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_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_free_sgtables(cmd);
567 scsi_uninit_cmd(cmd);
568 scsi_del_cmd_from_list(cmd);
571 static void scsi_run_queue_async(struct scsi_device *sdev)
573 if (scsi_target(sdev)->single_lun ||
574 !list_empty(&sdev->host->starved_list))
575 kblockd_schedule_work(&sdev->requeue_work);
577 blk_mq_run_hw_queues(sdev->request_queue, true);
580 /* Returns false when no more bytes to process, true if there are more */
581 static bool scsi_end_request(struct request *req, blk_status_t error,
584 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
585 struct scsi_device *sdev = cmd->device;
586 struct request_queue *q = sdev->request_queue;
588 if (blk_update_request(req, error, bytes))
591 if (blk_queue_add_random(q))
592 add_disk_randomness(req->rq_disk);
594 if (!blk_rq_is_scsi(req)) {
595 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
596 cmd->flags &= ~SCMD_INITIALIZED;
600 * Calling rcu_barrier() is not necessary here because the
601 * SCSI error handler guarantees that the function called by
602 * call_rcu() has been called before scsi_end_request() is
605 destroy_rcu_head(&cmd->rcu);
608 * In the MQ case the command gets freed by __blk_mq_end_request,
609 * so we have to do all cleanup that depends on it earlier.
611 * We also can't kick the queues from irq context, so we
612 * will have to defer it to a workqueue.
614 scsi_mq_uninit_cmd(cmd);
617 * queue is still alive, so grab the ref for preventing it
618 * from being cleaned up during running queue.
620 percpu_ref_get(&q->q_usage_counter);
622 __blk_mq_end_request(req, error);
624 scsi_run_queue_async(sdev);
626 percpu_ref_put(&q->q_usage_counter);
631 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
633 * @result: scsi error code
635 * Translate a SCSI result code into a blk_status_t value. May reset the host
636 * byte of @cmd->result.
638 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
640 switch (host_byte(result)) {
643 * Also check the other bytes than the status byte in result
644 * to handle the case when a SCSI LLD sets result to
645 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
647 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
649 return BLK_STS_IOERR;
650 case DID_TRANSPORT_FAILFAST:
651 return BLK_STS_TRANSPORT;
652 case DID_TARGET_FAILURE:
653 set_host_byte(cmd, DID_OK);
654 return BLK_STS_TARGET;
655 case DID_NEXUS_FAILURE:
656 set_host_byte(cmd, DID_OK);
657 return BLK_STS_NEXUS;
658 case DID_ALLOC_FAILURE:
659 set_host_byte(cmd, DID_OK);
660 return BLK_STS_NOSPC;
661 case DID_MEDIUM_ERROR:
662 set_host_byte(cmd, DID_OK);
663 return BLK_STS_MEDIUM;
665 return BLK_STS_IOERR;
669 /* Helper for scsi_io_completion() when "reprep" action required. */
670 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
671 struct request_queue *q)
673 /* A new command will be prepared and issued. */
674 scsi_mq_requeue_cmd(cmd);
677 /* Helper for scsi_io_completion() when special action required. */
678 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
680 struct request_queue *q = cmd->device->request_queue;
681 struct request *req = cmd->request;
683 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
684 ACTION_DELAYED_RETRY} action;
685 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
686 struct scsi_sense_hdr sshdr;
688 bool sense_current = true; /* false implies "deferred sense" */
689 blk_status_t blk_stat;
691 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
693 sense_current = !scsi_sense_is_deferred(&sshdr);
695 blk_stat = scsi_result_to_blk_status(cmd, result);
697 if (host_byte(result) == DID_RESET) {
698 /* Third party bus reset or reset for error recovery
699 * reasons. Just retry the command and see what
702 action = ACTION_RETRY;
703 } else if (sense_valid && sense_current) {
704 switch (sshdr.sense_key) {
706 if (cmd->device->removable) {
707 /* Detected disc change. Set a bit
708 * and quietly refuse further access.
710 cmd->device->changed = 1;
711 action = ACTION_FAIL;
713 /* Must have been a power glitch, or a
714 * bus reset. Could not have been a
715 * media change, so we just retry the
716 * command and see what happens.
718 action = ACTION_RETRY;
721 case ILLEGAL_REQUEST:
722 /* If we had an ILLEGAL REQUEST returned, then
723 * we may have performed an unsupported
724 * command. The only thing this should be
725 * would be a ten byte read where only a six
726 * byte read was supported. Also, on a system
727 * where READ CAPACITY failed, we may have
728 * read past the end of the disk.
730 if ((cmd->device->use_10_for_rw &&
731 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
732 (cmd->cmnd[0] == READ_10 ||
733 cmd->cmnd[0] == WRITE_10)) {
734 /* This will issue a new 6-byte command. */
735 cmd->device->use_10_for_rw = 0;
736 action = ACTION_REPREP;
737 } else if (sshdr.asc == 0x10) /* DIX */ {
738 action = ACTION_FAIL;
739 blk_stat = BLK_STS_PROTECTION;
740 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
741 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
742 action = ACTION_FAIL;
743 blk_stat = BLK_STS_TARGET;
745 action = ACTION_FAIL;
747 case ABORTED_COMMAND:
748 action = ACTION_FAIL;
749 if (sshdr.asc == 0x10) /* DIF */
750 blk_stat = BLK_STS_PROTECTION;
753 /* If the device is in the process of becoming
754 * ready, or has a temporary blockage, retry.
756 if (sshdr.asc == 0x04) {
757 switch (sshdr.ascq) {
758 case 0x01: /* becoming ready */
759 case 0x04: /* format in progress */
760 case 0x05: /* rebuild in progress */
761 case 0x06: /* recalculation in progress */
762 case 0x07: /* operation in progress */
763 case 0x08: /* Long write in progress */
764 case 0x09: /* self test in progress */
765 case 0x11: /* notify (enable spinup) required */
766 case 0x14: /* space allocation in progress */
767 case 0x1a: /* start stop unit in progress */
768 case 0x1b: /* sanitize in progress */
769 case 0x1d: /* configuration in progress */
770 case 0x24: /* depopulation in progress */
771 action = ACTION_DELAYED_RETRY;
774 action = ACTION_FAIL;
778 action = ACTION_FAIL;
780 case VOLUME_OVERFLOW:
781 /* See SSC3rXX or current. */
782 action = ACTION_FAIL;
785 action = ACTION_FAIL;
789 action = ACTION_FAIL;
791 if (action != ACTION_FAIL &&
792 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
793 action = ACTION_FAIL;
797 /* Give up and fail the remainder of the request */
798 if (!(req->rq_flags & RQF_QUIET)) {
799 static DEFINE_RATELIMIT_STATE(_rs,
800 DEFAULT_RATELIMIT_INTERVAL,
801 DEFAULT_RATELIMIT_BURST);
803 if (unlikely(scsi_logging_level))
805 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
806 SCSI_LOG_MLCOMPLETE_BITS);
809 * if logging is enabled the failure will be printed
810 * in scsi_log_completion(), so avoid duplicate messages
812 if (!level && __ratelimit(&_rs)) {
813 scsi_print_result(cmd, NULL, FAILED);
814 if (driver_byte(result) == DRIVER_SENSE)
815 scsi_print_sense(cmd);
816 scsi_print_command(cmd);
819 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
823 scsi_io_completion_reprep(cmd, q);
826 /* Retry the same command immediately */
827 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
829 case ACTION_DELAYED_RETRY:
830 /* Retry the same command after a delay */
831 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
837 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
838 * new result that may suppress further error checking. Also modifies
839 * *blk_statp in some cases.
841 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
842 blk_status_t *blk_statp)
845 bool sense_current = true; /* false implies "deferred sense" */
846 struct request *req = cmd->request;
847 struct scsi_sense_hdr sshdr;
849 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
851 sense_current = !scsi_sense_is_deferred(&sshdr);
853 if (blk_rq_is_passthrough(req)) {
856 * SG_IO wants current and deferred errors
858 scsi_req(req)->sense_len =
859 min(8 + cmd->sense_buffer[7],
860 SCSI_SENSE_BUFFERSIZE);
863 *blk_statp = scsi_result_to_blk_status(cmd, result);
864 } else if (blk_rq_bytes(req) == 0 && sense_current) {
866 * Flush commands do not transfers any data, and thus cannot use
867 * good_bytes != blk_rq_bytes(req) as the signal for an error.
868 * This sets *blk_statp explicitly for the problem case.
870 *blk_statp = scsi_result_to_blk_status(cmd, result);
873 * Recovered errors need reporting, but they're always treated as
874 * success, so fiddle the result code here. For passthrough requests
875 * we already took a copy of the original into sreq->result which
876 * is what gets returned to the user
878 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
879 bool do_print = true;
881 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
882 * skip print since caller wants ATA registers. Only occurs
883 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
885 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
887 else if (req->rq_flags & RQF_QUIET)
890 scsi_print_sense(cmd);
892 /* for passthrough, *blk_statp may be set */
893 *blk_statp = BLK_STS_OK;
896 * Another corner case: the SCSI status byte is non-zero but 'good'.
897 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
898 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
899 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
900 * intermediate statuses (both obsolete in SAM-4) as good.
902 if (status_byte(result) && scsi_status_is_good(result)) {
904 *blk_statp = BLK_STS_OK;
910 * Function: scsi_io_completion()
912 * Purpose: Completion processing for block device I/O requests.
914 * Arguments: cmd - command that is finished.
916 * Lock status: Assumed that no lock is held upon entry.
920 * Notes: We will finish off the specified number of sectors. If we
921 * are done, the command block will be released and the queue
922 * function will be goosed. If we are not done then we have to
923 * figure out what to do next:
925 * a) We can call scsi_requeue_command(). The request
926 * will be unprepared and put back on the queue. Then
927 * a new command will be created for it. This should
928 * be used if we made forward progress, or if we want
929 * to switch from READ(10) to READ(6) for example.
931 * b) We can call __scsi_queue_insert(). The request will
932 * be put back on the queue and retried using the same
933 * command as before, possibly after a delay.
935 * c) We can call scsi_end_request() with blk_stat other than
936 * BLK_STS_OK, to fail the remainder of the request.
938 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
940 int result = cmd->result;
941 struct request_queue *q = cmd->device->request_queue;
942 struct request *req = cmd->request;
943 blk_status_t blk_stat = BLK_STS_OK;
945 if (unlikely(result)) /* a nz result may or may not be an error */
946 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
948 if (unlikely(blk_rq_is_passthrough(req))) {
950 * scsi_result_to_blk_status may have reset the host_byte
952 scsi_req(req)->result = cmd->result;
956 * Next deal with any sectors which we were able to correctly
959 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
960 "%u sectors total, %d bytes done.\n",
961 blk_rq_sectors(req), good_bytes));
964 * Next deal with any sectors which we were able to correctly
965 * handle. Failed, zero length commands always need to drop down
966 * to retry code. Fast path should return in this block.
968 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
969 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
970 return; /* no bytes remaining */
973 /* Kill remainder if no retries. */
974 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
975 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
977 "Bytes remaining after failed, no-retry command");
982 * If there had been no error, but we have leftover bytes in the
983 * requeues just queue the command up again.
985 if (likely(result == 0))
986 scsi_io_completion_reprep(cmd, q);
988 scsi_io_completion_action(cmd, result);
991 static blk_status_t scsi_init_sgtable(struct request *req,
992 struct scsi_data_buffer *sdb)
997 * If sg table allocation fails, requeue request later.
999 if (unlikely(sg_alloc_table_chained(&sdb->table,
1000 blk_rq_nr_phys_segments(req), sdb->table.sgl,
1001 SCSI_INLINE_SG_CNT)))
1002 return BLK_STS_RESOURCE;
1005 * Next, walk the list, and fill in the addresses and sizes of
1008 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1009 BUG_ON(count > sdb->table.nents);
1010 sdb->table.nents = count;
1011 sdb->length = blk_rq_payload_bytes(req);
1016 * Function: scsi_init_io()
1018 * Purpose: SCSI I/O initialize function.
1020 * Arguments: cmd - Command descriptor we wish to initialize
1022 * Returns: BLK_STS_OK on success
1023 * BLK_STS_RESOURCE if the failure is retryable
1024 * BLK_STS_IOERR if the failure is fatal
1026 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1028 struct request *rq = cmd->request;
1031 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1032 return BLK_STS_IOERR;
1034 ret = scsi_init_sgtable(rq, &cmd->sdb);
1038 if (blk_integrity_rq(rq)) {
1039 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1042 if (WARN_ON_ONCE(!prot_sdb)) {
1044 * This can happen if someone (e.g. multipath)
1045 * queues a command to a device on an adapter
1046 * that does not support DIX.
1048 ret = BLK_STS_IOERR;
1049 goto out_free_sgtables;
1052 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1054 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1055 prot_sdb->table.sgl,
1056 SCSI_INLINE_PROT_SG_CNT)) {
1057 ret = BLK_STS_RESOURCE;
1058 goto out_free_sgtables;
1061 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1062 prot_sdb->table.sgl);
1063 BUG_ON(count > ivecs);
1064 BUG_ON(count > queue_max_integrity_segments(rq->q));
1066 cmd->prot_sdb = prot_sdb;
1067 cmd->prot_sdb->table.nents = count;
1072 scsi_free_sgtables(cmd);
1075 EXPORT_SYMBOL(scsi_init_io);
1078 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1079 * @rq: Request associated with the SCSI command to be initialized.
1081 * This function initializes the members of struct scsi_cmnd that must be
1082 * initialized before request processing starts and that won't be
1083 * reinitialized if a SCSI command is requeued.
1085 * Called from inside blk_get_request() for pass-through requests and from
1086 * inside scsi_init_command() for filesystem requests.
1088 static void scsi_initialize_rq(struct request *rq)
1090 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1092 scsi_req_init(&cmd->req);
1093 init_rcu_head(&cmd->rcu);
1094 cmd->jiffies_at_alloc = jiffies;
1099 * Only called when the request isn't completed by SCSI, and not freed by
1102 static void scsi_cleanup_rq(struct request *rq)
1104 if (rq->rq_flags & RQF_DONTPREP) {
1105 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1106 rq->rq_flags &= ~RQF_DONTPREP;
1110 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1111 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1113 struct scsi_device *sdev = cmd->device;
1114 struct Scsi_Host *shost = sdev->host;
1115 unsigned long flags;
1117 if (shost->use_cmd_list) {
1118 spin_lock_irqsave(&sdev->list_lock, flags);
1119 list_add_tail(&cmd->list, &sdev->cmd_list);
1120 spin_unlock_irqrestore(&sdev->list_lock, flags);
1124 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1125 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1127 struct scsi_device *sdev = cmd->device;
1128 struct Scsi_Host *shost = sdev->host;
1129 unsigned long flags;
1131 if (shost->use_cmd_list) {
1132 spin_lock_irqsave(&sdev->list_lock, flags);
1133 BUG_ON(list_empty(&cmd->list));
1134 list_del_init(&cmd->list);
1135 spin_unlock_irqrestore(&sdev->list_lock, flags);
1139 /* Called after a request has been started. */
1140 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1142 void *buf = cmd->sense_buffer;
1143 void *prot = cmd->prot_sdb;
1144 struct request *rq = blk_mq_rq_from_pdu(cmd);
1145 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1146 unsigned long jiffies_at_alloc;
1149 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1150 flags |= SCMD_INITIALIZED;
1151 scsi_initialize_rq(rq);
1154 jiffies_at_alloc = cmd->jiffies_at_alloc;
1155 retries = cmd->retries;
1156 /* zero out the cmd, except for the embedded scsi_request */
1157 memset((char *)cmd + sizeof(cmd->req), 0,
1158 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1161 cmd->sense_buffer = buf;
1162 cmd->prot_sdb = prot;
1164 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1165 cmd->jiffies_at_alloc = jiffies_at_alloc;
1166 cmd->retries = retries;
1168 scsi_add_cmd_to_list(cmd);
1171 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1172 struct request *req)
1174 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1177 * Passthrough requests may transfer data, in which case they must
1178 * a bio attached to them. Or they might contain a SCSI command
1179 * that does not transfer data, in which case they may optionally
1180 * submit a request without an attached bio.
1183 blk_status_t ret = scsi_init_io(cmd);
1184 if (unlikely(ret != BLK_STS_OK))
1187 BUG_ON(blk_rq_bytes(req));
1189 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1192 cmd->cmd_len = scsi_req(req)->cmd_len;
1193 cmd->cmnd = scsi_req(req)->cmd;
1194 cmd->transfersize = blk_rq_bytes(req);
1195 cmd->allowed = scsi_req(req)->retries;
1200 * Setup a normal block command. These are simple request from filesystems
1201 * that still need to be translated to SCSI CDBs from the ULD.
1203 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1204 struct request *req)
1206 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1208 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1209 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1210 if (ret != BLK_STS_OK)
1214 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1215 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1216 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1219 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1220 struct request *req)
1222 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1225 if (!blk_rq_bytes(req))
1226 cmd->sc_data_direction = DMA_NONE;
1227 else if (rq_data_dir(req) == WRITE)
1228 cmd->sc_data_direction = DMA_TO_DEVICE;
1230 cmd->sc_data_direction = DMA_FROM_DEVICE;
1232 if (blk_rq_is_scsi(req))
1233 ret = scsi_setup_scsi_cmnd(sdev, req);
1235 ret = scsi_setup_fs_cmnd(sdev, req);
1237 if (ret != BLK_STS_OK)
1238 scsi_free_sgtables(cmd);
1244 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1246 switch (sdev->sdev_state) {
1248 case SDEV_TRANSPORT_OFFLINE:
1250 * If the device is offline we refuse to process any
1251 * commands. The device must be brought online
1252 * before trying any recovery commands.
1254 sdev_printk(KERN_ERR, sdev,
1255 "rejecting I/O to offline device\n");
1256 return BLK_STS_IOERR;
1259 * If the device is fully deleted, we refuse to
1260 * process any commands as well.
1262 sdev_printk(KERN_ERR, sdev,
1263 "rejecting I/O to dead device\n");
1264 return BLK_STS_IOERR;
1266 case SDEV_CREATED_BLOCK:
1267 return BLK_STS_RESOURCE;
1270 * If the devices is blocked we defer normal commands.
1272 if (req && !(req->rq_flags & RQF_PREEMPT))
1273 return BLK_STS_RESOURCE;
1277 * For any other not fully online state we only allow
1278 * special commands. In particular any user initiated
1279 * command is not allowed.
1281 if (req && !(req->rq_flags & RQF_PREEMPT))
1282 return BLK_STS_IOERR;
1288 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1291 * Called with the queue_lock held.
1293 static inline int scsi_dev_queue_ready(struct request_queue *q,
1294 struct scsi_device *sdev)
1298 busy = atomic_inc_return(&sdev->device_busy) - 1;
1299 if (atomic_read(&sdev->device_blocked)) {
1304 * unblock after device_blocked iterates to zero
1306 if (atomic_dec_return(&sdev->device_blocked) > 0)
1308 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1309 "unblocking device at zero depth\n"));
1312 if (busy >= sdev->queue_depth)
1317 atomic_dec(&sdev->device_busy);
1322 * scsi_target_queue_ready: checks if there we can send commands to target
1323 * @sdev: scsi device on starget to check.
1325 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1326 struct scsi_device *sdev)
1328 struct scsi_target *starget = scsi_target(sdev);
1331 if (starget->single_lun) {
1332 spin_lock_irq(shost->host_lock);
1333 if (starget->starget_sdev_user &&
1334 starget->starget_sdev_user != sdev) {
1335 spin_unlock_irq(shost->host_lock);
1338 starget->starget_sdev_user = sdev;
1339 spin_unlock_irq(shost->host_lock);
1342 if (starget->can_queue <= 0)
1345 busy = atomic_inc_return(&starget->target_busy) - 1;
1346 if (atomic_read(&starget->target_blocked) > 0) {
1351 * unblock after target_blocked iterates to zero
1353 if (atomic_dec_return(&starget->target_blocked) > 0)
1356 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1357 "unblocking target at zero depth\n"));
1360 if (busy >= starget->can_queue)
1366 spin_lock_irq(shost->host_lock);
1367 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1368 spin_unlock_irq(shost->host_lock);
1370 if (starget->can_queue > 0)
1371 atomic_dec(&starget->target_busy);
1376 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1377 * return 0. We must end up running the queue again whenever 0 is
1378 * returned, else IO can hang.
1380 static inline int scsi_host_queue_ready(struct request_queue *q,
1381 struct Scsi_Host *shost,
1382 struct scsi_device *sdev)
1386 if (scsi_host_in_recovery(shost))
1389 busy = atomic_inc_return(&shost->host_busy) - 1;
1390 if (atomic_read(&shost->host_blocked) > 0) {
1395 * unblock after host_blocked iterates to zero
1397 if (atomic_dec_return(&shost->host_blocked) > 0)
1401 shost_printk(KERN_INFO, shost,
1402 "unblocking host at zero depth\n"));
1405 if (shost->can_queue > 0 && busy >= shost->can_queue)
1407 if (shost->host_self_blocked)
1410 /* We're OK to process the command, so we can't be starved */
1411 if (!list_empty(&sdev->starved_entry)) {
1412 spin_lock_irq(shost->host_lock);
1413 if (!list_empty(&sdev->starved_entry))
1414 list_del_init(&sdev->starved_entry);
1415 spin_unlock_irq(shost->host_lock);
1421 spin_lock_irq(shost->host_lock);
1422 if (list_empty(&sdev->starved_entry))
1423 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1424 spin_unlock_irq(shost->host_lock);
1426 scsi_dec_host_busy(shost);
1431 * Busy state exporting function for request stacking drivers.
1433 * For efficiency, no lock is taken to check the busy state of
1434 * shost/starget/sdev, since the returned value is not guaranteed and
1435 * may be changed after request stacking drivers call the function,
1436 * regardless of taking lock or not.
1438 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1439 * needs to return 'not busy'. Otherwise, request stacking drivers
1440 * may hold requests forever.
1442 static bool scsi_mq_lld_busy(struct request_queue *q)
1444 struct scsi_device *sdev = q->queuedata;
1445 struct Scsi_Host *shost;
1447 if (blk_queue_dying(q))
1453 * Ignore host/starget busy state.
1454 * Since block layer does not have a concept of fairness across
1455 * multiple queues, congestion of host/starget needs to be handled
1458 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1464 static void scsi_softirq_done(struct request *rq)
1466 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1467 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1470 INIT_LIST_HEAD(&cmd->eh_entry);
1472 atomic_inc(&cmd->device->iodone_cnt);
1474 atomic_inc(&cmd->device->ioerr_cnt);
1476 disposition = scsi_decide_disposition(cmd);
1477 if (disposition != SUCCESS &&
1478 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1479 scmd_printk(KERN_ERR, cmd,
1480 "timing out command, waited %lus\n",
1482 disposition = SUCCESS;
1485 scsi_log_completion(cmd, disposition);
1487 switch (disposition) {
1489 scsi_finish_command(cmd);
1492 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1494 case ADD_TO_MLQUEUE:
1495 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1498 scsi_eh_scmd_add(cmd);
1504 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1505 * @cmd: command block we are dispatching.
1507 * Return: nonzero return request was rejected and device's queue needs to be
1510 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1512 struct Scsi_Host *host = cmd->device->host;
1515 atomic_inc(&cmd->device->iorequest_cnt);
1517 /* check if the device is still usable */
1518 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1519 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1520 * returns an immediate error upwards, and signals
1521 * that the device is no longer present */
1522 cmd->result = DID_NO_CONNECT << 16;
1526 /* Check to see if the scsi lld made this device blocked. */
1527 if (unlikely(scsi_device_blocked(cmd->device))) {
1529 * in blocked state, the command is just put back on
1530 * the device queue. The suspend state has already
1531 * blocked the queue so future requests should not
1532 * occur until the device transitions out of the
1535 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1536 "queuecommand : device blocked\n"));
1537 return SCSI_MLQUEUE_DEVICE_BUSY;
1540 /* Store the LUN value in cmnd, if needed. */
1541 if (cmd->device->lun_in_cdb)
1542 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1543 (cmd->device->lun << 5 & 0xe0);
1548 * Before we queue this command, check if the command
1549 * length exceeds what the host adapter can handle.
1551 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1552 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1553 "queuecommand : command too long. "
1554 "cdb_size=%d host->max_cmd_len=%d\n",
1555 cmd->cmd_len, cmd->device->host->max_cmd_len));
1556 cmd->result = (DID_ABORT << 16);
1560 if (unlikely(host->shost_state == SHOST_DEL)) {
1561 cmd->result = (DID_NO_CONNECT << 16);
1566 trace_scsi_dispatch_cmd_start(cmd);
1567 rtn = host->hostt->queuecommand(host, cmd);
1569 trace_scsi_dispatch_cmd_error(cmd, rtn);
1570 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1571 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1572 rtn = SCSI_MLQUEUE_HOST_BUSY;
1574 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1575 "queuecommand : request rejected\n"));
1580 cmd->scsi_done(cmd);
1584 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1585 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1587 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1588 sizeof(struct scatterlist);
1591 static blk_status_t scsi_mq_prep_fn(struct request *req)
1593 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1594 struct scsi_device *sdev = req->q->queuedata;
1595 struct Scsi_Host *shost = sdev->host;
1596 struct scatterlist *sg;
1598 scsi_init_command(sdev, cmd);
1601 cmd->tag = req->tag;
1602 cmd->prot_op = SCSI_PROT_NORMAL;
1604 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1605 cmd->sdb.table.sgl = sg;
1607 if (scsi_host_get_prot(shost)) {
1608 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1610 cmd->prot_sdb->table.sgl =
1611 (struct scatterlist *)(cmd->prot_sdb + 1);
1614 blk_mq_start_request(req);
1616 return scsi_setup_cmnd(sdev, req);
1619 static void scsi_mq_done(struct scsi_cmnd *cmd)
1621 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1623 trace_scsi_dispatch_cmd_done(cmd);
1626 * If the block layer didn't complete the request due to a timeout
1627 * injection, scsi must clear its internal completed state so that the
1628 * timeout handler will see it needs to escalate its own error
1631 if (unlikely(!blk_mq_complete_request(cmd->request)))
1632 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1635 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1637 struct request_queue *q = hctx->queue;
1638 struct scsi_device *sdev = q->queuedata;
1640 atomic_dec(&sdev->device_busy);
1643 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1645 struct request_queue *q = hctx->queue;
1646 struct scsi_device *sdev = q->queuedata;
1648 if (scsi_dev_queue_ready(q, sdev))
1651 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1652 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1656 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1657 const struct blk_mq_queue_data *bd)
1659 struct request *req = bd->rq;
1660 struct request_queue *q = req->q;
1661 struct scsi_device *sdev = q->queuedata;
1662 struct Scsi_Host *shost = sdev->host;
1663 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1668 * If the device is not in running state we will reject some or all
1671 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1672 ret = scsi_prep_state_check(sdev, req);
1673 if (ret != BLK_STS_OK)
1674 goto out_put_budget;
1677 ret = BLK_STS_RESOURCE;
1678 if (!scsi_target_queue_ready(shost, sdev))
1679 goto out_put_budget;
1680 if (!scsi_host_queue_ready(q, shost, sdev))
1681 goto out_dec_target_busy;
1683 if (!(req->rq_flags & RQF_DONTPREP)) {
1684 ret = scsi_mq_prep_fn(req);
1685 if (ret != BLK_STS_OK)
1686 goto out_dec_host_busy;
1687 req->rq_flags |= RQF_DONTPREP;
1689 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1690 blk_mq_start_request(req);
1693 cmd->flags &= SCMD_PRESERVED_FLAGS;
1694 if (sdev->simple_tags)
1695 cmd->flags |= SCMD_TAGGED;
1697 cmd->flags |= SCMD_LAST;
1699 scsi_init_cmd_errh(cmd);
1700 cmd->scsi_done = scsi_mq_done;
1702 reason = scsi_dispatch_cmd(cmd);
1704 scsi_set_blocked(cmd, reason);
1705 ret = BLK_STS_RESOURCE;
1706 goto out_dec_host_busy;
1712 scsi_dec_host_busy(shost);
1713 out_dec_target_busy:
1714 if (scsi_target(sdev)->can_queue > 0)
1715 atomic_dec(&scsi_target(sdev)->target_busy);
1717 scsi_mq_put_budget(hctx);
1721 case BLK_STS_RESOURCE:
1722 if (atomic_read(&sdev->device_busy) ||
1723 scsi_device_blocked(sdev))
1724 ret = BLK_STS_DEV_RESOURCE;
1727 if (unlikely(!scsi_device_online(sdev)))
1728 scsi_req(req)->result = DID_NO_CONNECT << 16;
1730 scsi_req(req)->result = DID_ERROR << 16;
1732 * Make sure to release all allocated resources when
1733 * we hit an error, as we will never see this command
1736 if (req->rq_flags & RQF_DONTPREP)
1737 scsi_mq_uninit_cmd(cmd);
1738 scsi_run_queue_async(sdev);
1744 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1748 return BLK_EH_RESET_TIMER;
1749 return scsi_times_out(req);
1752 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1753 unsigned int hctx_idx, unsigned int numa_node)
1755 struct Scsi_Host *shost = set->driver_data;
1756 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1757 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1758 struct scatterlist *sg;
1760 if (unchecked_isa_dma)
1761 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1762 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1763 GFP_KERNEL, numa_node);
1764 if (!cmd->sense_buffer)
1766 cmd->req.sense = cmd->sense_buffer;
1768 if (scsi_host_get_prot(shost)) {
1769 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1770 shost->hostt->cmd_size;
1771 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1777 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1778 unsigned int hctx_idx)
1780 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1782 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1786 static int scsi_map_queues(struct blk_mq_tag_set *set)
1788 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1790 if (shost->hostt->map_queues)
1791 return shost->hostt->map_queues(shost);
1792 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1795 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1797 struct device *dev = shost->dma_dev;
1800 * this limit is imposed by hardware restrictions
1802 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1805 if (scsi_host_prot_dma(shost)) {
1806 shost->sg_prot_tablesize =
1807 min_not_zero(shost->sg_prot_tablesize,
1808 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1809 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1810 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1813 if (dev->dma_mask) {
1814 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1815 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1817 blk_queue_max_hw_sectors(q, shost->max_sectors);
1818 if (shost->unchecked_isa_dma)
1819 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1820 blk_queue_segment_boundary(q, shost->dma_boundary);
1821 dma_set_seg_boundary(dev, shost->dma_boundary);
1823 blk_queue_max_segment_size(q, shost->max_segment_size);
1824 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1825 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1828 * Set a reasonable default alignment: The larger of 32-byte (dword),
1829 * which is a common minimum for HBAs, and the minimum DMA alignment,
1830 * which is set by the platform.
1832 * Devices that require a bigger alignment can increase it later.
1834 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1836 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1838 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1839 .get_budget = scsi_mq_get_budget,
1840 .put_budget = scsi_mq_put_budget,
1841 .queue_rq = scsi_queue_rq,
1842 .complete = scsi_softirq_done,
1843 .timeout = scsi_timeout,
1844 #ifdef CONFIG_BLK_DEBUG_FS
1845 .show_rq = scsi_show_rq,
1847 .init_request = scsi_mq_init_request,
1848 .exit_request = scsi_mq_exit_request,
1849 .initialize_rq_fn = scsi_initialize_rq,
1850 .cleanup_rq = scsi_cleanup_rq,
1851 .busy = scsi_mq_lld_busy,
1852 .map_queues = scsi_map_queues,
1856 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1858 struct request_queue *q = hctx->queue;
1859 struct scsi_device *sdev = q->queuedata;
1860 struct Scsi_Host *shost = sdev->host;
1862 shost->hostt->commit_rqs(shost, hctx->queue_num);
1865 static const struct blk_mq_ops scsi_mq_ops = {
1866 .get_budget = scsi_mq_get_budget,
1867 .put_budget = scsi_mq_put_budget,
1868 .queue_rq = scsi_queue_rq,
1869 .commit_rqs = scsi_commit_rqs,
1870 .complete = scsi_softirq_done,
1871 .timeout = scsi_timeout,
1872 #ifdef CONFIG_BLK_DEBUG_FS
1873 .show_rq = scsi_show_rq,
1875 .init_request = scsi_mq_init_request,
1876 .exit_request = scsi_mq_exit_request,
1877 .initialize_rq_fn = scsi_initialize_rq,
1878 .cleanup_rq = scsi_cleanup_rq,
1879 .busy = scsi_mq_lld_busy,
1880 .map_queues = scsi_map_queues,
1883 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1885 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1886 if (IS_ERR(sdev->request_queue))
1889 sdev->request_queue->queuedata = sdev;
1890 __scsi_init_queue(sdev->host, sdev->request_queue);
1891 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1892 return sdev->request_queue;
1895 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1897 unsigned int cmd_size, sgl_size;
1899 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1900 scsi_mq_inline_sgl_size(shost));
1901 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1902 if (scsi_host_get_prot(shost))
1903 cmd_size += sizeof(struct scsi_data_buffer) +
1904 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1906 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1907 if (shost->hostt->commit_rqs)
1908 shost->tag_set.ops = &scsi_mq_ops;
1910 shost->tag_set.ops = &scsi_mq_ops_no_commit;
1911 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1912 shost->tag_set.queue_depth = shost->can_queue;
1913 shost->tag_set.cmd_size = cmd_size;
1914 shost->tag_set.numa_node = NUMA_NO_NODE;
1915 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1916 shost->tag_set.flags |=
1917 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1918 shost->tag_set.driver_data = shost;
1920 return blk_mq_alloc_tag_set(&shost->tag_set);
1923 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1925 blk_mq_free_tag_set(&shost->tag_set);
1929 * scsi_device_from_queue - return sdev associated with a request_queue
1930 * @q: The request queue to return the sdev from
1932 * Return the sdev associated with a request queue or NULL if the
1933 * request_queue does not reference a SCSI device.
1935 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1937 struct scsi_device *sdev = NULL;
1939 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1940 q->mq_ops == &scsi_mq_ops)
1941 sdev = q->queuedata;
1942 if (!sdev || !get_device(&sdev->sdev_gendev))
1947 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1950 * Function: scsi_block_requests()
1952 * Purpose: Utility function used by low-level drivers to prevent further
1953 * commands from being queued to the device.
1955 * Arguments: shost - Host in question
1959 * Lock status: No locks are assumed held.
1961 * Notes: There is no timer nor any other means by which the requests
1962 * get unblocked other than the low-level driver calling
1963 * scsi_unblock_requests().
1965 void scsi_block_requests(struct Scsi_Host *shost)
1967 shost->host_self_blocked = 1;
1969 EXPORT_SYMBOL(scsi_block_requests);
1972 * Function: scsi_unblock_requests()
1974 * Purpose: Utility function used by low-level drivers to allow further
1975 * commands from being queued to the device.
1977 * Arguments: shost - Host in question
1981 * Lock status: No locks are assumed held.
1983 * Notes: There is no timer nor any other means by which the requests
1984 * get unblocked other than the low-level driver calling
1985 * scsi_unblock_requests().
1987 * This is done as an API function so that changes to the
1988 * internals of the scsi mid-layer won't require wholesale
1989 * changes to drivers that use this feature.
1991 void scsi_unblock_requests(struct Scsi_Host *shost)
1993 shost->host_self_blocked = 0;
1994 scsi_run_host_queues(shost);
1996 EXPORT_SYMBOL(scsi_unblock_requests);
1998 int __init scsi_init_queue(void)
2000 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2001 sizeof(struct scsi_data_buffer),
2003 if (!scsi_sdb_cache) {
2004 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2011 void scsi_exit_queue(void)
2013 kmem_cache_destroy(scsi_sense_cache);
2014 kmem_cache_destroy(scsi_sense_isadma_cache);
2015 kmem_cache_destroy(scsi_sdb_cache);
2019 * scsi_mode_select - issue a mode select
2020 * @sdev: SCSI device to be queried
2021 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2022 * @sp: Save page bit (0 == don't save, 1 == save)
2023 * @modepage: mode page being requested
2024 * @buffer: request buffer (may not be smaller than eight bytes)
2025 * @len: length of request buffer.
2026 * @timeout: command timeout
2027 * @retries: number of retries before failing
2028 * @data: returns a structure abstracting the mode header data
2029 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2030 * must be SCSI_SENSE_BUFFERSIZE big.
2032 * Returns zero if successful; negative error number or scsi
2037 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2038 unsigned char *buffer, int len, int timeout, int retries,
2039 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2041 unsigned char cmd[10];
2042 unsigned char *real_buffer;
2045 memset(cmd, 0, sizeof(cmd));
2046 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2048 if (sdev->use_10_for_ms) {
2051 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2054 memcpy(real_buffer + 8, buffer, len);
2058 real_buffer[2] = data->medium_type;
2059 real_buffer[3] = data->device_specific;
2060 real_buffer[4] = data->longlba ? 0x01 : 0;
2062 real_buffer[6] = data->block_descriptor_length >> 8;
2063 real_buffer[7] = data->block_descriptor_length;
2065 cmd[0] = MODE_SELECT_10;
2069 if (len > 255 || data->block_descriptor_length > 255 ||
2073 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2076 memcpy(real_buffer + 4, buffer, len);
2079 real_buffer[1] = data->medium_type;
2080 real_buffer[2] = data->device_specific;
2081 real_buffer[3] = data->block_descriptor_length;
2084 cmd[0] = MODE_SELECT;
2088 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2089 sshdr, timeout, retries, NULL);
2093 EXPORT_SYMBOL_GPL(scsi_mode_select);
2096 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2097 * @sdev: SCSI device to be queried
2098 * @dbd: set if mode sense will allow block descriptors to be returned
2099 * @modepage: mode page being requested
2100 * @buffer: request buffer (may not be smaller than eight bytes)
2101 * @len: length of request buffer.
2102 * @timeout: command timeout
2103 * @retries: number of retries before failing
2104 * @data: returns a structure abstracting the mode header data
2105 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2106 * must be SCSI_SENSE_BUFFERSIZE big.
2108 * Returns zero if unsuccessful, or the header offset (either 4
2109 * or 8 depending on whether a six or ten byte command was
2110 * issued) if successful.
2113 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2114 unsigned char *buffer, int len, int timeout, int retries,
2115 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2117 unsigned char cmd[12];
2120 int result, retry_count = retries;
2121 struct scsi_sense_hdr my_sshdr;
2123 memset(data, 0, sizeof(*data));
2124 memset(&cmd[0], 0, 12);
2125 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2128 /* caller might not be interested in sense, but we need it */
2133 use_10_for_ms = sdev->use_10_for_ms;
2135 if (use_10_for_ms) {
2139 cmd[0] = MODE_SENSE_10;
2146 cmd[0] = MODE_SENSE;
2151 memset(buffer, 0, len);
2153 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2154 sshdr, timeout, retries, NULL);
2156 /* This code looks awful: what it's doing is making sure an
2157 * ILLEGAL REQUEST sense return identifies the actual command
2158 * byte as the problem. MODE_SENSE commands can return
2159 * ILLEGAL REQUEST if the code page isn't supported */
2161 if (use_10_for_ms && !scsi_status_is_good(result) &&
2162 driver_byte(result) == DRIVER_SENSE) {
2163 if (scsi_sense_valid(sshdr)) {
2164 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2165 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2167 * Invalid command operation code
2169 sdev->use_10_for_ms = 0;
2175 if(scsi_status_is_good(result)) {
2176 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2177 (modepage == 6 || modepage == 8))) {
2178 /* Initio breakage? */
2181 data->medium_type = 0;
2182 data->device_specific = 0;
2184 data->block_descriptor_length = 0;
2185 } else if(use_10_for_ms) {
2186 data->length = buffer[0]*256 + buffer[1] + 2;
2187 data->medium_type = buffer[2];
2188 data->device_specific = buffer[3];
2189 data->longlba = buffer[4] & 0x01;
2190 data->block_descriptor_length = buffer[6]*256
2193 data->length = buffer[0] + 1;
2194 data->medium_type = buffer[1];
2195 data->device_specific = buffer[2];
2196 data->block_descriptor_length = buffer[3];
2198 data->header_length = header_length;
2199 } else if ((status_byte(result) == CHECK_CONDITION) &&
2200 scsi_sense_valid(sshdr) &&
2201 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2208 EXPORT_SYMBOL(scsi_mode_sense);
2211 * scsi_test_unit_ready - test if unit is ready
2212 * @sdev: scsi device to change the state of.
2213 * @timeout: command timeout
2214 * @retries: number of retries before failing
2215 * @sshdr: outpout pointer for decoded sense information.
2217 * Returns zero if unsuccessful or an error if TUR failed. For
2218 * removable media, UNIT_ATTENTION sets ->changed flag.
2221 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2222 struct scsi_sense_hdr *sshdr)
2225 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2229 /* try to eat the UNIT_ATTENTION if there are enough retries */
2231 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2233 if (sdev->removable && scsi_sense_valid(sshdr) &&
2234 sshdr->sense_key == UNIT_ATTENTION)
2236 } while (scsi_sense_valid(sshdr) &&
2237 sshdr->sense_key == UNIT_ATTENTION && --retries);
2241 EXPORT_SYMBOL(scsi_test_unit_ready);
2244 * scsi_device_set_state - Take the given device through the device state model.
2245 * @sdev: scsi device to change the state of.
2246 * @state: state to change to.
2248 * Returns zero if successful or an error if the requested
2249 * transition is illegal.
2252 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2254 enum scsi_device_state oldstate = sdev->sdev_state;
2256 if (state == oldstate)
2262 case SDEV_CREATED_BLOCK:
2273 case SDEV_TRANSPORT_OFFLINE:
2286 case SDEV_TRANSPORT_OFFLINE:
2294 case SDEV_TRANSPORT_OFFLINE:
2309 case SDEV_CREATED_BLOCK:
2317 case SDEV_CREATED_BLOCK:
2332 case SDEV_TRANSPORT_OFFLINE:
2344 case SDEV_TRANSPORT_OFFLINE:
2347 case SDEV_CREATED_BLOCK:
2355 sdev->sdev_state = state;
2359 SCSI_LOG_ERROR_RECOVERY(1,
2360 sdev_printk(KERN_ERR, sdev,
2361 "Illegal state transition %s->%s",
2362 scsi_device_state_name(oldstate),
2363 scsi_device_state_name(state))
2367 EXPORT_SYMBOL(scsi_device_set_state);
2370 * sdev_evt_emit - emit a single SCSI device uevent
2371 * @sdev: associated SCSI device
2372 * @evt: event to emit
2374 * Send a single uevent (scsi_event) to the associated scsi_device.
2376 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2381 switch (evt->evt_type) {
2382 case SDEV_EVT_MEDIA_CHANGE:
2383 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2385 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2386 scsi_rescan_device(&sdev->sdev_gendev);
2387 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2389 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2390 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2392 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2393 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2395 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2396 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2398 case SDEV_EVT_LUN_CHANGE_REPORTED:
2399 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2401 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2402 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2404 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2405 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2414 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2418 * sdev_evt_thread - send a uevent for each scsi event
2419 * @work: work struct for scsi_device
2421 * Dispatch queued events to their associated scsi_device kobjects
2424 void scsi_evt_thread(struct work_struct *work)
2426 struct scsi_device *sdev;
2427 enum scsi_device_event evt_type;
2428 LIST_HEAD(event_list);
2430 sdev = container_of(work, struct scsi_device, event_work);
2432 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2433 if (test_and_clear_bit(evt_type, sdev->pending_events))
2434 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2437 struct scsi_event *evt;
2438 struct list_head *this, *tmp;
2439 unsigned long flags;
2441 spin_lock_irqsave(&sdev->list_lock, flags);
2442 list_splice_init(&sdev->event_list, &event_list);
2443 spin_unlock_irqrestore(&sdev->list_lock, flags);
2445 if (list_empty(&event_list))
2448 list_for_each_safe(this, tmp, &event_list) {
2449 evt = list_entry(this, struct scsi_event, node);
2450 list_del(&evt->node);
2451 scsi_evt_emit(sdev, evt);
2458 * sdev_evt_send - send asserted event to uevent thread
2459 * @sdev: scsi_device event occurred on
2460 * @evt: event to send
2462 * Assert scsi device event asynchronously.
2464 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2466 unsigned long flags;
2469 /* FIXME: currently this check eliminates all media change events
2470 * for polled devices. Need to update to discriminate between AN
2471 * and polled events */
2472 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2478 spin_lock_irqsave(&sdev->list_lock, flags);
2479 list_add_tail(&evt->node, &sdev->event_list);
2480 schedule_work(&sdev->event_work);
2481 spin_unlock_irqrestore(&sdev->list_lock, flags);
2483 EXPORT_SYMBOL_GPL(sdev_evt_send);
2486 * sdev_evt_alloc - allocate a new scsi event
2487 * @evt_type: type of event to allocate
2488 * @gfpflags: GFP flags for allocation
2490 * Allocates and returns a new scsi_event.
2492 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2495 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2499 evt->evt_type = evt_type;
2500 INIT_LIST_HEAD(&evt->node);
2502 /* evt_type-specific initialization, if any */
2504 case SDEV_EVT_MEDIA_CHANGE:
2505 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2506 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2507 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2508 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2509 case SDEV_EVT_LUN_CHANGE_REPORTED:
2510 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2511 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2519 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2522 * sdev_evt_send_simple - send asserted event to uevent thread
2523 * @sdev: scsi_device event occurred on
2524 * @evt_type: type of event to send
2525 * @gfpflags: GFP flags for allocation
2527 * Assert scsi device event asynchronously, given an event type.
2529 void sdev_evt_send_simple(struct scsi_device *sdev,
2530 enum scsi_device_event evt_type, gfp_t gfpflags)
2532 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2534 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2539 sdev_evt_send(sdev, evt);
2541 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2544 * scsi_device_quiesce - Block user issued commands.
2545 * @sdev: scsi device to quiesce.
2547 * This works by trying to transition to the SDEV_QUIESCE state
2548 * (which must be a legal transition). When the device is in this
2549 * state, only special requests will be accepted, all others will
2550 * be deferred. Since special requests may also be requeued requests,
2551 * a successful return doesn't guarantee the device will be
2552 * totally quiescent.
2554 * Must be called with user context, may sleep.
2556 * Returns zero if unsuccessful or an error if not.
2559 scsi_device_quiesce(struct scsi_device *sdev)
2561 struct request_queue *q = sdev->request_queue;
2565 * It is allowed to call scsi_device_quiesce() multiple times from
2566 * the same context but concurrent scsi_device_quiesce() calls are
2569 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2571 if (sdev->quiesced_by == current)
2576 blk_mq_freeze_queue(q);
2578 * Ensure that the effect of blk_set_pm_only() will be visible
2579 * for percpu_ref_tryget() callers that occur after the queue
2580 * unfreeze even if the queue was already frozen before this function
2581 * was called. See also https://lwn.net/Articles/573497/.
2584 blk_mq_unfreeze_queue(q);
2586 mutex_lock(&sdev->state_mutex);
2587 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2589 sdev->quiesced_by = current;
2591 blk_clear_pm_only(q);
2592 mutex_unlock(&sdev->state_mutex);
2596 EXPORT_SYMBOL(scsi_device_quiesce);
2599 * scsi_device_resume - Restart user issued commands to a quiesced device.
2600 * @sdev: scsi device to resume.
2602 * Moves the device from quiesced back to running and restarts the
2605 * Must be called with user context, may sleep.
2607 void scsi_device_resume(struct scsi_device *sdev)
2609 /* check if the device state was mutated prior to resume, and if
2610 * so assume the state is being managed elsewhere (for example
2611 * device deleted during suspend)
2613 mutex_lock(&sdev->state_mutex);
2614 if (sdev->quiesced_by) {
2615 sdev->quiesced_by = NULL;
2616 blk_clear_pm_only(sdev->request_queue);
2618 if (sdev->sdev_state == SDEV_QUIESCE)
2619 scsi_device_set_state(sdev, SDEV_RUNNING);
2620 mutex_unlock(&sdev->state_mutex);
2622 EXPORT_SYMBOL(scsi_device_resume);
2625 device_quiesce_fn(struct scsi_device *sdev, void *data)
2627 scsi_device_quiesce(sdev);
2631 scsi_target_quiesce(struct scsi_target *starget)
2633 starget_for_each_device(starget, NULL, device_quiesce_fn);
2635 EXPORT_SYMBOL(scsi_target_quiesce);
2638 device_resume_fn(struct scsi_device *sdev, void *data)
2640 scsi_device_resume(sdev);
2644 scsi_target_resume(struct scsi_target *starget)
2646 starget_for_each_device(starget, NULL, device_resume_fn);
2648 EXPORT_SYMBOL(scsi_target_resume);
2651 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2652 * @sdev: device to block
2654 * Pause SCSI command processing on the specified device. Does not sleep.
2656 * Returns zero if successful or a negative error code upon failure.
2659 * This routine transitions the device to the SDEV_BLOCK state (which must be
2660 * a legal transition). When the device is in this state, command processing
2661 * is paused until the device leaves the SDEV_BLOCK state. See also
2662 * scsi_internal_device_unblock_nowait().
2664 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2666 struct request_queue *q = sdev->request_queue;
2669 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2671 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2678 * The device has transitioned to SDEV_BLOCK. Stop the
2679 * block layer from calling the midlayer with this device's
2682 blk_mq_quiesce_queue_nowait(q);
2685 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2688 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2689 * @sdev: device to block
2691 * Pause SCSI command processing on the specified device and wait until all
2692 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2694 * Returns zero if successful or a negative error code upon failure.
2697 * This routine transitions the device to the SDEV_BLOCK state (which must be
2698 * a legal transition). When the device is in this state, command processing
2699 * is paused until the device leaves the SDEV_BLOCK state. See also
2700 * scsi_internal_device_unblock().
2702 static int scsi_internal_device_block(struct scsi_device *sdev)
2704 struct request_queue *q = sdev->request_queue;
2707 mutex_lock(&sdev->state_mutex);
2708 err = scsi_internal_device_block_nowait(sdev);
2710 blk_mq_quiesce_queue(q);
2711 mutex_unlock(&sdev->state_mutex);
2716 void scsi_start_queue(struct scsi_device *sdev)
2718 struct request_queue *q = sdev->request_queue;
2720 blk_mq_unquiesce_queue(q);
2724 * scsi_internal_device_unblock_nowait - resume a device after a block request
2725 * @sdev: device to resume
2726 * @new_state: state to set the device to after unblocking
2728 * Restart the device queue for a previously suspended SCSI device. Does not
2731 * Returns zero if successful or a negative error code upon failure.
2734 * This routine transitions the device to the SDEV_RUNNING state or to one of
2735 * the offline states (which must be a legal transition) allowing the midlayer
2736 * to goose the queue for this device.
2738 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2739 enum scsi_device_state new_state)
2741 switch (new_state) {
2743 case SDEV_TRANSPORT_OFFLINE:
2750 * Try to transition the scsi device to SDEV_RUNNING or one of the
2751 * offlined states and goose the device queue if successful.
2753 switch (sdev->sdev_state) {
2755 case SDEV_TRANSPORT_OFFLINE:
2756 sdev->sdev_state = new_state;
2758 case SDEV_CREATED_BLOCK:
2759 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2760 new_state == SDEV_OFFLINE)
2761 sdev->sdev_state = new_state;
2763 sdev->sdev_state = SDEV_CREATED;
2771 scsi_start_queue(sdev);
2775 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2778 * scsi_internal_device_unblock - resume a device after a block request
2779 * @sdev: device to resume
2780 * @new_state: state to set the device to after unblocking
2782 * Restart the device queue for a previously suspended SCSI device. May sleep.
2784 * Returns zero if successful or a negative error code upon failure.
2787 * This routine transitions the device to the SDEV_RUNNING state or to one of
2788 * the offline states (which must be a legal transition) allowing the midlayer
2789 * to goose the queue for this device.
2791 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2792 enum scsi_device_state new_state)
2796 mutex_lock(&sdev->state_mutex);
2797 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2798 mutex_unlock(&sdev->state_mutex);
2804 device_block(struct scsi_device *sdev, void *data)
2808 ret = scsi_internal_device_block(sdev);
2810 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2811 dev_name(&sdev->sdev_gendev), ret);
2815 target_block(struct device *dev, void *data)
2817 if (scsi_is_target_device(dev))
2818 starget_for_each_device(to_scsi_target(dev), NULL,
2824 scsi_target_block(struct device *dev)
2826 if (scsi_is_target_device(dev))
2827 starget_for_each_device(to_scsi_target(dev), NULL,
2830 device_for_each_child(dev, NULL, target_block);
2832 EXPORT_SYMBOL_GPL(scsi_target_block);
2835 device_unblock(struct scsi_device *sdev, void *data)
2837 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2841 target_unblock(struct device *dev, void *data)
2843 if (scsi_is_target_device(dev))
2844 starget_for_each_device(to_scsi_target(dev), data,
2850 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2852 if (scsi_is_target_device(dev))
2853 starget_for_each_device(to_scsi_target(dev), &new_state,
2856 device_for_each_child(dev, &new_state, target_unblock);
2858 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2861 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2862 * @sgl: scatter-gather list
2863 * @sg_count: number of segments in sg
2864 * @offset: offset in bytes into sg, on return offset into the mapped area
2865 * @len: bytes to map, on return number of bytes mapped
2867 * Returns virtual address of the start of the mapped page
2869 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2870 size_t *offset, size_t *len)
2873 size_t sg_len = 0, len_complete = 0;
2874 struct scatterlist *sg;
2877 WARN_ON(!irqs_disabled());
2879 for_each_sg(sgl, sg, sg_count, i) {
2880 len_complete = sg_len; /* Complete sg-entries */
2881 sg_len += sg->length;
2882 if (sg_len > *offset)
2886 if (unlikely(i == sg_count)) {
2887 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2889 __func__, sg_len, *offset, sg_count);
2894 /* Offset starting from the beginning of first page in this sg-entry */
2895 *offset = *offset - len_complete + sg->offset;
2897 /* Assumption: contiguous pages can be accessed as "page + i" */
2898 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2899 *offset &= ~PAGE_MASK;
2901 /* Bytes in this sg-entry from *offset to the end of the page */
2902 sg_len = PAGE_SIZE - *offset;
2906 return kmap_atomic(page);
2908 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2911 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2912 * @virt: virtual address to be unmapped
2914 void scsi_kunmap_atomic_sg(void *virt)
2916 kunmap_atomic(virt);
2918 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2920 void sdev_disable_disk_events(struct scsi_device *sdev)
2922 atomic_inc(&sdev->disk_events_disable_depth);
2924 EXPORT_SYMBOL(sdev_disable_disk_events);
2926 void sdev_enable_disk_events(struct scsi_device *sdev)
2928 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2930 atomic_dec(&sdev->disk_events_disable_depth);
2932 EXPORT_SYMBOL(sdev_enable_disk_events);
2934 static unsigned char designator_prio(const unsigned char *d)
2937 /* not associated with LUN */
2941 /* invalid length */
2945 * Order of preference for lun descriptor:
2946 * - SCSI name string
2947 * - NAA IEEE Registered Extended
2948 * - EUI-64 based 16-byte
2949 * - EUI-64 based 12-byte
2950 * - NAA IEEE Registered
2951 * - NAA IEEE Extended
2952 * - EUI-64 based 8-byte
2953 * - SCSI name string (truncated)
2955 * as longer descriptors reduce the likelyhood
2956 * of identification clashes.
2959 switch (d[1] & 0xf) {
2961 /* SCSI name string, variable-length UTF-8 */
2964 switch (d[4] >> 4) {
2966 /* NAA registered extended */
2969 /* NAA registered */
2975 /* NAA locally assigned */
2984 /* EUI64-based, 16 byte */
2987 /* EUI64-based, 12 byte */
2990 /* EUI64-based, 8 byte */
3007 * scsi_vpd_lun_id - return a unique device identification
3008 * @sdev: SCSI device
3009 * @id: buffer for the identification
3010 * @id_len: length of the buffer
3012 * Copies a unique device identification into @id based
3013 * on the information in the VPD page 0x83 of the device.
3014 * The string will be formatted as a SCSI name string.
3016 * Returns the length of the identification or error on failure.
3017 * If the identifier is longer than the supplied buffer the actual
3018 * identifier length is returned and the buffer is not zero-padded.
3020 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3024 const unsigned char *d, *cur_id_str;
3025 const struct scsi_vpd *vpd_pg83;
3026 int id_size = -EINVAL;
3029 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3035 /* The id string must be at least 20 bytes + terminating NULL byte */
3041 memset(id, 0, id_len);
3042 d = vpd_pg83->data + 4;
3043 while (d < vpd_pg83->data + vpd_pg83->len) {
3044 u8 prio = designator_prio(d);
3046 if (prio == 0 || cur_id_prio > prio)
3049 switch (d[1] & 0xf) {
3052 if (cur_id_size > d[3])
3056 if (cur_id_size + 4 > id_len)
3057 cur_id_size = id_len - 4;
3059 id_size = snprintf(id, id_len, "t10.%*pE",
3060 cur_id_size, cur_id_str);
3067 switch (cur_id_size) {
3069 id_size = snprintf(id, id_len,
3074 id_size = snprintf(id, id_len,
3079 id_size = snprintf(id, id_len,
3092 switch (cur_id_size) {
3094 id_size = snprintf(id, id_len,
3099 id_size = snprintf(id, id_len,
3108 /* SCSI name string */
3109 if (cur_id_size > d[3])
3111 /* Prefer others for truncated descriptor */
3112 if (d[3] > id_len) {
3114 if (cur_id_prio > prio)
3118 cur_id_size = id_size = d[3];
3120 if (cur_id_size >= id_len)
3121 cur_id_size = id_len - 1;
3122 memcpy(id, cur_id_str, cur_id_size);
3134 EXPORT_SYMBOL(scsi_vpd_lun_id);
3137 * scsi_vpd_tpg_id - return a target port group identifier
3138 * @sdev: SCSI device
3140 * Returns the Target Port Group identifier from the information
3141 * froom VPD page 0x83 of the device.
3143 * Returns the identifier or error on failure.
3145 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3147 const unsigned char *d;
3148 const struct scsi_vpd *vpd_pg83;
3149 int group_id = -EAGAIN, rel_port = -1;
3152 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3158 d = vpd_pg83->data + 4;
3159 while (d < vpd_pg83->data + vpd_pg83->len) {
3160 switch (d[1] & 0xf) {
3162 /* Relative target port */
3163 rel_port = get_unaligned_be16(&d[6]);
3166 /* Target port group */
3167 group_id = get_unaligned_be16(&d[6]);
3176 if (group_id >= 0 && rel_id && rel_port != -1)
3181 EXPORT_SYMBOL(scsi_vpd_tpg_id);