 data->device_specific = buffer[2];
2181 data->block_descriptor_length = buffer[3];
2183 data->header_length = header_length;
2184 } else if ((status_byte(result) == CHECK_CONDITION) &&
2185 scsi_sense_valid(sshdr) &&
2186 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2193 EXPORT_SYMBOL(scsi_mode_sense);
2196 * scsi_test_unit_ready - test if unit is ready
2197 * @sdev: scsi device to change the state of.
2198 * @timeout: command timeout
2199 * @retries: number of retries before failing
2200 * @sshdr: outpout pointer for decoded sense information.
2202 * Returns zero if unsuccessful or an error if TUR failed. For
2203 * removable media, UNIT_ATTENTION sets ->changed flag.
2206 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2207 struct scsi_sense_hdr *sshdr)
2210 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2214 /* try to eat the UNIT_ATTENTION if there are enough retries */
2216 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2218 if (sdev->removable && scsi_sense_valid(sshdr) &&
2219 sshdr->sense_key == UNIT_ATTENTION)
2221 } while (scsi_sense_valid(sshdr) &&
2222 sshdr->sense_key == UNIT_ATTENTION && --retries);
2226 EXPORT_SYMBOL(scsi_test_unit_ready);
2229 * scsi_device_set_state - Take the given device through the device state model.
2230 * @sdev: scsi device to change the state of.
2231 * @state: state to change to.
2233 * Returns zero if successful or an error if the requested
2234 * transition is illegal.
2237 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2239 enum scsi_device_state oldstate = sdev->sdev_state;
2241 if (state == oldstate)
2247 case SDEV_CREATED_BLOCK:
2258 case SDEV_TRANSPORT_OFFLINE:
2271 case SDEV_TRANSPORT_OFFLINE:
2279 case SDEV_TRANSPORT_OFFLINE:
2294 case SDEV_CREATED_BLOCK:
2302 case SDEV_CREATED_BLOCK:
2317 case SDEV_TRANSPORT_OFFLINE:
2329 case SDEV_TRANSPORT_OFFLINE:
2332 case SDEV_CREATED_BLOCK:
2340 sdev->sdev_state = state;
2344 SCSI_LOG_ERROR_RECOVERY(1,
2345 sdev_printk(KERN_ERR, sdev,
2346 "Illegal state transition %s->%s",
2347 scsi_device_state_name(oldstate),
2348 scsi_device_state_name(state))
2352 EXPORT_SYMBOL(scsi_device_set_state);
2355 * sdev_evt_emit - emit a single SCSI device uevent
2356 * @sdev: associated SCSI device
2357 * @evt: event to emit
2359 * Send a single uevent (scsi_event) to the associated scsi_device.
2361 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2366 switch (evt->evt_type) {
2367 case SDEV_EVT_MEDIA_CHANGE:
2368 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2370 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2371 scsi_rescan_device(&sdev->sdev_gendev);
2372 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2374 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2375 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2377 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2378 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2380 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2381 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2383 case SDEV_EVT_LUN_CHANGE_REPORTED:
2384 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2386 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2387 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2389 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2390 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2399 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2403 * sdev_evt_thread - send a uevent for each scsi event
2404 * @work: work struct for scsi_device
2406 * Dispatch queued events to their associated scsi_device kobjects
2409 void scsi_evt_thread(struct work_struct *work)
2411 struct scsi_device *sdev;
2412 enum scsi_device_event evt_type;
2413 LIST_HEAD(event_list);
2415 sdev = container_of(work, struct scsi_device, event_work);
2417 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2418 if (test_and_clear_bit(evt_type, sdev->pending_events))
2419 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2422 struct scsi_event *evt;
2423 struct list_head *this, *tmp;
2424 unsigned long flags;
2426 spin_lock_irqsave(&sdev->list_lock, flags);
2427 list_splice_init(&sdev->event_list, &event_list);
2428 spin_unlock_irqrestore(&sdev->list_lock, flags);
2430 if (list_empty(&event_list))
2433 list_for_each_safe(this, tmp, &event_list) {
2434 evt = list_entry(this, struct scsi_event, node);
2435 list_del(&evt->node);
2436 scsi_evt_emit(sdev, evt);
2443 * sdev_evt_send - send asserted event to uevent thread
2444 * @sdev: scsi_device event occurred on
2445 * @evt: event to send
2447 * Assert scsi device event asynchronously.
2449 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2451 unsigned long flags;
2454 /* FIXME: currently this check eliminates all media change events
2455 * for polled devices. Need to update to discriminate between AN
2456 * and polled events */
2457 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2463 spin_lock_irqsave(&sdev->list_lock, flags);
2464 list_add_tail(&evt->node, &sdev->event_list);
2465 schedule_work(&sdev->event_work);
2466 spin_unlock_irqrestore(&sdev->list_lock, flags);
2468 EXPORT_SYMBOL_GPL(sdev_evt_send);
2471 * sdev_evt_alloc - allocate a new scsi event
2472 * @evt_type: type of event to allocate
2473 * @gfpflags: GFP flags for allocation
2475 * Allocates and returns a new scsi_event.
2477 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2480 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2484 evt->evt_type = evt_type;
2485 INIT_LIST_HEAD(&evt->node);
2487 /* evt_type-specific initialization, if any */
2489 case SDEV_EVT_MEDIA_CHANGE:
2490 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2491 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2492 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2493 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2494 case SDEV_EVT_LUN_CHANGE_REPORTED:
2495 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2496 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2504 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2507 * sdev_evt_send_simple - send asserted event to uevent thread
2508 * @sdev: scsi_device event occurred on
2509 * @evt_type: type of event to send
2510 * @gfpflags: GFP flags for allocation
2512 * Assert scsi device event asynchronously, given an event type.
2514 void sdev_evt_send_simple(struct scsi_device *sdev,
2515 enum scsi_device_event evt_type, gfp_t gfpflags)
2517 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2519 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2524 sdev_evt_send(sdev, evt);
2526 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2529 * scsi_device_quiesce - Block user issued commands.
2530 * @sdev: scsi device to quiesce.
2532 * This works by trying to transition to the SDEV_QUIESCE state
2533 * (which must be a legal transition). When the device is in this
2534 * state, only special requests will be accepted, all others will
2535 * be deferred. Since special requests may also be requeued requests,
2536 * a successful return doesn't guarantee the device will be
2537 * totally quiescent.
2539 * Must be called with user context, may sleep.
2541 * Returns zero if unsuccessful or an error if not.
2544 scsi_device_quiesce(struct scsi_device *sdev)
2546 struct request_queue *q = sdev->request_queue;
2550 * It is allowed to call scsi_device_quiesce() multiple times from
2551 * the same context but concurrent scsi_device_quiesce() calls are
2554 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2556 if (sdev->quiesced_by == current)
2561 blk_mq_freeze_queue(q);
2563 * Ensure that the effect of blk_set_pm_only() will be visible
2564 * for percpu_ref_tryget() callers that occur after the queue
2565 * unfreeze even if the queue was already frozen before this function
2566 * was called. See also https://lwn.net/Articles/573497/.
2569 blk_mq_unfreeze_queue(q);
2571 mutex_lock(&sdev->state_mutex);
2572 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2574 sdev->quiesced_by = current;
2576 blk_clear_pm_only(q);
2577 mutex_unlock(&sdev->state_mutex);
2581 EXPORT_SYMBOL(scsi_device_quiesce);
2584 * scsi_device_resume - Restart user issued commands to a quiesced device.
2585 * @sdev: scsi device to resume.
2587 * Moves the device from quiesced back to running and restarts the
2590 * Must be called with user context, may sleep.
2592 void scsi_device_resume(struct scsi_device *sdev)
2594 /* check if the device state was mutated prior to resume, and if
2595 * so assume the state is being managed elsewhere (for example
2596 * device deleted during suspend)
2598 mutex_lock(&sdev->state_mutex);
2599 if (sdev->quiesced_by) {
2600 sdev->quiesced_by = NULL;
2601 blk_clear_pm_only(sdev->request_queue);
2603 if (sdev->sdev_state == SDEV_QUIESCE)
2604 scsi_device_set_state(sdev, SDEV_RUNNING);
2605 mutex_unlock(&sdev->state_mutex);
2607 EXPORT_SYMBOL(scsi_device_resume);
2610 device_quiesce_fn(struct scsi_device *sdev, void *data)
2612 scsi_device_quiesce(sdev);
2616 scsi_target_quiesce(struct scsi_target *starget)
2618 starget_for_each_device(starget, NULL, device_quiesce_fn);
2620 EXPORT_SYMBOL(scsi_target_quiesce);
2623 device_resume_fn(struct scsi_device *sdev, void *data)
2625 scsi_device_resume(sdev);
2629 scsi_target_resume(struct scsi_target *starget)
2631 starget_for_each_device(starget, NULL, device_resume_fn);
2633 EXPORT_SYMBOL(scsi_target_resume);
2636 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2637 * @sdev: device to block
2639 * Pause SCSI command processing on the specified device. Does not sleep.
2641 * Returns zero if successful or a negative error code upon failure.
2644 * This routine transitions the device to the SDEV_BLOCK state (which must be
2645 * a legal transition). When the device is in this state, command processing
2646 * is paused until the device leaves the SDEV_BLOCK state. See also
2647 * scsi_internal_device_unblock_nowait().
2649 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2651 struct request_queue *q = sdev->request_queue;
2654 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2656 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2663 * The device has transitioned to SDEV_BLOCK. Stop the
2664 * block layer from calling the midlayer with this device's
2667 blk_mq_quiesce_queue_nowait(q);
2670 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2673 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2674 * @sdev: device to block
2676 * Pause SCSI command processing on the specified device and wait until all
2677 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2679 * Returns zero if successful or a negative error code upon failure.
2682 * This routine transitions the device to the SDEV_BLOCK state (which must be
2683 * a legal transition). When the device is in this state, command processing
2684 * is paused until the device leaves the SDEV_BLOCK state. See also
2685 * scsi_internal_device_unblock().
2687 static int scsi_internal_device_block(struct scsi_device *sdev)
2689 struct request_queue *q = sdev->request_queue;
2692 mutex_lock(&sdev->state_mutex);
2693 err = scsi_internal_device_block_nowait(sdev);
2695 blk_mq_quiesce_queue(q);
2696 mutex_unlock(&sdev->state_mutex);
2701 void scsi_start_queue(struct scsi_device *sdev)
2703 struct request_queue *q = sdev->request_queue;
2705 blk_mq_unquiesce_queue(q);
2709 * scsi_internal_device_unblock_nowait - resume a device after a block request
2710 * @sdev: device to resume
2711 * @new_state: state to set the device to after unblocking
2713 * Restart the device queue for a previously suspended SCSI device. Does not
2716 * Returns zero if successful or a negative error code upon failure.
2719 * This routine transitions the device to the SDEV_RUNNING state or to one of
2720 * the offline states (which must be a legal transition) allowing the midlayer
2721 * to goose the queue for this device.
2723 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2724 enum scsi_device_state new_state)
2726 switch (new_state) {
2728 case SDEV_TRANSPORT_OFFLINE:
2735 * Try to transition the scsi device to SDEV_RUNNING or one of the
2736 * offlined states and goose the device queue if successful.
2738 switch (sdev->sdev_state) {
2740 case SDEV_TRANSPORT_OFFLINE:
2741 sdev->sdev_state = new_state;
2743 case SDEV_CREATED_BLOCK:
2744 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2745 new_state == SDEV_OFFLINE)
2746 sdev->sdev_state = new_state;
2748 sdev->sdev_state = SDEV_CREATED;
2756 scsi_start_queue(sdev);
2760 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2763 * scsi_internal_device_unblock - resume a device after a block request
2764 * @sdev: device to resume
2765 * @new_state: state to set the device to after unblocking
2767 * Restart the device queue for a previously suspended SCSI device. May sleep.
2769 * Returns zero if successful or a negative error code upon failure.
2772 * This routine transitions the device to the SDEV_RUNNING state or to one of
2773 * the offline states (which must be a legal transition) allowing the midlayer
2774 * to goose the queue for this device.
2776 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2777 enum scsi_device_state new_state)
2781 mutex_lock(&sdev->state_mutex);
2782 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2783 mutex_unlock(&sdev->state_mutex);
2789 device_block(struct scsi_device *sdev, void *data)
2793 ret = scsi_internal_device_block(sdev);
2795 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2796 dev_name(&sdev->sdev_gendev), ret);
2800 target_block(struct device *dev, void *data)
2802 if (scsi_is_target_device(dev))
2803 starget_for_each_device(to_scsi_target(dev), NULL,
2809 scsi_target_block(struct device *dev)
2811 if (scsi_is_target_device(dev))
2812 starget_for_each_device(to_scsi_target(dev), NULL,
2815 device_for_each_child(dev, NULL, target_block);
2817 EXPORT_SYMBOL_GPL(scsi_target_block);
2820 device_unblock(struct scsi_device *sdev, void *data)
2822 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2826 target_unblock(struct device *dev, void *data)
2828 if (scsi_is_target_device(dev))
2829 starget_for_each_device(to_scsi_target(dev), data,
2835 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2837 if (scsi_is_target_device(dev))
2838 starget_for_each_device(to_scsi_target(dev), &new_state,
2841 device_for_each_child(dev, &new_state, target_unblock);
2843 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2846 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2847 * @sgl: scatter-gather list
2848 * @sg_count: number of segments in sg
2849 * @offset: offset in bytes into sg, on return offset into the mapped area
2850 * @len: bytes to map, on return number of bytes mapped
2852 * Returns virtual address of the start of the mapped page
2854 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2855 size_t *offset, size_t *len)
2858 size_t sg_len = 0, len_complete = 0;
2859 struct scatterlist *sg;
2862 WARN_ON(!irqs_disabled());
2864 for_each_sg(sgl, sg, sg_count, i) {
2865 len_complete = sg_len; /* Complete sg-entries */
2866 sg_len += sg->length;
2867 if (sg_len > *offset)
2871 if (unlikely(i == sg_count)) {
2872 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2874 __func__, sg_len, *offset, sg_count);
2879 /* Offset starting from the beginning of first page in this sg-entry */
2880 *offset = *offset - len_complete + sg->offset;
2882 /* Assumption: contiguous pages can be accessed as "page + i" */
2883 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2884 *offset &= ~PAGE_MASK;
2886 /* Bytes in this sg-entry from *offset to the end of the page */
2887 sg_len = PAGE_SIZE - *offset;
2891 return kmap_atomic(page);
2893 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2896 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2897 * @virt: virtual address to be unmapped
2899 void scsi_kunmap_atomic_sg(void *virt)
2901 kunmap_atomic(virt);
2903 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2905 void sdev_disable_disk_events(struct scsi_device *sdev)
2907 atomic_inc(&sdev->disk_events_disable_depth);
2909 EXPORT_SYMBOL(sdev_disable_disk_events);
2911 void sdev_enable_disk_events(struct scsi_device *sdev)
2913 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2915 atomic_dec(&sdev->disk_events_disable_depth);
2917 EXPORT_SYMBOL(sdev_enable_disk_events);
2920 * scsi_vpd_lun_id - return a unique device identification
2921 * @sdev: SCSI device
2922 * @id: buffer for the identification
2923 * @id_len: length of the buffer
2925 * Copies a unique device identification into @id based
2926 * on the information in the VPD page 0x83 of the device.
2927 * The string will be formatted as a SCSI name string.
2929 * Returns the length of the identification or error on failure.
2930 * If the identifier is longer than the supplied buffer the actual
2931 * identifier length is returned and the buffer is not zero-padded.
2933 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2935 u8 cur_id_type = 0xff;
2937 const unsigned char *d, *cur_id_str;
2938 const struct scsi_vpd *vpd_pg83;
2939 int id_size = -EINVAL;
2942 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2949 * Look for the correct descriptor.
2950 * Order of preference for lun descriptor:
2951 * - SCSI name string
2952 * - NAA IEEE Registered Extended
2953 * - EUI-64 based 16-byte
2954 * - EUI-64 based 12-byte
2955 * - NAA IEEE Registered
2956 * - NAA IEEE Extended
2958 * as longer descriptors reduce the likelyhood
2959 * of identification clashes.
2962 /* The id string must be at least 20 bytes + terminating NULL byte */
2968 memset(id, 0, id_len);
2969 d = vpd_pg83->data + 4;
2970 while (d < vpd_pg83->data + vpd_pg83->len) {
2971 /* Skip designators not referring to the LUN */
2972 if ((d[1] & 0x30) != 0x00)
2975 switch (d[1] & 0xf) {
2978 if (cur_id_size > d[3])
2980 /* Prefer anything */
2981 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2984 if (cur_id_size + 4 > id_len)
2985 cur_id_size = id_len - 4;
2987 cur_id_type = d[1] & 0xf;
2988 id_size = snprintf(id, id_len, "t10.%*pE",
2989 cur_id_size, cur_id_str);
2993 if (cur_id_size > d[3])
2995 /* Prefer NAA IEEE Registered Extended */
2996 if (cur_id_type == 0x3 &&
2997 cur_id_size == d[3])
3001 cur_id_type = d[1] & 0xf;
3002 switch (cur_id_size) {
3004 id_size = snprintf(id, id_len,
3009 id_size = snprintf(id, id_len,
3014 id_size = snprintf(id, id_len,
3025 if (cur_id_size > d[3])
3029 cur_id_type = d[1] & 0xf;
3030 switch (cur_id_size) {
3032 id_size = snprintf(id, id_len,
3037 id_size = snprintf(id, id_len,
3047 /* SCSI name string */
3048 if (cur_id_size + 4 > d[3])
3050 /* Prefer others for truncated descriptor */
3051 if (cur_id_size && d[3] > id_len)
3053 cur_id_size = id_size = d[3];
3055 cur_id_type = d[1] & 0xf;
3056 if (cur_id_size >= id_len)
3057 cur_id_size = id_len - 1;
3058 memcpy(id, cur_id_str, cur_id_size);
3059 /* Decrease priority for truncated descriptor */
3060 if (cur_id_size != id_size)
3073 EXPORT_SYMBOL(scsi_vpd_lun_id);
3076 * scsi_vpd_tpg_id - return a target port group identifier
3077 * @sdev: SCSI device
3079 * Returns the Target Port Group identifier from the information
3080 * froom VPD page 0x83 of the device.
3082 * Returns the identifier or error on failure.
3084 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3086 const unsigned char *d;
3087 const struct scsi_vpd *vpd_pg83;
3088 int group_id = -EAGAIN, rel_port = -1;
3091 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3097 d = vpd_pg83->data + 4;
3098 while (d < vpd_pg83->data + vpd_pg83->len) {
3099 switch (d[1] & 0xf) {
3101 /* Relative target port */
3102 rel_port = get_unaligned_be16(&d[6]);
3105 /* Target port group */
3106 group_id = get_unaligned_be16(&d[6]);
3115 if (group_id >= 0 && rel_id && rel_port != -1)
3120 EXPORT_SYMBOL(scsi_vpd_tpg_id);