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/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
36 #include <scsi/scsi_dh.h>
38 #include <trace/events/scsi.h>
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
45 * Size of integrity metadata is usually small, 1 inline sg should
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define SCSI_INLINE_PROT_SG_CNT 0
50 #define SCSI_INLINE_SG_CNT 0
52 #define SCSI_INLINE_PROT_SG_CNT 1
53 #define SCSI_INLINE_SG_CNT 2
56 static struct kmem_cache *scsi_sense_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 int scsi_init_sense_cache(struct Scsi_Host *shost)
65 mutex_lock(&scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
74 mutex_unlock(&scsi_sense_cache_mutex);
79 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
80 * not change behaviour from the previous unplug mechanism, experimentation
81 * may prove this needs changing.
83 #define SCSI_QUEUE_DELAY 3
86 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
88 struct Scsi_Host *host = cmd->device->host;
89 struct scsi_device *device = cmd->device;
90 struct scsi_target *starget = scsi_target(device);
93 * Set the appropriate busy bit for the device/host.
95 * If the host/device isn't busy, assume that something actually
96 * completed, and that we should be able to queue a command now.
98 * Note that the prior mid-layer assumption that any host could
99 * always queue at least one command is now broken. The mid-layer
100 * will implement a user specifiable stall (see
101 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
102 * if a command is requeued with no other commands outstanding
103 * either for the device or for the host.
106 case SCSI_MLQUEUE_HOST_BUSY:
107 atomic_set(&host->host_blocked, host->max_host_blocked);
109 case SCSI_MLQUEUE_DEVICE_BUSY:
110 case SCSI_MLQUEUE_EH_RETRY:
111 atomic_set(&device->device_blocked,
112 device->max_device_blocked);
114 case SCSI_MLQUEUE_TARGET_BUSY:
115 atomic_set(&starget->target_blocked,
116 starget->max_target_blocked);
121 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
123 struct request *rq = scsi_cmd_to_rq(cmd);
125 if (rq->rq_flags & RQF_DONTPREP) {
126 rq->rq_flags &= ~RQF_DONTPREP;
127 scsi_mq_uninit_cmd(cmd);
131 blk_mq_requeue_request(rq, true);
135 * __scsi_queue_insert - private queue insertion
136 * @cmd: The SCSI command being requeued
137 * @reason: The reason for the requeue
138 * @unbusy: Whether the queue should be unbusied
140 * This is a private queue insertion. The public interface
141 * scsi_queue_insert() always assumes the queue should be unbusied
142 * because it's always called before the completion. This function is
143 * for a requeue after completion, which should only occur in this
146 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
148 struct scsi_device *device = cmd->device;
150 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
151 "Inserting command %p into mlqueue\n", cmd));
153 scsi_set_blocked(cmd, reason);
156 * Decrement the counters, since these commands are no longer
157 * active on the host/device.
160 scsi_device_unbusy(device, cmd);
163 * Requeue this command. It will go before all other commands
164 * that are already in the queue. Schedule requeue work under
165 * lock such that the kblockd_schedule_work() call happens
166 * before blk_cleanup_queue() finishes.
170 blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
174 * scsi_queue_insert - Reinsert a command in the queue.
175 * @cmd: command that we are adding to queue.
176 * @reason: why we are inserting command to queue.
178 * We do this for one of two cases. Either the host is busy and it cannot accept
179 * any more commands for the time being, or the device returned QUEUE_FULL and
180 * can accept no more commands.
182 * Context: This could be called either from an interrupt context or a normal
185 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
187 __scsi_queue_insert(cmd, reason, true);
192 * __scsi_execute - insert request and wait for the result
195 * @data_direction: data direction
196 * @buffer: data buffer
197 * @bufflen: len of buffer
198 * @sense: optional sense buffer
199 * @sshdr: optional decoded sense header
200 * @timeout: request timeout in HZ
201 * @retries: number of times to retry request
202 * @flags: flags for ->cmd_flags
203 * @rq_flags: flags for ->rq_flags
204 * @resid: optional residual length
206 * Returns the scsi_cmnd result field if a command was executed, or a negative
207 * Linux error code if we didn't get that far.
209 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
210 int data_direction, void *buffer, unsigned bufflen,
211 unsigned char *sense, struct scsi_sense_hdr *sshdr,
212 int timeout, int retries, u64 flags, req_flags_t rq_flags,
216 struct scsi_cmnd *scmd;
219 req = scsi_alloc_request(sdev->request_queue,
220 data_direction == DMA_TO_DEVICE ?
221 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
222 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
227 ret = blk_rq_map_kern(sdev->request_queue, req,
228 buffer, bufflen, GFP_NOIO);
232 scmd = blk_mq_rq_to_pdu(req);
233 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
234 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
235 scmd->allowed = retries;
236 req->timeout = timeout;
237 req->cmd_flags |= flags;
238 req->rq_flags |= rq_flags | RQF_QUIET;
241 * head injection *required* here otherwise quiesce won't work
243 blk_execute_rq(req, true);
246 * Some devices (USB mass-storage in particular) may transfer
247 * garbage data together with a residue indicating that the data
248 * is invalid. Prevent the garbage from being misinterpreted
249 * and prevent security leaks by zeroing out the excess data.
251 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
252 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
255 *resid = scmd->resid_len;
256 if (sense && scmd->sense_len)
257 memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
259 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
263 blk_mq_free_request(req);
267 EXPORT_SYMBOL(__scsi_execute);
270 * Wake up the error handler if necessary. Avoid as follows that the error
271 * handler is not woken up if host in-flight requests number ==
272 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
273 * with an RCU read lock in this function to ensure that this function in
274 * its entirety either finishes before scsi_eh_scmd_add() increases the
275 * host_failed counter or that it notices the shost state change made by
276 * scsi_eh_scmd_add().
278 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
283 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
284 if (unlikely(scsi_host_in_recovery(shost))) {
285 spin_lock_irqsave(shost->host_lock, flags);
286 if (shost->host_failed || shost->host_eh_scheduled)
287 scsi_eh_wakeup(shost);
288 spin_unlock_irqrestore(shost->host_lock, flags);
293 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
295 struct Scsi_Host *shost = sdev->host;
296 struct scsi_target *starget = scsi_target(sdev);
298 scsi_dec_host_busy(shost, cmd);
300 if (starget->can_queue > 0)
301 atomic_dec(&starget->target_busy);
303 sbitmap_put(&sdev->budget_map, cmd->budget_token);
304 cmd->budget_token = -1;
307 static void scsi_kick_queue(struct request_queue *q)
309 blk_mq_run_hw_queues(q, false);
313 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
314 * and call blk_run_queue for all the scsi_devices on the target -
315 * including current_sdev first.
317 * Called with *no* scsi locks held.
319 static void scsi_single_lun_run(struct scsi_device *current_sdev)
321 struct Scsi_Host *shost = current_sdev->host;
322 struct scsi_device *sdev, *tmp;
323 struct scsi_target *starget = scsi_target(current_sdev);
326 spin_lock_irqsave(shost->host_lock, flags);
327 starget->starget_sdev_user = NULL;
328 spin_unlock_irqrestore(shost->host_lock, flags);
331 * Call blk_run_queue for all LUNs on the target, starting with
332 * current_sdev. We race with others (to set starget_sdev_user),
333 * but in most cases, we will be first. Ideally, each LU on the
334 * target would get some limited time or requests on the target.
336 scsi_kick_queue(current_sdev->request_queue);
338 spin_lock_irqsave(shost->host_lock, flags);
339 if (starget->starget_sdev_user)
341 list_for_each_entry_safe(sdev, tmp, &starget->devices,
342 same_target_siblings) {
343 if (sdev == current_sdev)
345 if (scsi_device_get(sdev))
348 spin_unlock_irqrestore(shost->host_lock, flags);
349 scsi_kick_queue(sdev->request_queue);
350 spin_lock_irqsave(shost->host_lock, flags);
352 scsi_device_put(sdev);
355 spin_unlock_irqrestore(shost->host_lock, flags);
358 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
360 if (scsi_device_busy(sdev) >= sdev->queue_depth)
362 if (atomic_read(&sdev->device_blocked) > 0)
367 static inline bool scsi_target_is_busy(struct scsi_target *starget)
369 if (starget->can_queue > 0) {
370 if (atomic_read(&starget->target_busy) >= starget->can_queue)
372 if (atomic_read(&starget->target_blocked) > 0)
378 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
380 if (atomic_read(&shost->host_blocked) > 0)
382 if (shost->host_self_blocked)
387 static void scsi_starved_list_run(struct Scsi_Host *shost)
389 LIST_HEAD(starved_list);
390 struct scsi_device *sdev;
393 spin_lock_irqsave(shost->host_lock, flags);
394 list_splice_init(&shost->starved_list, &starved_list);
396 while (!list_empty(&starved_list)) {
397 struct request_queue *slq;
400 * As long as shost is accepting commands and we have
401 * starved queues, call blk_run_queue. scsi_request_fn
402 * drops the queue_lock and can add us back to the
405 * host_lock protects the starved_list and starved_entry.
406 * scsi_request_fn must get the host_lock before checking
407 * or modifying starved_list or starved_entry.
409 if (scsi_host_is_busy(shost))
412 sdev = list_entry(starved_list.next,
413 struct scsi_device, starved_entry);
414 list_del_init(&sdev->starved_entry);
415 if (scsi_target_is_busy(scsi_target(sdev))) {
416 list_move_tail(&sdev->starved_entry,
417 &shost->starved_list);
422 * Once we drop the host lock, a racing scsi_remove_device()
423 * call may remove the sdev from the starved list and destroy
424 * it and the queue. Mitigate by taking a reference to the
425 * queue and never touching the sdev again after we drop the
426 * host lock. Note: if __scsi_remove_device() invokes
427 * blk_cleanup_queue() before the queue is run from this
428 * function then blk_run_queue() will return immediately since
429 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
431 slq = sdev->request_queue;
432 if (!blk_get_queue(slq))
434 spin_unlock_irqrestore(shost->host_lock, flags);
436 scsi_kick_queue(slq);
439 spin_lock_irqsave(shost->host_lock, flags);
441 /* put any unprocessed entries back */
442 list_splice(&starved_list, &shost->starved_list);
443 spin_unlock_irqrestore(shost->host_lock, flags);
447 * scsi_run_queue - Select a proper request queue to serve next.
448 * @q: last request's queue
450 * The previous command was completely finished, start a new one if possible.
452 static void scsi_run_queue(struct request_queue *q)
454 struct scsi_device *sdev = q->queuedata;
456 if (scsi_target(sdev)->single_lun)
457 scsi_single_lun_run(sdev);
458 if (!list_empty(&sdev->host->starved_list))
459 scsi_starved_list_run(sdev->host);
461 blk_mq_run_hw_queues(q, false);
464 void scsi_requeue_run_queue(struct work_struct *work)
466 struct scsi_device *sdev;
467 struct request_queue *q;
469 sdev = container_of(work, struct scsi_device, requeue_work);
470 q = sdev->request_queue;
474 void scsi_run_host_queues(struct Scsi_Host *shost)
476 struct scsi_device *sdev;
478 shost_for_each_device(sdev, shost)
479 scsi_run_queue(sdev->request_queue);
482 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
484 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
485 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
487 if (drv->uninit_command)
488 drv->uninit_command(cmd);
492 void scsi_free_sgtables(struct scsi_cmnd *cmd)
494 if (cmd->sdb.table.nents)
495 sg_free_table_chained(&cmd->sdb.table,
497 if (scsi_prot_sg_count(cmd))
498 sg_free_table_chained(&cmd->prot_sdb->table,
499 SCSI_INLINE_PROT_SG_CNT);
501 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
503 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
505 scsi_free_sgtables(cmd);
506 scsi_uninit_cmd(cmd);
509 static void scsi_run_queue_async(struct scsi_device *sdev)
511 if (scsi_target(sdev)->single_lun ||
512 !list_empty(&sdev->host->starved_list)) {
513 kblockd_schedule_work(&sdev->requeue_work);
516 * smp_mb() present in sbitmap_queue_clear() or implied in
517 * .end_io is for ordering writing .device_busy in
518 * scsi_device_unbusy() and reading sdev->restarts.
520 int old = atomic_read(&sdev->restarts);
523 * ->restarts has to be kept as non-zero if new budget
526 * No need to run queue when either another re-run
527 * queue wins in updating ->restarts or a new budget
530 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
531 blk_mq_run_hw_queues(sdev->request_queue, true);
535 /* Returns false when no more bytes to process, true if there are more */
536 static bool scsi_end_request(struct request *req, blk_status_t error,
539 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
540 struct scsi_device *sdev = cmd->device;
541 struct request_queue *q = sdev->request_queue;
543 if (blk_update_request(req, error, bytes))
547 if (blk_queue_add_random(q))
548 add_disk_randomness(req->q->disk);
550 if (!blk_rq_is_passthrough(req)) {
551 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
552 cmd->flags &= ~SCMD_INITIALIZED;
556 * Calling rcu_barrier() is not necessary here because the
557 * SCSI error handler guarantees that the function called by
558 * call_rcu() has been called before scsi_end_request() is
561 destroy_rcu_head(&cmd->rcu);
564 * In the MQ case the command gets freed by __blk_mq_end_request,
565 * so we have to do all cleanup that depends on it earlier.
567 * We also can't kick the queues from irq context, so we
568 * will have to defer it to a workqueue.
570 scsi_mq_uninit_cmd(cmd);
573 * queue is still alive, so grab the ref for preventing it
574 * from being cleaned up during running queue.
576 percpu_ref_get(&q->q_usage_counter);
578 __blk_mq_end_request(req, error);
580 scsi_run_queue_async(sdev);
582 percpu_ref_put(&q->q_usage_counter);
587 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
589 * @result: scsi error code
591 * Translate a SCSI result code into a blk_status_t value. May reset the host
592 * byte of @cmd->result.
594 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
596 switch (host_byte(result)) {
598 if (scsi_status_is_good(result))
600 return BLK_STS_IOERR;
601 case DID_TRANSPORT_FAILFAST:
602 case DID_TRANSPORT_MARGINAL:
603 return BLK_STS_TRANSPORT;
604 case DID_TARGET_FAILURE:
605 set_host_byte(cmd, DID_OK);
606 return BLK_STS_TARGET;
607 case DID_NEXUS_FAILURE:
608 set_host_byte(cmd, DID_OK);
609 return BLK_STS_NEXUS;
610 case DID_ALLOC_FAILURE:
611 set_host_byte(cmd, DID_OK);
612 return BLK_STS_NOSPC;
613 case DID_MEDIUM_ERROR:
614 set_host_byte(cmd, DID_OK);
615 return BLK_STS_MEDIUM;
617 return BLK_STS_IOERR;
622 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
623 * @rq: request to examine
626 * A request could be merge of IOs which require different failure
627 * handling. This function determines the number of bytes which
628 * can be failed from the beginning of the request without
629 * crossing into area which need to be retried further.
632 * The number of bytes to fail.
634 static unsigned int scsi_rq_err_bytes(const struct request *rq)
636 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
637 unsigned int bytes = 0;
640 if (!(rq->rq_flags & RQF_MIXED_MERGE))
641 return blk_rq_bytes(rq);
644 * Currently the only 'mixing' which can happen is between
645 * different fastfail types. We can safely fail portions
646 * which have all the failfast bits that the first one has -
647 * the ones which are at least as eager to fail as the first
650 for (bio = rq->bio; bio; bio = bio->bi_next) {
651 if ((bio->bi_opf & ff) != ff)
653 bytes += bio->bi_iter.bi_size;
656 /* this could lead to infinite loop */
657 BUG_ON(blk_rq_bytes(rq) && !bytes);
661 /* Helper for scsi_io_completion() when "reprep" action required. */
662 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
663 struct request_queue *q)
665 /* A new command will be prepared and issued. */
666 scsi_mq_requeue_cmd(cmd);
669 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
671 struct request *req = scsi_cmd_to_rq(cmd);
672 unsigned long wait_for;
674 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
677 wait_for = (cmd->allowed + 1) * req->timeout;
678 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
679 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
686 /* Helper for scsi_io_completion() when special action required. */
687 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
689 struct request_queue *q = cmd->device->request_queue;
690 struct request *req = scsi_cmd_to_rq(cmd);
692 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
693 ACTION_DELAYED_RETRY} action;
694 struct scsi_sense_hdr sshdr;
696 bool sense_current = true; /* false implies "deferred sense" */
697 blk_status_t blk_stat;
699 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
701 sense_current = !scsi_sense_is_deferred(&sshdr);
703 blk_stat = scsi_result_to_blk_status(cmd, result);
705 if (host_byte(result) == DID_RESET) {
706 /* Third party bus reset or reset for error recovery
707 * reasons. Just retry the command and see what
710 action = ACTION_RETRY;
711 } else if (sense_valid && sense_current) {
712 switch (sshdr.sense_key) {
714 if (cmd->device->removable) {
715 /* Detected disc change. Set a bit
716 * and quietly refuse further access.
718 cmd->device->changed = 1;
719 action = ACTION_FAIL;
721 /* Must have been a power glitch, or a
722 * bus reset. Could not have been a
723 * media change, so we just retry the
724 * command and see what happens.
726 action = ACTION_RETRY;
729 case ILLEGAL_REQUEST:
730 /* If we had an ILLEGAL REQUEST returned, then
731 * we may have performed an unsupported
732 * command. The only thing this should be
733 * would be a ten byte read where only a six
734 * byte read was supported. Also, on a system
735 * where READ CAPACITY failed, we may have
736 * read past the end of the disk.
738 if ((cmd->device->use_10_for_rw &&
739 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
740 (cmd->cmnd[0] == READ_10 ||
741 cmd->cmnd[0] == WRITE_10)) {
742 /* This will issue a new 6-byte command. */
743 cmd->device->use_10_for_rw = 0;
744 action = ACTION_REPREP;
745 } else if (sshdr.asc == 0x10) /* DIX */ {
746 action = ACTION_FAIL;
747 blk_stat = BLK_STS_PROTECTION;
748 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
749 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
750 action = ACTION_FAIL;
751 blk_stat = BLK_STS_TARGET;
753 action = ACTION_FAIL;
755 case ABORTED_COMMAND:
756 action = ACTION_FAIL;
757 if (sshdr.asc == 0x10) /* DIF */
758 blk_stat = BLK_STS_PROTECTION;
761 /* If the device is in the process of becoming
762 * ready, or has a temporary blockage, retry.
764 if (sshdr.asc == 0x04) {
765 switch (sshdr.ascq) {
766 case 0x01: /* becoming ready */
767 case 0x04: /* format in progress */
768 case 0x05: /* rebuild in progress */
769 case 0x06: /* recalculation in progress */
770 case 0x07: /* operation in progress */
771 case 0x08: /* Long write in progress */
772 case 0x09: /* self test in progress */
773 case 0x11: /* notify (enable spinup) required */
774 case 0x14: /* space allocation in progress */
775 case 0x1a: /* start stop unit in progress */
776 case 0x1b: /* sanitize in progress */
777 case 0x1d: /* configuration in progress */
778 case 0x24: /* depopulation in progress */
779 action = ACTION_DELAYED_RETRY;
781 case 0x0a: /* ALUA state transition */
782 blk_stat = BLK_STS_AGAIN;
785 action = ACTION_FAIL;
789 action = ACTION_FAIL;
791 case VOLUME_OVERFLOW:
792 /* See SSC3rXX or current. */
793 action = ACTION_FAIL;
796 action = ACTION_FAIL;
797 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
798 (sshdr.asc == 0x55 &&
799 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
800 /* Insufficient zone resources */
801 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
805 action = ACTION_FAIL;
809 action = ACTION_FAIL;
811 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
812 action = ACTION_FAIL;
816 /* Give up and fail the remainder of the request */
817 if (!(req->rq_flags & RQF_QUIET)) {
818 static DEFINE_RATELIMIT_STATE(_rs,
819 DEFAULT_RATELIMIT_INTERVAL,
820 DEFAULT_RATELIMIT_BURST);
822 if (unlikely(scsi_logging_level))
824 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
825 SCSI_LOG_MLCOMPLETE_BITS);
828 * if logging is enabled the failure will be printed
829 * in scsi_log_completion(), so avoid duplicate messages
831 if (!level && __ratelimit(&_rs)) {
832 scsi_print_result(cmd, NULL, FAILED);
834 scsi_print_sense(cmd);
835 scsi_print_command(cmd);
838 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
842 scsi_io_completion_reprep(cmd, q);
845 /* Retry the same command immediately */
846 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
848 case ACTION_DELAYED_RETRY:
849 /* Retry the same command after a delay */
850 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
856 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
857 * new result that may suppress further error checking. Also modifies
858 * *blk_statp in some cases.
860 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
861 blk_status_t *blk_statp)
864 bool sense_current = true; /* false implies "deferred sense" */
865 struct request *req = scsi_cmd_to_rq(cmd);
866 struct scsi_sense_hdr sshdr;
868 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
870 sense_current = !scsi_sense_is_deferred(&sshdr);
872 if (blk_rq_is_passthrough(req)) {
875 * SG_IO wants current and deferred errors
877 cmd->sense_len = min(8 + cmd->sense_buffer[7],
878 SCSI_SENSE_BUFFERSIZE);
881 *blk_statp = scsi_result_to_blk_status(cmd, result);
882 } else if (blk_rq_bytes(req) == 0 && sense_current) {
884 * Flush commands do not transfers any data, and thus cannot use
885 * good_bytes != blk_rq_bytes(req) as the signal for an error.
886 * This sets *blk_statp explicitly for the problem case.
888 *blk_statp = scsi_result_to_blk_status(cmd, result);
891 * Recovered errors need reporting, but they're always treated as
892 * success, so fiddle the result code here. For passthrough requests
893 * we already took a copy of the original into sreq->result which
894 * is what gets returned to the user
896 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
897 bool do_print = true;
899 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
900 * skip print since caller wants ATA registers. Only occurs
901 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
903 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
905 else if (req->rq_flags & RQF_QUIET)
908 scsi_print_sense(cmd);
910 /* for passthrough, *blk_statp may be set */
911 *blk_statp = BLK_STS_OK;
914 * Another corner case: the SCSI status byte is non-zero but 'good'.
915 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
916 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
917 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
918 * intermediate statuses (both obsolete in SAM-4) as good.
920 if ((result & 0xff) && scsi_status_is_good(result)) {
922 *blk_statp = BLK_STS_OK;
928 * scsi_io_completion - Completion processing for SCSI commands.
929 * @cmd: command that is finished.
930 * @good_bytes: number of processed bytes.
932 * We will finish off the specified number of sectors. If we are done, the
933 * command block will be released and the queue function will be goosed. If we
934 * are not done then we have to figure out what to do next:
936 * a) We can call scsi_io_completion_reprep(). The request will be
937 * unprepared and put back on the queue. Then a new command will
938 * be created for it. This should be used if we made forward
939 * progress, or if we want to switch from READ(10) to READ(6) for
942 * b) We can call scsi_io_completion_action(). The request will be
943 * put back on the queue and retried using the same command as
944 * before, possibly after a delay.
946 * c) We can call scsi_end_request() with blk_stat other than
947 * BLK_STS_OK, to fail the remainder of the request.
949 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
951 int result = cmd->result;
952 struct request_queue *q = cmd->device->request_queue;
953 struct request *req = scsi_cmd_to_rq(cmd);
954 blk_status_t blk_stat = BLK_STS_OK;
956 if (unlikely(result)) /* a nz result may or may not be an error */
957 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
960 * Next deal with any sectors which we were able to correctly
963 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
964 "%u sectors total, %d bytes done.\n",
965 blk_rq_sectors(req), good_bytes));
968 * Failed, zero length commands always need to drop down
969 * to retry code. Fast path should return in this block.
971 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
972 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
973 return; /* no bytes remaining */
976 /* Kill remainder if no retries. */
977 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
978 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
980 "Bytes remaining after failed, no-retry command");
985 * If there had been no error, but we have leftover bytes in the
986 * request just queue the command up again.
988 if (likely(result == 0))
989 scsi_io_completion_reprep(cmd, q);
991 scsi_io_completion_action(cmd, result);
994 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
997 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
998 !op_is_write(req_op(rq)) &&
999 sdev->host->hostt->dma_need_drain(rq);
1003 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1004 * @cmd: SCSI command data structure to initialize.
1006 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1010 * * BLK_STS_OK - on success
1011 * * BLK_STS_RESOURCE - if the failure is retryable
1012 * * BLK_STS_IOERR - if the failure is fatal
1014 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1016 struct scsi_device *sdev = cmd->device;
1017 struct request *rq = scsi_cmd_to_rq(cmd);
1018 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1019 struct scatterlist *last_sg = NULL;
1021 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1024 if (WARN_ON_ONCE(!nr_segs))
1025 return BLK_STS_IOERR;
1028 * Make sure there is space for the drain. The driver must adjust
1029 * max_hw_segments to be prepared for this.
1035 * If sg table allocation fails, requeue request later.
1037 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1038 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1039 return BLK_STS_RESOURCE;
1042 * Next, walk the list, and fill in the addresses and sizes of
1045 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1047 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1048 unsigned int pad_len =
1049 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1051 last_sg->length += pad_len;
1052 cmd->extra_len += pad_len;
1056 sg_unmark_end(last_sg);
1057 last_sg = sg_next(last_sg);
1058 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1059 sg_mark_end(last_sg);
1061 cmd->extra_len += sdev->dma_drain_len;
1065 BUG_ON(count > cmd->sdb.table.nents);
1066 cmd->sdb.table.nents = count;
1067 cmd->sdb.length = blk_rq_payload_bytes(rq);
1069 if (blk_integrity_rq(rq)) {
1070 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1073 if (WARN_ON_ONCE(!prot_sdb)) {
1075 * This can happen if someone (e.g. multipath)
1076 * queues a command to a device on an adapter
1077 * that does not support DIX.
1079 ret = BLK_STS_IOERR;
1080 goto out_free_sgtables;
1083 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1085 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1086 prot_sdb->table.sgl,
1087 SCSI_INLINE_PROT_SG_CNT)) {
1088 ret = BLK_STS_RESOURCE;
1089 goto out_free_sgtables;
1092 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1093 prot_sdb->table.sgl);
1094 BUG_ON(count > ivecs);
1095 BUG_ON(count > queue_max_integrity_segments(rq->q));
1097 cmd->prot_sdb = prot_sdb;
1098 cmd->prot_sdb->table.nents = count;
1103 scsi_free_sgtables(cmd);
1106 EXPORT_SYMBOL(scsi_alloc_sgtables);
1109 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1110 * @rq: Request associated with the SCSI command to be initialized.
1112 * This function initializes the members of struct scsi_cmnd that must be
1113 * initialized before request processing starts and that won't be
1114 * reinitialized if a SCSI command is requeued.
1116 static void scsi_initialize_rq(struct request *rq)
1118 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1120 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1121 cmd->cmd_len = MAX_COMMAND_SIZE;
1123 init_rcu_head(&cmd->rcu);
1124 cmd->jiffies_at_alloc = jiffies;
1128 struct request *scsi_alloc_request(struct request_queue *q,
1129 unsigned int op, blk_mq_req_flags_t flags)
1133 rq = blk_mq_alloc_request(q, op, flags);
1135 scsi_initialize_rq(rq);
1138 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1141 * Only called when the request isn't completed by SCSI, and not freed by
1144 static void scsi_cleanup_rq(struct request *rq)
1146 if (rq->rq_flags & RQF_DONTPREP) {
1147 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1148 rq->rq_flags &= ~RQF_DONTPREP;
1152 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1153 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1155 struct request *rq = scsi_cmd_to_rq(cmd);
1157 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1158 cmd->flags |= SCMD_INITIALIZED;
1159 scsi_initialize_rq(rq);
1163 INIT_LIST_HEAD(&cmd->eh_entry);
1164 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1167 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1168 struct request *req)
1170 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1173 * Passthrough requests may transfer data, in which case they must
1174 * a bio attached to them. Or they might contain a SCSI command
1175 * that does not transfer data, in which case they may optionally
1176 * submit a request without an attached bio.
1179 blk_status_t ret = scsi_alloc_sgtables(cmd);
1180 if (unlikely(ret != BLK_STS_OK))
1183 BUG_ON(blk_rq_bytes(req));
1185 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1188 cmd->transfersize = blk_rq_bytes(req);
1193 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1195 switch (sdev->sdev_state) {
1199 case SDEV_TRANSPORT_OFFLINE:
1201 * If the device is offline we refuse to process any
1202 * commands. The device must be brought online
1203 * before trying any recovery commands.
1205 if (!sdev->offline_already) {
1206 sdev->offline_already = true;
1207 sdev_printk(KERN_ERR, sdev,
1208 "rejecting I/O to offline device\n");
1210 return BLK_STS_IOERR;
1213 * If the device is fully deleted, we refuse to
1214 * process any commands as well.
1216 sdev_printk(KERN_ERR, sdev,
1217 "rejecting I/O to dead device\n");
1218 return BLK_STS_IOERR;
1220 case SDEV_CREATED_BLOCK:
1221 return BLK_STS_RESOURCE;
1224 * If the device is blocked we only accept power management
1227 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1228 return BLK_STS_RESOURCE;
1232 * For any other not fully online state we only allow
1233 * power management commands.
1235 if (req && !(req->rq_flags & RQF_PM))
1236 return BLK_STS_OFFLINE;
1242 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1243 * and return the token else return -1.
1245 static inline int scsi_dev_queue_ready(struct request_queue *q,
1246 struct scsi_device *sdev)
1250 token = sbitmap_get(&sdev->budget_map);
1251 if (atomic_read(&sdev->device_blocked)) {
1255 if (scsi_device_busy(sdev) > 1)
1259 * unblock after device_blocked iterates to zero
1261 if (atomic_dec_return(&sdev->device_blocked) > 0)
1263 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1264 "unblocking device at zero depth\n"));
1270 sbitmap_put(&sdev->budget_map, token);
1276 * scsi_target_queue_ready: checks if there we can send commands to target
1277 * @sdev: scsi device on starget to check.
1279 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1280 struct scsi_device *sdev)
1282 struct scsi_target *starget = scsi_target(sdev);
1285 if (starget->single_lun) {
1286 spin_lock_irq(shost->host_lock);
1287 if (starget->starget_sdev_user &&
1288 starget->starget_sdev_user != sdev) {
1289 spin_unlock_irq(shost->host_lock);
1292 starget->starget_sdev_user = sdev;
1293 spin_unlock_irq(shost->host_lock);
1296 if (starget->can_queue <= 0)
1299 busy = atomic_inc_return(&starget->target_busy) - 1;
1300 if (atomic_read(&starget->target_blocked) > 0) {
1305 * unblock after target_blocked iterates to zero
1307 if (atomic_dec_return(&starget->target_blocked) > 0)
1310 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1311 "unblocking target at zero depth\n"));
1314 if (busy >= starget->can_queue)
1320 spin_lock_irq(shost->host_lock);
1321 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1322 spin_unlock_irq(shost->host_lock);
1324 if (starget->can_queue > 0)
1325 atomic_dec(&starget->target_busy);
1330 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1331 * return 0. We must end up running the queue again whenever 0 is
1332 * returned, else IO can hang.
1334 static inline int scsi_host_queue_ready(struct request_queue *q,
1335 struct Scsi_Host *shost,
1336 struct scsi_device *sdev,
1337 struct scsi_cmnd *cmd)
1339 if (scsi_host_in_recovery(shost))
1342 if (atomic_read(&shost->host_blocked) > 0) {
1343 if (scsi_host_busy(shost) > 0)
1347 * unblock after host_blocked iterates to zero
1349 if (atomic_dec_return(&shost->host_blocked) > 0)
1353 shost_printk(KERN_INFO, shost,
1354 "unblocking host at zero depth\n"));
1357 if (shost->host_self_blocked)
1360 /* We're OK to process the command, so we can't be starved */
1361 if (!list_empty(&sdev->starved_entry)) {
1362 spin_lock_irq(shost->host_lock);
1363 if (!list_empty(&sdev->starved_entry))
1364 list_del_init(&sdev->starved_entry);
1365 spin_unlock_irq(shost->host_lock);
1368 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1373 spin_lock_irq(shost->host_lock);
1374 if (list_empty(&sdev->starved_entry))
1375 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1376 spin_unlock_irq(shost->host_lock);
1378 scsi_dec_host_busy(shost, cmd);
1383 * Busy state exporting function for request stacking drivers.
1385 * For efficiency, no lock is taken to check the busy state of
1386 * shost/starget/sdev, since the returned value is not guaranteed and
1387 * may be changed after request stacking drivers call the function,
1388 * regardless of taking lock or not.
1390 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1391 * needs to return 'not busy'. Otherwise, request stacking drivers
1392 * may hold requests forever.
1394 static bool scsi_mq_lld_busy(struct request_queue *q)
1396 struct scsi_device *sdev = q->queuedata;
1397 struct Scsi_Host *shost;
1399 if (blk_queue_dying(q))
1405 * Ignore host/starget busy state.
1406 * Since block layer does not have a concept of fairness across
1407 * multiple queues, congestion of host/starget needs to be handled
1410 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1417 * Block layer request completion callback. May be called from interrupt
1420 static void scsi_complete(struct request *rq)
1422 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1423 enum scsi_disposition disposition;
1425 INIT_LIST_HEAD(&cmd->eh_entry);
1427 atomic_inc(&cmd->device->iodone_cnt);
1429 atomic_inc(&cmd->device->ioerr_cnt);
1431 disposition = scsi_decide_disposition(cmd);
1432 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1433 disposition = SUCCESS;
1435 scsi_log_completion(cmd, disposition);
1437 switch (disposition) {
1439 scsi_finish_command(cmd);
1442 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1444 case ADD_TO_MLQUEUE:
1445 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1448 scsi_eh_scmd_add(cmd);
1454 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1455 * @cmd: command block we are dispatching.
1457 * Return: nonzero return request was rejected and device's queue needs to be
1460 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1462 struct Scsi_Host *host = cmd->device->host;
1465 atomic_inc(&cmd->device->iorequest_cnt);
1467 /* check if the device is still usable */
1468 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1469 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1470 * returns an immediate error upwards, and signals
1471 * that the device is no longer present */
1472 cmd->result = DID_NO_CONNECT << 16;
1476 /* Check to see if the scsi lld made this device blocked. */
1477 if (unlikely(scsi_device_blocked(cmd->device))) {
1479 * in blocked state, the command is just put back on
1480 * the device queue. The suspend state has already
1481 * blocked the queue so future requests should not
1482 * occur until the device transitions out of the
1485 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1486 "queuecommand : device blocked\n"));
1487 return SCSI_MLQUEUE_DEVICE_BUSY;
1490 /* Store the LUN value in cmnd, if needed. */
1491 if (cmd->device->lun_in_cdb)
1492 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1493 (cmd->device->lun << 5 & 0xe0);
1498 * Before we queue this command, check if the command
1499 * length exceeds what the host adapter can handle.
1501 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1502 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1503 "queuecommand : command too long. "
1504 "cdb_size=%d host->max_cmd_len=%d\n",
1505 cmd->cmd_len, cmd->device->host->max_cmd_len));
1506 cmd->result = (DID_ABORT << 16);
1510 if (unlikely(host->shost_state == SHOST_DEL)) {
1511 cmd->result = (DID_NO_CONNECT << 16);
1516 trace_scsi_dispatch_cmd_start(cmd);
1517 rtn = host->hostt->queuecommand(host, cmd);
1519 trace_scsi_dispatch_cmd_error(cmd, rtn);
1520 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1521 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1522 rtn = SCSI_MLQUEUE_HOST_BUSY;
1524 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1525 "queuecommand : request rejected\n"));
1534 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1535 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1537 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1538 sizeof(struct scatterlist);
1541 static blk_status_t scsi_prepare_cmd(struct request *req)
1543 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1544 struct scsi_device *sdev = req->q->queuedata;
1545 struct Scsi_Host *shost = sdev->host;
1546 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1547 struct scatterlist *sg;
1549 scsi_init_command(sdev, cmd);
1554 cmd->prot_flags = 0;
1556 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1558 cmd->transfersize = 0;
1559 cmd->host_scribble = NULL;
1564 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1567 * Only clear the driver-private command data if the LLD does not supply
1568 * a function to initialize that data.
1570 if (!shost->hostt->init_cmd_priv)
1571 memset(cmd + 1, 0, shost->hostt->cmd_size);
1573 cmd->prot_op = SCSI_PROT_NORMAL;
1574 if (blk_rq_bytes(req))
1575 cmd->sc_data_direction = rq_dma_dir(req);
1577 cmd->sc_data_direction = DMA_NONE;
1579 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 cmd->sdb.table.sgl = sg;
1582 if (scsi_host_get_prot(shost)) {
1583 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1585 cmd->prot_sdb->table.sgl =
1586 (struct scatterlist *)(cmd->prot_sdb + 1);
1590 * Special handling for passthrough commands, which don't go to the ULP
1593 if (blk_rq_is_passthrough(req))
1594 return scsi_setup_scsi_cmnd(sdev, req);
1596 if (sdev->handler && sdev->handler->prep_fn) {
1597 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1599 if (ret != BLK_STS_OK)
1603 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1604 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1607 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1609 struct request *req = scsi_cmd_to_rq(cmd);
1611 switch (cmd->submitter) {
1612 case SUBMITTED_BY_BLOCK_LAYER:
1614 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1615 return scsi_eh_done(cmd);
1616 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1620 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1622 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1624 trace_scsi_dispatch_cmd_done(cmd);
1626 if (complete_directly)
1627 blk_mq_complete_request_direct(req, scsi_complete);
1629 blk_mq_complete_request(req);
1632 void scsi_done(struct scsi_cmnd *cmd)
1634 scsi_done_internal(cmd, false);
1636 EXPORT_SYMBOL(scsi_done);
1638 void scsi_done_direct(struct scsi_cmnd *cmd)
1640 scsi_done_internal(cmd, true);
1642 EXPORT_SYMBOL(scsi_done_direct);
1644 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1646 struct scsi_device *sdev = q->queuedata;
1648 sbitmap_put(&sdev->budget_map, budget_token);
1651 static int scsi_mq_get_budget(struct request_queue *q)
1653 struct scsi_device *sdev = q->queuedata;
1654 int token = scsi_dev_queue_ready(q, sdev);
1659 atomic_inc(&sdev->restarts);
1662 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1663 * .restarts must be incremented before .device_busy is read because the
1664 * code in scsi_run_queue_async() depends on the order of these operations.
1666 smp_mb__after_atomic();
1669 * If all in-flight requests originated from this LUN are completed
1670 * before reading .device_busy, sdev->device_busy will be observed as
1671 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1672 * soon. Otherwise, completion of one of these requests will observe
1673 * the .restarts flag, and the request queue will be run for handling
1674 * this request, see scsi_end_request().
1676 if (unlikely(scsi_device_busy(sdev) == 0 &&
1677 !scsi_device_blocked(sdev)))
1678 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1682 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1684 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1686 cmd->budget_token = token;
1689 static int scsi_mq_get_rq_budget_token(struct request *req)
1691 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1693 return cmd->budget_token;
1696 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1697 const struct blk_mq_queue_data *bd)
1699 struct request *req = bd->rq;
1700 struct request_queue *q = req->q;
1701 struct scsi_device *sdev = q->queuedata;
1702 struct Scsi_Host *shost = sdev->host;
1703 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1707 WARN_ON_ONCE(cmd->budget_token < 0);
1710 * If the device is not in running state we will reject some or all
1713 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1714 ret = scsi_device_state_check(sdev, req);
1715 if (ret != BLK_STS_OK)
1716 goto out_put_budget;
1719 ret = BLK_STS_RESOURCE;
1720 if (!scsi_target_queue_ready(shost, sdev))
1721 goto out_put_budget;
1722 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1723 goto out_dec_target_busy;
1725 if (!(req->rq_flags & RQF_DONTPREP)) {
1726 ret = scsi_prepare_cmd(req);
1727 if (ret != BLK_STS_OK)
1728 goto out_dec_host_busy;
1729 req->rq_flags |= RQF_DONTPREP;
1731 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1734 cmd->flags &= SCMD_PRESERVED_FLAGS;
1735 if (sdev->simple_tags)
1736 cmd->flags |= SCMD_TAGGED;
1738 cmd->flags |= SCMD_LAST;
1740 scsi_set_resid(cmd, 0);
1741 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1742 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1744 blk_mq_start_request(req);
1745 reason = scsi_dispatch_cmd(cmd);
1747 scsi_set_blocked(cmd, reason);
1748 ret = BLK_STS_RESOURCE;
1749 goto out_dec_host_busy;
1755 scsi_dec_host_busy(shost, cmd);
1756 out_dec_target_busy:
1757 if (scsi_target(sdev)->can_queue > 0)
1758 atomic_dec(&scsi_target(sdev)->target_busy);
1760 scsi_mq_put_budget(q, cmd->budget_token);
1761 cmd->budget_token = -1;
1765 case BLK_STS_RESOURCE:
1766 case BLK_STS_ZONE_RESOURCE:
1767 if (scsi_device_blocked(sdev))
1768 ret = BLK_STS_DEV_RESOURCE;
1771 cmd->result = DID_BUS_BUSY << 16;
1772 if (req->rq_flags & RQF_DONTPREP)
1773 scsi_mq_uninit_cmd(cmd);
1776 if (unlikely(!scsi_device_online(sdev)))
1777 cmd->result = DID_NO_CONNECT << 16;
1779 cmd->result = DID_ERROR << 16;
1781 * Make sure to release all allocated resources when
1782 * we hit an error, as we will never see this command
1785 if (req->rq_flags & RQF_DONTPREP)
1786 scsi_mq_uninit_cmd(cmd);
1787 scsi_run_queue_async(sdev);
1793 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1797 return BLK_EH_RESET_TIMER;
1798 return scsi_times_out(req);
1801 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1802 unsigned int hctx_idx, unsigned int numa_node)
1804 struct Scsi_Host *shost = set->driver_data;
1805 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1806 struct scatterlist *sg;
1810 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1811 if (!cmd->sense_buffer)
1814 if (scsi_host_get_prot(shost)) {
1815 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1816 shost->hostt->cmd_size;
1817 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1820 if (shost->hostt->init_cmd_priv) {
1821 ret = shost->hostt->init_cmd_priv(shost, cmd);
1823 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1829 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1830 unsigned int hctx_idx)
1832 struct Scsi_Host *shost = set->driver_data;
1833 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1835 if (shost->hostt->exit_cmd_priv)
1836 shost->hostt->exit_cmd_priv(shost, cmd);
1837 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1841 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1843 struct Scsi_Host *shost = hctx->driver_data;
1845 if (shost->hostt->mq_poll)
1846 return shost->hostt->mq_poll(shost, hctx->queue_num);
1851 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1852 unsigned int hctx_idx)
1854 struct Scsi_Host *shost = data;
1856 hctx->driver_data = shost;
1860 static int scsi_map_queues(struct blk_mq_tag_set *set)
1862 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1864 if (shost->hostt->map_queues)
1865 return shost->hostt->map_queues(shost);
1866 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1869 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1871 struct device *dev = shost->dma_dev;
1874 * this limit is imposed by hardware restrictions
1876 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1879 if (scsi_host_prot_dma(shost)) {
1880 shost->sg_prot_tablesize =
1881 min_not_zero(shost->sg_prot_tablesize,
1882 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1883 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1884 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1887 if (dev->dma_mask) {
1888 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1889 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1891 blk_queue_max_hw_sectors(q, shost->max_sectors);
1892 blk_queue_segment_boundary(q, shost->dma_boundary);
1893 dma_set_seg_boundary(dev, shost->dma_boundary);
1895 blk_queue_max_segment_size(q, shost->max_segment_size);
1896 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1897 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1900 * Set a reasonable default alignment: The larger of 32-byte (dword),
1901 * which is a common minimum for HBAs, and the minimum DMA alignment,
1902 * which is set by the platform.
1904 * Devices that require a bigger alignment can increase it later.
1906 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1908 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1910 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1911 .get_budget = scsi_mq_get_budget,
1912 .put_budget = scsi_mq_put_budget,
1913 .queue_rq = scsi_queue_rq,
1914 .complete = scsi_complete,
1915 .timeout = scsi_timeout,
1916 #ifdef CONFIG_BLK_DEBUG_FS
1917 .show_rq = scsi_show_rq,
1919 .init_request = scsi_mq_init_request,
1920 .exit_request = scsi_mq_exit_request,
1921 .cleanup_rq = scsi_cleanup_rq,
1922 .busy = scsi_mq_lld_busy,
1923 .map_queues = scsi_map_queues,
1924 .init_hctx = scsi_init_hctx,
1925 .poll = scsi_mq_poll,
1926 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1927 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1931 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1933 struct Scsi_Host *shost = hctx->driver_data;
1935 shost->hostt->commit_rqs(shost, hctx->queue_num);
1938 static const struct blk_mq_ops scsi_mq_ops = {
1939 .get_budget = scsi_mq_get_budget,
1940 .put_budget = scsi_mq_put_budget,
1941 .queue_rq = scsi_queue_rq,
1942 .commit_rqs = scsi_commit_rqs,
1943 .complete = scsi_complete,
1944 .timeout = scsi_timeout,
1945 #ifdef CONFIG_BLK_DEBUG_FS
1946 .show_rq = scsi_show_rq,
1948 .init_request = scsi_mq_init_request,
1949 .exit_request = scsi_mq_exit_request,
1950 .cleanup_rq = scsi_cleanup_rq,
1951 .busy = scsi_mq_lld_busy,
1952 .map_queues = scsi_map_queues,
1953 .init_hctx = scsi_init_hctx,
1954 .poll = scsi_mq_poll,
1955 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1956 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1959 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1961 unsigned int cmd_size, sgl_size;
1962 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1964 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1965 scsi_mq_inline_sgl_size(shost));
1966 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1967 if (scsi_host_get_prot(shost))
1968 cmd_size += sizeof(struct scsi_data_buffer) +
1969 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1971 memset(tag_set, 0, sizeof(*tag_set));
1972 if (shost->hostt->commit_rqs)
1973 tag_set->ops = &scsi_mq_ops;
1975 tag_set->ops = &scsi_mq_ops_no_commit;
1976 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1977 tag_set->nr_maps = shost->nr_maps ? : 1;
1978 tag_set->queue_depth = shost->can_queue;
1979 tag_set->cmd_size = cmd_size;
1980 tag_set->numa_node = NUMA_NO_NODE;
1981 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1983 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1984 tag_set->driver_data = shost;
1985 if (shost->host_tagset)
1986 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1988 return blk_mq_alloc_tag_set(tag_set);
1991 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1993 blk_mq_free_tag_set(&shost->tag_set);
1997 * scsi_device_from_queue - return sdev associated with a request_queue
1998 * @q: The request queue to return the sdev from
2000 * Return the sdev associated with a request queue or NULL if the
2001 * request_queue does not reference a SCSI device.
2003 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2005 struct scsi_device *sdev = NULL;
2007 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2008 q->mq_ops == &scsi_mq_ops)
2009 sdev = q->queuedata;
2010 if (!sdev || !get_device(&sdev->sdev_gendev))
2016 * pktcdvd should have been integrated into the SCSI layers, but for historical
2017 * reasons like the old IDE driver it isn't. This export allows it to safely
2018 * probe if a given device is a SCSI one and only attach to that.
2020 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2021 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2025 * scsi_block_requests - Utility function used by low-level drivers to prevent
2026 * further commands from being queued to the device.
2027 * @shost: host in question
2029 * There is no timer nor any other means by which the requests get unblocked
2030 * other than the low-level driver calling scsi_unblock_requests().
2032 void scsi_block_requests(struct Scsi_Host *shost)
2034 shost->host_self_blocked = 1;
2036 EXPORT_SYMBOL(scsi_block_requests);
2039 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2040 * further commands to be queued to the device.
2041 * @shost: host in question
2043 * There is no timer nor any other means by which the requests get unblocked
2044 * other than the low-level driver calling scsi_unblock_requests(). This is done
2045 * as an API function so that changes to the internals of the scsi mid-layer
2046 * won't require wholesale changes to drivers that use this feature.
2048 void scsi_unblock_requests(struct Scsi_Host *shost)
2050 shost->host_self_blocked = 0;
2051 scsi_run_host_queues(shost);
2053 EXPORT_SYMBOL(scsi_unblock_requests);
2055 void scsi_exit_queue(void)
2057 kmem_cache_destroy(scsi_sense_cache);
2061 * scsi_mode_select - issue a mode select
2062 * @sdev: SCSI device to be queried
2063 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2064 * @sp: Save page bit (0 == don't save, 1 == save)
2065 * @buffer: request buffer (may not be smaller than eight bytes)
2066 * @len: length of request buffer.
2067 * @timeout: command timeout
2068 * @retries: number of retries before failing
2069 * @data: returns a structure abstracting the mode header data
2070 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2071 * must be SCSI_SENSE_BUFFERSIZE big.
2073 * Returns zero if successful; negative error number or scsi
2077 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2078 unsigned char *buffer, int len, int timeout, int retries,
2079 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2081 unsigned char cmd[10];
2082 unsigned char *real_buffer;
2085 memset(cmd, 0, sizeof(cmd));
2086 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2089 * Use MODE SELECT(10) if the device asked for it or if the mode page
2090 * and the mode select header cannot fit within the maximumm 255 bytes
2091 * of the MODE SELECT(6) command.
2093 if (sdev->use_10_for_ms ||
2095 data->block_descriptor_length > 255) {
2096 if (len > 65535 - 8)
2098 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2101 memcpy(real_buffer + 8, buffer, len);
2105 real_buffer[2] = data->medium_type;
2106 real_buffer[3] = data->device_specific;
2107 real_buffer[4] = data->longlba ? 0x01 : 0;
2109 put_unaligned_be16(data->block_descriptor_length,
2112 cmd[0] = MODE_SELECT_10;
2113 put_unaligned_be16(len, &cmd[7]);
2118 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2121 memcpy(real_buffer + 4, buffer, len);
2124 real_buffer[1] = data->medium_type;
2125 real_buffer[2] = data->device_specific;
2126 real_buffer[3] = data->block_descriptor_length;
2128 cmd[0] = MODE_SELECT;
2132 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2133 sshdr, timeout, retries, NULL);
2137 EXPORT_SYMBOL_GPL(scsi_mode_select);
2140 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2141 * @sdev: SCSI device to be queried
2142 * @dbd: set to prevent mode sense from returning block descriptors
2143 * @modepage: mode page being requested
2144 * @buffer: request buffer (may not be smaller than eight bytes)
2145 * @len: length of request buffer.
2146 * @timeout: command timeout
2147 * @retries: number of retries before failing
2148 * @data: returns a structure abstracting the mode header data
2149 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2150 * must be SCSI_SENSE_BUFFERSIZE big.
2152 * Returns zero if successful, or a negative error number on failure
2155 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2156 unsigned char *buffer, int len, int timeout, int retries,
2157 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2159 unsigned char cmd[12];
2162 int result, retry_count = retries;
2163 struct scsi_sense_hdr my_sshdr;
2165 memset(data, 0, sizeof(*data));
2166 memset(&cmd[0], 0, 12);
2168 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2169 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2172 /* caller might not be interested in sense, but we need it */
2177 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2179 if (use_10_for_ms) {
2180 if (len < 8 || len > 65535)
2183 cmd[0] = MODE_SENSE_10;
2184 put_unaligned_be16(len, &cmd[7]);
2190 cmd[0] = MODE_SENSE;
2195 memset(buffer, 0, len);
2197 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2198 sshdr, timeout, retries, NULL);
2202 /* This code looks awful: what it's doing is making sure an
2203 * ILLEGAL REQUEST sense return identifies the actual command
2204 * byte as the problem. MODE_SENSE commands can return
2205 * ILLEGAL REQUEST if the code page isn't supported */
2207 if (!scsi_status_is_good(result)) {
2208 if (scsi_sense_valid(sshdr)) {
2209 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2210 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2212 * Invalid command operation code: retry using
2213 * MODE SENSE(6) if this was a MODE SENSE(10)
2214 * request, except if the request mode page is
2215 * too large for MODE SENSE single byte
2216 * allocation length field.
2218 if (use_10_for_ms) {
2221 sdev->use_10_for_ms = 0;
2225 if (scsi_status_is_check_condition(result) &&
2226 sshdr->sense_key == UNIT_ATTENTION &&
2234 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2235 (modepage == 6 || modepage == 8))) {
2236 /* Initio breakage? */
2239 data->medium_type = 0;
2240 data->device_specific = 0;
2242 data->block_descriptor_length = 0;
2243 } else if (use_10_for_ms) {
2244 data->length = get_unaligned_be16(&buffer[0]) + 2;
2245 data->medium_type = buffer[2];
2246 data->device_specific = buffer[3];
2247 data->longlba = buffer[4] & 0x01;
2248 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2250 data->length = buffer[0] + 1;
2251 data->medium_type = buffer[1];
2252 data->device_specific = buffer[2];
2253 data->block_descriptor_length = buffer[3];
2255 data->header_length = header_length;
2259 EXPORT_SYMBOL(scsi_mode_sense);
2262 * scsi_test_unit_ready - test if unit is ready
2263 * @sdev: scsi device to change the state of.
2264 * @timeout: command timeout
2265 * @retries: number of retries before failing
2266 * @sshdr: outpout pointer for decoded sense information.
2268 * Returns zero if unsuccessful or an error if TUR failed. For
2269 * removable media, UNIT_ATTENTION sets ->changed flag.
2272 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2273 struct scsi_sense_hdr *sshdr)
2276 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2280 /* try to eat the UNIT_ATTENTION if there are enough retries */
2282 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2284 if (sdev->removable && scsi_sense_valid(sshdr) &&
2285 sshdr->sense_key == UNIT_ATTENTION)
2287 } while (scsi_sense_valid(sshdr) &&
2288 sshdr->sense_key == UNIT_ATTENTION && --retries);
2292 EXPORT_SYMBOL(scsi_test_unit_ready);
2295 * scsi_device_set_state - Take the given device through the device state model.
2296 * @sdev: scsi device to change the state of.
2297 * @state: state to change to.
2299 * Returns zero if successful or an error if the requested
2300 * transition is illegal.
2303 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2305 enum scsi_device_state oldstate = sdev->sdev_state;
2307 if (state == oldstate)
2313 case SDEV_CREATED_BLOCK:
2324 case SDEV_TRANSPORT_OFFLINE:
2337 case SDEV_TRANSPORT_OFFLINE:
2345 case SDEV_TRANSPORT_OFFLINE:
2360 case SDEV_CREATED_BLOCK:
2369 case SDEV_CREATED_BLOCK:
2384 case SDEV_TRANSPORT_OFFLINE:
2396 case SDEV_TRANSPORT_OFFLINE:
2399 case SDEV_CREATED_BLOCK:
2407 sdev->offline_already = false;
2408 sdev->sdev_state = state;
2412 SCSI_LOG_ERROR_RECOVERY(1,
2413 sdev_printk(KERN_ERR, sdev,
2414 "Illegal state transition %s->%s",
2415 scsi_device_state_name(oldstate),
2416 scsi_device_state_name(state))
2420 EXPORT_SYMBOL(scsi_device_set_state);
2423 * scsi_evt_emit - emit a single SCSI device uevent
2424 * @sdev: associated SCSI device
2425 * @evt: event to emit
2427 * Send a single uevent (scsi_event) to the associated scsi_device.
2429 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2434 switch (evt->evt_type) {
2435 case SDEV_EVT_MEDIA_CHANGE:
2436 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2438 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2439 scsi_rescan_device(&sdev->sdev_gendev);
2440 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2442 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2443 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2445 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2446 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2448 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2449 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2451 case SDEV_EVT_LUN_CHANGE_REPORTED:
2452 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2454 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2455 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2457 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2458 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2467 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2471 * scsi_evt_thread - send a uevent for each scsi event
2472 * @work: work struct for scsi_device
2474 * Dispatch queued events to their associated scsi_device kobjects
2477 void scsi_evt_thread(struct work_struct *work)
2479 struct scsi_device *sdev;
2480 enum scsi_device_event evt_type;
2481 LIST_HEAD(event_list);
2483 sdev = container_of(work, struct scsi_device, event_work);
2485 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2486 if (test_and_clear_bit(evt_type, sdev->pending_events))
2487 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2490 struct scsi_event *evt;
2491 struct list_head *this, *tmp;
2492 unsigned long flags;
2494 spin_lock_irqsave(&sdev->list_lock, flags);
2495 list_splice_init(&sdev->event_list, &event_list);
2496 spin_unlock_irqrestore(&sdev->list_lock, flags);
2498 if (list_empty(&event_list))
2501 list_for_each_safe(this, tmp, &event_list) {
2502 evt = list_entry(this, struct scsi_event, node);
2503 list_del(&evt->node);
2504 scsi_evt_emit(sdev, evt);
2511 * sdev_evt_send - send asserted event to uevent thread
2512 * @sdev: scsi_device event occurred on
2513 * @evt: event to send
2515 * Assert scsi device event asynchronously.
2517 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2519 unsigned long flags;
2522 /* FIXME: currently this check eliminates all media change events
2523 * for polled devices. Need to update to discriminate between AN
2524 * and polled events */
2525 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2531 spin_lock_irqsave(&sdev->list_lock, flags);
2532 list_add_tail(&evt->node, &sdev->event_list);
2533 schedule_work(&sdev->event_work);
2534 spin_unlock_irqrestore(&sdev->list_lock, flags);
2536 EXPORT_SYMBOL_GPL(sdev_evt_send);
2539 * sdev_evt_alloc - allocate a new scsi event
2540 * @evt_type: type of event to allocate
2541 * @gfpflags: GFP flags for allocation
2543 * Allocates and returns a new scsi_event.
2545 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2548 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2552 evt->evt_type = evt_type;
2553 INIT_LIST_HEAD(&evt->node);
2555 /* evt_type-specific initialization, if any */
2557 case SDEV_EVT_MEDIA_CHANGE:
2558 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2559 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2560 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2561 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2562 case SDEV_EVT_LUN_CHANGE_REPORTED:
2563 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2564 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2572 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2575 * sdev_evt_send_simple - send asserted event to uevent thread
2576 * @sdev: scsi_device event occurred on
2577 * @evt_type: type of event to send
2578 * @gfpflags: GFP flags for allocation
2580 * Assert scsi device event asynchronously, given an event type.
2582 void sdev_evt_send_simple(struct scsi_device *sdev,
2583 enum scsi_device_event evt_type, gfp_t gfpflags)
2585 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2587 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2592 sdev_evt_send(sdev, evt);
2594 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2597 * scsi_device_quiesce - Block all commands except power management.
2598 * @sdev: scsi device to quiesce.
2600 * This works by trying to transition to the SDEV_QUIESCE state
2601 * (which must be a legal transition). When the device is in this
2602 * state, only power management requests will be accepted, all others will
2605 * Must be called with user context, may sleep.
2607 * Returns zero if unsuccessful or an error if not.
2610 scsi_device_quiesce(struct scsi_device *sdev)
2612 struct request_queue *q = sdev->request_queue;
2616 * It is allowed to call scsi_device_quiesce() multiple times from
2617 * the same context but concurrent scsi_device_quiesce() calls are
2620 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2622 if (sdev->quiesced_by == current)
2627 blk_mq_freeze_queue(q);
2629 * Ensure that the effect of blk_set_pm_only() will be visible
2630 * for percpu_ref_tryget() callers that occur after the queue
2631 * unfreeze even if the queue was already frozen before this function
2632 * was called. See also https://lwn.net/Articles/573497/.
2635 blk_mq_unfreeze_queue(q);
2637 mutex_lock(&sdev->state_mutex);
2638 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2640 sdev->quiesced_by = current;
2642 blk_clear_pm_only(q);
2643 mutex_unlock(&sdev->state_mutex);
2647 EXPORT_SYMBOL(scsi_device_quiesce);
2650 * scsi_device_resume - Restart user issued commands to a quiesced device.
2651 * @sdev: scsi device to resume.
2653 * Moves the device from quiesced back to running and restarts the
2656 * Must be called with user context, may sleep.
2658 void scsi_device_resume(struct scsi_device *sdev)
2660 /* check if the device state was mutated prior to resume, and if
2661 * so assume the state is being managed elsewhere (for example
2662 * device deleted during suspend)
2664 mutex_lock(&sdev->state_mutex);
2665 if (sdev->sdev_state == SDEV_QUIESCE)
2666 scsi_device_set_state(sdev, SDEV_RUNNING);
2667 if (sdev->quiesced_by) {
2668 sdev->quiesced_by = NULL;
2669 blk_clear_pm_only(sdev->request_queue);
2671 mutex_unlock(&sdev->state_mutex);
2673 EXPORT_SYMBOL(scsi_device_resume);
2676 device_quiesce_fn(struct scsi_device *sdev, void *data)
2678 scsi_device_quiesce(sdev);
2682 scsi_target_quiesce(struct scsi_target *starget)
2684 starget_for_each_device(starget, NULL, device_quiesce_fn);
2686 EXPORT_SYMBOL(scsi_target_quiesce);
2689 device_resume_fn(struct scsi_device *sdev, void *data)
2691 scsi_device_resume(sdev);
2695 scsi_target_resume(struct scsi_target *starget)
2697 starget_for_each_device(starget, NULL, device_resume_fn);
2699 EXPORT_SYMBOL(scsi_target_resume);
2701 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2703 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2704 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2709 void scsi_start_queue(struct scsi_device *sdev)
2711 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2712 blk_mq_unquiesce_queue(sdev->request_queue);
2715 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2718 * The atomic variable of ->queue_stopped covers that
2719 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2721 * However, we still need to wait until quiesce is done
2722 * in case that queue has been stopped.
2724 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2726 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2728 blk_mq_quiesce_queue(sdev->request_queue);
2731 blk_mq_wait_quiesce_done(sdev->request_queue);
2736 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2737 * @sdev: device to block
2739 * Pause SCSI command processing on the specified device. Does not sleep.
2741 * Returns zero if successful or a negative error code upon failure.
2744 * This routine transitions the device to the SDEV_BLOCK state (which must be
2745 * a legal transition). When the device is in this state, command processing
2746 * is paused until the device leaves the SDEV_BLOCK state. See also
2747 * scsi_internal_device_unblock_nowait().
2749 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2751 int ret = __scsi_internal_device_block_nowait(sdev);
2754 * The device has transitioned to SDEV_BLOCK. Stop the
2755 * block layer from calling the midlayer with this device's
2759 scsi_stop_queue(sdev, true);
2762 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2765 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2766 * @sdev: device to block
2768 * Pause SCSI command processing on the specified device and wait until all
2769 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2771 * Returns zero if successful or a negative error code upon failure.
2774 * This routine transitions the device to the SDEV_BLOCK state (which must be
2775 * a legal transition). When the device is in this state, command processing
2776 * is paused until the device leaves the SDEV_BLOCK state. See also
2777 * scsi_internal_device_unblock().
2779 static int scsi_internal_device_block(struct scsi_device *sdev)
2783 mutex_lock(&sdev->state_mutex);
2784 err = __scsi_internal_device_block_nowait(sdev);
2786 scsi_stop_queue(sdev, false);
2787 mutex_unlock(&sdev->state_mutex);
2793 * scsi_internal_device_unblock_nowait - resume a device after a block request
2794 * @sdev: device to resume
2795 * @new_state: state to set the device to after unblocking
2797 * Restart the device queue for a previously suspended SCSI device. Does not
2800 * Returns zero if successful or a negative error code upon failure.
2803 * This routine transitions the device to the SDEV_RUNNING state or to one of
2804 * the offline states (which must be a legal transition) allowing the midlayer
2805 * to goose the queue for this device.
2807 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2808 enum scsi_device_state new_state)
2810 switch (new_state) {
2812 case SDEV_TRANSPORT_OFFLINE:
2819 * Try to transition the scsi device to SDEV_RUNNING or one of the
2820 * offlined states and goose the device queue if successful.
2822 switch (sdev->sdev_state) {
2824 case SDEV_TRANSPORT_OFFLINE:
2825 sdev->sdev_state = new_state;
2827 case SDEV_CREATED_BLOCK:
2828 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2829 new_state == SDEV_OFFLINE)
2830 sdev->sdev_state = new_state;
2832 sdev->sdev_state = SDEV_CREATED;
2840 scsi_start_queue(sdev);
2844 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2847 * scsi_internal_device_unblock - resume a device after a block request
2848 * @sdev: device to resume
2849 * @new_state: state to set the device to after unblocking
2851 * Restart the device queue for a previously suspended SCSI device. May sleep.
2853 * Returns zero if successful or a negative error code upon failure.
2856 * This routine transitions the device to the SDEV_RUNNING state or to one of
2857 * the offline states (which must be a legal transition) allowing the midlayer
2858 * to goose the queue for this device.
2860 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2861 enum scsi_device_state new_state)
2865 mutex_lock(&sdev->state_mutex);
2866 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2867 mutex_unlock(&sdev->state_mutex);
2873 device_block(struct scsi_device *sdev, void *data)
2877 ret = scsi_internal_device_block(sdev);
2879 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2880 dev_name(&sdev->sdev_gendev), ret);
2884 target_block(struct device *dev, void *data)
2886 if (scsi_is_target_device(dev))
2887 starget_for_each_device(to_scsi_target(dev), NULL,
2893 scsi_target_block(struct device *dev)
2895 if (scsi_is_target_device(dev))
2896 starget_for_each_device(to_scsi_target(dev), NULL,
2899 device_for_each_child(dev, NULL, target_block);
2901 EXPORT_SYMBOL_GPL(scsi_target_block);
2904 device_unblock(struct scsi_device *sdev, void *data)
2906 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2910 target_unblock(struct device *dev, void *data)
2912 if (scsi_is_target_device(dev))
2913 starget_for_each_device(to_scsi_target(dev), data,
2919 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2921 if (scsi_is_target_device(dev))
2922 starget_for_each_device(to_scsi_target(dev), &new_state,
2925 device_for_each_child(dev, &new_state, target_unblock);
2927 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2930 scsi_host_block(struct Scsi_Host *shost)
2932 struct scsi_device *sdev;
2936 * Call scsi_internal_device_block_nowait so we can avoid
2937 * calling synchronize_rcu() for each LUN.
2939 shost_for_each_device(sdev, shost) {
2940 mutex_lock(&sdev->state_mutex);
2941 ret = scsi_internal_device_block_nowait(sdev);
2942 mutex_unlock(&sdev->state_mutex);
2944 scsi_device_put(sdev);
2950 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2951 * calling synchronize_rcu() once is enough.
2953 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2960 EXPORT_SYMBOL_GPL(scsi_host_block);
2963 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2965 struct scsi_device *sdev;
2968 shost_for_each_device(sdev, shost) {
2969 ret = scsi_internal_device_unblock(sdev, new_state);
2971 scsi_device_put(sdev);
2977 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2980 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2981 * @sgl: scatter-gather list
2982 * @sg_count: number of segments in sg
2983 * @offset: offset in bytes into sg, on return offset into the mapped area
2984 * @len: bytes to map, on return number of bytes mapped
2986 * Returns virtual address of the start of the mapped page
2988 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2989 size_t *offset, size_t *len)
2992 size_t sg_len = 0, len_complete = 0;
2993 struct scatterlist *sg;
2996 WARN_ON(!irqs_disabled());
2998 for_each_sg(sgl, sg, sg_count, i) {
2999 len_complete = sg_len; /* Complete sg-entries */
3000 sg_len += sg->length;
3001 if (sg_len > *offset)
3005 if (unlikely(i == sg_count)) {
3006 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3008 __func__, sg_len, *offset, sg_count);
3013 /* Offset starting from the beginning of first page in this sg-entry */
3014 *offset = *offset - len_complete + sg->offset;
3016 /* Assumption: contiguous pages can be accessed as "page + i" */
3017 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3018 *offset &= ~PAGE_MASK;
3020 /* Bytes in this sg-entry from *offset to the end of the page */
3021 sg_len = PAGE_SIZE - *offset;
3025 return kmap_atomic(page);
3027 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3030 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3031 * @virt: virtual address to be unmapped
3033 void scsi_kunmap_atomic_sg(void *virt)
3035 kunmap_atomic(virt);
3037 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3039 void sdev_disable_disk_events(struct scsi_device *sdev)
3041 atomic_inc(&sdev->disk_events_disable_depth);
3043 EXPORT_SYMBOL(sdev_disable_disk_events);
3045 void sdev_enable_disk_events(struct scsi_device *sdev)
3047 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3049 atomic_dec(&sdev->disk_events_disable_depth);
3051 EXPORT_SYMBOL(sdev_enable_disk_events);
3053 static unsigned char designator_prio(const unsigned char *d)
3056 /* not associated with LUN */
3060 /* invalid length */
3064 * Order of preference for lun descriptor:
3065 * - SCSI name string
3066 * - NAA IEEE Registered Extended
3067 * - EUI-64 based 16-byte
3068 * - EUI-64 based 12-byte
3069 * - NAA IEEE Registered
3070 * - NAA IEEE Extended
3071 * - EUI-64 based 8-byte
3072 * - SCSI name string (truncated)
3074 * as longer descriptors reduce the likelyhood
3075 * of identification clashes.
3078 switch (d[1] & 0xf) {
3080 /* SCSI name string, variable-length UTF-8 */
3083 switch (d[4] >> 4) {
3085 /* NAA registered extended */
3088 /* NAA registered */
3094 /* NAA locally assigned */
3103 /* EUI64-based, 16 byte */
3106 /* EUI64-based, 12 byte */
3109 /* EUI64-based, 8 byte */
3126 * scsi_vpd_lun_id - return a unique device identification
3127 * @sdev: SCSI device
3128 * @id: buffer for the identification
3129 * @id_len: length of the buffer
3131 * Copies a unique device identification into @id based
3132 * on the information in the VPD page 0x83 of the device.
3133 * The string will be formatted as a SCSI name string.
3135 * Returns the length of the identification or error on failure.
3136 * If the identifier is longer than the supplied buffer the actual
3137 * identifier length is returned and the buffer is not zero-padded.
3139 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3143 const unsigned char *d, *cur_id_str;
3144 const struct scsi_vpd *vpd_pg83;
3145 int id_size = -EINVAL;
3148 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3154 /* The id string must be at least 20 bytes + terminating NULL byte */
3160 memset(id, 0, id_len);
3161 for (d = vpd_pg83->data + 4;
3162 d < vpd_pg83->data + vpd_pg83->len;
3164 u8 prio = designator_prio(d);
3166 if (prio == 0 || cur_id_prio > prio)
3169 switch (d[1] & 0xf) {
3172 if (cur_id_size > d[3])
3176 if (cur_id_size + 4 > id_len)
3177 cur_id_size = id_len - 4;
3179 id_size = snprintf(id, id_len, "t10.%*pE",
3180 cur_id_size, cur_id_str);
3187 switch (cur_id_size) {
3189 id_size = snprintf(id, id_len,
3194 id_size = snprintf(id, id_len,
3199 id_size = snprintf(id, id_len,
3212 switch (cur_id_size) {
3214 id_size = snprintf(id, id_len,
3219 id_size = snprintf(id, id_len,
3228 /* SCSI name string */
3229 if (cur_id_size > d[3])
3231 /* Prefer others for truncated descriptor */
3232 if (d[3] > id_len) {
3234 if (cur_id_prio > prio)
3238 cur_id_size = id_size = d[3];
3240 if (cur_id_size >= id_len)
3241 cur_id_size = id_len - 1;
3242 memcpy(id, cur_id_str, cur_id_size);
3252 EXPORT_SYMBOL(scsi_vpd_lun_id);
3255 * scsi_vpd_tpg_id - return a target port group identifier
3256 * @sdev: SCSI device
3258 * Returns the Target Port Group identifier from the information
3259 * froom VPD page 0x83 of the device.
3261 * Returns the identifier or error on failure.
3263 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3265 const unsigned char *d;
3266 const struct scsi_vpd *vpd_pg83;
3267 int group_id = -EAGAIN, rel_port = -1;
3270 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3276 d = vpd_pg83->data + 4;
3277 while (d < vpd_pg83->data + vpd_pg83->len) {
3278 switch (d[1] & 0xf) {
3280 /* Relative target port */
3281 rel_port = get_unaligned_be16(&d[6]);
3284 /* Target port group */
3285 group_id = get_unaligned_be16(&d[6]);
3294 if (group_id >= 0 && rel_id && rel_port != -1)
3299 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3302 * scsi_build_sense - build sense data for a command
3303 * @scmd: scsi command for which the sense should be formatted
3304 * @desc: Sense format (non-zero == descriptor format,
3305 * 0 == fixed format)
3307 * @asc: Additional sense code
3308 * @ascq: Additional sense code qualifier
3311 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3313 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3314 scmd->result = SAM_STAT_CHECK_CONDITION;
3316 EXPORT_SYMBOL_GPL(scsi_build_sense);