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_sense_cache;
56 static DEFINE_MUTEX(scsi_sense_cache_mutex);
58 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60 int scsi_init_sense_cache(struct Scsi_Host *shost)
64 mutex_lock(&scsi_sense_cache_mutex);
65 if (!scsi_sense_cache) {
67 kmem_cache_create_usercopy("scsi_sense_cache",
68 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
69 0, SCSI_SENSE_BUFFERSIZE, NULL);
70 if (!scsi_sense_cache)
73 mutex_unlock(&scsi_sense_cache_mutex);
78 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
79 * not change behaviour from the previous unplug mechanism, experimentation
80 * may prove this needs changing.
82 #define SCSI_QUEUE_DELAY 3
85 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
87 struct Scsi_Host *host = cmd->device->host;
88 struct scsi_device *device = cmd->device;
89 struct scsi_target *starget = scsi_target(device);
92 * Set the appropriate busy bit for the device/host.
94 * If the host/device isn't busy, assume that something actually
95 * completed, and that we should be able to queue a command now.
97 * Note that the prior mid-layer assumption that any host could
98 * always queue at least one command is now broken. The mid-layer
99 * will implement a user specifiable stall (see
100 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
101 * if a command is requeued with no other commands outstanding
102 * either for the device or for the host.
105 case SCSI_MLQUEUE_HOST_BUSY:
106 atomic_set(&host->host_blocked, host->max_host_blocked);
108 case SCSI_MLQUEUE_DEVICE_BUSY:
109 case SCSI_MLQUEUE_EH_RETRY:
110 atomic_set(&device->device_blocked,
111 device->max_device_blocked);
113 case SCSI_MLQUEUE_TARGET_BUSY:
114 atomic_set(&starget->target_blocked,
115 starget->max_target_blocked);
120 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
122 if (cmd->request->rq_flags & RQF_DONTPREP) {
123 cmd->request->rq_flags &= ~RQF_DONTPREP;
124 scsi_mq_uninit_cmd(cmd);
128 blk_mq_requeue_request(cmd->request, true);
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
143 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
145 struct scsi_device *device = cmd->device;
147 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
148 "Inserting command %p into mlqueue\n", cmd));
150 scsi_set_blocked(cmd, reason);
153 * Decrement the counters, since these commands are no longer
154 * active on the host/device.
157 scsi_device_unbusy(device, cmd);
160 * Requeue this command. It will go before all other commands
161 * that are already in the queue. Schedule requeue work under
162 * lock such that the kblockd_schedule_work() call happens
163 * before blk_cleanup_queue() finishes.
167 blk_mq_requeue_request(cmd->request, true);
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
179 * Context: This could be called either from an interrupt context or a normal
182 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
184 __scsi_queue_insert(cmd, reason, true);
189 * __scsi_execute - insert request and wait for the result
192 * @data_direction: data direction
193 * @buffer: data buffer
194 * @bufflen: len of buffer
195 * @sense: optional sense buffer
196 * @sshdr: optional decoded sense header
197 * @timeout: request timeout in seconds
198 * @retries: number of times to retry request
199 * @flags: flags for ->cmd_flags
200 * @rq_flags: flags for ->rq_flags
201 * @resid: optional residual length
203 * Returns the scsi_cmnd result field if a command was executed, or a negative
204 * Linux error code if we didn't get that far.
206 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
207 int data_direction, void *buffer, unsigned bufflen,
208 unsigned char *sense, struct scsi_sense_hdr *sshdr,
209 int timeout, int retries, u64 flags, req_flags_t rq_flags,
213 struct scsi_request *rq;
214 int ret = DRIVER_ERROR << 24;
216 req = blk_get_request(sdev->request_queue,
217 data_direction == DMA_TO_DEVICE ?
218 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
219 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
224 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
225 buffer, bufflen, GFP_NOIO))
228 rq->cmd_len = COMMAND_SIZE(cmd[0]);
229 memcpy(rq->cmd, cmd, rq->cmd_len);
230 rq->retries = retries;
231 req->timeout = timeout;
232 req->cmd_flags |= flags;
233 req->rq_flags |= rq_flags | RQF_QUIET;
236 * head injection *required* here otherwise quiesce won't work
238 blk_execute_rq(NULL, req, 1);
241 * Some devices (USB mass-storage in particular) may transfer
242 * garbage data together with a residue indicating that the data
243 * is invalid. Prevent the garbage from being misinterpreted
244 * and prevent security leaks by zeroing out the excess data.
246 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
247 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
250 *resid = rq->resid_len;
251 if (sense && rq->sense_len)
252 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
254 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
257 blk_put_request(req);
261 EXPORT_SYMBOL(__scsi_execute);
264 * Wake up the error handler if necessary. Avoid as follows that the error
265 * handler is not woken up if host in-flight requests number ==
266 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
267 * with an RCU read lock in this function to ensure that this function in
268 * its entirety either finishes before scsi_eh_scmd_add() increases the
269 * host_failed counter or that it notices the shost state change made by
270 * scsi_eh_scmd_add().
272 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
277 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
278 if (unlikely(scsi_host_in_recovery(shost))) {
279 spin_lock_irqsave(shost->host_lock, flags);
280 if (shost->host_failed || shost->host_eh_scheduled)
281 scsi_eh_wakeup(shost);
282 spin_unlock_irqrestore(shost->host_lock, flags);
287 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
289 struct Scsi_Host *shost = sdev->host;
290 struct scsi_target *starget = scsi_target(sdev);
292 scsi_dec_host_busy(shost, cmd);
294 if (starget->can_queue > 0)
295 atomic_dec(&starget->target_busy);
297 sbitmap_put(&sdev->budget_map, cmd->budget_token);
298 cmd->budget_token = -1;
301 static void scsi_kick_queue(struct request_queue *q)
303 blk_mq_run_hw_queues(q, false);
307 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
308 * and call blk_run_queue for all the scsi_devices on the target -
309 * including current_sdev first.
311 * Called with *no* scsi locks held.
313 static void scsi_single_lun_run(struct scsi_device *current_sdev)
315 struct Scsi_Host *shost = current_sdev->host;
316 struct scsi_device *sdev, *tmp;
317 struct scsi_target *starget = scsi_target(current_sdev);
320 spin_lock_irqsave(shost->host_lock, flags);
321 starget->starget_sdev_user = NULL;
322 spin_unlock_irqrestore(shost->host_lock, flags);
325 * Call blk_run_queue for all LUNs on the target, starting with
326 * current_sdev. We race with others (to set starget_sdev_user),
327 * but in most cases, we will be first. Ideally, each LU on the
328 * target would get some limited time or requests on the target.
330 scsi_kick_queue(current_sdev->request_queue);
332 spin_lock_irqsave(shost->host_lock, flags);
333 if (starget->starget_sdev_user)
335 list_for_each_entry_safe(sdev, tmp, &starget->devices,
336 same_target_siblings) {
337 if (sdev == current_sdev)
339 if (scsi_device_get(sdev))
342 spin_unlock_irqrestore(shost->host_lock, flags);
343 scsi_kick_queue(sdev->request_queue);
344 spin_lock_irqsave(shost->host_lock, flags);
346 scsi_device_put(sdev);
349 spin_unlock_irqrestore(shost->host_lock, flags);
352 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
354 if (scsi_device_busy(sdev) >= sdev->queue_depth)
356 if (atomic_read(&sdev->device_blocked) > 0)
361 static inline bool scsi_target_is_busy(struct scsi_target *starget)
363 if (starget->can_queue > 0) {
364 if (atomic_read(&starget->target_busy) >= starget->can_queue)
366 if (atomic_read(&starget->target_blocked) > 0)
372 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
374 if (atomic_read(&shost->host_blocked) > 0)
376 if (shost->host_self_blocked)
381 static void scsi_starved_list_run(struct Scsi_Host *shost)
383 LIST_HEAD(starved_list);
384 struct scsi_device *sdev;
387 spin_lock_irqsave(shost->host_lock, flags);
388 list_splice_init(&shost->starved_list, &starved_list);
390 while (!list_empty(&starved_list)) {
391 struct request_queue *slq;
394 * As long as shost is accepting commands and we have
395 * starved queues, call blk_run_queue. scsi_request_fn
396 * drops the queue_lock and can add us back to the
399 * host_lock protects the starved_list and starved_entry.
400 * scsi_request_fn must get the host_lock before checking
401 * or modifying starved_list or starved_entry.
403 if (scsi_host_is_busy(shost))
406 sdev = list_entry(starved_list.next,
407 struct scsi_device, starved_entry);
408 list_del_init(&sdev->starved_entry);
409 if (scsi_target_is_busy(scsi_target(sdev))) {
410 list_move_tail(&sdev->starved_entry,
411 &shost->starved_list);
416 * Once we drop the host lock, a racing scsi_remove_device()
417 * call may remove the sdev from the starved list and destroy
418 * it and the queue. Mitigate by taking a reference to the
419 * queue and never touching the sdev again after we drop the
420 * host lock. Note: if __scsi_remove_device() invokes
421 * blk_cleanup_queue() before the queue is run from this
422 * function then blk_run_queue() will return immediately since
423 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
425 slq = sdev->request_queue;
426 if (!blk_get_queue(slq))
428 spin_unlock_irqrestore(shost->host_lock, flags);
430 scsi_kick_queue(slq);
433 spin_lock_irqsave(shost->host_lock, flags);
435 /* put any unprocessed entries back */
436 list_splice(&starved_list, &shost->starved_list);
437 spin_unlock_irqrestore(shost->host_lock, flags);
441 * scsi_run_queue - Select a proper request queue to serve next.
442 * @q: last request's queue
444 * The previous command was completely finished, start a new one if possible.
446 static void scsi_run_queue(struct request_queue *q)
448 struct scsi_device *sdev = q->queuedata;
450 if (scsi_target(sdev)->single_lun)
451 scsi_single_lun_run(sdev);
452 if (!list_empty(&sdev->host->starved_list))
453 scsi_starved_list_run(sdev->host);
455 blk_mq_run_hw_queues(q, false);
458 void scsi_requeue_run_queue(struct work_struct *work)
460 struct scsi_device *sdev;
461 struct request_queue *q;
463 sdev = container_of(work, struct scsi_device, requeue_work);
464 q = sdev->request_queue;
468 void scsi_run_host_queues(struct Scsi_Host *shost)
470 struct scsi_device *sdev;
472 shost_for_each_device(sdev, shost)
473 scsi_run_queue(sdev->request_queue);
476 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
478 if (!blk_rq_is_passthrough(cmd->request)) {
479 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
481 if (drv->uninit_command)
482 drv->uninit_command(cmd);
486 void scsi_free_sgtables(struct scsi_cmnd *cmd)
488 if (cmd->sdb.table.nents)
489 sg_free_table_chained(&cmd->sdb.table,
491 if (scsi_prot_sg_count(cmd))
492 sg_free_table_chained(&cmd->prot_sdb->table,
493 SCSI_INLINE_PROT_SG_CNT);
495 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
497 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
499 scsi_free_sgtables(cmd);
500 scsi_uninit_cmd(cmd);
503 static void scsi_run_queue_async(struct scsi_device *sdev)
505 if (scsi_target(sdev)->single_lun ||
506 !list_empty(&sdev->host->starved_list)) {
507 kblockd_schedule_work(&sdev->requeue_work);
510 * smp_mb() present in sbitmap_queue_clear() or implied in
511 * .end_io is for ordering writing .device_busy in
512 * scsi_device_unbusy() and reading sdev->restarts.
514 int old = atomic_read(&sdev->restarts);
517 * ->restarts has to be kept as non-zero if new budget
520 * No need to run queue when either another re-run
521 * queue wins in updating ->restarts or a new budget
524 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
525 blk_mq_run_hw_queues(sdev->request_queue, true);
529 /* Returns false when no more bytes to process, true if there are more */
530 static bool scsi_end_request(struct request *req, blk_status_t error,
533 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
534 struct scsi_device *sdev = cmd->device;
535 struct request_queue *q = sdev->request_queue;
537 if (blk_update_request(req, error, bytes))
540 if (blk_queue_add_random(q))
541 add_disk_randomness(req->rq_disk);
543 if (!blk_rq_is_passthrough(req)) {
544 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
545 cmd->flags &= ~SCMD_INITIALIZED;
549 * Calling rcu_barrier() is not necessary here because the
550 * SCSI error handler guarantees that the function called by
551 * call_rcu() has been called before scsi_end_request() is
554 destroy_rcu_head(&cmd->rcu);
557 * In the MQ case the command gets freed by __blk_mq_end_request,
558 * so we have to do all cleanup that depends on it earlier.
560 * We also can't kick the queues from irq context, so we
561 * will have to defer it to a workqueue.
563 scsi_mq_uninit_cmd(cmd);
566 * queue is still alive, so grab the ref for preventing it
567 * from being cleaned up during running queue.
569 percpu_ref_get(&q->q_usage_counter);
571 __blk_mq_end_request(req, error);
573 scsi_run_queue_async(sdev);
575 percpu_ref_put(&q->q_usage_counter);
580 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
582 * @result: scsi error code
584 * Translate a SCSI result code into a blk_status_t value. May reset the host
585 * byte of @cmd->result.
587 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
589 switch (host_byte(result)) {
592 * Also check the other bytes than the status byte in result
593 * to handle the case when a SCSI LLD sets result to
594 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
596 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
598 return BLK_STS_IOERR;
599 case DID_TRANSPORT_FAILFAST:
600 case DID_TRANSPORT_MARGINAL:
601 return BLK_STS_TRANSPORT;
602 case DID_TARGET_FAILURE:
603 set_host_byte(cmd, DID_OK);
604 return BLK_STS_TARGET;
605 case DID_NEXUS_FAILURE:
606 set_host_byte(cmd, DID_OK);
607 return BLK_STS_NEXUS;
608 case DID_ALLOC_FAILURE:
609 set_host_byte(cmd, DID_OK);
610 return BLK_STS_NOSPC;
611 case DID_MEDIUM_ERROR:
612 set_host_byte(cmd, DID_OK);
613 return BLK_STS_MEDIUM;
615 return BLK_STS_IOERR;
619 /* Helper for scsi_io_completion() when "reprep" action required. */
620 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
621 struct request_queue *q)
623 /* A new command will be prepared and issued. */
624 scsi_mq_requeue_cmd(cmd);
627 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
629 struct request *req = cmd->request;
630 unsigned long wait_for;
632 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
635 wait_for = (cmd->allowed + 1) * req->timeout;
636 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
637 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
644 /* Helper for scsi_io_completion() when special action required. */
645 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
647 struct request_queue *q = cmd->device->request_queue;
648 struct request *req = cmd->request;
650 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
651 ACTION_DELAYED_RETRY} action;
652 struct scsi_sense_hdr sshdr;
654 bool sense_current = true; /* false implies "deferred sense" */
655 blk_status_t blk_stat;
657 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
659 sense_current = !scsi_sense_is_deferred(&sshdr);
661 blk_stat = scsi_result_to_blk_status(cmd, result);
663 if (host_byte(result) == DID_RESET) {
664 /* Third party bus reset or reset for error recovery
665 * reasons. Just retry the command and see what
668 action = ACTION_RETRY;
669 } else if (sense_valid && sense_current) {
670 switch (sshdr.sense_key) {
672 if (cmd->device->removable) {
673 /* Detected disc change. Set a bit
674 * and quietly refuse further access.
676 cmd->device->changed = 1;
677 action = ACTION_FAIL;
679 /* Must have been a power glitch, or a
680 * bus reset. Could not have been a
681 * media change, so we just retry the
682 * command and see what happens.
684 action = ACTION_RETRY;
687 case ILLEGAL_REQUEST:
688 /* If we had an ILLEGAL REQUEST returned, then
689 * we may have performed an unsupported
690 * command. The only thing this should be
691 * would be a ten byte read where only a six
692 * byte read was supported. Also, on a system
693 * where READ CAPACITY failed, we may have
694 * read past the end of the disk.
696 if ((cmd->device->use_10_for_rw &&
697 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
698 (cmd->cmnd[0] == READ_10 ||
699 cmd->cmnd[0] == WRITE_10)) {
700 /* This will issue a new 6-byte command. */
701 cmd->device->use_10_for_rw = 0;
702 action = ACTION_REPREP;
703 } else if (sshdr.asc == 0x10) /* DIX */ {
704 action = ACTION_FAIL;
705 blk_stat = BLK_STS_PROTECTION;
706 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
707 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
708 action = ACTION_FAIL;
709 blk_stat = BLK_STS_TARGET;
711 action = ACTION_FAIL;
713 case ABORTED_COMMAND:
714 action = ACTION_FAIL;
715 if (sshdr.asc == 0x10) /* DIF */
716 blk_stat = BLK_STS_PROTECTION;
719 /* If the device is in the process of becoming
720 * ready, or has a temporary blockage, retry.
722 if (sshdr.asc == 0x04) {
723 switch (sshdr.ascq) {
724 case 0x01: /* becoming ready */
725 case 0x04: /* format in progress */
726 case 0x05: /* rebuild in progress */
727 case 0x06: /* recalculation in progress */
728 case 0x07: /* operation in progress */
729 case 0x08: /* Long write in progress */
730 case 0x09: /* self test in progress */
731 case 0x14: /* space allocation in progress */
732 case 0x1a: /* start stop unit in progress */
733 case 0x1b: /* sanitize in progress */
734 case 0x1d: /* configuration in progress */
735 case 0x24: /* depopulation in progress */
736 action = ACTION_DELAYED_RETRY;
738 case 0x0a: /* ALUA state transition */
739 blk_stat = BLK_STS_AGAIN;
742 action = ACTION_FAIL;
746 action = ACTION_FAIL;
748 case VOLUME_OVERFLOW:
749 /* See SSC3rXX or current. */
750 action = ACTION_FAIL;
753 action = ACTION_FAIL;
754 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
755 (sshdr.asc == 0x55 &&
756 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
757 /* Insufficient zone resources */
758 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
762 action = ACTION_FAIL;
766 action = ACTION_FAIL;
768 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
769 action = ACTION_FAIL;
773 /* Give up and fail the remainder of the request */
774 if (!(req->rq_flags & RQF_QUIET)) {
775 static DEFINE_RATELIMIT_STATE(_rs,
776 DEFAULT_RATELIMIT_INTERVAL,
777 DEFAULT_RATELIMIT_BURST);
779 if (unlikely(scsi_logging_level))
781 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
782 SCSI_LOG_MLCOMPLETE_BITS);
785 * if logging is enabled the failure will be printed
786 * in scsi_log_completion(), so avoid duplicate messages
788 if (!level && __ratelimit(&_rs)) {
789 scsi_print_result(cmd, NULL, FAILED);
790 if (driver_byte(result) == DRIVER_SENSE)
791 scsi_print_sense(cmd);
792 scsi_print_command(cmd);
795 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
799 scsi_io_completion_reprep(cmd, q);
802 /* Retry the same command immediately */
803 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
805 case ACTION_DELAYED_RETRY:
806 /* Retry the same command after a delay */
807 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
813 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
814 * new result that may suppress further error checking. Also modifies
815 * *blk_statp in some cases.
817 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
818 blk_status_t *blk_statp)
821 bool sense_current = true; /* false implies "deferred sense" */
822 struct request *req = cmd->request;
823 struct scsi_sense_hdr sshdr;
825 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
827 sense_current = !scsi_sense_is_deferred(&sshdr);
829 if (blk_rq_is_passthrough(req)) {
832 * SG_IO wants current and deferred errors
834 scsi_req(req)->sense_len =
835 min(8 + cmd->sense_buffer[7],
836 SCSI_SENSE_BUFFERSIZE);
839 *blk_statp = scsi_result_to_blk_status(cmd, result);
840 } else if (blk_rq_bytes(req) == 0 && sense_current) {
842 * Flush commands do not transfers any data, and thus cannot use
843 * good_bytes != blk_rq_bytes(req) as the signal for an error.
844 * This sets *blk_statp explicitly for the problem case.
846 *blk_statp = scsi_result_to_blk_status(cmd, result);
849 * Recovered errors need reporting, but they're always treated as
850 * success, so fiddle the result code here. For passthrough requests
851 * we already took a copy of the original into sreq->result which
852 * is what gets returned to the user
854 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
855 bool do_print = true;
857 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
858 * skip print since caller wants ATA registers. Only occurs
859 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
861 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
863 else if (req->rq_flags & RQF_QUIET)
866 scsi_print_sense(cmd);
868 /* for passthrough, *blk_statp may be set */
869 *blk_statp = BLK_STS_OK;
872 * Another corner case: the SCSI status byte is non-zero but 'good'.
873 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
874 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
875 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
876 * intermediate statuses (both obsolete in SAM-4) as good.
878 if (status_byte(result) && scsi_status_is_good(result)) {
880 *blk_statp = BLK_STS_OK;
886 * scsi_io_completion - Completion processing for SCSI commands.
887 * @cmd: command that is finished.
888 * @good_bytes: number of processed bytes.
890 * We will finish off the specified number of sectors. If we are done, the
891 * command block will be released and the queue function will be goosed. If we
892 * are not done then we have to figure out what to do next:
894 * a) We can call scsi_io_completion_reprep(). The request will be
895 * unprepared and put back on the queue. Then a new command will
896 * be created for it. This should be used if we made forward
897 * progress, or if we want to switch from READ(10) to READ(6) for
900 * b) We can call scsi_io_completion_action(). The request will be
901 * put back on the queue and retried using the same command as
902 * before, possibly after a delay.
904 * c) We can call scsi_end_request() with blk_stat other than
905 * BLK_STS_OK, to fail the remainder of the request.
907 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
909 int result = cmd->result;
910 struct request_queue *q = cmd->device->request_queue;
911 struct request *req = cmd->request;
912 blk_status_t blk_stat = BLK_STS_OK;
914 if (unlikely(result)) /* a nz result may or may not be an error */
915 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
917 if (unlikely(blk_rq_is_passthrough(req))) {
919 * scsi_result_to_blk_status may have reset the host_byte
921 scsi_req(req)->result = cmd->result;
925 * Next deal with any sectors which we were able to correctly
928 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
929 "%u sectors total, %d bytes done.\n",
930 blk_rq_sectors(req), good_bytes));
933 * Failed, zero length commands always need to drop down
934 * to retry code. Fast path should return in this block.
936 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
937 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
938 return; /* no bytes remaining */
941 /* Kill remainder if no retries. */
942 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
943 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
945 "Bytes remaining after failed, no-retry command");
950 * If there had been no error, but we have leftover bytes in the
951 * requeues just queue the command up again.
953 if (likely(result == 0))
954 scsi_io_completion_reprep(cmd, q);
956 scsi_io_completion_action(cmd, result);
959 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
962 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
963 !op_is_write(req_op(rq)) &&
964 sdev->host->hostt->dma_need_drain(rq);
968 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
969 * @cmd: SCSI command data structure to initialize.
971 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
975 * * BLK_STS_OK - on success
976 * * BLK_STS_RESOURCE - if the failure is retryable
977 * * BLK_STS_IOERR - if the failure is fatal
979 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
981 struct scsi_device *sdev = cmd->device;
982 struct request *rq = cmd->request;
983 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
984 struct scatterlist *last_sg = NULL;
986 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
989 if (WARN_ON_ONCE(!nr_segs))
990 return BLK_STS_IOERR;
993 * Make sure there is space for the drain. The driver must adjust
994 * max_hw_segments to be prepared for this.
1000 * If sg table allocation fails, requeue request later.
1002 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1003 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1004 return BLK_STS_RESOURCE;
1007 * Next, walk the list, and fill in the addresses and sizes of
1010 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1012 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1013 unsigned int pad_len =
1014 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1016 last_sg->length += pad_len;
1017 cmd->extra_len += pad_len;
1021 sg_unmark_end(last_sg);
1022 last_sg = sg_next(last_sg);
1023 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1024 sg_mark_end(last_sg);
1026 cmd->extra_len += sdev->dma_drain_len;
1030 BUG_ON(count > cmd->sdb.table.nents);
1031 cmd->sdb.table.nents = count;
1032 cmd->sdb.length = blk_rq_payload_bytes(rq);
1034 if (blk_integrity_rq(rq)) {
1035 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1038 if (WARN_ON_ONCE(!prot_sdb)) {
1040 * This can happen if someone (e.g. multipath)
1041 * queues a command to a device on an adapter
1042 * that does not support DIX.
1044 ret = BLK_STS_IOERR;
1045 goto out_free_sgtables;
1048 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1050 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1051 prot_sdb->table.sgl,
1052 SCSI_INLINE_PROT_SG_CNT)) {
1053 ret = BLK_STS_RESOURCE;
1054 goto out_free_sgtables;
1057 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1058 prot_sdb->table.sgl);
1059 BUG_ON(count > ivecs);
1060 BUG_ON(count > queue_max_integrity_segments(rq->q));
1062 cmd->prot_sdb = prot_sdb;
1063 cmd->prot_sdb->table.nents = count;
1068 scsi_free_sgtables(cmd);
1071 EXPORT_SYMBOL(scsi_alloc_sgtables);
1074 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1075 * @rq: Request associated with the SCSI command to be initialized.
1077 * This function initializes the members of struct scsi_cmnd that must be
1078 * initialized before request processing starts and that won't be
1079 * reinitialized if a SCSI command is requeued.
1081 * Called from inside blk_get_request() for pass-through requests and from
1082 * inside scsi_init_command() for filesystem requests.
1084 static void scsi_initialize_rq(struct request *rq)
1086 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1088 scsi_req_init(&cmd->req);
1089 init_rcu_head(&cmd->rcu);
1090 cmd->jiffies_at_alloc = jiffies;
1095 * Only called when the request isn't completed by SCSI, and not freed by
1098 static void scsi_cleanup_rq(struct request *rq)
1100 if (rq->rq_flags & RQF_DONTPREP) {
1101 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1102 rq->rq_flags &= ~RQF_DONTPREP;
1106 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1107 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1109 void *buf = cmd->sense_buffer;
1110 void *prot = cmd->prot_sdb;
1111 struct request *rq = blk_mq_rq_from_pdu(cmd);
1112 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1113 unsigned long jiffies_at_alloc;
1114 int retries, to_clear;
1116 int budget_token = cmd->budget_token;
1118 if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
1119 flags |= SCMD_INITIALIZED;
1120 scsi_initialize_rq(rq);
1123 jiffies_at_alloc = cmd->jiffies_at_alloc;
1124 retries = cmd->retries;
1125 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1127 * Zero out the cmd, except for the embedded scsi_request. Only clear
1128 * the driver-private command data if the LLD does not supply a
1129 * function to initialize that data.
1131 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1132 if (!dev->host->hostt->init_cmd_priv)
1133 to_clear += dev->host->hostt->cmd_size;
1134 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1137 cmd->sense_buffer = buf;
1138 cmd->prot_sdb = prot;
1140 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1141 cmd->jiffies_at_alloc = jiffies_at_alloc;
1142 cmd->retries = retries;
1144 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1145 cmd->budget_token = budget_token;
1149 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1150 struct request *req)
1152 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1155 * Passthrough requests may transfer data, in which case they must
1156 * a bio attached to them. Or they might contain a SCSI command
1157 * that does not transfer data, in which case they may optionally
1158 * submit a request without an attached bio.
1161 blk_status_t ret = scsi_alloc_sgtables(cmd);
1162 if (unlikely(ret != BLK_STS_OK))
1165 BUG_ON(blk_rq_bytes(req));
1167 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1170 cmd->cmd_len = scsi_req(req)->cmd_len;
1171 if (cmd->cmd_len == 0)
1172 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1173 cmd->cmnd = scsi_req(req)->cmd;
1174 cmd->transfersize = blk_rq_bytes(req);
1175 cmd->allowed = scsi_req(req)->retries;
1180 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1182 switch (sdev->sdev_state) {
1186 case SDEV_TRANSPORT_OFFLINE:
1188 * If the device is offline we refuse to process any
1189 * commands. The device must be brought online
1190 * before trying any recovery commands.
1192 if (!sdev->offline_already) {
1193 sdev->offline_already = true;
1194 sdev_printk(KERN_ERR, sdev,
1195 "rejecting I/O to offline device\n");
1197 return BLK_STS_IOERR;
1200 * If the device is fully deleted, we refuse to
1201 * process any commands as well.
1203 sdev_printk(KERN_ERR, sdev,
1204 "rejecting I/O to dead device\n");
1205 return BLK_STS_IOERR;
1207 case SDEV_CREATED_BLOCK:
1208 return BLK_STS_RESOURCE;
1211 * If the device is blocked we only accept power management
1214 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1215 return BLK_STS_RESOURCE;
1219 * For any other not fully online state we only allow
1220 * power management commands.
1222 if (req && !(req->rq_flags & RQF_PM))
1223 return BLK_STS_IOERR;
1229 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1230 * and return the token else return -1.
1232 static inline int scsi_dev_queue_ready(struct request_queue *q,
1233 struct scsi_device *sdev)
1237 token = sbitmap_get(&sdev->budget_map);
1238 if (atomic_read(&sdev->device_blocked)) {
1242 if (scsi_device_busy(sdev) > 1)
1246 * unblock after device_blocked iterates to zero
1248 if (atomic_dec_return(&sdev->device_blocked) > 0)
1250 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1251 "unblocking device at zero depth\n"));
1257 sbitmap_put(&sdev->budget_map, token);
1263 * scsi_target_queue_ready: checks if there we can send commands to target
1264 * @sdev: scsi device on starget to check.
1266 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1267 struct scsi_device *sdev)
1269 struct scsi_target *starget = scsi_target(sdev);
1272 if (starget->single_lun) {
1273 spin_lock_irq(shost->host_lock);
1274 if (starget->starget_sdev_user &&
1275 starget->starget_sdev_user != sdev) {
1276 spin_unlock_irq(shost->host_lock);
1279 starget->starget_sdev_user = sdev;
1280 spin_unlock_irq(shost->host_lock);
1283 if (starget->can_queue <= 0)
1286 busy = atomic_inc_return(&starget->target_busy) - 1;
1287 if (atomic_read(&starget->target_blocked) > 0) {
1292 * unblock after target_blocked iterates to zero
1294 if (atomic_dec_return(&starget->target_blocked) > 0)
1297 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1298 "unblocking target at zero depth\n"));
1301 if (busy >= starget->can_queue)
1307 spin_lock_irq(shost->host_lock);
1308 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1309 spin_unlock_irq(shost->host_lock);
1311 if (starget->can_queue > 0)
1312 atomic_dec(&starget->target_busy);
1317 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1318 * return 0. We must end up running the queue again whenever 0 is
1319 * returned, else IO can hang.
1321 static inline int scsi_host_queue_ready(struct request_queue *q,
1322 struct Scsi_Host *shost,
1323 struct scsi_device *sdev,
1324 struct scsi_cmnd *cmd)
1326 if (scsi_host_in_recovery(shost))
1329 if (atomic_read(&shost->host_blocked) > 0) {
1330 if (scsi_host_busy(shost) > 0)
1334 * unblock after host_blocked iterates to zero
1336 if (atomic_dec_return(&shost->host_blocked) > 0)
1340 shost_printk(KERN_INFO, shost,
1341 "unblocking host at zero depth\n"));
1344 if (shost->host_self_blocked)
1347 /* We're OK to process the command, so we can't be starved */
1348 if (!list_empty(&sdev->starved_entry)) {
1349 spin_lock_irq(shost->host_lock);
1350 if (!list_empty(&sdev->starved_entry))
1351 list_del_init(&sdev->starved_entry);
1352 spin_unlock_irq(shost->host_lock);
1355 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1360 spin_lock_irq(shost->host_lock);
1361 if (list_empty(&sdev->starved_entry))
1362 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1363 spin_unlock_irq(shost->host_lock);
1365 scsi_dec_host_busy(shost, cmd);
1370 * Busy state exporting function for request stacking drivers.
1372 * For efficiency, no lock is taken to check the busy state of
1373 * shost/starget/sdev, since the returned value is not guaranteed and
1374 * may be changed after request stacking drivers call the function,
1375 * regardless of taking lock or not.
1377 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1378 * needs to return 'not busy'. Otherwise, request stacking drivers
1379 * may hold requests forever.
1381 static bool scsi_mq_lld_busy(struct request_queue *q)
1383 struct scsi_device *sdev = q->queuedata;
1384 struct Scsi_Host *shost;
1386 if (blk_queue_dying(q))
1392 * Ignore host/starget busy state.
1393 * Since block layer does not have a concept of fairness across
1394 * multiple queues, congestion of host/starget needs to be handled
1397 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1404 * Block layer request completion callback. May be called from interrupt
1407 static void scsi_complete(struct request *rq)
1409 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1410 enum scsi_disposition disposition;
1412 INIT_LIST_HEAD(&cmd->eh_entry);
1414 atomic_inc(&cmd->device->iodone_cnt);
1416 atomic_inc(&cmd->device->ioerr_cnt);
1418 disposition = scsi_decide_disposition(cmd);
1419 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1420 disposition = SUCCESS;
1422 scsi_log_completion(cmd, disposition);
1424 switch (disposition) {
1426 scsi_finish_command(cmd);
1429 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1431 case ADD_TO_MLQUEUE:
1432 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1435 scsi_eh_scmd_add(cmd);
1441 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1442 * @cmd: command block we are dispatching.
1444 * Return: nonzero return request was rejected and device's queue needs to be
1447 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1449 struct Scsi_Host *host = cmd->device->host;
1452 atomic_inc(&cmd->device->iorequest_cnt);
1454 /* check if the device is still usable */
1455 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1456 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1457 * returns an immediate error upwards, and signals
1458 * that the device is no longer present */
1459 cmd->result = DID_NO_CONNECT << 16;
1463 /* Check to see if the scsi lld made this device blocked. */
1464 if (unlikely(scsi_device_blocked(cmd->device))) {
1466 * in blocked state, the command is just put back on
1467 * the device queue. The suspend state has already
1468 * blocked the queue so future requests should not
1469 * occur until the device transitions out of the
1472 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1473 "queuecommand : device blocked\n"));
1474 return SCSI_MLQUEUE_DEVICE_BUSY;
1477 /* Store the LUN value in cmnd, if needed. */
1478 if (cmd->device->lun_in_cdb)
1479 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1480 (cmd->device->lun << 5 & 0xe0);
1485 * Before we queue this command, check if the command
1486 * length exceeds what the host adapter can handle.
1488 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1489 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1490 "queuecommand : command too long. "
1491 "cdb_size=%d host->max_cmd_len=%d\n",
1492 cmd->cmd_len, cmd->device->host->max_cmd_len));
1493 cmd->result = (DID_ABORT << 16);
1497 if (unlikely(host->shost_state == SHOST_DEL)) {
1498 cmd->result = (DID_NO_CONNECT << 16);
1503 trace_scsi_dispatch_cmd_start(cmd);
1504 rtn = host->hostt->queuecommand(host, cmd);
1506 trace_scsi_dispatch_cmd_error(cmd, rtn);
1507 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1508 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1509 rtn = SCSI_MLQUEUE_HOST_BUSY;
1511 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1512 "queuecommand : request rejected\n"));
1517 cmd->scsi_done(cmd);
1521 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1522 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1524 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1525 sizeof(struct scatterlist);
1528 static blk_status_t scsi_prepare_cmd(struct request *req)
1530 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1531 struct scsi_device *sdev = req->q->queuedata;
1532 struct Scsi_Host *shost = sdev->host;
1533 struct scatterlist *sg;
1535 scsi_init_command(sdev, cmd);
1538 cmd->tag = req->tag;
1539 cmd->prot_op = SCSI_PROT_NORMAL;
1540 if (blk_rq_bytes(req))
1541 cmd->sc_data_direction = rq_dma_dir(req);
1543 cmd->sc_data_direction = DMA_NONE;
1545 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1546 cmd->sdb.table.sgl = sg;
1548 if (scsi_host_get_prot(shost)) {
1549 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1551 cmd->prot_sdb->table.sgl =
1552 (struct scatterlist *)(cmd->prot_sdb + 1);
1556 * Special handling for passthrough commands, which don't go to the ULP
1559 if (blk_rq_is_passthrough(req))
1560 return scsi_setup_scsi_cmnd(sdev, req);
1562 if (sdev->handler && sdev->handler->prep_fn) {
1563 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1565 if (ret != BLK_STS_OK)
1569 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1570 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1571 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1574 static void scsi_mq_done(struct scsi_cmnd *cmd)
1576 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1578 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1580 trace_scsi_dispatch_cmd_done(cmd);
1581 blk_mq_complete_request(cmd->request);
1584 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1586 struct scsi_device *sdev = q->queuedata;
1588 sbitmap_put(&sdev->budget_map, budget_token);
1591 static int scsi_mq_get_budget(struct request_queue *q)
1593 struct scsi_device *sdev = q->queuedata;
1594 int token = scsi_dev_queue_ready(q, sdev);
1599 atomic_inc(&sdev->restarts);
1602 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1603 * .restarts must be incremented before .device_busy is read because the
1604 * code in scsi_run_queue_async() depends on the order of these operations.
1606 smp_mb__after_atomic();
1609 * If all in-flight requests originated from this LUN are completed
1610 * before reading .device_busy, sdev->device_busy will be observed as
1611 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1612 * soon. Otherwise, completion of one of these requests will observe
1613 * the .restarts flag, and the request queue will be run for handling
1614 * this request, see scsi_end_request().
1616 if (unlikely(scsi_device_busy(sdev) == 0 &&
1617 !scsi_device_blocked(sdev)))
1618 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1622 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1624 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1626 cmd->budget_token = token;
1629 static int scsi_mq_get_rq_budget_token(struct request *req)
1631 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1633 return cmd->budget_token;
1636 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1637 const struct blk_mq_queue_data *bd)
1639 struct request *req = bd->rq;
1640 struct request_queue *q = req->q;
1641 struct scsi_device *sdev = q->queuedata;
1642 struct Scsi_Host *shost = sdev->host;
1643 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1647 WARN_ON_ONCE(cmd->budget_token < 0);
1650 * If the device is not in running state we will reject some or all
1653 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1654 ret = scsi_device_state_check(sdev, req);
1655 if (ret != BLK_STS_OK)
1656 goto out_put_budget;
1659 ret = BLK_STS_RESOURCE;
1660 if (!scsi_target_queue_ready(shost, sdev))
1661 goto out_put_budget;
1662 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1663 goto out_dec_target_busy;
1665 if (!(req->rq_flags & RQF_DONTPREP)) {
1666 ret = scsi_prepare_cmd(req);
1667 if (ret != BLK_STS_OK)
1668 goto out_dec_host_busy;
1669 req->rq_flags |= RQF_DONTPREP;
1671 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1674 cmd->flags &= SCMD_PRESERVED_FLAGS;
1675 if (sdev->simple_tags)
1676 cmd->flags |= SCMD_TAGGED;
1678 cmd->flags |= SCMD_LAST;
1680 scsi_set_resid(cmd, 0);
1681 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1682 cmd->scsi_done = scsi_mq_done;
1684 blk_mq_start_request(req);
1685 reason = scsi_dispatch_cmd(cmd);
1687 scsi_set_blocked(cmd, reason);
1688 ret = BLK_STS_RESOURCE;
1689 goto out_dec_host_busy;
1695 scsi_dec_host_busy(shost, cmd);
1696 out_dec_target_busy:
1697 if (scsi_target(sdev)->can_queue > 0)
1698 atomic_dec(&scsi_target(sdev)->target_busy);
1700 scsi_mq_put_budget(q, cmd->budget_token);
1701 cmd->budget_token = -1;
1705 case BLK_STS_RESOURCE:
1706 case BLK_STS_ZONE_RESOURCE:
1707 if (scsi_device_blocked(sdev))
1708 ret = BLK_STS_DEV_RESOURCE;
1711 scsi_req(req)->result = DID_BUS_BUSY << 16;
1712 if (req->rq_flags & RQF_DONTPREP)
1713 scsi_mq_uninit_cmd(cmd);
1716 if (unlikely(!scsi_device_online(sdev)))
1717 scsi_req(req)->result = DID_NO_CONNECT << 16;
1719 scsi_req(req)->result = DID_ERROR << 16;
1721 * Make sure to release all allocated resources when
1722 * we hit an error, as we will never see this command
1725 if (req->rq_flags & RQF_DONTPREP)
1726 scsi_mq_uninit_cmd(cmd);
1727 scsi_run_queue_async(sdev);
1733 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1737 return BLK_EH_RESET_TIMER;
1738 return scsi_times_out(req);
1741 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1742 unsigned int hctx_idx, unsigned int numa_node)
1744 struct Scsi_Host *shost = set->driver_data;
1745 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1746 struct scatterlist *sg;
1750 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1751 if (!cmd->sense_buffer)
1753 cmd->req.sense = cmd->sense_buffer;
1755 if (scsi_host_get_prot(shost)) {
1756 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1757 shost->hostt->cmd_size;
1758 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1761 if (shost->hostt->init_cmd_priv) {
1762 ret = shost->hostt->init_cmd_priv(shost, cmd);
1764 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1770 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1771 unsigned int hctx_idx)
1773 struct Scsi_Host *shost = set->driver_data;
1774 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1776 if (shost->hostt->exit_cmd_priv)
1777 shost->hostt->exit_cmd_priv(shost, cmd);
1778 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1782 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
1784 struct Scsi_Host *shost = hctx->driver_data;
1786 if (shost->hostt->mq_poll)
1787 return shost->hostt->mq_poll(shost, hctx->queue_num);
1792 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1793 unsigned int hctx_idx)
1795 struct Scsi_Host *shost = data;
1797 hctx->driver_data = shost;
1801 static int scsi_map_queues(struct blk_mq_tag_set *set)
1803 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1805 if (shost->hostt->map_queues)
1806 return shost->hostt->map_queues(shost);
1807 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1810 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1812 struct device *dev = shost->dma_dev;
1815 * this limit is imposed by hardware restrictions
1817 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1820 if (scsi_host_prot_dma(shost)) {
1821 shost->sg_prot_tablesize =
1822 min_not_zero(shost->sg_prot_tablesize,
1823 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1824 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1825 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1828 if (dev->dma_mask) {
1829 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1830 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1832 blk_queue_max_hw_sectors(q, shost->max_sectors);
1833 blk_queue_segment_boundary(q, shost->dma_boundary);
1834 dma_set_seg_boundary(dev, shost->dma_boundary);
1836 blk_queue_max_segment_size(q, shost->max_segment_size);
1837 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1838 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1841 * Set a reasonable default alignment: The larger of 32-byte (dword),
1842 * which is a common minimum for HBAs, and the minimum DMA alignment,
1843 * which is set by the platform.
1845 * Devices that require a bigger alignment can increase it later.
1847 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1849 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1851 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1852 .get_budget = scsi_mq_get_budget,
1853 .put_budget = scsi_mq_put_budget,
1854 .queue_rq = scsi_queue_rq,
1855 .complete = scsi_complete,
1856 .timeout = scsi_timeout,
1857 #ifdef CONFIG_BLK_DEBUG_FS
1858 .show_rq = scsi_show_rq,
1860 .init_request = scsi_mq_init_request,
1861 .exit_request = scsi_mq_exit_request,
1862 .initialize_rq_fn = scsi_initialize_rq,
1863 .cleanup_rq = scsi_cleanup_rq,
1864 .busy = scsi_mq_lld_busy,
1865 .map_queues = scsi_map_queues,
1866 .init_hctx = scsi_init_hctx,
1867 .poll = scsi_mq_poll,
1868 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1869 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1873 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1875 struct Scsi_Host *shost = hctx->driver_data;
1877 shost->hostt->commit_rqs(shost, hctx->queue_num);
1880 static const struct blk_mq_ops scsi_mq_ops = {
1881 .get_budget = scsi_mq_get_budget,
1882 .put_budget = scsi_mq_put_budget,
1883 .queue_rq = scsi_queue_rq,
1884 .commit_rqs = scsi_commit_rqs,
1885 .complete = scsi_complete,
1886 .timeout = scsi_timeout,
1887 #ifdef CONFIG_BLK_DEBUG_FS
1888 .show_rq = scsi_show_rq,
1890 .init_request = scsi_mq_init_request,
1891 .exit_request = scsi_mq_exit_request,
1892 .initialize_rq_fn = scsi_initialize_rq,
1893 .cleanup_rq = scsi_cleanup_rq,
1894 .busy = scsi_mq_lld_busy,
1895 .map_queues = scsi_map_queues,
1896 .init_hctx = scsi_init_hctx,
1897 .poll = scsi_mq_poll,
1898 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1899 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1902 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1904 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1905 if (IS_ERR(sdev->request_queue))
1908 sdev->request_queue->queuedata = sdev;
1909 __scsi_init_queue(sdev->host, sdev->request_queue);
1910 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1911 return sdev->request_queue;
1914 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1916 unsigned int cmd_size, sgl_size;
1917 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1919 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1920 scsi_mq_inline_sgl_size(shost));
1921 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1922 if (scsi_host_get_prot(shost))
1923 cmd_size += sizeof(struct scsi_data_buffer) +
1924 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1926 memset(tag_set, 0, sizeof(*tag_set));
1927 if (shost->hostt->commit_rqs)
1928 tag_set->ops = &scsi_mq_ops;
1930 tag_set->ops = &scsi_mq_ops_no_commit;
1931 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1932 tag_set->nr_maps = shost->nr_maps ? : 1;
1933 tag_set->queue_depth = shost->can_queue;
1934 tag_set->cmd_size = cmd_size;
1935 tag_set->numa_node = NUMA_NO_NODE;
1936 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1938 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1939 tag_set->driver_data = shost;
1940 if (shost->host_tagset)
1941 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1943 return blk_mq_alloc_tag_set(tag_set);
1946 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1948 blk_mq_free_tag_set(&shost->tag_set);
1952 * scsi_device_from_queue - return sdev associated with a request_queue
1953 * @q: The request queue to return the sdev from
1955 * Return the sdev associated with a request queue or NULL if the
1956 * request_queue does not reference a SCSI device.
1958 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1960 struct scsi_device *sdev = NULL;
1962 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1963 q->mq_ops == &scsi_mq_ops)
1964 sdev = q->queuedata;
1965 if (!sdev || !get_device(&sdev->sdev_gendev))
1972 * scsi_block_requests - Utility function used by low-level drivers to prevent
1973 * further commands from being queued to the device.
1974 * @shost: host in question
1976 * There is no timer nor any other means by which the requests get unblocked
1977 * other than the low-level driver calling scsi_unblock_requests().
1979 void scsi_block_requests(struct Scsi_Host *shost)
1981 shost->host_self_blocked = 1;
1983 EXPORT_SYMBOL(scsi_block_requests);
1986 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1987 * further commands to be queued to the device.
1988 * @shost: host in question
1990 * There is no timer nor any other means by which the requests get unblocked
1991 * other than the low-level driver calling scsi_unblock_requests(). This is done
1992 * as an API function so that changes to the internals of the scsi mid-layer
1993 * won't require wholesale changes to drivers that use this feature.
1995 void scsi_unblock_requests(struct Scsi_Host *shost)
1997 shost->host_self_blocked = 0;
1998 scsi_run_host_queues(shost);
2000 EXPORT_SYMBOL(scsi_unblock_requests);
2002 void scsi_exit_queue(void)
2004 kmem_cache_destroy(scsi_sense_cache);
2008 * scsi_mode_select - issue a mode select
2009 * @sdev: SCSI device to be queried
2010 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2011 * @sp: Save page bit (0 == don't save, 1 == save)
2012 * @modepage: mode page being requested
2013 * @buffer: request buffer (may not be smaller than eight bytes)
2014 * @len: length of request buffer.
2015 * @timeout: command timeout
2016 * @retries: number of retries before failing
2017 * @data: returns a structure abstracting the mode header data
2018 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2019 * must be SCSI_SENSE_BUFFERSIZE big.
2021 * Returns zero if successful; negative error number or scsi
2026 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2027 unsigned char *buffer, int len, int timeout, int retries,
2028 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2030 unsigned char cmd[10];
2031 unsigned char *real_buffer;
2034 memset(cmd, 0, sizeof(cmd));
2035 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2037 if (sdev->use_10_for_ms) {
2040 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2043 memcpy(real_buffer + 8, buffer, len);
2047 real_buffer[2] = data->medium_type;
2048 real_buffer[3] = data->device_specific;
2049 real_buffer[4] = data->longlba ? 0x01 : 0;
2051 real_buffer[6] = data->block_descriptor_length >> 8;
2052 real_buffer[7] = data->block_descriptor_length;
2054 cmd[0] = MODE_SELECT_10;
2058 if (len > 255 || data->block_descriptor_length > 255 ||
2062 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2065 memcpy(real_buffer + 4, buffer, len);
2068 real_buffer[1] = data->medium_type;
2069 real_buffer[2] = data->device_specific;
2070 real_buffer[3] = data->block_descriptor_length;
2072 cmd[0] = MODE_SELECT;
2076 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2077 sshdr, timeout, retries, NULL);
2081 EXPORT_SYMBOL_GPL(scsi_mode_select);
2084 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2085 * @sdev: SCSI device to be queried
2086 * @dbd: set if mode sense will allow block descriptors to be returned
2087 * @modepage: mode page being requested
2088 * @buffer: request buffer (may not be smaller than eight bytes)
2089 * @len: length of request buffer.
2090 * @timeout: command timeout
2091 * @retries: number of retries before failing
2092 * @data: returns a structure abstracting the mode header data
2093 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2094 * must be SCSI_SENSE_BUFFERSIZE big.
2096 * Returns zero if unsuccessful, or the header offset (either 4
2097 * or 8 depending on whether a six or ten byte command was
2098 * issued) if successful.
2101 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2102 unsigned char *buffer, int len, int timeout, int retries,
2103 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2105 unsigned char cmd[12];
2108 int result, retry_count = retries;
2109 struct scsi_sense_hdr my_sshdr;
2111 memset(data, 0, sizeof(*data));
2112 memset(&cmd[0], 0, 12);
2114 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2115 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2118 /* caller might not be interested in sense, but we need it */
2123 use_10_for_ms = sdev->use_10_for_ms;
2125 if (use_10_for_ms) {
2129 cmd[0] = MODE_SENSE_10;
2136 cmd[0] = MODE_SENSE;
2141 memset(buffer, 0, len);
2143 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2144 sshdr, timeout, retries, NULL);
2146 /* This code looks awful: what it's doing is making sure an
2147 * ILLEGAL REQUEST sense return identifies the actual command
2148 * byte as the problem. MODE_SENSE commands can return
2149 * ILLEGAL REQUEST if the code page isn't supported */
2151 if (use_10_for_ms && !scsi_status_is_good(result) &&
2152 driver_byte(result) == DRIVER_SENSE) {
2153 if (scsi_sense_valid(sshdr)) {
2154 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2155 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2157 * Invalid command operation code
2159 sdev->use_10_for_ms = 0;
2165 if (scsi_status_is_good(result)) {
2166 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2167 (modepage == 6 || modepage == 8))) {
2168 /* Initio breakage? */
2171 data->medium_type = 0;
2172 data->device_specific = 0;
2174 data->block_descriptor_length = 0;
2175 } else if (use_10_for_ms) {
2176 data->length = buffer[0]*256 + buffer[1] + 2;
2177 data->medium_type = buffer[2];
2178 data->device_specific = buffer[3];
2179 data->longlba = buffer[4] & 0x01;
2180 data->block_descriptor_length = buffer[6]*256
2183 data->length = buffer[0] + 1;
2184 data->medium_type = buffer[1];
2185 data->device_specific = buffer[2];
2186 data->block_descriptor_length = buffer[3];
2188 data->header_length = header_length;
2189 } else if ((status_byte(result) == CHECK_CONDITION) &&
2190 scsi_sense_valid(sshdr) &&
2191 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2198 EXPORT_SYMBOL(scsi_mode_sense);
2201 * scsi_test_unit_ready - test if unit is ready
2202 * @sdev: scsi device to change the state of.
2203 * @timeout: command timeout
2204 * @retries: number of retries before failing
2205 * @sshdr: outpout pointer for decoded sense information.
2207 * Returns zero if unsuccessful or an error if TUR failed. For
2208 * removable media, UNIT_ATTENTION sets ->changed flag.
2211 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2212 struct scsi_sense_hdr *sshdr)
2215 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2219 /* try to eat the UNIT_ATTENTION if there are enough retries */
2221 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2223 if (sdev->removable && scsi_sense_valid(sshdr) &&
2224 sshdr->sense_key == UNIT_ATTENTION)
2226 } while (scsi_sense_valid(sshdr) &&
2227 sshdr->sense_key == UNIT_ATTENTION && --retries);
2231 EXPORT_SYMBOL(scsi_test_unit_ready);
2234 * scsi_device_set_state - Take the given device through the device state model.
2235 * @sdev: scsi device to change the state of.
2236 * @state: state to change to.
2238 * Returns zero if successful or an error if the requested
2239 * transition is illegal.
2242 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2244 enum scsi_device_state oldstate = sdev->sdev_state;
2246 if (state == oldstate)
2252 case SDEV_CREATED_BLOCK:
2263 case SDEV_TRANSPORT_OFFLINE:
2276 case SDEV_TRANSPORT_OFFLINE:
2284 case SDEV_TRANSPORT_OFFLINE:
2299 case SDEV_CREATED_BLOCK:
2308 case SDEV_CREATED_BLOCK:
2323 case SDEV_TRANSPORT_OFFLINE:
2335 case SDEV_TRANSPORT_OFFLINE:
2338 case SDEV_CREATED_BLOCK:
2346 sdev->offline_already = false;
2347 sdev->sdev_state = state;
2351 SCSI_LOG_ERROR_RECOVERY(1,
2352 sdev_printk(KERN_ERR, sdev,
2353 "Illegal state transition %s->%s",
2354 scsi_device_state_name(oldstate),
2355 scsi_device_state_name(state))
2359 EXPORT_SYMBOL(scsi_device_set_state);
2362 * scsi_evt_emit - emit a single SCSI device uevent
2363 * @sdev: associated SCSI device
2364 * @evt: event to emit
2366 * Send a single uevent (scsi_event) to the associated scsi_device.
2368 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2373 switch (evt->evt_type) {
2374 case SDEV_EVT_MEDIA_CHANGE:
2375 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2377 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2378 scsi_rescan_device(&sdev->sdev_gendev);
2379 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2381 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2382 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2384 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2385 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2387 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2388 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2390 case SDEV_EVT_LUN_CHANGE_REPORTED:
2391 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2393 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2394 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2396 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2397 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2406 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2410 * scsi_evt_thread - send a uevent for each scsi event
2411 * @work: work struct for scsi_device
2413 * Dispatch queued events to their associated scsi_device kobjects
2416 void scsi_evt_thread(struct work_struct *work)
2418 struct scsi_device *sdev;
2419 enum scsi_device_event evt_type;
2420 LIST_HEAD(event_list);
2422 sdev = container_of(work, struct scsi_device, event_work);
2424 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2425 if (test_and_clear_bit(evt_type, sdev->pending_events))
2426 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2429 struct scsi_event *evt;
2430 struct list_head *this, *tmp;
2431 unsigned long flags;
2433 spin_lock_irqsave(&sdev->list_lock, flags);
2434 list_splice_init(&sdev->event_list, &event_list);
2435 spin_unlock_irqrestore(&sdev->list_lock, flags);
2437 if (list_empty(&event_list))
2440 list_for_each_safe(this, tmp, &event_list) {
2441 evt = list_entry(this, struct scsi_event, node);
2442 list_del(&evt->node);
2443 scsi_evt_emit(sdev, evt);
2450 * sdev_evt_send - send asserted event to uevent thread
2451 * @sdev: scsi_device event occurred on
2452 * @evt: event to send
2454 * Assert scsi device event asynchronously.
2456 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2458 unsigned long flags;
2461 /* FIXME: currently this check eliminates all media change events
2462 * for polled devices. Need to update to discriminate between AN
2463 * and polled events */
2464 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2470 spin_lock_irqsave(&sdev->list_lock, flags);
2471 list_add_tail(&evt->node, &sdev->event_list);
2472 schedule_work(&sdev->event_work);
2473 spin_unlock_irqrestore(&sdev->list_lock, flags);
2475 EXPORT_SYMBOL_GPL(sdev_evt_send);
2478 * sdev_evt_alloc - allocate a new scsi event
2479 * @evt_type: type of event to allocate
2480 * @gfpflags: GFP flags for allocation
2482 * Allocates and returns a new scsi_event.
2484 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2487 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2491 evt->evt_type = evt_type;
2492 INIT_LIST_HEAD(&evt->node);
2494 /* evt_type-specific initialization, if any */
2496 case SDEV_EVT_MEDIA_CHANGE:
2497 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2498 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2499 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2500 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2501 case SDEV_EVT_LUN_CHANGE_REPORTED:
2502 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2503 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2511 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2514 * sdev_evt_send_simple - send asserted event to uevent thread
2515 * @sdev: scsi_device event occurred on
2516 * @evt_type: type of event to send
2517 * @gfpflags: GFP flags for allocation
2519 * Assert scsi device event asynchronously, given an event type.
2521 void sdev_evt_send_simple(struct scsi_device *sdev,
2522 enum scsi_device_event evt_type, gfp_t gfpflags)
2524 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2526 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2531 sdev_evt_send(sdev, evt);
2533 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2536 * scsi_device_quiesce - Block all commands except power management.
2537 * @sdev: scsi device to quiesce.
2539 * This works by trying to transition to the SDEV_QUIESCE state
2540 * (which must be a legal transition). When the device is in this
2541 * state, only power management requests will be accepted, all others will
2544 * Must be called with user context, may sleep.
2546 * Returns zero if unsuccessful or an error if not.
2549 scsi_device_quiesce(struct scsi_device *sdev)
2551 struct request_queue *q = sdev->request_queue;
2555 * It is allowed to call scsi_device_quiesce() multiple times from
2556 * the same context but concurrent scsi_device_quiesce() calls are
2559 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2561 if (sdev->quiesced_by == current)
2566 blk_mq_freeze_queue(q);
2568 * Ensure that the effect of blk_set_pm_only() will be visible
2569 * for percpu_ref_tryget() callers that occur after the queue
2570 * unfreeze even if the queue was already frozen before this function
2571 * was called. See also https://lwn.net/Articles/573497/.
2574 blk_mq_unfreeze_queue(q);
2576 mutex_lock(&sdev->state_mutex);
2577 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2579 sdev->quiesced_by = current;
2581 blk_clear_pm_only(q);
2582 mutex_unlock(&sdev->state_mutex);
2586 EXPORT_SYMBOL(scsi_device_quiesce);
2589 * scsi_device_resume - Restart user issued commands to a quiesced device.
2590 * @sdev: scsi device to resume.
2592 * Moves the device from quiesced back to running and restarts the
2595 * Must be called with user context, may sleep.
2597 void scsi_device_resume(struct scsi_device *sdev)
2599 /* check if the device state was mutated prior to resume, and if
2600 * so assume the state is being managed elsewhere (for example
2601 * device deleted during suspend)
2603 mutex_lock(&sdev->state_mutex);
2604 if (sdev->sdev_state == SDEV_QUIESCE)
2605 scsi_device_set_state(sdev, SDEV_RUNNING);
2606 if (sdev->quiesced_by) {
2607 sdev->quiesced_by = NULL;
2608 blk_clear_pm_only(sdev->request_queue);
2610 mutex_unlock(&sdev->state_mutex);
2612 EXPORT_SYMBOL(scsi_device_resume);
2615 device_quiesce_fn(struct scsi_device *sdev, void *data)
2617 scsi_device_quiesce(sdev);
2621 scsi_target_quiesce(struct scsi_target *starget)
2623 starget_for_each_device(starget, NULL, device_quiesce_fn);
2625 EXPORT_SYMBOL(scsi_target_quiesce);
2628 device_resume_fn(struct scsi_device *sdev, void *data)
2630 scsi_device_resume(sdev);
2634 scsi_target_resume(struct scsi_target *starget)
2636 starget_for_each_device(starget, NULL, device_resume_fn);
2638 EXPORT_SYMBOL(scsi_target_resume);
2641 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2642 * @sdev: device to block
2644 * Pause SCSI command processing on the specified device. Does not sleep.
2646 * Returns zero if successful or a negative error code upon failure.
2649 * This routine transitions the device to the SDEV_BLOCK state (which must be
2650 * a legal transition). When the device is in this state, command processing
2651 * is paused until the device leaves the SDEV_BLOCK state. See also
2652 * scsi_internal_device_unblock_nowait().
2654 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2656 struct request_queue *q = sdev->request_queue;
2659 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2661 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2668 * The device has transitioned to SDEV_BLOCK. Stop the
2669 * block layer from calling the midlayer with this device's
2672 blk_mq_quiesce_queue_nowait(q);
2675 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2678 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2679 * @sdev: device to block
2681 * Pause SCSI command processing on the specified device and wait until all
2682 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2684 * Returns zero if successful or a negative error code upon failure.
2687 * This routine transitions the device to the SDEV_BLOCK state (which must be
2688 * a legal transition). When the device is in this state, command processing
2689 * is paused until the device leaves the SDEV_BLOCK state. See also
2690 * scsi_internal_device_unblock().
2692 static int scsi_internal_device_block(struct scsi_device *sdev)
2694 struct request_queue *q = sdev->request_queue;
2697 mutex_lock(&sdev->state_mutex);
2698 err = scsi_internal_device_block_nowait(sdev);
2700 blk_mq_quiesce_queue(q);
2701 mutex_unlock(&sdev->state_mutex);
2706 void scsi_start_queue(struct scsi_device *sdev)
2708 struct request_queue *q = sdev->request_queue;
2710 blk_mq_unquiesce_queue(q);
2714 * scsi_internal_device_unblock_nowait - resume a device after a block request
2715 * @sdev: device to resume
2716 * @new_state: state to set the device to after unblocking
2718 * Restart the device queue for a previously suspended SCSI device. Does not
2721 * Returns zero if successful or a negative error code upon failure.
2724 * This routine transitions the device to the SDEV_RUNNING state or to one of
2725 * the offline states (which must be a legal transition) allowing the midlayer
2726 * to goose the queue for this device.
2728 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2729 enum scsi_device_state new_state)
2731 switch (new_state) {
2733 case SDEV_TRANSPORT_OFFLINE:
2740 * Try to transition the scsi device to SDEV_RUNNING or one of the
2741 * offlined states and goose the device queue if successful.
2743 switch (sdev->sdev_state) {
2745 case SDEV_TRANSPORT_OFFLINE:
2746 sdev->sdev_state = new_state;
2748 case SDEV_CREATED_BLOCK:
2749 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2750 new_state == SDEV_OFFLINE)
2751 sdev->sdev_state = new_state;
2753 sdev->sdev_state = SDEV_CREATED;
2761 scsi_start_queue(sdev);
2765 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2768 * scsi_internal_device_unblock - resume a device after a block request
2769 * @sdev: device to resume
2770 * @new_state: state to set the device to after unblocking
2772 * Restart the device queue for a previously suspended SCSI device. May sleep.
2774 * Returns zero if successful or a negative error code upon failure.
2777 * This routine transitions the device to the SDEV_RUNNING state or to one of
2778 * the offline states (which must be a legal transition) allowing the midlayer
2779 * to goose the queue for this device.
2781 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2782 enum scsi_device_state new_state)
2786 mutex_lock(&sdev->state_mutex);
2787 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2788 mutex_unlock(&sdev->state_mutex);
2794 device_block(struct scsi_device *sdev, void *data)
2798 ret = scsi_internal_device_block(sdev);
2800 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2801 dev_name(&sdev->sdev_gendev), ret);
2805 target_block(struct device *dev, void *data)
2807 if (scsi_is_target_device(dev))
2808 starget_for_each_device(to_scsi_target(dev), NULL,
2814 scsi_target_block(struct device *dev)
2816 if (scsi_is_target_device(dev))
2817 starget_for_each_device(to_scsi_target(dev), NULL,
2820 device_for_each_child(dev, NULL, target_block);
2822 EXPORT_SYMBOL_GPL(scsi_target_block);
2825 device_unblock(struct scsi_device *sdev, void *data)
2827 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2831 target_unblock(struct device *dev, void *data)
2833 if (scsi_is_target_device(dev))
2834 starget_for_each_device(to_scsi_target(dev), data,
2840 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2842 if (scsi_is_target_device(dev))
2843 starget_for_each_device(to_scsi_target(dev), &new_state,
2846 device_for_each_child(dev, &new_state, target_unblock);
2848 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2851 scsi_host_block(struct Scsi_Host *shost)
2853 struct scsi_device *sdev;
2857 * Call scsi_internal_device_block_nowait so we can avoid
2858 * calling synchronize_rcu() for each LUN.
2860 shost_for_each_device(sdev, shost) {
2861 mutex_lock(&sdev->state_mutex);
2862 ret = scsi_internal_device_block_nowait(sdev);
2863 mutex_unlock(&sdev->state_mutex);
2865 scsi_device_put(sdev);
2871 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2872 * calling synchronize_rcu() once is enough.
2874 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2881 EXPORT_SYMBOL_GPL(scsi_host_block);
2884 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2886 struct scsi_device *sdev;
2889 shost_for_each_device(sdev, shost) {
2890 ret = scsi_internal_device_unblock(sdev, new_state);
2892 scsi_device_put(sdev);
2898 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2901 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2902 * @sgl: scatter-gather list
2903 * @sg_count: number of segments in sg
2904 * @offset: offset in bytes into sg, on return offset into the mapped area
2905 * @len: bytes to map, on return number of bytes mapped
2907 * Returns virtual address of the start of the mapped page
2909 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2910 size_t *offset, size_t *len)
2913 size_t sg_len = 0, len_complete = 0;
2914 struct scatterlist *sg;
2917 WARN_ON(!irqs_disabled());
2919 for_each_sg(sgl, sg, sg_count, i) {
2920 len_complete = sg_len; /* Complete sg-entries */
2921 sg_len += sg->length;
2922 if (sg_len > *offset)
2926 if (unlikely(i == sg_count)) {
2927 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2929 __func__, sg_len, *offset, sg_count);
2934 /* Offset starting from the beginning of first page in this sg-entry */
2935 *offset = *offset - len_complete + sg->offset;
2937 /* Assumption: contiguous pages can be accessed as "page + i" */
2938 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2939 *offset &= ~PAGE_MASK;
2941 /* Bytes in this sg-entry from *offset to the end of the page */
2942 sg_len = PAGE_SIZE - *offset;
2946 return kmap_atomic(page);
2948 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2951 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2952 * @virt: virtual address to be unmapped
2954 void scsi_kunmap_atomic_sg(void *virt)
2956 kunmap_atomic(virt);
2958 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2960 void sdev_disable_disk_events(struct scsi_device *sdev)
2962 atomic_inc(&sdev->disk_events_disable_depth);
2964 EXPORT_SYMBOL(sdev_disable_disk_events);
2966 void sdev_enable_disk_events(struct scsi_device *sdev)
2968 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2970 atomic_dec(&sdev->disk_events_disable_depth);
2972 EXPORT_SYMBOL(sdev_enable_disk_events);
2974 static unsigned char designator_prio(const unsigned char *d)
2977 /* not associated with LUN */
2981 /* invalid length */
2985 * Order of preference for lun descriptor:
2986 * - SCSI name string
2987 * - NAA IEEE Registered Extended
2988 * - EUI-64 based 16-byte
2989 * - EUI-64 based 12-byte
2990 * - NAA IEEE Registered
2991 * - NAA IEEE Extended
2992 * - EUI-64 based 8-byte
2993 * - SCSI name string (truncated)
2995 * as longer descriptors reduce the likelyhood
2996 * of identification clashes.
2999 switch (d[1] & 0xf) {
3001 /* SCSI name string, variable-length UTF-8 */
3004 switch (d[4] >> 4) {
3006 /* NAA registered extended */
3009 /* NAA registered */
3015 /* NAA locally assigned */
3024 /* EUI64-based, 16 byte */
3027 /* EUI64-based, 12 byte */
3030 /* EUI64-based, 8 byte */
3047 * scsi_vpd_lun_id - return a unique device identification
3048 * @sdev: SCSI device
3049 * @id: buffer for the identification
3050 * @id_len: length of the buffer
3052 * Copies a unique device identification into @id based
3053 * on the information in the VPD page 0x83 of the device.
3054 * The string will be formatted as a SCSI name string.
3056 * Returns the length of the identification or error on failure.
3057 * If the identifier is longer than the supplied buffer the actual
3058 * identifier length is returned and the buffer is not zero-padded.
3060 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3064 const unsigned char *d, *cur_id_str;
3065 const struct scsi_vpd *vpd_pg83;
3066 int id_size = -EINVAL;
3069 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3075 /* The id string must be at least 20 bytes + terminating NULL byte */
3081 memset(id, 0, id_len);
3082 for (d = vpd_pg83->data + 4;
3083 d < vpd_pg83->data + vpd_pg83->len;
3085 u8 prio = designator_prio(d);
3087 if (prio == 0 || cur_id_prio > prio)
3090 switch (d[1] & 0xf) {
3093 if (cur_id_size > d[3])
3097 if (cur_id_size + 4 > id_len)
3098 cur_id_size = id_len - 4;
3100 id_size = snprintf(id, id_len, "t10.%*pE",
3101 cur_id_size, cur_id_str);
3108 switch (cur_id_size) {
3110 id_size = snprintf(id, id_len,
3115 id_size = snprintf(id, id_len,
3120 id_size = snprintf(id, id_len,
3133 switch (cur_id_size) {
3135 id_size = snprintf(id, id_len,
3140 id_size = snprintf(id, id_len,
3149 /* SCSI name string */
3150 if (cur_id_size > d[3])
3152 /* Prefer others for truncated descriptor */
3153 if (d[3] > id_len) {
3155 if (cur_id_prio > prio)
3159 cur_id_size = id_size = d[3];
3161 if (cur_id_size >= id_len)
3162 cur_id_size = id_len - 1;
3163 memcpy(id, cur_id_str, cur_id_size);
3173 EXPORT_SYMBOL(scsi_vpd_lun_id);
3176 * scsi_vpd_tpg_id - return a target port group identifier
3177 * @sdev: SCSI device
3179 * Returns the Target Port Group identifier from the information
3180 * froom VPD page 0x83 of the device.
3182 * Returns the identifier or error on failure.
3184 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3186 const unsigned char *d;
3187 const struct scsi_vpd *vpd_pg83;
3188 int group_id = -EAGAIN, rel_port = -1;
3191 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3197 d = vpd_pg83->data + 4;
3198 while (d < vpd_pg83->data + vpd_pg83->len) {
3199 switch (d[1] & 0xf) {
3201 /* Relative target port */
3202 rel_port = get_unaligned_be16(&d[6]);
3205 /* Target port group */
3206 group_id = get_unaligned_be16(&d[6]);
3215 if (group_id >= 0 && rel_id && rel_port != -1)
3220 EXPORT_SYMBOL(scsi_vpd_tpg_id);