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;
216 req = blk_get_request(sdev->request_queue,
217 data_direction == DMA_TO_DEVICE ?
218 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN,
219 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
226 ret = blk_rq_map_kern(sdev->request_queue, req,
227 buffer, bufflen, GFP_NOIO);
231 rq->cmd_len = COMMAND_SIZE(cmd[0]);
232 memcpy(rq->cmd, cmd, rq->cmd_len);
233 rq->retries = retries;
234 req->timeout = timeout;
235 req->cmd_flags |= flags;
236 req->rq_flags |= rq_flags | RQF_QUIET;
239 * head injection *required* here otherwise quiesce won't work
241 blk_execute_rq(NULL, req, 1);
244 * Some devices (USB mass-storage in particular) may transfer
245 * garbage data together with a residue indicating that the data
246 * is invalid. Prevent the garbage from being misinterpreted
247 * and prevent security leaks by zeroing out the excess data.
249 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
250 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
253 *resid = rq->resid_len;
254 if (sense && rq->sense_len)
255 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
257 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
260 blk_put_request(req);
264 EXPORT_SYMBOL(__scsi_execute);
267 * Wake up the error handler if necessary. Avoid as follows that the error
268 * handler is not woken up if host in-flight requests number ==
269 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
270 * with an RCU read lock in this function to ensure that this function in
271 * its entirety either finishes before scsi_eh_scmd_add() increases the
272 * host_failed counter or that it notices the shost state change made by
273 * scsi_eh_scmd_add().
275 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
280 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
281 if (unlikely(scsi_host_in_recovery(shost))) {
282 spin_lock_irqsave(shost->host_lock, flags);
283 if (shost->host_failed || shost->host_eh_scheduled)
284 scsi_eh_wakeup(shost);
285 spin_unlock_irqrestore(shost->host_lock, flags);
290 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
292 struct Scsi_Host *shost = sdev->host;
293 struct scsi_target *starget = scsi_target(sdev);
295 scsi_dec_host_busy(shost, cmd);
297 if (starget->can_queue > 0)
298 atomic_dec(&starget->target_busy);
300 sbitmap_put(&sdev->budget_map, cmd->budget_token);
301 cmd->budget_token = -1;
304 static void scsi_kick_queue(struct request_queue *q)
306 blk_mq_run_hw_queues(q, false);
310 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
311 * and call blk_run_queue for all the scsi_devices on the target -
312 * including current_sdev first.
314 * Called with *no* scsi locks held.
316 static void scsi_single_lun_run(struct scsi_device *current_sdev)
318 struct Scsi_Host *shost = current_sdev->host;
319 struct scsi_device *sdev, *tmp;
320 struct scsi_target *starget = scsi_target(current_sdev);
323 spin_lock_irqsave(shost->host_lock, flags);
324 starget->starget_sdev_user = NULL;
325 spin_unlock_irqrestore(shost->host_lock, flags);
328 * Call blk_run_queue for all LUNs on the target, starting with
329 * current_sdev. We race with others (to set starget_sdev_user),
330 * but in most cases, we will be first. Ideally, each LU on the
331 * target would get some limited time or requests on the target.
333 scsi_kick_queue(current_sdev->request_queue);
335 spin_lock_irqsave(shost->host_lock, flags);
336 if (starget->starget_sdev_user)
338 list_for_each_entry_safe(sdev, tmp, &starget->devices,
339 same_target_siblings) {
340 if (sdev == current_sdev)
342 if (scsi_device_get(sdev))
345 spin_unlock_irqrestore(shost->host_lock, flags);
346 scsi_kick_queue(sdev->request_queue);
347 spin_lock_irqsave(shost->host_lock, flags);
349 scsi_device_put(sdev);
352 spin_unlock_irqrestore(shost->host_lock, flags);
355 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
357 if (scsi_device_busy(sdev) >= sdev->queue_depth)
359 if (atomic_read(&sdev->device_blocked) > 0)
364 static inline bool scsi_target_is_busy(struct scsi_target *starget)
366 if (starget->can_queue > 0) {
367 if (atomic_read(&starget->target_busy) >= starget->can_queue)
369 if (atomic_read(&starget->target_blocked) > 0)
375 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
377 if (atomic_read(&shost->host_blocked) > 0)
379 if (shost->host_self_blocked)
384 static void scsi_starved_list_run(struct Scsi_Host *shost)
386 LIST_HEAD(starved_list);
387 struct scsi_device *sdev;
390 spin_lock_irqsave(shost->host_lock, flags);
391 list_splice_init(&shost->starved_list, &starved_list);
393 while (!list_empty(&starved_list)) {
394 struct request_queue *slq;
397 * As long as shost is accepting commands and we have
398 * starved queues, call blk_run_queue. scsi_request_fn
399 * drops the queue_lock and can add us back to the
402 * host_lock protects the starved_list and starved_entry.
403 * scsi_request_fn must get the host_lock before checking
404 * or modifying starved_list or starved_entry.
406 if (scsi_host_is_busy(shost))
409 sdev = list_entry(starved_list.next,
410 struct scsi_device, starved_entry);
411 list_del_init(&sdev->starved_entry);
412 if (scsi_target_is_busy(scsi_target(sdev))) {
413 list_move_tail(&sdev->starved_entry,
414 &shost->starved_list);
419 * Once we drop the host lock, a racing scsi_remove_device()
420 * call may remove the sdev from the starved list and destroy
421 * it and the queue. Mitigate by taking a reference to the
422 * queue and never touching the sdev again after we drop the
423 * host lock. Note: if __scsi_remove_device() invokes
424 * blk_cleanup_queue() before the queue is run from this
425 * function then blk_run_queue() will return immediately since
426 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
428 slq = sdev->request_queue;
429 if (!blk_get_queue(slq))
431 spin_unlock_irqrestore(shost->host_lock, flags);
433 scsi_kick_queue(slq);
436 spin_lock_irqsave(shost->host_lock, flags);
438 /* put any unprocessed entries back */
439 list_splice(&starved_list, &shost->starved_list);
440 spin_unlock_irqrestore(shost->host_lock, flags);
444 * scsi_run_queue - Select a proper request queue to serve next.
445 * @q: last request's queue
447 * The previous command was completely finished, start a new one if possible.
449 static void scsi_run_queue(struct request_queue *q)
451 struct scsi_device *sdev = q->queuedata;
453 if (scsi_target(sdev)->single_lun)
454 scsi_single_lun_run(sdev);
455 if (!list_empty(&sdev->host->starved_list))
456 scsi_starved_list_run(sdev->host);
458 blk_mq_run_hw_queues(q, false);
461 void scsi_requeue_run_queue(struct work_struct *work)
463 struct scsi_device *sdev;
464 struct request_queue *q;
466 sdev = container_of(work, struct scsi_device, requeue_work);
467 q = sdev->request_queue;
471 void scsi_run_host_queues(struct Scsi_Host *shost)
473 struct scsi_device *sdev;
475 shost_for_each_device(sdev, shost)
476 scsi_run_queue(sdev->request_queue);
479 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
481 if (!blk_rq_is_passthrough(cmd->request)) {
482 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
484 if (drv->uninit_command)
485 drv->uninit_command(cmd);
489 void scsi_free_sgtables(struct scsi_cmnd *cmd)
491 if (cmd->sdb.table.nents)
492 sg_free_table_chained(&cmd->sdb.table,
494 if (scsi_prot_sg_count(cmd))
495 sg_free_table_chained(&cmd->prot_sdb->table,
496 SCSI_INLINE_PROT_SG_CNT);
498 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
500 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
502 scsi_free_sgtables(cmd);
503 scsi_uninit_cmd(cmd);
506 static void scsi_run_queue_async(struct scsi_device *sdev)
508 if (scsi_target(sdev)->single_lun ||
509 !list_empty(&sdev->host->starved_list)) {
510 kblockd_schedule_work(&sdev->requeue_work);
513 * smp_mb() present in sbitmap_queue_clear() or implied in
514 * .end_io is for ordering writing .device_busy in
515 * scsi_device_unbusy() and reading sdev->restarts.
517 int old = atomic_read(&sdev->restarts);
520 * ->restarts has to be kept as non-zero if new budget
523 * No need to run queue when either another re-run
524 * queue wins in updating ->restarts or a new budget
527 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
528 blk_mq_run_hw_queues(sdev->request_queue, true);
532 /* Returns false when no more bytes to process, true if there are more */
533 static bool scsi_end_request(struct request *req, blk_status_t error,
536 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
537 struct scsi_device *sdev = cmd->device;
538 struct request_queue *q = sdev->request_queue;
540 if (blk_update_request(req, error, bytes))
543 if (blk_queue_add_random(q))
544 add_disk_randomness(req->rq_disk);
546 if (!blk_rq_is_scsi(req)) {
547 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
548 cmd->flags &= ~SCMD_INITIALIZED;
552 * Calling rcu_barrier() is not necessary here because the
553 * SCSI error handler guarantees that the function called by
554 * call_rcu() has been called before scsi_end_request() is
557 destroy_rcu_head(&cmd->rcu);
560 * In the MQ case the command gets freed by __blk_mq_end_request,
561 * so we have to do all cleanup that depends on it earlier.
563 * We also can't kick the queues from irq context, so we
564 * will have to defer it to a workqueue.
566 scsi_mq_uninit_cmd(cmd);
569 * queue is still alive, so grab the ref for preventing it
570 * from being cleaned up during running queue.
572 percpu_ref_get(&q->q_usage_counter);
574 __blk_mq_end_request(req, error);
576 scsi_run_queue_async(sdev);
578 percpu_ref_put(&q->q_usage_counter);
583 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
585 * @result: scsi error code
587 * Translate a SCSI result code into a blk_status_t value. May reset the host
588 * byte of @cmd->result.
590 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
592 switch (host_byte(result)) {
594 if (scsi_status_is_good(result))
596 return BLK_STS_IOERR;
597 case DID_TRANSPORT_FAILFAST:
598 case DID_TRANSPORT_MARGINAL:
599 return BLK_STS_TRANSPORT;
600 case DID_TARGET_FAILURE:
601 set_host_byte(cmd, DID_OK);
602 return BLK_STS_TARGET;
603 case DID_NEXUS_FAILURE:
604 set_host_byte(cmd, DID_OK);
605 return BLK_STS_NEXUS;
606 case DID_ALLOC_FAILURE:
607 set_host_byte(cmd, DID_OK);
608 return BLK_STS_NOSPC;
609 case DID_MEDIUM_ERROR:
610 set_host_byte(cmd, DID_OK);
611 return BLK_STS_MEDIUM;
613 return BLK_STS_IOERR;
617 /* Helper for scsi_io_completion() when "reprep" action required. */
618 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
619 struct request_queue *q)
621 /* A new command will be prepared and issued. */
622 scsi_mq_requeue_cmd(cmd);
625 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
627 struct request *req = cmd->request;
628 unsigned long wait_for;
630 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
633 wait_for = (cmd->allowed + 1) * req->timeout;
634 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
635 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
642 /* Helper for scsi_io_completion() when special action required. */
643 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
645 struct request_queue *q = cmd->device->request_queue;
646 struct request *req = cmd->request;
648 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
649 ACTION_DELAYED_RETRY} action;
650 struct scsi_sense_hdr sshdr;
652 bool sense_current = true; /* false implies "deferred sense" */
653 blk_status_t blk_stat;
655 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
657 sense_current = !scsi_sense_is_deferred(&sshdr);
659 blk_stat = scsi_result_to_blk_status(cmd, result);
661 if (host_byte(result) == DID_RESET) {
662 /* Third party bus reset or reset for error recovery
663 * reasons. Just retry the command and see what
666 action = ACTION_RETRY;
667 } else if (sense_valid && sense_current) {
668 switch (sshdr.sense_key) {
670 if (cmd->device->removable) {
671 /* Detected disc change. Set a bit
672 * and quietly refuse further access.
674 cmd->device->changed = 1;
675 action = ACTION_FAIL;
677 /* Must have been a power glitch, or a
678 * bus reset. Could not have been a
679 * media change, so we just retry the
680 * command and see what happens.
682 action = ACTION_RETRY;
685 case ILLEGAL_REQUEST:
686 /* If we had an ILLEGAL REQUEST returned, then
687 * we may have performed an unsupported
688 * command. The only thing this should be
689 * would be a ten byte read where only a six
690 * byte read was supported. Also, on a system
691 * where READ CAPACITY failed, we may have
692 * read past the end of the disk.
694 if ((cmd->device->use_10_for_rw &&
695 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
696 (cmd->cmnd[0] == READ_10 ||
697 cmd->cmnd[0] == WRITE_10)) {
698 /* This will issue a new 6-byte command. */
699 cmd->device->use_10_for_rw = 0;
700 action = ACTION_REPREP;
701 } else if (sshdr.asc == 0x10) /* DIX */ {
702 action = ACTION_FAIL;
703 blk_stat = BLK_STS_PROTECTION;
704 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
705 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
706 action = ACTION_FAIL;
707 blk_stat = BLK_STS_TARGET;
709 action = ACTION_FAIL;
711 case ABORTED_COMMAND:
712 action = ACTION_FAIL;
713 if (sshdr.asc == 0x10) /* DIF */
714 blk_stat = BLK_STS_PROTECTION;
717 /* If the device is in the process of becoming
718 * ready, or has a temporary blockage, retry.
720 if (sshdr.asc == 0x04) {
721 switch (sshdr.ascq) {
722 case 0x01: /* becoming ready */
723 case 0x04: /* format in progress */
724 case 0x05: /* rebuild in progress */
725 case 0x06: /* recalculation in progress */
726 case 0x07: /* operation in progress */
727 case 0x08: /* Long write in progress */
728 case 0x09: /* self test in progress */
729 case 0x14: /* space allocation in progress */
730 case 0x1a: /* start stop unit in progress */
731 case 0x1b: /* sanitize in progress */
732 case 0x1d: /* configuration in progress */
733 case 0x24: /* depopulation in progress */
734 action = ACTION_DELAYED_RETRY;
736 case 0x0a: /* ALUA state transition */
737 blk_stat = BLK_STS_AGAIN;
740 action = ACTION_FAIL;
744 action = ACTION_FAIL;
746 case VOLUME_OVERFLOW:
747 /* See SSC3rXX or current. */
748 action = ACTION_FAIL;
751 action = ACTION_FAIL;
752 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
753 (sshdr.asc == 0x55 &&
754 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
755 /* Insufficient zone resources */
756 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
760 action = ACTION_FAIL;
764 action = ACTION_FAIL;
766 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
767 action = ACTION_FAIL;
771 /* Give up and fail the remainder of the request */
772 if (!(req->rq_flags & RQF_QUIET)) {
773 static DEFINE_RATELIMIT_STATE(_rs,
774 DEFAULT_RATELIMIT_INTERVAL,
775 DEFAULT_RATELIMIT_BURST);
777 if (unlikely(scsi_logging_level))
779 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
780 SCSI_LOG_MLCOMPLETE_BITS);
783 * if logging is enabled the failure will be printed
784 * in scsi_log_completion(), so avoid duplicate messages
786 if (!level && __ratelimit(&_rs)) {
787 scsi_print_result(cmd, NULL, FAILED);
789 scsi_print_sense(cmd);
790 scsi_print_command(cmd);
793 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
797 scsi_io_completion_reprep(cmd, q);
800 /* Retry the same command immediately */
801 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
803 case ACTION_DELAYED_RETRY:
804 /* Retry the same command after a delay */
805 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
811 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
812 * new result that may suppress further error checking. Also modifies
813 * *blk_statp in some cases.
815 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
816 blk_status_t *blk_statp)
819 bool sense_current = true; /* false implies "deferred sense" */
820 struct request *req = cmd->request;
821 struct scsi_sense_hdr sshdr;
823 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
825 sense_current = !scsi_sense_is_deferred(&sshdr);
827 if (blk_rq_is_passthrough(req)) {
830 * SG_IO wants current and deferred errors
832 scsi_req(req)->sense_len =
833 min(8 + cmd->sense_buffer[7],
834 SCSI_SENSE_BUFFERSIZE);
837 *blk_statp = scsi_result_to_blk_status(cmd, result);
838 } else if (blk_rq_bytes(req) == 0 && sense_current) {
840 * Flush commands do not transfers any data, and thus cannot use
841 * good_bytes != blk_rq_bytes(req) as the signal for an error.
842 * This sets *blk_statp explicitly for the problem case.
844 *blk_statp = scsi_result_to_blk_status(cmd, result);
847 * Recovered errors need reporting, but they're always treated as
848 * success, so fiddle the result code here. For passthrough requests
849 * we already took a copy of the original into sreq->result which
850 * is what gets returned to the user
852 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
853 bool do_print = true;
855 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
856 * skip print since caller wants ATA registers. Only occurs
857 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
859 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
861 else if (req->rq_flags & RQF_QUIET)
864 scsi_print_sense(cmd);
866 /* for passthrough, *blk_statp may be set */
867 *blk_statp = BLK_STS_OK;
870 * Another corner case: the SCSI status byte is non-zero but 'good'.
871 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
872 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
873 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
874 * intermediate statuses (both obsolete in SAM-4) as good.
876 if ((result & 0xff) && scsi_status_is_good(result)) {
878 *blk_statp = BLK_STS_OK;
884 * scsi_io_completion - Completion processing for SCSI commands.
885 * @cmd: command that is finished.
886 * @good_bytes: number of processed bytes.
888 * We will finish off the specified number of sectors. If we are done, the
889 * command block will be released and the queue function will be goosed. If we
890 * are not done then we have to figure out what to do next:
892 * a) We can call scsi_io_completion_reprep(). The request will be
893 * unprepared and put back on the queue. Then a new command will
894 * be created for it. This should be used if we made forward
895 * progress, or if we want to switch from READ(10) to READ(6) for
898 * b) We can call scsi_io_completion_action(). The request will be
899 * put back on the queue and retried using the same command as
900 * before, possibly after a delay.
902 * c) We can call scsi_end_request() with blk_stat other than
903 * BLK_STS_OK, to fail the remainder of the request.
905 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
907 int result = cmd->result;
908 struct request_queue *q = cmd->device->request_queue;
909 struct request *req = cmd->request;
910 blk_status_t blk_stat = BLK_STS_OK;
912 if (unlikely(result)) /* a nz result may or may not be an error */
913 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
915 if (unlikely(blk_rq_is_passthrough(req))) {
917 * scsi_result_to_blk_status may have reset the host_byte
919 scsi_req(req)->result = cmd->result;
923 * Next deal with any sectors which we were able to correctly
926 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
927 "%u sectors total, %d bytes done.\n",
928 blk_rq_sectors(req), good_bytes));
931 * Failed, zero length commands always need to drop down
932 * to retry code. Fast path should return in this block.
934 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
935 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
936 return; /* no bytes remaining */
939 /* Kill remainder if no retries. */
940 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
941 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
943 "Bytes remaining after failed, no-retry command");
948 * If there had been no error, but we have leftover bytes in the
949 * requeues just queue the command up again.
951 if (likely(result == 0))
952 scsi_io_completion_reprep(cmd, q);
954 scsi_io_completion_action(cmd, result);
957 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
960 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
961 !op_is_write(req_op(rq)) &&
962 sdev->host->hostt->dma_need_drain(rq);
966 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
967 * @cmd: SCSI command data structure to initialize.
969 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
973 * * BLK_STS_OK - on success
974 * * BLK_STS_RESOURCE - if the failure is retryable
975 * * BLK_STS_IOERR - if the failure is fatal
977 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
979 struct scsi_device *sdev = cmd->device;
980 struct request *rq = cmd->request;
981 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
982 struct scatterlist *last_sg = NULL;
984 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
987 if (WARN_ON_ONCE(!nr_segs))
988 return BLK_STS_IOERR;
991 * Make sure there is space for the drain. The driver must adjust
992 * max_hw_segments to be prepared for this.
998 * If sg table allocation fails, requeue request later.
1000 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1001 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1002 return BLK_STS_RESOURCE;
1005 * Next, walk the list, and fill in the addresses and sizes of
1008 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1010 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1011 unsigned int pad_len =
1012 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1014 last_sg->length += pad_len;
1015 cmd->extra_len += pad_len;
1019 sg_unmark_end(last_sg);
1020 last_sg = sg_next(last_sg);
1021 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1022 sg_mark_end(last_sg);
1024 cmd->extra_len += sdev->dma_drain_len;
1028 BUG_ON(count > cmd->sdb.table.nents);
1029 cmd->sdb.table.nents = count;
1030 cmd->sdb.length = blk_rq_payload_bytes(rq);
1032 if (blk_integrity_rq(rq)) {
1033 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1036 if (WARN_ON_ONCE(!prot_sdb)) {
1038 * This can happen if someone (e.g. multipath)
1039 * queues a command to a device on an adapter
1040 * that does not support DIX.
1042 ret = BLK_STS_IOERR;
1043 goto out_free_sgtables;
1046 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1048 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049 prot_sdb->table.sgl,
1050 SCSI_INLINE_PROT_SG_CNT)) {
1051 ret = BLK_STS_RESOURCE;
1052 goto out_free_sgtables;
1055 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056 prot_sdb->table.sgl);
1057 BUG_ON(count > ivecs);
1058 BUG_ON(count > queue_max_integrity_segments(rq->q));
1060 cmd->prot_sdb = prot_sdb;
1061 cmd->prot_sdb->table.nents = count;
1066 scsi_free_sgtables(cmd);
1069 EXPORT_SYMBOL(scsi_alloc_sgtables);
1072 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073 * @rq: Request associated with the SCSI command to be initialized.
1075 * This function initializes the members of struct scsi_cmnd that must be
1076 * initialized before request processing starts and that won't be
1077 * reinitialized if a SCSI command is requeued.
1079 * Called from inside blk_get_request() for pass-through requests and from
1080 * inside scsi_init_command() for filesystem requests.
1082 static void scsi_initialize_rq(struct request *rq)
1084 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1086 scsi_req_init(&cmd->req);
1087 init_rcu_head(&cmd->rcu);
1088 cmd->jiffies_at_alloc = jiffies;
1093 * Only called when the request isn't completed by SCSI, and not freed by
1096 static void scsi_cleanup_rq(struct request *rq)
1098 if (rq->rq_flags & RQF_DONTPREP) {
1099 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1100 rq->rq_flags &= ~RQF_DONTPREP;
1104 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1105 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1107 void *buf = cmd->sense_buffer;
1108 void *prot = cmd->prot_sdb;
1109 struct request *rq = blk_mq_rq_from_pdu(cmd);
1110 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1111 unsigned long jiffies_at_alloc;
1112 int retries, to_clear;
1114 int budget_token = cmd->budget_token;
1116 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1117 flags |= SCMD_INITIALIZED;
1118 scsi_initialize_rq(rq);
1121 jiffies_at_alloc = cmd->jiffies_at_alloc;
1122 retries = cmd->retries;
1123 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1125 * Zero out the cmd, except for the embedded scsi_request. Only clear
1126 * the driver-private command data if the LLD does not supply a
1127 * function to initialize that data.
1129 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1130 if (!dev->host->hostt->init_cmd_priv)
1131 to_clear += dev->host->hostt->cmd_size;
1132 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1135 cmd->sense_buffer = buf;
1136 cmd->prot_sdb = prot;
1138 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1139 cmd->jiffies_at_alloc = jiffies_at_alloc;
1140 cmd->retries = retries;
1142 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1143 cmd->budget_token = budget_token;
1147 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1148 struct request *req)
1150 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1153 * Passthrough requests may transfer data, in which case they must
1154 * a bio attached to them. Or they might contain a SCSI command
1155 * that does not transfer data, in which case they may optionally
1156 * submit a request without an attached bio.
1159 blk_status_t ret = scsi_alloc_sgtables(cmd);
1160 if (unlikely(ret != BLK_STS_OK))
1163 BUG_ON(blk_rq_bytes(req));
1165 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1168 cmd->cmd_len = scsi_req(req)->cmd_len;
1169 if (cmd->cmd_len == 0)
1170 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1171 cmd->cmnd = scsi_req(req)->cmd;
1172 cmd->transfersize = blk_rq_bytes(req);
1173 cmd->allowed = scsi_req(req)->retries;
1178 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1180 switch (sdev->sdev_state) {
1184 case SDEV_TRANSPORT_OFFLINE:
1186 * If the device is offline we refuse to process any
1187 * commands. The device must be brought online
1188 * before trying any recovery commands.
1190 if (!sdev->offline_already) {
1191 sdev->offline_already = true;
1192 sdev_printk(KERN_ERR, sdev,
1193 "rejecting I/O to offline device\n");
1195 return BLK_STS_IOERR;
1198 * If the device is fully deleted, we refuse to
1199 * process any commands as well.
1201 sdev_printk(KERN_ERR, sdev,
1202 "rejecting I/O to dead device\n");
1203 return BLK_STS_IOERR;
1205 case SDEV_CREATED_BLOCK:
1206 return BLK_STS_RESOURCE;
1209 * If the device is blocked we only accept power management
1212 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1213 return BLK_STS_RESOURCE;
1217 * For any other not fully online state we only allow
1218 * power management commands.
1220 if (req && !(req->rq_flags & RQF_PM))
1221 return BLK_STS_IOERR;
1227 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1228 * and return the token else return -1.
1230 static inline int scsi_dev_queue_ready(struct request_queue *q,
1231 struct scsi_device *sdev)
1235 token = sbitmap_get(&sdev->budget_map);
1236 if (atomic_read(&sdev->device_blocked)) {
1240 if (scsi_device_busy(sdev) > 1)
1244 * unblock after device_blocked iterates to zero
1246 if (atomic_dec_return(&sdev->device_blocked) > 0)
1248 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1249 "unblocking device at zero depth\n"));
1255 sbitmap_put(&sdev->budget_map, token);
1261 * scsi_target_queue_ready: checks if there we can send commands to target
1262 * @sdev: scsi device on starget to check.
1264 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1265 struct scsi_device *sdev)
1267 struct scsi_target *starget = scsi_target(sdev);
1270 if (starget->single_lun) {
1271 spin_lock_irq(shost->host_lock);
1272 if (starget->starget_sdev_user &&
1273 starget->starget_sdev_user != sdev) {
1274 spin_unlock_irq(shost->host_lock);
1277 starget->starget_sdev_user = sdev;
1278 spin_unlock_irq(shost->host_lock);
1281 if (starget->can_queue <= 0)
1284 busy = atomic_inc_return(&starget->target_busy) - 1;
1285 if (atomic_read(&starget->target_blocked) > 0) {
1290 * unblock after target_blocked iterates to zero
1292 if (atomic_dec_return(&starget->target_blocked) > 0)
1295 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1296 "unblocking target at zero depth\n"));
1299 if (busy >= starget->can_queue)
1305 spin_lock_irq(shost->host_lock);
1306 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1307 spin_unlock_irq(shost->host_lock);
1309 if (starget->can_queue > 0)
1310 atomic_dec(&starget->target_busy);
1315 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1316 * return 0. We must end up running the queue again whenever 0 is
1317 * returned, else IO can hang.
1319 static inline int scsi_host_queue_ready(struct request_queue *q,
1320 struct Scsi_Host *shost,
1321 struct scsi_device *sdev,
1322 struct scsi_cmnd *cmd)
1324 if (scsi_host_in_recovery(shost))
1327 if (atomic_read(&shost->host_blocked) > 0) {
1328 if (scsi_host_busy(shost) > 0)
1332 * unblock after host_blocked iterates to zero
1334 if (atomic_dec_return(&shost->host_blocked) > 0)
1338 shost_printk(KERN_INFO, shost,
1339 "unblocking host at zero depth\n"));
1342 if (shost->host_self_blocked)
1345 /* We're OK to process the command, so we can't be starved */
1346 if (!list_empty(&sdev->starved_entry)) {
1347 spin_lock_irq(shost->host_lock);
1348 if (!list_empty(&sdev->starved_entry))
1349 list_del_init(&sdev->starved_entry);
1350 spin_unlock_irq(shost->host_lock);
1353 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1358 spin_lock_irq(shost->host_lock);
1359 if (list_empty(&sdev->starved_entry))
1360 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1361 spin_unlock_irq(shost->host_lock);
1363 scsi_dec_host_busy(shost, cmd);
1368 * Busy state exporting function for request stacking drivers.
1370 * For efficiency, no lock is taken to check the busy state of
1371 * shost/starget/sdev, since the returned value is not guaranteed and
1372 * may be changed after request stacking drivers call the function,
1373 * regardless of taking lock or not.
1375 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1376 * needs to return 'not busy'. Otherwise, request stacking drivers
1377 * may hold requests forever.
1379 static bool scsi_mq_lld_busy(struct request_queue *q)
1381 struct scsi_device *sdev = q->queuedata;
1382 struct Scsi_Host *shost;
1384 if (blk_queue_dying(q))
1390 * Ignore host/starget busy state.
1391 * Since block layer does not have a concept of fairness across
1392 * multiple queues, congestion of host/starget needs to be handled
1395 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1402 * Block layer request completion callback. May be called from interrupt
1405 static void scsi_complete(struct request *rq)
1407 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1408 enum scsi_disposition disposition;
1410 INIT_LIST_HEAD(&cmd->eh_entry);
1412 atomic_inc(&cmd->device->iodone_cnt);
1414 atomic_inc(&cmd->device->ioerr_cnt);
1416 disposition = scsi_decide_disposition(cmd);
1417 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1418 disposition = SUCCESS;
1420 scsi_log_completion(cmd, disposition);
1422 switch (disposition) {
1424 scsi_finish_command(cmd);
1427 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1429 case ADD_TO_MLQUEUE:
1430 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1433 scsi_eh_scmd_add(cmd);
1439 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1440 * @cmd: command block we are dispatching.
1442 * Return: nonzero return request was rejected and device's queue needs to be
1445 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1447 struct Scsi_Host *host = cmd->device->host;
1450 atomic_inc(&cmd->device->iorequest_cnt);
1452 /* check if the device is still usable */
1453 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1454 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1455 * returns an immediate error upwards, and signals
1456 * that the device is no longer present */
1457 cmd->result = DID_NO_CONNECT << 16;
1461 /* Check to see if the scsi lld made this device blocked. */
1462 if (unlikely(scsi_device_blocked(cmd->device))) {
1464 * in blocked state, the command is just put back on
1465 * the device queue. The suspend state has already
1466 * blocked the queue so future requests should not
1467 * occur until the device transitions out of the
1470 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1471 "queuecommand : device blocked\n"));
1472 return SCSI_MLQUEUE_DEVICE_BUSY;
1475 /* Store the LUN value in cmnd, if needed. */
1476 if (cmd->device->lun_in_cdb)
1477 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1478 (cmd->device->lun << 5 & 0xe0);
1483 * Before we queue this command, check if the command
1484 * length exceeds what the host adapter can handle.
1486 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1487 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1488 "queuecommand : command too long. "
1489 "cdb_size=%d host->max_cmd_len=%d\n",
1490 cmd->cmd_len, cmd->device->host->max_cmd_len));
1491 cmd->result = (DID_ABORT << 16);
1495 if (unlikely(host->shost_state == SHOST_DEL)) {
1496 cmd->result = (DID_NO_CONNECT << 16);
1501 trace_scsi_dispatch_cmd_start(cmd);
1502 rtn = host->hostt->queuecommand(host, cmd);
1504 trace_scsi_dispatch_cmd_error(cmd, rtn);
1505 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1506 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1507 rtn = SCSI_MLQUEUE_HOST_BUSY;
1509 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1510 "queuecommand : request rejected\n"));
1515 cmd->scsi_done(cmd);
1519 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1520 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1522 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1523 sizeof(struct scatterlist);
1526 static blk_status_t scsi_prepare_cmd(struct request *req)
1528 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1529 struct scsi_device *sdev = req->q->queuedata;
1530 struct Scsi_Host *shost = sdev->host;
1531 struct scatterlist *sg;
1533 scsi_init_command(sdev, cmd);
1536 cmd->tag = req->tag;
1537 cmd->prot_op = SCSI_PROT_NORMAL;
1538 if (blk_rq_bytes(req))
1539 cmd->sc_data_direction = rq_dma_dir(req);
1541 cmd->sc_data_direction = DMA_NONE;
1543 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1544 cmd->sdb.table.sgl = sg;
1546 if (scsi_host_get_prot(shost)) {
1547 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1549 cmd->prot_sdb->table.sgl =
1550 (struct scatterlist *)(cmd->prot_sdb + 1);
1554 * Special handling for passthrough commands, which don't go to the ULP
1557 if (blk_rq_is_scsi(req))
1558 return scsi_setup_scsi_cmnd(sdev, req);
1560 if (sdev->handler && sdev->handler->prep_fn) {
1561 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1563 if (ret != BLK_STS_OK)
1567 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1568 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1569 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1572 static void scsi_mq_done(struct scsi_cmnd *cmd)
1574 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1576 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1578 trace_scsi_dispatch_cmd_done(cmd);
1579 blk_mq_complete_request(cmd->request);
1582 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1584 struct scsi_device *sdev = q->queuedata;
1586 sbitmap_put(&sdev->budget_map, budget_token);
1589 static int scsi_mq_get_budget(struct request_queue *q)
1591 struct scsi_device *sdev = q->queuedata;
1592 int token = scsi_dev_queue_ready(q, sdev);
1597 atomic_inc(&sdev->restarts);
1600 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1601 * .restarts must be incremented before .device_busy is read because the
1602 * code in scsi_run_queue_async() depends on the order of these operations.
1604 smp_mb__after_atomic();
1607 * If all in-flight requests originated from this LUN are completed
1608 * before reading .device_busy, sdev->device_busy will be observed as
1609 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1610 * soon. Otherwise, completion of one of these requests will observe
1611 * the .restarts flag, and the request queue will be run for handling
1612 * this request, see scsi_end_request().
1614 if (unlikely(scsi_device_busy(sdev) == 0 &&
1615 !scsi_device_blocked(sdev)))
1616 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1620 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1622 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1624 cmd->budget_token = token;
1627 static int scsi_mq_get_rq_budget_token(struct request *req)
1629 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1631 return cmd->budget_token;
1634 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1635 const struct blk_mq_queue_data *bd)
1637 struct request *req = bd->rq;
1638 struct request_queue *q = req->q;
1639 struct scsi_device *sdev = q->queuedata;
1640 struct Scsi_Host *shost = sdev->host;
1641 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1645 WARN_ON_ONCE(cmd->budget_token < 0);
1648 * If the device is not in running state we will reject some or all
1651 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1652 ret = scsi_device_state_check(sdev, req);
1653 if (ret != BLK_STS_OK)
1654 goto out_put_budget;
1657 ret = BLK_STS_RESOURCE;
1658 if (!scsi_target_queue_ready(shost, sdev))
1659 goto out_put_budget;
1660 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1661 goto out_dec_target_busy;
1663 if (!(req->rq_flags & RQF_DONTPREP)) {
1664 ret = scsi_prepare_cmd(req);
1665 if (ret != BLK_STS_OK)
1666 goto out_dec_host_busy;
1667 req->rq_flags |= RQF_DONTPREP;
1669 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1672 cmd->flags &= SCMD_PRESERVED_FLAGS;
1673 if (sdev->simple_tags)
1674 cmd->flags |= SCMD_TAGGED;
1676 cmd->flags |= SCMD_LAST;
1678 scsi_set_resid(cmd, 0);
1679 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1680 cmd->scsi_done = scsi_mq_done;
1682 blk_mq_start_request(req);
1683 reason = scsi_dispatch_cmd(cmd);
1685 scsi_set_blocked(cmd, reason);
1686 ret = BLK_STS_RESOURCE;
1687 goto out_dec_host_busy;
1693 scsi_dec_host_busy(shost, cmd);
1694 out_dec_target_busy:
1695 if (scsi_target(sdev)->can_queue > 0)
1696 atomic_dec(&scsi_target(sdev)->target_busy);
1698 scsi_mq_put_budget(q, cmd->budget_token);
1699 cmd->budget_token = -1;
1703 case BLK_STS_RESOURCE:
1704 case BLK_STS_ZONE_RESOURCE:
1705 if (scsi_device_blocked(sdev))
1706 ret = BLK_STS_DEV_RESOURCE;
1709 scsi_req(req)->result = DID_BUS_BUSY << 16;
1710 if (req->rq_flags & RQF_DONTPREP)
1711 scsi_mq_uninit_cmd(cmd);
1714 if (unlikely(!scsi_device_online(sdev)))
1715 scsi_req(req)->result = DID_NO_CONNECT << 16;
1717 scsi_req(req)->result = DID_ERROR << 16;
1719 * Make sure to release all allocated resources when
1720 * we hit an error, as we will never see this command
1723 if (req->rq_flags & RQF_DONTPREP)
1724 scsi_mq_uninit_cmd(cmd);
1725 scsi_run_queue_async(sdev);
1731 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1735 return BLK_EH_RESET_TIMER;
1736 return scsi_times_out(req);
1739 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1740 unsigned int hctx_idx, unsigned int numa_node)
1742 struct Scsi_Host *shost = set->driver_data;
1743 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1744 struct scatterlist *sg;
1748 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1749 if (!cmd->sense_buffer)
1751 cmd->req.sense = cmd->sense_buffer;
1753 if (scsi_host_get_prot(shost)) {
1754 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1755 shost->hostt->cmd_size;
1756 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1759 if (shost->hostt->init_cmd_priv) {
1760 ret = shost->hostt->init_cmd_priv(shost, cmd);
1762 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1768 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1769 unsigned int hctx_idx)
1771 struct Scsi_Host *shost = set->driver_data;
1772 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1774 if (shost->hostt->exit_cmd_priv)
1775 shost->hostt->exit_cmd_priv(shost, cmd);
1776 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1780 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
1782 struct Scsi_Host *shost = hctx->driver_data;
1784 if (shost->hostt->mq_poll)
1785 return shost->hostt->mq_poll(shost, hctx->queue_num);
1790 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1791 unsigned int hctx_idx)
1793 struct Scsi_Host *shost = data;
1795 hctx->driver_data = shost;
1799 static int scsi_map_queues(struct blk_mq_tag_set *set)
1801 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1803 if (shost->hostt->map_queues)
1804 return shost->hostt->map_queues(shost);
1805 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1808 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1810 struct device *dev = shost->dma_dev;
1813 * this limit is imposed by hardware restrictions
1815 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1818 if (scsi_host_prot_dma(shost)) {
1819 shost->sg_prot_tablesize =
1820 min_not_zero(shost->sg_prot_tablesize,
1821 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1822 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1823 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1826 if (dev->dma_mask) {
1827 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1828 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1830 blk_queue_max_hw_sectors(q, shost->max_sectors);
1831 blk_queue_segment_boundary(q, shost->dma_boundary);
1832 dma_set_seg_boundary(dev, shost->dma_boundary);
1834 blk_queue_max_segment_size(q, shost->max_segment_size);
1835 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1836 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1839 * Set a reasonable default alignment: The larger of 32-byte (dword),
1840 * which is a common minimum for HBAs, and the minimum DMA alignment,
1841 * which is set by the platform.
1843 * Devices that require a bigger alignment can increase it later.
1845 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1847 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1849 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1850 .get_budget = scsi_mq_get_budget,
1851 .put_budget = scsi_mq_put_budget,
1852 .queue_rq = scsi_queue_rq,
1853 .complete = scsi_complete,
1854 .timeout = scsi_timeout,
1855 #ifdef CONFIG_BLK_DEBUG_FS
1856 .show_rq = scsi_show_rq,
1858 .init_request = scsi_mq_init_request,
1859 .exit_request = scsi_mq_exit_request,
1860 .initialize_rq_fn = scsi_initialize_rq,
1861 .cleanup_rq = scsi_cleanup_rq,
1862 .busy = scsi_mq_lld_busy,
1863 .map_queues = scsi_map_queues,
1864 .init_hctx = scsi_init_hctx,
1865 .poll = scsi_mq_poll,
1866 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1867 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1871 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1873 struct Scsi_Host *shost = hctx->driver_data;
1875 shost->hostt->commit_rqs(shost, hctx->queue_num);
1878 static const struct blk_mq_ops scsi_mq_ops = {
1879 .get_budget = scsi_mq_get_budget,
1880 .put_budget = scsi_mq_put_budget,
1881 .queue_rq = scsi_queue_rq,
1882 .commit_rqs = scsi_commit_rqs,
1883 .complete = scsi_complete,
1884 .timeout = scsi_timeout,
1885 #ifdef CONFIG_BLK_DEBUG_FS
1886 .show_rq = scsi_show_rq,
1888 .init_request = scsi_mq_init_request,
1889 .exit_request = scsi_mq_exit_request,
1890 .initialize_rq_fn = scsi_initialize_rq,
1891 .cleanup_rq = scsi_cleanup_rq,
1892 .busy = scsi_mq_lld_busy,
1893 .map_queues = scsi_map_queues,
1894 .init_hctx = scsi_init_hctx,
1895 .poll = scsi_mq_poll,
1896 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1897 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1900 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1902 unsigned int cmd_size, sgl_size;
1903 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1905 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1906 scsi_mq_inline_sgl_size(shost));
1907 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1908 if (scsi_host_get_prot(shost))
1909 cmd_size += sizeof(struct scsi_data_buffer) +
1910 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1912 memset(tag_set, 0, sizeof(*tag_set));
1913 if (shost->hostt->commit_rqs)
1914 tag_set->ops = &scsi_mq_ops;
1916 tag_set->ops = &scsi_mq_ops_no_commit;
1917 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1918 tag_set->nr_maps = shost->nr_maps ? : 1;
1919 tag_set->queue_depth = shost->can_queue;
1920 tag_set->cmd_size = cmd_size;
1921 tag_set->numa_node = NUMA_NO_NODE;
1922 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1924 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1925 tag_set->driver_data = shost;
1926 if (shost->host_tagset)
1927 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1929 return blk_mq_alloc_tag_set(tag_set);
1932 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1934 blk_mq_free_tag_set(&shost->tag_set);
1938 * scsi_device_from_queue - return sdev associated with a request_queue
1939 * @q: The request queue to return the sdev from
1941 * Return the sdev associated with a request queue or NULL if the
1942 * request_queue does not reference a SCSI device.
1944 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1946 struct scsi_device *sdev = NULL;
1948 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1949 q->mq_ops == &scsi_mq_ops)
1950 sdev = q->queuedata;
1951 if (!sdev || !get_device(&sdev->sdev_gendev))
1958 * scsi_block_requests - Utility function used by low-level drivers to prevent
1959 * further commands from being queued to the device.
1960 * @shost: host in question
1962 * There is no timer nor any other means by which the requests get unblocked
1963 * other than the low-level driver calling scsi_unblock_requests().
1965 void scsi_block_requests(struct Scsi_Host *shost)
1967 shost->host_self_blocked = 1;
1969 EXPORT_SYMBOL(scsi_block_requests);
1972 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1973 * further commands to be 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(). This is done
1978 * as an API function so that changes to the internals of the scsi mid-layer
1979 * won't require wholesale changes to drivers that use this feature.
1981 void scsi_unblock_requests(struct Scsi_Host *shost)
1983 shost->host_self_blocked = 0;
1984 scsi_run_host_queues(shost);
1986 EXPORT_SYMBOL(scsi_unblock_requests);
1988 void scsi_exit_queue(void)
1990 kmem_cache_destroy(scsi_sense_cache);
1994 * scsi_mode_select - issue a mode select
1995 * @sdev: SCSI device to be queried
1996 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1997 * @sp: Save page bit (0 == don't save, 1 == save)
1998 * @modepage: mode page being requested
1999 * @buffer: request buffer (may not be smaller than eight bytes)
2000 * @len: length of request buffer.
2001 * @timeout: command timeout
2002 * @retries: number of retries before failing
2003 * @data: returns a structure abstracting the mode header data
2004 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2005 * must be SCSI_SENSE_BUFFERSIZE big.
2007 * Returns zero if successful; negative error number or scsi
2012 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2013 unsigned char *buffer, int len, int timeout, int retries,
2014 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2016 unsigned char cmd[10];
2017 unsigned char *real_buffer;
2020 memset(cmd, 0, sizeof(cmd));
2021 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2023 if (sdev->use_10_for_ms) {
2026 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2029 memcpy(real_buffer + 8, buffer, len);
2033 real_buffer[2] = data->medium_type;
2034 real_buffer[3] = data->device_specific;
2035 real_buffer[4] = data->longlba ? 0x01 : 0;
2037 real_buffer[6] = data->block_descriptor_length >> 8;
2038 real_buffer[7] = data->block_descriptor_length;
2040 cmd[0] = MODE_SELECT_10;
2044 if (len > 255 || data->block_descriptor_length > 255 ||
2048 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2051 memcpy(real_buffer + 4, buffer, len);
2054 real_buffer[1] = data->medium_type;
2055 real_buffer[2] = data->device_specific;
2056 real_buffer[3] = data->block_descriptor_length;
2058 cmd[0] = MODE_SELECT;
2062 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2063 sshdr, timeout, retries, NULL);
2067 EXPORT_SYMBOL_GPL(scsi_mode_select);
2070 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2071 * @sdev: SCSI device to be queried
2072 * @dbd: set if mode sense will allow block descriptors to be returned
2073 * @modepage: mode page being requested
2074 * @buffer: request buffer (may not be smaller than eight bytes)
2075 * @len: length of request buffer.
2076 * @timeout: command timeout
2077 * @retries: number of retries before failing
2078 * @data: returns a structure abstracting the mode header data
2079 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2080 * must be SCSI_SENSE_BUFFERSIZE big.
2082 * Returns zero if successful, or a negative error number on failure
2085 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2086 unsigned char *buffer, int len, int timeout, int retries,
2087 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2089 unsigned char cmd[12];
2092 int result, retry_count = retries;
2093 struct scsi_sense_hdr my_sshdr;
2095 memset(data, 0, sizeof(*data));
2096 memset(&cmd[0], 0, 12);
2098 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2099 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2102 /* caller might not be interested in sense, but we need it */
2107 use_10_for_ms = sdev->use_10_for_ms;
2109 if (use_10_for_ms) {
2113 cmd[0] = MODE_SENSE_10;
2120 cmd[0] = MODE_SENSE;
2125 memset(buffer, 0, len);
2127 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2128 sshdr, timeout, retries, NULL);
2132 /* This code looks awful: what it's doing is making sure an
2133 * ILLEGAL REQUEST sense return identifies the actual command
2134 * byte as the problem. MODE_SENSE commands can return
2135 * ILLEGAL REQUEST if the code page isn't supported */
2137 if (!scsi_status_is_good(result)) {
2138 if (scsi_sense_valid(sshdr)) {
2139 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2140 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2142 * Invalid command operation code
2144 if (use_10_for_ms) {
2145 sdev->use_10_for_ms = 0;
2149 if (scsi_status_is_check_condition(result) &&
2150 sshdr->sense_key == UNIT_ATTENTION &&
2158 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2159 (modepage == 6 || modepage == 8))) {
2160 /* Initio breakage? */
2163 data->medium_type = 0;
2164 data->device_specific = 0;
2166 data->block_descriptor_length = 0;
2167 } else if (use_10_for_ms) {
2168 data->length = buffer[0]*256 + buffer[1] + 2;
2169 data->medium_type = buffer[2];
2170 data->device_specific = buffer[3];
2171 data->longlba = buffer[4] & 0x01;
2172 data->block_descriptor_length = buffer[6]*256
2175 data->length = buffer[0] + 1;
2176 data->medium_type = buffer[1];
2177 data->device_specific = buffer[2];
2178 data->block_descriptor_length = buffer[3];
2180 data->header_length = header_length;
2184 EXPORT_SYMBOL(scsi_mode_sense);
2187 * scsi_test_unit_ready - test if unit is ready
2188 * @sdev: scsi device to change the state of.
2189 * @timeout: command timeout
2190 * @retries: number of retries before failing
2191 * @sshdr: outpout pointer for decoded sense information.
2193 * Returns zero if unsuccessful or an error if TUR failed. For
2194 * removable media, UNIT_ATTENTION sets ->changed flag.
2197 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2198 struct scsi_sense_hdr *sshdr)
2201 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2205 /* try to eat the UNIT_ATTENTION if there are enough retries */
2207 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2209 if (sdev->removable && scsi_sense_valid(sshdr) &&
2210 sshdr->sense_key == UNIT_ATTENTION)
2212 } while (scsi_sense_valid(sshdr) &&
2213 sshdr->sense_key == UNIT_ATTENTION && --retries);
2217 EXPORT_SYMBOL(scsi_test_unit_ready);
2220 * scsi_device_set_state - Take the given device through the device state model.
2221 * @sdev: scsi device to change the state of.
2222 * @state: state to change to.
2224 * Returns zero if successful or an error if the requested
2225 * transition is illegal.
2228 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2230 enum scsi_device_state oldstate = sdev->sdev_state;
2232 if (state == oldstate)
2238 case SDEV_CREATED_BLOCK:
2249 case SDEV_TRANSPORT_OFFLINE:
2262 case SDEV_TRANSPORT_OFFLINE:
2270 case SDEV_TRANSPORT_OFFLINE:
2285 case SDEV_CREATED_BLOCK:
2294 case SDEV_CREATED_BLOCK:
2309 case SDEV_TRANSPORT_OFFLINE:
2321 case SDEV_TRANSPORT_OFFLINE:
2324 case SDEV_CREATED_BLOCK:
2332 sdev->offline_already = false;
2333 sdev->sdev_state = state;
2337 SCSI_LOG_ERROR_RECOVERY(1,
2338 sdev_printk(KERN_ERR, sdev,
2339 "Illegal state transition %s->%s",
2340 scsi_device_state_name(oldstate),
2341 scsi_device_state_name(state))
2345 EXPORT_SYMBOL(scsi_device_set_state);
2348 * scsi_evt_emit - emit a single SCSI device uevent
2349 * @sdev: associated SCSI device
2350 * @evt: event to emit
2352 * Send a single uevent (scsi_event) to the associated scsi_device.
2354 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2359 switch (evt->evt_type) {
2360 case SDEV_EVT_MEDIA_CHANGE:
2361 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2363 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2364 scsi_rescan_device(&sdev->sdev_gendev);
2365 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2367 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2368 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2370 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2371 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2373 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2374 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2376 case SDEV_EVT_LUN_CHANGE_REPORTED:
2377 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2379 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2380 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2382 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2383 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2392 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2396 * scsi_evt_thread - send a uevent for each scsi event
2397 * @work: work struct for scsi_device
2399 * Dispatch queued events to their associated scsi_device kobjects
2402 void scsi_evt_thread(struct work_struct *work)
2404 struct scsi_device *sdev;
2405 enum scsi_device_event evt_type;
2406 LIST_HEAD(event_list);
2408 sdev = container_of(work, struct scsi_device, event_work);
2410 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2411 if (test_and_clear_bit(evt_type, sdev->pending_events))
2412 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2415 struct scsi_event *evt;
2416 struct list_head *this, *tmp;
2417 unsigned long flags;
2419 spin_lock_irqsave(&sdev->list_lock, flags);
2420 list_splice_init(&sdev->event_list, &event_list);
2421 spin_unlock_irqrestore(&sdev->list_lock, flags);
2423 if (list_empty(&event_list))
2426 list_for_each_safe(this, tmp, &event_list) {
2427 evt = list_entry(this, struct scsi_event, node);
2428 list_del(&evt->node);
2429 scsi_evt_emit(sdev, evt);
2436 * sdev_evt_send - send asserted event to uevent thread
2437 * @sdev: scsi_device event occurred on
2438 * @evt: event to send
2440 * Assert scsi device event asynchronously.
2442 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2444 unsigned long flags;
2447 /* FIXME: currently this check eliminates all media change events
2448 * for polled devices. Need to update to discriminate between AN
2449 * and polled events */
2450 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2456 spin_lock_irqsave(&sdev->list_lock, flags);
2457 list_add_tail(&evt->node, &sdev->event_list);
2458 schedule_work(&sdev->event_work);
2459 spin_unlock_irqrestore(&sdev->list_lock, flags);
2461 EXPORT_SYMBOL_GPL(sdev_evt_send);
2464 * sdev_evt_alloc - allocate a new scsi event
2465 * @evt_type: type of event to allocate
2466 * @gfpflags: GFP flags for allocation
2468 * Allocates and returns a new scsi_event.
2470 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2473 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2477 evt->evt_type = evt_type;
2478 INIT_LIST_HEAD(&evt->node);
2480 /* evt_type-specific initialization, if any */
2482 case SDEV_EVT_MEDIA_CHANGE:
2483 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2484 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2485 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2486 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2487 case SDEV_EVT_LUN_CHANGE_REPORTED:
2488 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2489 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2497 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2500 * sdev_evt_send_simple - send asserted event to uevent thread
2501 * @sdev: scsi_device event occurred on
2502 * @evt_type: type of event to send
2503 * @gfpflags: GFP flags for allocation
2505 * Assert scsi device event asynchronously, given an event type.
2507 void sdev_evt_send_simple(struct scsi_device *sdev,
2508 enum scsi_device_event evt_type, gfp_t gfpflags)
2510 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2512 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2517 sdev_evt_send(sdev, evt);
2519 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2522 * scsi_device_quiesce - Block all commands except power management.
2523 * @sdev: scsi device to quiesce.
2525 * This works by trying to transition to the SDEV_QUIESCE state
2526 * (which must be a legal transition). When the device is in this
2527 * state, only power management requests will be accepted, all others will
2530 * Must be called with user context, may sleep.
2532 * Returns zero if unsuccessful or an error if not.
2535 scsi_device_quiesce(struct scsi_device *sdev)
2537 struct request_queue *q = sdev->request_queue;
2541 * It is allowed to call scsi_device_quiesce() multiple times from
2542 * the same context but concurrent scsi_device_quiesce() calls are
2545 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2547 if (sdev->quiesced_by == current)
2552 blk_mq_freeze_queue(q);
2554 * Ensure that the effect of blk_set_pm_only() will be visible
2555 * for percpu_ref_tryget() callers that occur after the queue
2556 * unfreeze even if the queue was already frozen before this function
2557 * was called. See also https://lwn.net/Articles/573497/.
2560 blk_mq_unfreeze_queue(q);
2562 mutex_lock(&sdev->state_mutex);
2563 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2565 sdev->quiesced_by = current;
2567 blk_clear_pm_only(q);
2568 mutex_unlock(&sdev->state_mutex);
2572 EXPORT_SYMBOL(scsi_device_quiesce);
2575 * scsi_device_resume - Restart user issued commands to a quiesced device.
2576 * @sdev: scsi device to resume.
2578 * Moves the device from quiesced back to running and restarts the
2581 * Must be called with user context, may sleep.
2583 void scsi_device_resume(struct scsi_device *sdev)
2585 /* check if the device state was mutated prior to resume, and if
2586 * so assume the state is being managed elsewhere (for example
2587 * device deleted during suspend)
2589 mutex_lock(&sdev->state_mutex);
2590 if (sdev->sdev_state == SDEV_QUIESCE)
2591 scsi_device_set_state(sdev, SDEV_RUNNING);
2592 if (sdev->quiesced_by) {
2593 sdev->quiesced_by = NULL;
2594 blk_clear_pm_only(sdev->request_queue);
2596 mutex_unlock(&sdev->state_mutex);
2598 EXPORT_SYMBOL(scsi_device_resume);
2601 device_quiesce_fn(struct scsi_device *sdev, void *data)
2603 scsi_device_quiesce(sdev);
2607 scsi_target_quiesce(struct scsi_target *starget)
2609 starget_for_each_device(starget, NULL, device_quiesce_fn);
2611 EXPORT_SYMBOL(scsi_target_quiesce);
2614 device_resume_fn(struct scsi_device *sdev, void *data)
2616 scsi_device_resume(sdev);
2620 scsi_target_resume(struct scsi_target *starget)
2622 starget_for_each_device(starget, NULL, device_resume_fn);
2624 EXPORT_SYMBOL(scsi_target_resume);
2627 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2628 * @sdev: device to block
2630 * Pause SCSI command processing on the specified device. Does not sleep.
2632 * Returns zero if successful or a negative error code upon failure.
2635 * This routine transitions the device to the SDEV_BLOCK state (which must be
2636 * a legal transition). When the device is in this state, command processing
2637 * is paused until the device leaves the SDEV_BLOCK state. See also
2638 * scsi_internal_device_unblock_nowait().
2640 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2642 struct request_queue *q = sdev->request_queue;
2645 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2647 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2654 * The device has transitioned to SDEV_BLOCK. Stop the
2655 * block layer from calling the midlayer with this device's
2658 blk_mq_quiesce_queue_nowait(q);
2661 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2664 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2665 * @sdev: device to block
2667 * Pause SCSI command processing on the specified device and wait until all
2668 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2670 * Returns zero if successful or a negative error code upon failure.
2673 * This routine transitions the device to the SDEV_BLOCK state (which must be
2674 * a legal transition). When the device is in this state, command processing
2675 * is paused until the device leaves the SDEV_BLOCK state. See also
2676 * scsi_internal_device_unblock().
2678 static int scsi_internal_device_block(struct scsi_device *sdev)
2680 struct request_queue *q = sdev->request_queue;
2683 mutex_lock(&sdev->state_mutex);
2684 err = scsi_internal_device_block_nowait(sdev);
2686 blk_mq_quiesce_queue(q);
2687 mutex_unlock(&sdev->state_mutex);
2692 void scsi_start_queue(struct scsi_device *sdev)
2694 struct request_queue *q = sdev->request_queue;
2696 blk_mq_unquiesce_queue(q);
2700 * scsi_internal_device_unblock_nowait - resume a device after a block request
2701 * @sdev: device to resume
2702 * @new_state: state to set the device to after unblocking
2704 * Restart the device queue for a previously suspended SCSI device. Does not
2707 * Returns zero if successful or a negative error code upon failure.
2710 * This routine transitions the device to the SDEV_RUNNING state or to one of
2711 * the offline states (which must be a legal transition) allowing the midlayer
2712 * to goose the queue for this device.
2714 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2715 enum scsi_device_state new_state)
2717 switch (new_state) {
2719 case SDEV_TRANSPORT_OFFLINE:
2726 * Try to transition the scsi device to SDEV_RUNNING or one of the
2727 * offlined states and goose the device queue if successful.
2729 switch (sdev->sdev_state) {
2731 case SDEV_TRANSPORT_OFFLINE:
2732 sdev->sdev_state = new_state;
2734 case SDEV_CREATED_BLOCK:
2735 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2736 new_state == SDEV_OFFLINE)
2737 sdev->sdev_state = new_state;
2739 sdev->sdev_state = SDEV_CREATED;
2747 scsi_start_queue(sdev);
2751 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2754 * scsi_internal_device_unblock - resume a device after a block request
2755 * @sdev: device to resume
2756 * @new_state: state to set the device to after unblocking
2758 * Restart the device queue for a previously suspended SCSI device. May sleep.
2760 * Returns zero if successful or a negative error code upon failure.
2763 * This routine transitions the device to the SDEV_RUNNING state or to one of
2764 * the offline states (which must be a legal transition) allowing the midlayer
2765 * to goose the queue for this device.
2767 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2768 enum scsi_device_state new_state)
2772 mutex_lock(&sdev->state_mutex);
2773 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2774 mutex_unlock(&sdev->state_mutex);
2780 device_block(struct scsi_device *sdev, void *data)
2784 ret = scsi_internal_device_block(sdev);
2786 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2787 dev_name(&sdev->sdev_gendev), ret);
2791 target_block(struct device *dev, void *data)
2793 if (scsi_is_target_device(dev))
2794 starget_for_each_device(to_scsi_target(dev), NULL,
2800 scsi_target_block(struct device *dev)
2802 if (scsi_is_target_device(dev))
2803 starget_for_each_device(to_scsi_target(dev), NULL,
2806 device_for_each_child(dev, NULL, target_block);
2808 EXPORT_SYMBOL_GPL(scsi_target_block);
2811 device_unblock(struct scsi_device *sdev, void *data)
2813 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2817 target_unblock(struct device *dev, void *data)
2819 if (scsi_is_target_device(dev))
2820 starget_for_each_device(to_scsi_target(dev), data,
2826 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2828 if (scsi_is_target_device(dev))
2829 starget_for_each_device(to_scsi_target(dev), &new_state,
2832 device_for_each_child(dev, &new_state, target_unblock);
2834 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2837 scsi_host_block(struct Scsi_Host *shost)
2839 struct scsi_device *sdev;
2843 * Call scsi_internal_device_block_nowait so we can avoid
2844 * calling synchronize_rcu() for each LUN.
2846 shost_for_each_device(sdev, shost) {
2847 mutex_lock(&sdev->state_mutex);
2848 ret = scsi_internal_device_block_nowait(sdev);
2849 mutex_unlock(&sdev->state_mutex);
2851 scsi_device_put(sdev);
2857 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2858 * calling synchronize_rcu() once is enough.
2860 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2867 EXPORT_SYMBOL_GPL(scsi_host_block);
2870 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2872 struct scsi_device *sdev;
2875 shost_for_each_device(sdev, shost) {
2876 ret = scsi_internal_device_unblock(sdev, new_state);
2878 scsi_device_put(sdev);
2884 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2887 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2888 * @sgl: scatter-gather list
2889 * @sg_count: number of segments in sg
2890 * @offset: offset in bytes into sg, on return offset into the mapped area
2891 * @len: bytes to map, on return number of bytes mapped
2893 * Returns virtual address of the start of the mapped page
2895 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2896 size_t *offset, size_t *len)
2899 size_t sg_len = 0, len_complete = 0;
2900 struct scatterlist *sg;
2903 WARN_ON(!irqs_disabled());
2905 for_each_sg(sgl, sg, sg_count, i) {
2906 len_complete = sg_len; /* Complete sg-entries */
2907 sg_len += sg->length;
2908 if (sg_len > *offset)
2912 if (unlikely(i == sg_count)) {
2913 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2915 __func__, sg_len, *offset, sg_count);
2920 /* Offset starting from the beginning of first page in this sg-entry */
2921 *offset = *offset - len_complete + sg->offset;
2923 /* Assumption: contiguous pages can be accessed as "page + i" */
2924 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2925 *offset &= ~PAGE_MASK;
2927 /* Bytes in this sg-entry from *offset to the end of the page */
2928 sg_len = PAGE_SIZE - *offset;
2932 return kmap_atomic(page);
2934 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2937 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2938 * @virt: virtual address to be unmapped
2940 void scsi_kunmap_atomic_sg(void *virt)
2942 kunmap_atomic(virt);
2944 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2946 void sdev_disable_disk_events(struct scsi_device *sdev)
2948 atomic_inc(&sdev->disk_events_disable_depth);
2950 EXPORT_SYMBOL(sdev_disable_disk_events);
2952 void sdev_enable_disk_events(struct scsi_device *sdev)
2954 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2956 atomic_dec(&sdev->disk_events_disable_depth);
2958 EXPORT_SYMBOL(sdev_enable_disk_events);
2960 static unsigned char designator_prio(const unsigned char *d)
2963 /* not associated with LUN */
2967 /* invalid length */
2971 * Order of preference for lun descriptor:
2972 * - SCSI name string
2973 * - NAA IEEE Registered Extended
2974 * - EUI-64 based 16-byte
2975 * - EUI-64 based 12-byte
2976 * - NAA IEEE Registered
2977 * - NAA IEEE Extended
2978 * - EUI-64 based 8-byte
2979 * - SCSI name string (truncated)
2981 * as longer descriptors reduce the likelyhood
2982 * of identification clashes.
2985 switch (d[1] & 0xf) {
2987 /* SCSI name string, variable-length UTF-8 */
2990 switch (d[4] >> 4) {
2992 /* NAA registered extended */
2995 /* NAA registered */
3001 /* NAA locally assigned */
3010 /* EUI64-based, 16 byte */
3013 /* EUI64-based, 12 byte */
3016 /* EUI64-based, 8 byte */
3033 * scsi_vpd_lun_id - return a unique device identification
3034 * @sdev: SCSI device
3035 * @id: buffer for the identification
3036 * @id_len: length of the buffer
3038 * Copies a unique device identification into @id based
3039 * on the information in the VPD page 0x83 of the device.
3040 * The string will be formatted as a SCSI name string.
3042 * Returns the length of the identification or error on failure.
3043 * If the identifier is longer than the supplied buffer the actual
3044 * identifier length is returned and the buffer is not zero-padded.
3046 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3050 const unsigned char *d, *cur_id_str;
3051 const struct scsi_vpd *vpd_pg83;
3052 int id_size = -EINVAL;
3055 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3061 /* The id string must be at least 20 bytes + terminating NULL byte */
3067 memset(id, 0, id_len);
3068 for (d = vpd_pg83->data + 4;
3069 d < vpd_pg83->data + vpd_pg83->len;
3071 u8 prio = designator_prio(d);
3073 if (prio == 0 || cur_id_prio > prio)
3076 switch (d[1] & 0xf) {
3079 if (cur_id_size > d[3])
3083 if (cur_id_size + 4 > id_len)
3084 cur_id_size = id_len - 4;
3086 id_size = snprintf(id, id_len, "t10.%*pE",
3087 cur_id_size, cur_id_str);
3094 switch (cur_id_size) {
3096 id_size = snprintf(id, id_len,
3101 id_size = snprintf(id, id_len,
3106 id_size = snprintf(id, id_len,
3119 switch (cur_id_size) {
3121 id_size = snprintf(id, id_len,
3126 id_size = snprintf(id, id_len,
3135 /* SCSI name string */
3136 if (cur_id_size > d[3])
3138 /* Prefer others for truncated descriptor */
3139 if (d[3] > id_len) {
3141 if (cur_id_prio > prio)
3145 cur_id_size = id_size = d[3];
3147 if (cur_id_size >= id_len)
3148 cur_id_size = id_len - 1;
3149 memcpy(id, cur_id_str, cur_id_size);
3159 EXPORT_SYMBOL(scsi_vpd_lun_id);
3162 * scsi_vpd_tpg_id - return a target port group identifier
3163 * @sdev: SCSI device
3165 * Returns the Target Port Group identifier from the information
3166 * froom VPD page 0x83 of the device.
3168 * Returns the identifier or error on failure.
3170 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3172 const unsigned char *d;
3173 const struct scsi_vpd *vpd_pg83;
3174 int group_id = -EAGAIN, rel_port = -1;
3177 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3183 d = vpd_pg83->data + 4;
3184 while (d < vpd_pg83->data + vpd_pg83->len) {
3185 switch (d[1] & 0xf) {
3187 /* Relative target port */
3188 rel_port = get_unaligned_be16(&d[6]);
3191 /* Target port group */
3192 group_id = get_unaligned_be16(&d[6]);
3201 if (group_id >= 0 && rel_id && rel_port != -1)
3206 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3209 * scsi_build_sense - build sense data for a command
3210 * @scmd: scsi command for which the sense should be formatted
3211 * @desc: Sense format (non-zero == descriptor format,
3212 * 0 == fixed format)
3214 * @asc: Additional sense code
3215 * @ascq: Additional sense code qualifier
3218 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3220 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3221 scmd->result = SAM_STAT_CHECK_CONDITION;
3223 EXPORT_SYMBOL_GPL(scsi_build_sense);