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 struct kmem_cache *scsi_sense_isadma_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 static inline struct kmem_cache *
62 scsi_select_sense_cache(bool unchecked_isa_dma)
64 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
67 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
68 unsigned char *sense_buffer)
70 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
74 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
75 gfp_t gfp_mask, int numa_node)
77 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
81 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 struct kmem_cache *cache;
86 mutex_lock(&scsi_sense_cache_mutex);
87 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
91 if (shost->unchecked_isa_dma) {
92 scsi_sense_isadma_cache =
93 kmem_cache_create("scsi_sense_cache(DMA)",
94 SCSI_SENSE_BUFFERSIZE, 0,
95 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
96 if (!scsi_sense_isadma_cache)
100 kmem_cache_create_usercopy("scsi_sense_cache",
101 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
102 0, SCSI_SENSE_BUFFERSIZE, NULL);
103 if (!scsi_sense_cache)
107 mutex_unlock(&scsi_sense_cache_mutex);
112 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
113 * not change behaviour from the previous unplug mechanism, experimentation
114 * may prove this needs changing.
116 #define SCSI_QUEUE_DELAY 3
119 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 struct Scsi_Host *host = cmd->device->host;
122 struct scsi_device *device = cmd->device;
123 struct scsi_target *starget = scsi_target(device);
126 * Set the appropriate busy bit for the device/host.
128 * If the host/device isn't busy, assume that something actually
129 * completed, and that we should be able to queue a command now.
131 * Note that the prior mid-layer assumption that any host could
132 * always queue at least one command is now broken. The mid-layer
133 * will implement a user specifiable stall (see
134 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
135 * if a command is requeued with no other commands outstanding
136 * either for the device or for the host.
139 case SCSI_MLQUEUE_HOST_BUSY:
140 atomic_set(&host->host_blocked, host->max_host_blocked);
142 case SCSI_MLQUEUE_DEVICE_BUSY:
143 case SCSI_MLQUEUE_EH_RETRY:
144 atomic_set(&device->device_blocked,
145 device->max_device_blocked);
147 case SCSI_MLQUEUE_TARGET_BUSY:
148 atomic_set(&starget->target_blocked,
149 starget->max_target_blocked);
154 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 if (cmd->request->rq_flags & RQF_DONTPREP) {
157 cmd->request->rq_flags &= ~RQF_DONTPREP;
158 scsi_mq_uninit_cmd(cmd);
162 blk_mq_requeue_request(cmd->request, true);
166 * __scsi_queue_insert - private queue insertion
167 * @cmd: The SCSI command being requeued
168 * @reason: The reason for the requeue
169 * @unbusy: Whether the queue should be unbusied
171 * This is a private queue insertion. The public interface
172 * scsi_queue_insert() always assumes the queue should be unbusied
173 * because it's always called before the completion. This function is
174 * for a requeue after completion, which should only occur in this
177 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 struct scsi_device *device = cmd->device;
181 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
182 "Inserting command %p into mlqueue\n", cmd));
184 scsi_set_blocked(cmd, reason);
187 * Decrement the counters, since these commands are no longer
188 * active on the host/device.
191 scsi_device_unbusy(device, cmd);
194 * Requeue this command. It will go before all other commands
195 * that are already in the queue. Schedule requeue work under
196 * lock such that the kblockd_schedule_work() call happens
197 * before blk_cleanup_queue() finishes.
201 blk_mq_requeue_request(cmd->request, true);
205 * scsi_queue_insert - Reinsert a command in the queue.
206 * @cmd: command that we are adding to queue.
207 * @reason: why we are inserting command to queue.
209 * We do this for one of two cases. Either the host is busy and it cannot accept
210 * any more commands for the time being, or the device returned QUEUE_FULL and
211 * can accept no more commands.
213 * Context: This could be called either from an interrupt context or a normal
216 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
218 __scsi_queue_insert(cmd, reason, true);
223 * __scsi_execute - insert request and wait for the result
226 * @data_direction: data direction
227 * @buffer: data buffer
228 * @bufflen: len of buffer
229 * @sense: optional sense buffer
230 * @sshdr: optional decoded sense header
231 * @timeout: request timeout in seconds
232 * @retries: number of times to retry request
233 * @flags: flags for ->cmd_flags
234 * @rq_flags: flags for ->rq_flags
235 * @resid: optional residual length
237 * Returns the scsi_cmnd result field if a command was executed, or a negative
238 * Linux error code if we didn't get that far.
240 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
241 int data_direction, void *buffer, unsigned bufflen,
242 unsigned char *sense, struct scsi_sense_hdr *sshdr,
243 int timeout, int retries, u64 flags, req_flags_t rq_flags,
247 struct scsi_request *rq;
248 int ret = DRIVER_ERROR << 24;
250 req = blk_get_request(sdev->request_queue,
251 data_direction == DMA_TO_DEVICE ?
252 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
257 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
258 buffer, bufflen, GFP_NOIO))
261 rq->cmd_len = COMMAND_SIZE(cmd[0]);
262 memcpy(rq->cmd, cmd, rq->cmd_len);
263 rq->retries = retries;
264 req->timeout = timeout;
265 req->cmd_flags |= flags;
266 req->rq_flags |= rq_flags | RQF_QUIET;
269 * head injection *required* here otherwise quiesce won't work
271 blk_execute_rq(req->q, NULL, req, 1);
274 * Some devices (USB mass-storage in particular) may transfer
275 * garbage data together with a residue indicating that the data
276 * is invalid. Prevent the garbage from being misinterpreted
277 * and prevent security leaks by zeroing out the excess data.
279 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
280 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
283 *resid = rq->resid_len;
284 if (sense && rq->sense_len)
285 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
287 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
290 blk_put_request(req);
294 EXPORT_SYMBOL(__scsi_execute);
297 * scsi_init_cmd_errh - Initialize cmd fields related to error handling.
298 * @cmd: command that is ready to be queued.
300 * This function has the job of initializing a number of fields related to error
301 * handling. Typically this will be called once for each command, as required.
303 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
305 scsi_set_resid(cmd, 0);
306 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
307 if (cmd->cmd_len == 0)
308 cmd->cmd_len = scsi_command_size(cmd->cmnd);
312 * Wake up the error handler if necessary. Avoid as follows that the error
313 * handler is not woken up if host in-flight requests number ==
314 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
315 * with an RCU read lock in this function to ensure that this function in
316 * its entirety either finishes before scsi_eh_scmd_add() increases the
317 * host_failed counter or that it notices the shost state change made by
318 * scsi_eh_scmd_add().
320 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
325 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
326 if (unlikely(scsi_host_in_recovery(shost))) {
327 spin_lock_irqsave(shost->host_lock, flags);
328 if (shost->host_failed || shost->host_eh_scheduled)
329 scsi_eh_wakeup(shost);
330 spin_unlock_irqrestore(shost->host_lock, flags);
335 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
337 struct Scsi_Host *shost = sdev->host;
338 struct scsi_target *starget = scsi_target(sdev);
340 scsi_dec_host_busy(shost, cmd);
342 if (starget->can_queue > 0)
343 atomic_dec(&starget->target_busy);
345 atomic_dec(&sdev->device_busy);
348 static void scsi_kick_queue(struct request_queue *q)
350 blk_mq_run_hw_queues(q, false);
354 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
355 * and call blk_run_queue for all the scsi_devices on the target -
356 * including current_sdev first.
358 * Called with *no* scsi locks held.
360 static void scsi_single_lun_run(struct scsi_device *current_sdev)
362 struct Scsi_Host *shost = current_sdev->host;
363 struct scsi_device *sdev, *tmp;
364 struct scsi_target *starget = scsi_target(current_sdev);
367 spin_lock_irqsave(shost->host_lock, flags);
368 starget->starget_sdev_user = NULL;
369 spin_unlock_irqrestore(shost->host_lock, flags);
372 * Call blk_run_queue for all LUNs on the target, starting with
373 * current_sdev. We race with others (to set starget_sdev_user),
374 * but in most cases, we will be first. Ideally, each LU on the
375 * target would get some limited time or requests on the target.
377 scsi_kick_queue(current_sdev->request_queue);
379 spin_lock_irqsave(shost->host_lock, flags);
380 if (starget->starget_sdev_user)
382 list_for_each_entry_safe(sdev, tmp, &starget->devices,
383 same_target_siblings) {
384 if (sdev == current_sdev)
386 if (scsi_device_get(sdev))
389 spin_unlock_irqrestore(shost->host_lock, flags);
390 scsi_kick_queue(sdev->request_queue);
391 spin_lock_irqsave(shost->host_lock, flags);
393 scsi_device_put(sdev);
396 spin_unlock_irqrestore(shost->host_lock, flags);
399 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
401 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
403 if (atomic_read(&sdev->device_blocked) > 0)
408 static inline bool scsi_target_is_busy(struct scsi_target *starget)
410 if (starget->can_queue > 0) {
411 if (atomic_read(&starget->target_busy) >= starget->can_queue)
413 if (atomic_read(&starget->target_blocked) > 0)
419 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
421 if (atomic_read(&shost->host_blocked) > 0)
423 if (shost->host_self_blocked)
428 static void scsi_starved_list_run(struct Scsi_Host *shost)
430 LIST_HEAD(starved_list);
431 struct scsi_device *sdev;
434 spin_lock_irqsave(shost->host_lock, flags);
435 list_splice_init(&shost->starved_list, &starved_list);
437 while (!list_empty(&starved_list)) {
438 struct request_queue *slq;
441 * As long as shost is accepting commands and we have
442 * starved queues, call blk_run_queue. scsi_request_fn
443 * drops the queue_lock and can add us back to the
446 * host_lock protects the starved_list and starved_entry.
447 * scsi_request_fn must get the host_lock before checking
448 * or modifying starved_list or starved_entry.
450 if (scsi_host_is_busy(shost))
453 sdev = list_entry(starved_list.next,
454 struct scsi_device, starved_entry);
455 list_del_init(&sdev->starved_entry);
456 if (scsi_target_is_busy(scsi_target(sdev))) {
457 list_move_tail(&sdev->starved_entry,
458 &shost->starved_list);
463 * Once we drop the host lock, a racing scsi_remove_device()
464 * call may remove the sdev from the starved list and destroy
465 * it and the queue. Mitigate by taking a reference to the
466 * queue and never touching the sdev again after we drop the
467 * host lock. Note: if __scsi_remove_device() invokes
468 * blk_cleanup_queue() before the queue is run from this
469 * function then blk_run_queue() will return immediately since
470 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
472 slq = sdev->request_queue;
473 if (!blk_get_queue(slq))
475 spin_unlock_irqrestore(shost->host_lock, flags);
477 scsi_kick_queue(slq);
480 spin_lock_irqsave(shost->host_lock, flags);
482 /* put any unprocessed entries back */
483 list_splice(&starved_list, &shost->starved_list);
484 spin_unlock_irqrestore(shost->host_lock, flags);
488 * scsi_run_queue - Select a proper request queue to serve next.
489 * @q: last request's queue
491 * The previous command was completely finished, start a new one if possible.
493 static void scsi_run_queue(struct request_queue *q)
495 struct scsi_device *sdev = q->queuedata;
497 if (scsi_target(sdev)->single_lun)
498 scsi_single_lun_run(sdev);
499 if (!list_empty(&sdev->host->starved_list))
500 scsi_starved_list_run(sdev->host);
502 blk_mq_run_hw_queues(q, false);
505 void scsi_requeue_run_queue(struct work_struct *work)
507 struct scsi_device *sdev;
508 struct request_queue *q;
510 sdev = container_of(work, struct scsi_device, requeue_work);
511 q = sdev->request_queue;
515 void scsi_run_host_queues(struct Scsi_Host *shost)
517 struct scsi_device *sdev;
519 shost_for_each_device(sdev, shost)
520 scsi_run_queue(sdev->request_queue);
523 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
525 if (!blk_rq_is_passthrough(cmd->request)) {
526 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
528 if (drv->uninit_command)
529 drv->uninit_command(cmd);
533 static void scsi_free_sgtables(struct scsi_cmnd *cmd)
535 if (cmd->sdb.table.nents)
536 sg_free_table_chained(&cmd->sdb.table,
538 if (scsi_prot_sg_count(cmd))
539 sg_free_table_chained(&cmd->prot_sdb->table,
540 SCSI_INLINE_PROT_SG_CNT);
543 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
545 scsi_free_sgtables(cmd);
546 scsi_uninit_cmd(cmd);
549 static void scsi_run_queue_async(struct scsi_device *sdev)
551 if (scsi_target(sdev)->single_lun ||
552 !list_empty(&sdev->host->starved_list)) {
553 kblockd_schedule_work(&sdev->requeue_work);
556 * smp_mb() present in sbitmap_queue_clear() or implied in
557 * .end_io is for ordering writing .device_busy in
558 * scsi_device_unbusy() and reading sdev->restarts.
560 int old = atomic_read(&sdev->restarts);
563 * ->restarts has to be kept as non-zero if new budget
566 * No need to run queue when either another re-run
567 * queue wins in updating ->restarts or a new budget
570 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
571 blk_mq_run_hw_queues(sdev->request_queue, true);
575 /* Returns false when no more bytes to process, true if there are more */
576 static bool scsi_end_request(struct request *req, blk_status_t error,
579 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
580 struct scsi_device *sdev = cmd->device;
581 struct request_queue *q = sdev->request_queue;
583 if (blk_update_request(req, error, bytes))
586 if (blk_queue_add_random(q))
587 add_disk_randomness(req->rq_disk);
589 if (!blk_rq_is_scsi(req)) {
590 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
591 cmd->flags &= ~SCMD_INITIALIZED;
595 * Calling rcu_barrier() is not necessary here because the
596 * SCSI error handler guarantees that the function called by
597 * call_rcu() has been called before scsi_end_request() is
600 destroy_rcu_head(&cmd->rcu);
603 * In the MQ case the command gets freed by __blk_mq_end_request,
604 * so we have to do all cleanup that depends on it earlier.
606 * We also can't kick the queues from irq context, so we
607 * will have to defer it to a workqueue.
609 scsi_mq_uninit_cmd(cmd);
612 * queue is still alive, so grab the ref for preventing it
613 * from being cleaned up during running queue.
615 percpu_ref_get(&q->q_usage_counter);
617 __blk_mq_end_request(req, error);
619 scsi_run_queue_async(sdev);
621 percpu_ref_put(&q->q_usage_counter);
626 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
628 * @result: scsi error code
630 * Translate a SCSI result code into a blk_status_t value. May reset the host
631 * byte of @cmd->result.
633 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
635 switch (host_byte(result)) {
638 * Also check the other bytes than the status byte in result
639 * to handle the case when a SCSI LLD sets result to
640 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
642 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
644 return BLK_STS_IOERR;
645 case DID_TRANSPORT_FAILFAST:
646 return BLK_STS_TRANSPORT;
647 case DID_TARGET_FAILURE:
648 set_host_byte(cmd, DID_OK);
649 return BLK_STS_TARGET;
650 case DID_NEXUS_FAILURE:
651 set_host_byte(cmd, DID_OK);
652 return BLK_STS_NEXUS;
653 case DID_ALLOC_FAILURE:
654 set_host_byte(cmd, DID_OK);
655 return BLK_STS_NOSPC;
656 case DID_MEDIUM_ERROR:
657 set_host_byte(cmd, DID_OK);
658 return BLK_STS_MEDIUM;
660 return BLK_STS_IOERR;
664 /* Helper for scsi_io_completion() when "reprep" action required. */
665 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
666 struct request_queue *q)
668 /* A new command will be prepared and issued. */
669 scsi_mq_requeue_cmd(cmd);
672 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
674 struct request *req = cmd->request;
675 unsigned long wait_for;
677 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
680 wait_for = (cmd->allowed + 1) * req->timeout;
681 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
682 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
689 /* Helper for scsi_io_completion() when special action required. */
690 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
692 struct request_queue *q = cmd->device->request_queue;
693 struct request *req = cmd->request;
695 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
696 ACTION_DELAYED_RETRY} action;
697 struct scsi_sense_hdr sshdr;
699 bool sense_current = true; /* false implies "deferred sense" */
700 blk_status_t blk_stat;
702 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
704 sense_current = !scsi_sense_is_deferred(&sshdr);
706 blk_stat = scsi_result_to_blk_status(cmd, result);
708 if (host_byte(result) == DID_RESET) {
709 /* Third party bus reset or reset for error recovery
710 * reasons. Just retry the command and see what
713 action = ACTION_RETRY;
714 } else if (sense_valid && sense_current) {
715 switch (sshdr.sense_key) {
717 if (cmd->device->removable) {
718 /* Detected disc change. Set a bit
719 * and quietly refuse further access.
721 cmd->device->changed = 1;
722 action = ACTION_FAIL;
724 /* Must have been a power glitch, or a
725 * bus reset. Could not have been a
726 * media change, so we just retry the
727 * command and see what happens.
729 action = ACTION_RETRY;
732 case ILLEGAL_REQUEST:
733 /* If we had an ILLEGAL REQUEST returned, then
734 * we may have performed an unsupported
735 * command. The only thing this should be
736 * would be a ten byte read where only a six
737 * byte read was supported. Also, on a system
738 * where READ CAPACITY failed, we may have
739 * read past the end of the disk.
741 if ((cmd->device->use_10_for_rw &&
742 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
743 (cmd->cmnd[0] == READ_10 ||
744 cmd->cmnd[0] == WRITE_10)) {
745 /* This will issue a new 6-byte command. */
746 cmd->device->use_10_for_rw = 0;
747 action = ACTION_REPREP;
748 } else if (sshdr.asc == 0x10) /* DIX */ {
749 action = ACTION_FAIL;
750 blk_stat = BLK_STS_PROTECTION;
751 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
752 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
753 action = ACTION_FAIL;
754 blk_stat = BLK_STS_TARGET;
756 action = ACTION_FAIL;
758 case ABORTED_COMMAND:
759 action = ACTION_FAIL;
760 if (sshdr.asc == 0x10) /* DIF */
761 blk_stat = BLK_STS_PROTECTION;
764 /* If the device is in the process of becoming
765 * ready, or has a temporary blockage, retry.
767 if (sshdr.asc == 0x04) {
768 switch (sshdr.ascq) {
769 case 0x01: /* becoming ready */
770 case 0x04: /* format in progress */
771 case 0x05: /* rebuild in progress */
772 case 0x06: /* recalculation in progress */
773 case 0x07: /* operation in progress */
774 case 0x08: /* Long write in progress */
775 case 0x09: /* self test in progress */
776 case 0x14: /* space allocation in progress */
777 case 0x1a: /* start stop unit in progress */
778 case 0x1b: /* sanitize in progress */
779 case 0x1d: /* configuration in progress */
780 case 0x24: /* depopulation in progress */
781 action = ACTION_DELAYED_RETRY;
784 action = ACTION_FAIL;
788 action = ACTION_FAIL;
790 case VOLUME_OVERFLOW:
791 /* See SSC3rXX or current. */
792 action = ACTION_FAIL;
795 action = ACTION_FAIL;
799 action = ACTION_FAIL;
801 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
802 action = ACTION_FAIL;
806 /* Give up and fail the remainder of the request */
807 if (!(req->rq_flags & RQF_QUIET)) {
808 static DEFINE_RATELIMIT_STATE(_rs,
809 DEFAULT_RATELIMIT_INTERVAL,
810 DEFAULT_RATELIMIT_BURST);
812 if (unlikely(scsi_logging_level))
814 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
815 SCSI_LOG_MLCOMPLETE_BITS);
818 * if logging is enabled the failure will be printed
819 * in scsi_log_completion(), so avoid duplicate messages
821 if (!level && __ratelimit(&_rs)) {
822 scsi_print_result(cmd, NULL, FAILED);
823 if (driver_byte(result) == DRIVER_SENSE)
824 scsi_print_sense(cmd);
825 scsi_print_command(cmd);
828 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
832 scsi_io_completion_reprep(cmd, q);
835 /* Retry the same command immediately */
836 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
838 case ACTION_DELAYED_RETRY:
839 /* Retry the same command after a delay */
840 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
846 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
847 * new result that may suppress further error checking. Also modifies
848 * *blk_statp in some cases.
850 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
851 blk_status_t *blk_statp)
854 bool sense_current = true; /* false implies "deferred sense" */
855 struct request *req = cmd->request;
856 struct scsi_sense_hdr sshdr;
858 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
860 sense_current = !scsi_sense_is_deferred(&sshdr);
862 if (blk_rq_is_passthrough(req)) {
865 * SG_IO wants current and deferred errors
867 scsi_req(req)->sense_len =
868 min(8 + cmd->sense_buffer[7],
869 SCSI_SENSE_BUFFERSIZE);
872 *blk_statp = scsi_result_to_blk_status(cmd, result);
873 } else if (blk_rq_bytes(req) == 0 && sense_current) {
875 * Flush commands do not transfers any data, and thus cannot use
876 * good_bytes != blk_rq_bytes(req) as the signal for an error.
877 * This sets *blk_statp explicitly for the problem case.
879 *blk_statp = scsi_result_to_blk_status(cmd, result);
882 * Recovered errors need reporting, but they're always treated as
883 * success, so fiddle the result code here. For passthrough requests
884 * we already took a copy of the original into sreq->result which
885 * is what gets returned to the user
887 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
888 bool do_print = true;
890 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
891 * skip print since caller wants ATA registers. Only occurs
892 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
894 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
896 else if (req->rq_flags & RQF_QUIET)
899 scsi_print_sense(cmd);
901 /* for passthrough, *blk_statp may be set */
902 *blk_statp = BLK_STS_OK;
905 * Another corner case: the SCSI status byte is non-zero but 'good'.
906 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
907 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
908 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
909 * intermediate statuses (both obsolete in SAM-4) as good.
911 if (status_byte(result) && scsi_status_is_good(result)) {
913 *blk_statp = BLK_STS_OK;
919 * scsi_io_completion - Completion processing for SCSI commands.
920 * @cmd: command that is finished.
921 * @good_bytes: number of processed bytes.
923 * We will finish off the specified number of sectors. If we are done, the
924 * command block will be released and the queue function will be goosed. If we
925 * are not done then we have to figure out what to do next:
927 * a) We can call scsi_io_completion_reprep(). The request will be
928 * unprepared and put back on the queue. Then a new command will
929 * be created for it. This should be used if we made forward
930 * progress, or if we want to switch from READ(10) to READ(6) for
933 * b) We can call scsi_io_completion_action(). The request will be
934 * put back on the queue and retried using the same command as
935 * before, possibly after a delay.
937 * c) We can call scsi_end_request() with blk_stat other than
938 * BLK_STS_OK, to fail the remainder of the request.
940 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
942 int result = cmd->result;
943 struct request_queue *q = cmd->device->request_queue;
944 struct request *req = cmd->request;
945 blk_status_t blk_stat = BLK_STS_OK;
947 if (unlikely(result)) /* a nz result may or may not be an error */
948 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
950 if (unlikely(blk_rq_is_passthrough(req))) {
952 * scsi_result_to_blk_status may have reset the host_byte
954 scsi_req(req)->result = cmd->result;
958 * Next deal with any sectors which we were able to correctly
961 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
962 "%u sectors total, %d bytes done.\n",
963 blk_rq_sectors(req), good_bytes));
966 * Failed, zero length commands always need to drop down
967 * to retry code. Fast path should return in this block.
969 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
970 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
971 return; /* no bytes remaining */
974 /* Kill remainder if no retries. */
975 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
976 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
978 "Bytes remaining after failed, no-retry command");
983 * If there had been no error, but we have leftover bytes in the
984 * requeues just queue the command up again.
986 if (likely(result == 0))
987 scsi_io_completion_reprep(cmd, q);
989 scsi_io_completion_action(cmd, result);
992 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
995 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
996 !op_is_write(req_op(rq)) &&
997 sdev->host->hostt->dma_need_drain(rq);
1001 * scsi_init_io - SCSI I/O initialization function.
1002 * @cmd: command descriptor we wish to initialize
1005 * * BLK_STS_OK - on success
1006 * * BLK_STS_RESOURCE - if the failure is retryable
1007 * * BLK_STS_IOERR - if the failure is fatal
1009 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1011 struct scsi_device *sdev = cmd->device;
1012 struct request *rq = cmd->request;
1013 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1014 struct scatterlist *last_sg = NULL;
1016 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1019 if (WARN_ON_ONCE(!nr_segs))
1020 return BLK_STS_IOERR;
1023 * Make sure there is space for the drain. The driver must adjust
1024 * max_hw_segments to be prepared for this.
1030 * If sg table allocation fails, requeue request later.
1032 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1033 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1034 return BLK_STS_RESOURCE;
1037 * Next, walk the list, and fill in the addresses and sizes of
1040 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1042 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1043 unsigned int pad_len =
1044 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1046 last_sg->length += pad_len;
1047 cmd->extra_len += pad_len;
1051 sg_unmark_end(last_sg);
1052 last_sg = sg_next(last_sg);
1053 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1054 sg_mark_end(last_sg);
1056 cmd->extra_len += sdev->dma_drain_len;
1060 BUG_ON(count > cmd->sdb.table.nents);
1061 cmd->sdb.table.nents = count;
1062 cmd->sdb.length = blk_rq_payload_bytes(rq);
1064 if (blk_integrity_rq(rq)) {
1065 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1068 if (WARN_ON_ONCE(!prot_sdb)) {
1070 * This can happen if someone (e.g. multipath)
1071 * queues a command to a device on an adapter
1072 * that does not support DIX.
1074 ret = BLK_STS_IOERR;
1075 goto out_free_sgtables;
1078 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1080 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1081 prot_sdb->table.sgl,
1082 SCSI_INLINE_PROT_SG_CNT)) {
1083 ret = BLK_STS_RESOURCE;
1084 goto out_free_sgtables;
1087 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1088 prot_sdb->table.sgl);
1089 BUG_ON(count > ivecs);
1090 BUG_ON(count > queue_max_integrity_segments(rq->q));
1092 cmd->prot_sdb = prot_sdb;
1093 cmd->prot_sdb->table.nents = count;
1098 scsi_free_sgtables(cmd);
1101 EXPORT_SYMBOL(scsi_init_io);
1104 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1105 * @rq: Request associated with the SCSI command to be initialized.
1107 * This function initializes the members of struct scsi_cmnd that must be
1108 * initialized before request processing starts and that won't be
1109 * reinitialized if a SCSI command is requeued.
1111 * Called from inside blk_get_request() for pass-through requests and from
1112 * inside scsi_init_command() for filesystem requests.
1114 static void scsi_initialize_rq(struct request *rq)
1116 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1118 scsi_req_init(&cmd->req);
1119 init_rcu_head(&cmd->rcu);
1120 cmd->jiffies_at_alloc = jiffies;
1125 * Only called when the request isn't completed by SCSI, and not freed by
1128 static void scsi_cleanup_rq(struct request *rq)
1130 if (rq->rq_flags & RQF_DONTPREP) {
1131 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1132 rq->rq_flags &= ~RQF_DONTPREP;
1136 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1137 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1139 void *buf = cmd->sense_buffer;
1140 void *prot = cmd->prot_sdb;
1141 struct request *rq = blk_mq_rq_from_pdu(cmd);
1142 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1143 unsigned long jiffies_at_alloc;
1144 int retries, to_clear;
1147 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1148 flags |= SCMD_INITIALIZED;
1149 scsi_initialize_rq(rq);
1152 jiffies_at_alloc = cmd->jiffies_at_alloc;
1153 retries = cmd->retries;
1154 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1156 * Zero out the cmd, except for the embedded scsi_request. Only clear
1157 * the driver-private command data if the LLD does not supply a
1158 * function to initialize that data.
1160 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1161 if (!dev->host->hostt->init_cmd_priv)
1162 to_clear += dev->host->hostt->cmd_size;
1163 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1166 cmd->sense_buffer = buf;
1167 cmd->prot_sdb = prot;
1169 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1170 cmd->jiffies_at_alloc = jiffies_at_alloc;
1171 cmd->retries = retries;
1173 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1177 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1178 struct request *req)
1180 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1183 * Passthrough requests may transfer data, in which case they must
1184 * a bio attached to them. Or they might contain a SCSI command
1185 * that does not transfer data, in which case they may optionally
1186 * submit a request without an attached bio.
1189 blk_status_t ret = scsi_init_io(cmd);
1190 if (unlikely(ret != BLK_STS_OK))
1193 BUG_ON(blk_rq_bytes(req));
1195 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1198 cmd->cmd_len = scsi_req(req)->cmd_len;
1199 cmd->cmnd = scsi_req(req)->cmd;
1200 cmd->transfersize = blk_rq_bytes(req);
1201 cmd->allowed = scsi_req(req)->retries;
1206 * Setup a normal block command. These are simple request from filesystems
1207 * that still need to be translated to SCSI CDBs from the ULD.
1209 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1210 struct request *req)
1212 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1214 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1215 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1216 if (ret != BLK_STS_OK)
1220 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1221 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1222 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1225 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1226 struct request *req)
1228 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1231 if (!blk_rq_bytes(req))
1232 cmd->sc_data_direction = DMA_NONE;
1233 else if (rq_data_dir(req) == WRITE)
1234 cmd->sc_data_direction = DMA_TO_DEVICE;
1236 cmd->sc_data_direction = DMA_FROM_DEVICE;
1238 if (blk_rq_is_scsi(req))
1239 ret = scsi_setup_scsi_cmnd(sdev, req);
1241 ret = scsi_setup_fs_cmnd(sdev, req);
1243 if (ret != BLK_STS_OK)
1244 scsi_free_sgtables(cmd);
1250 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1252 switch (sdev->sdev_state) {
1254 case SDEV_TRANSPORT_OFFLINE:
1256 * If the device is offline we refuse to process any
1257 * commands. The device must be brought online
1258 * before trying any recovery commands.
1260 if (!sdev->offline_already) {
1261 sdev->offline_already = true;
1262 sdev_printk(KERN_ERR, sdev,
1263 "rejecting I/O to offline device\n");
1265 return BLK_STS_IOERR;
1268 * If the device is fully deleted, we refuse to
1269 * process any commands as well.
1271 sdev_printk(KERN_ERR, sdev,
1272 "rejecting I/O to dead device\n");
1273 return BLK_STS_IOERR;
1275 case SDEV_CREATED_BLOCK:
1276 return BLK_STS_RESOURCE;
1279 * If the devices is blocked we defer normal commands.
1281 if (req && !(req->rq_flags & RQF_PREEMPT))
1282 return BLK_STS_RESOURCE;
1286 * For any other not fully online state we only allow
1287 * special commands. In particular any user initiated
1288 * command is not allowed.
1290 if (req && !(req->rq_flags & RQF_PREEMPT))
1291 return BLK_STS_IOERR;
1297 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1300 * Called with the queue_lock held.
1302 static inline int scsi_dev_queue_ready(struct request_queue *q,
1303 struct scsi_device *sdev)
1307 busy = atomic_inc_return(&sdev->device_busy) - 1;
1308 if (atomic_read(&sdev->device_blocked)) {
1313 * unblock after device_blocked iterates to zero
1315 if (atomic_dec_return(&sdev->device_blocked) > 0)
1317 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1318 "unblocking device at zero depth\n"));
1321 if (busy >= sdev->queue_depth)
1326 atomic_dec(&sdev->device_busy);
1331 * scsi_target_queue_ready: checks if there we can send commands to target
1332 * @sdev: scsi device on starget to check.
1334 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1335 struct scsi_device *sdev)
1337 struct scsi_target *starget = scsi_target(sdev);
1340 if (starget->single_lun) {
1341 spin_lock_irq(shost->host_lock);
1342 if (starget->starget_sdev_user &&
1343 starget->starget_sdev_user != sdev) {
1344 spin_unlock_irq(shost->host_lock);
1347 starget->starget_sdev_user = sdev;
1348 spin_unlock_irq(shost->host_lock);
1351 if (starget->can_queue <= 0)
1354 busy = atomic_inc_return(&starget->target_busy) - 1;
1355 if (atomic_read(&starget->target_blocked) > 0) {
1360 * unblock after target_blocked iterates to zero
1362 if (atomic_dec_return(&starget->target_blocked) > 0)
1365 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1366 "unblocking target at zero depth\n"));
1369 if (busy >= starget->can_queue)
1375 spin_lock_irq(shost->host_lock);
1376 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1377 spin_unlock_irq(shost->host_lock);
1379 if (starget->can_queue > 0)
1380 atomic_dec(&starget->target_busy);
1385 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1386 * return 0. We must end up running the queue again whenever 0 is
1387 * returned, else IO can hang.
1389 static inline int scsi_host_queue_ready(struct request_queue *q,
1390 struct Scsi_Host *shost,
1391 struct scsi_device *sdev,
1392 struct scsi_cmnd *cmd)
1394 if (scsi_host_in_recovery(shost))
1397 if (atomic_read(&shost->host_blocked) > 0) {
1398 if (scsi_host_busy(shost) > 0)
1402 * unblock after host_blocked iterates to zero
1404 if (atomic_dec_return(&shost->host_blocked) > 0)
1408 shost_printk(KERN_INFO, shost,
1409 "unblocking host at zero depth\n"));
1412 if (shost->host_self_blocked)
1415 /* We're OK to process the command, so we can't be starved */
1416 if (!list_empty(&sdev->starved_entry)) {
1417 spin_lock_irq(shost->host_lock);
1418 if (!list_empty(&sdev->starved_entry))
1419 list_del_init(&sdev->starved_entry);
1420 spin_unlock_irq(shost->host_lock);
1423 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1428 spin_lock_irq(shost->host_lock);
1429 if (list_empty(&sdev->starved_entry))
1430 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1431 spin_unlock_irq(shost->host_lock);
1433 scsi_dec_host_busy(shost, cmd);
1438 * Busy state exporting function for request stacking drivers.
1440 * For efficiency, no lock is taken to check the busy state of
1441 * shost/starget/sdev, since the returned value is not guaranteed and
1442 * may be changed after request stacking drivers call the function,
1443 * regardless of taking lock or not.
1445 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1446 * needs to return 'not busy'. Otherwise, request stacking drivers
1447 * may hold requests forever.
1449 static bool scsi_mq_lld_busy(struct request_queue *q)
1451 struct scsi_device *sdev = q->queuedata;
1452 struct Scsi_Host *shost;
1454 if (blk_queue_dying(q))
1460 * Ignore host/starget busy state.
1461 * Since block layer does not have a concept of fairness across
1462 * multiple queues, congestion of host/starget needs to be handled
1465 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1471 static void scsi_softirq_done(struct request *rq)
1473 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1476 INIT_LIST_HEAD(&cmd->eh_entry);
1478 atomic_inc(&cmd->device->iodone_cnt);
1480 atomic_inc(&cmd->device->ioerr_cnt);
1482 disposition = scsi_decide_disposition(cmd);
1483 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1484 disposition = SUCCESS;
1486 scsi_log_completion(cmd, disposition);
1488 switch (disposition) {
1490 scsi_finish_command(cmd);
1493 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1495 case ADD_TO_MLQUEUE:
1496 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1499 scsi_eh_scmd_add(cmd);
1505 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1506 * @cmd: command block we are dispatching.
1508 * Return: nonzero return request was rejected and device's queue needs to be
1511 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1513 struct Scsi_Host *host = cmd->device->host;
1516 atomic_inc(&cmd->device->iorequest_cnt);
1518 /* check if the device is still usable */
1519 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1520 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1521 * returns an immediate error upwards, and signals
1522 * that the device is no longer present */
1523 cmd->result = DID_NO_CONNECT << 16;
1527 /* Check to see if the scsi lld made this device blocked. */
1528 if (unlikely(scsi_device_blocked(cmd->device))) {
1530 * in blocked state, the command is just put back on
1531 * the device queue. The suspend state has already
1532 * blocked the queue so future requests should not
1533 * occur until the device transitions out of the
1536 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1537 "queuecommand : device blocked\n"));
1538 return SCSI_MLQUEUE_DEVICE_BUSY;
1541 /* Store the LUN value in cmnd, if needed. */
1542 if (cmd->device->lun_in_cdb)
1543 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1544 (cmd->device->lun << 5 & 0xe0);
1549 * Before we queue this command, check if the command
1550 * length exceeds what the host adapter can handle.
1552 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1553 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1554 "queuecommand : command too long. "
1555 "cdb_size=%d host->max_cmd_len=%d\n",
1556 cmd->cmd_len, cmd->device->host->max_cmd_len));
1557 cmd->result = (DID_ABORT << 16);
1561 if (unlikely(host->shost_state == SHOST_DEL)) {
1562 cmd->result = (DID_NO_CONNECT << 16);
1567 trace_scsi_dispatch_cmd_start(cmd);
1568 rtn = host->hostt->queuecommand(host, cmd);
1570 trace_scsi_dispatch_cmd_error(cmd, rtn);
1571 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1572 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1573 rtn = SCSI_MLQUEUE_HOST_BUSY;
1575 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1576 "queuecommand : request rejected\n"));
1581 cmd->scsi_done(cmd);
1585 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1586 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1588 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1589 sizeof(struct scatterlist);
1592 static blk_status_t scsi_mq_prep_fn(struct request *req)
1594 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1595 struct scsi_device *sdev = req->q->queuedata;
1596 struct Scsi_Host *shost = sdev->host;
1597 struct scatterlist *sg;
1599 scsi_init_command(sdev, cmd);
1602 cmd->tag = req->tag;
1603 cmd->prot_op = SCSI_PROT_NORMAL;
1605 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1606 cmd->sdb.table.sgl = sg;
1608 if (scsi_host_get_prot(shost)) {
1609 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1611 cmd->prot_sdb->table.sgl =
1612 (struct scatterlist *)(cmd->prot_sdb + 1);
1615 blk_mq_start_request(req);
1617 return scsi_setup_cmnd(sdev, req);
1620 static void scsi_mq_done(struct scsi_cmnd *cmd)
1622 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1624 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1626 trace_scsi_dispatch_cmd_done(cmd);
1627 blk_mq_complete_request(cmd->request);
1630 static void scsi_mq_put_budget(struct request_queue *q)
1632 struct scsi_device *sdev = q->queuedata;
1634 atomic_dec(&sdev->device_busy);
1637 static bool scsi_mq_get_budget(struct request_queue *q)
1639 struct scsi_device *sdev = q->queuedata;
1641 if (scsi_dev_queue_ready(q, sdev))
1644 atomic_inc(&sdev->restarts);
1647 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1648 * .restarts must be incremented before .device_busy is read because the
1649 * code in scsi_run_queue_async() depends on the order of these operations.
1651 smp_mb__after_atomic();
1654 * If all in-flight requests originated from this LUN are completed
1655 * before reading .device_busy, sdev->device_busy will be observed as
1656 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1657 * soon. Otherwise, completion of one of these requests will observe
1658 * the .restarts flag, and the request queue will be run for handling
1659 * this request, see scsi_end_request().
1661 if (unlikely(atomic_read(&sdev->device_busy) == 0 &&
1662 !scsi_device_blocked(sdev)))
1663 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1667 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1668 const struct blk_mq_queue_data *bd)
1670 struct request *req = bd->rq;
1671 struct request_queue *q = req->q;
1672 struct scsi_device *sdev = q->queuedata;
1673 struct Scsi_Host *shost = sdev->host;
1674 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1679 * If the device is not in running state we will reject some or all
1682 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1683 ret = scsi_prep_state_check(sdev, req);
1684 if (ret != BLK_STS_OK)
1685 goto out_put_budget;
1688 ret = BLK_STS_RESOURCE;
1689 if (!scsi_target_queue_ready(shost, sdev))
1690 goto out_put_budget;
1691 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1692 goto out_dec_target_busy;
1694 if (!(req->rq_flags & RQF_DONTPREP)) {
1695 ret = scsi_mq_prep_fn(req);
1696 if (ret != BLK_STS_OK)
1697 goto out_dec_host_busy;
1698 req->rq_flags |= RQF_DONTPREP;
1700 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1701 blk_mq_start_request(req);
1704 cmd->flags &= SCMD_PRESERVED_FLAGS;
1705 if (sdev->simple_tags)
1706 cmd->flags |= SCMD_TAGGED;
1708 cmd->flags |= SCMD_LAST;
1710 scsi_init_cmd_errh(cmd);
1711 cmd->scsi_done = scsi_mq_done;
1713 reason = scsi_dispatch_cmd(cmd);
1715 scsi_set_blocked(cmd, reason);
1716 ret = BLK_STS_RESOURCE;
1717 goto out_dec_host_busy;
1723 scsi_dec_host_busy(shost, cmd);
1724 out_dec_target_busy:
1725 if (scsi_target(sdev)->can_queue > 0)
1726 atomic_dec(&scsi_target(sdev)->target_busy);
1728 scsi_mq_put_budget(q);
1732 case BLK_STS_RESOURCE:
1733 case BLK_STS_ZONE_RESOURCE:
1734 if (atomic_read(&sdev->device_busy) ||
1735 scsi_device_blocked(sdev))
1736 ret = BLK_STS_DEV_RESOURCE;
1739 if (unlikely(!scsi_device_online(sdev)))
1740 scsi_req(req)->result = DID_NO_CONNECT << 16;
1742 scsi_req(req)->result = DID_ERROR << 16;
1744 * Make sure to release all allocated resources when
1745 * we hit an error, as we will never see this command
1748 if (req->rq_flags & RQF_DONTPREP)
1749 scsi_mq_uninit_cmd(cmd);
1750 scsi_run_queue_async(sdev);
1756 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1760 return BLK_EH_RESET_TIMER;
1761 return scsi_times_out(req);
1764 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1765 unsigned int hctx_idx, unsigned int numa_node)
1767 struct Scsi_Host *shost = set->driver_data;
1768 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1769 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1770 struct scatterlist *sg;
1773 if (unchecked_isa_dma)
1774 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1775 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1776 GFP_KERNEL, numa_node);
1777 if (!cmd->sense_buffer)
1779 cmd->req.sense = cmd->sense_buffer;
1781 if (scsi_host_get_prot(shost)) {
1782 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1783 shost->hostt->cmd_size;
1784 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1787 if (shost->hostt->init_cmd_priv) {
1788 ret = shost->hostt->init_cmd_priv(shost, cmd);
1790 scsi_free_sense_buffer(unchecked_isa_dma,
1797 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1798 unsigned int hctx_idx)
1800 struct Scsi_Host *shost = set->driver_data;
1801 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1803 if (shost->hostt->exit_cmd_priv)
1804 shost->hostt->exit_cmd_priv(shost, cmd);
1805 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1809 static int scsi_map_queues(struct blk_mq_tag_set *set)
1811 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1813 if (shost->hostt->map_queues)
1814 return shost->hostt->map_queues(shost);
1815 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1818 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1820 struct device *dev = shost->dma_dev;
1823 * this limit is imposed by hardware restrictions
1825 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1828 if (scsi_host_prot_dma(shost)) {
1829 shost->sg_prot_tablesize =
1830 min_not_zero(shost->sg_prot_tablesize,
1831 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1832 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1833 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1836 if (dev->dma_mask) {
1837 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1838 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1840 blk_queue_max_hw_sectors(q, shost->max_sectors);
1841 if (shost->unchecked_isa_dma)
1842 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1843 blk_queue_segment_boundary(q, shost->dma_boundary);
1844 dma_set_seg_boundary(dev, shost->dma_boundary);
1846 blk_queue_max_segment_size(q, shost->max_segment_size);
1847 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1848 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1851 * Set a reasonable default alignment: The larger of 32-byte (dword),
1852 * which is a common minimum for HBAs, and the minimum DMA alignment,
1853 * which is set by the platform.
1855 * Devices that require a bigger alignment can increase it later.
1857 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1859 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1861 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1862 .get_budget = scsi_mq_get_budget,
1863 .put_budget = scsi_mq_put_budget,
1864 .queue_rq = scsi_queue_rq,
1865 .complete = scsi_softirq_done,
1866 .timeout = scsi_timeout,
1867 #ifdef CONFIG_BLK_DEBUG_FS
1868 .show_rq = scsi_show_rq,
1870 .init_request = scsi_mq_init_request,
1871 .exit_request = scsi_mq_exit_request,
1872 .initialize_rq_fn = scsi_initialize_rq,
1873 .cleanup_rq = scsi_cleanup_rq,
1874 .busy = scsi_mq_lld_busy,
1875 .map_queues = scsi_map_queues,
1879 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1881 struct request_queue *q = hctx->queue;
1882 struct scsi_device *sdev = q->queuedata;
1883 struct Scsi_Host *shost = sdev->host;
1885 shost->hostt->commit_rqs(shost, hctx->queue_num);
1888 static const struct blk_mq_ops scsi_mq_ops = {
1889 .get_budget = scsi_mq_get_budget,
1890 .put_budget = scsi_mq_put_budget,
1891 .queue_rq = scsi_queue_rq,
1892 .commit_rqs = scsi_commit_rqs,
1893 .complete = scsi_softirq_done,
1894 .timeout = scsi_timeout,
1895 #ifdef CONFIG_BLK_DEBUG_FS
1896 .show_rq = scsi_show_rq,
1898 .init_request = scsi_mq_init_request,
1899 .exit_request = scsi_mq_exit_request,
1900 .initialize_rq_fn = scsi_initialize_rq,
1901 .cleanup_rq = scsi_cleanup_rq,
1902 .busy = scsi_mq_lld_busy,
1903 .map_queues = scsi_map_queues,
1906 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1908 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1909 if (IS_ERR(sdev->request_queue))
1912 sdev->request_queue->queuedata = sdev;
1913 __scsi_init_queue(sdev->host, sdev->request_queue);
1914 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1915 return sdev->request_queue;
1918 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1920 unsigned int cmd_size, sgl_size;
1921 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1923 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1924 scsi_mq_inline_sgl_size(shost));
1925 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1926 if (scsi_host_get_prot(shost))
1927 cmd_size += sizeof(struct scsi_data_buffer) +
1928 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1930 memset(tag_set, 0, sizeof(*tag_set));
1931 if (shost->hostt->commit_rqs)
1932 tag_set->ops = &scsi_mq_ops;
1934 tag_set->ops = &scsi_mq_ops_no_commit;
1935 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1936 tag_set->queue_depth = shost->can_queue;
1937 tag_set->cmd_size = cmd_size;
1938 tag_set->numa_node = NUMA_NO_NODE;
1939 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1941 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1942 tag_set->driver_data = shost;
1943 if (shost->host_tagset)
1944 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1946 return blk_mq_alloc_tag_set(tag_set);
1949 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1951 blk_mq_free_tag_set(&shost->tag_set);
1955 * scsi_device_from_queue - return sdev associated with a request_queue
1956 * @q: The request queue to return the sdev from
1958 * Return the sdev associated with a request queue or NULL if the
1959 * request_queue does not reference a SCSI device.
1961 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1963 struct scsi_device *sdev = NULL;
1965 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1966 q->mq_ops == &scsi_mq_ops)
1967 sdev = q->queuedata;
1968 if (!sdev || !get_device(&sdev->sdev_gendev))
1973 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1976 * scsi_block_requests - Utility function used by low-level drivers to prevent
1977 * further commands from being queued to the device.
1978 * @shost: host in question
1980 * There is no timer nor any other means by which the requests get unblocked
1981 * other than the low-level driver calling scsi_unblock_requests().
1983 void scsi_block_requests(struct Scsi_Host *shost)
1985 shost->host_self_blocked = 1;
1987 EXPORT_SYMBOL(scsi_block_requests);
1990 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1991 * further commands to be queued to the device.
1992 * @shost: host in question
1994 * There is no timer nor any other means by which the requests get unblocked
1995 * other than the low-level driver calling scsi_unblock_requests(). This is done
1996 * as an API function so that changes to the internals of the scsi mid-layer
1997 * won't require wholesale changes to drivers that use this feature.
1999 void scsi_unblock_requests(struct Scsi_Host *shost)
2001 shost->host_self_blocked = 0;
2002 scsi_run_host_queues(shost);
2004 EXPORT_SYMBOL(scsi_unblock_requests);
2006 void scsi_exit_queue(void)
2008 kmem_cache_destroy(scsi_sense_cache);
2009 kmem_cache_destroy(scsi_sense_isadma_cache);
2013 * scsi_mode_select - issue a mode select
2014 * @sdev: SCSI device to be queried
2015 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2016 * @sp: Save page bit (0 == don't save, 1 == save)
2017 * @modepage: mode page being requested
2018 * @buffer: request buffer (may not be smaller than eight bytes)
2019 * @len: length of request buffer.
2020 * @timeout: command timeout
2021 * @retries: number of retries before failing
2022 * @data: returns a structure abstracting the mode header data
2023 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2024 * must be SCSI_SENSE_BUFFERSIZE big.
2026 * Returns zero if successful; negative error number or scsi
2031 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2032 unsigned char *buffer, int len, int timeout, int retries,
2033 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2035 unsigned char cmd[10];
2036 unsigned char *real_buffer;
2039 memset(cmd, 0, sizeof(cmd));
2040 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2042 if (sdev->use_10_for_ms) {
2045 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2048 memcpy(real_buffer + 8, buffer, len);
2052 real_buffer[2] = data->medium_type;
2053 real_buffer[3] = data->device_specific;
2054 real_buffer[4] = data->longlba ? 0x01 : 0;
2056 real_buffer[6] = data->block_descriptor_length >> 8;
2057 real_buffer[7] = data->block_descriptor_length;
2059 cmd[0] = MODE_SELECT_10;
2063 if (len > 255 || data->block_descriptor_length > 255 ||
2067 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2070 memcpy(real_buffer + 4, buffer, len);
2073 real_buffer[1] = data->medium_type;
2074 real_buffer[2] = data->device_specific;
2075 real_buffer[3] = data->block_descriptor_length;
2077 cmd[0] = MODE_SELECT;
2081 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2082 sshdr, timeout, retries, NULL);
2086 EXPORT_SYMBOL_GPL(scsi_mode_select);
2089 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2090 * @sdev: SCSI device to be queried
2091 * @dbd: set if mode sense will allow block descriptors to be returned
2092 * @modepage: mode page being requested
2093 * @buffer: request buffer (may not be smaller than eight bytes)
2094 * @len: length of request buffer.
2095 * @timeout: command timeout
2096 * @retries: number of retries before failing
2097 * @data: returns a structure abstracting the mode header data
2098 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2099 * must be SCSI_SENSE_BUFFERSIZE big.
2101 * Returns zero if unsuccessful, or the header offset (either 4
2102 * or 8 depending on whether a six or ten byte command was
2103 * issued) if successful.
2106 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2107 unsigned char *buffer, int len, int timeout, int retries,
2108 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2110 unsigned char cmd[12];
2113 int result, retry_count = retries;
2114 struct scsi_sense_hdr my_sshdr;
2116 memset(data, 0, sizeof(*data));
2117 memset(&cmd[0], 0, 12);
2119 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2120 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2123 /* caller might not be interested in sense, but we need it */
2128 use_10_for_ms = sdev->use_10_for_ms;
2130 if (use_10_for_ms) {
2134 cmd[0] = MODE_SENSE_10;
2141 cmd[0] = MODE_SENSE;
2146 memset(buffer, 0, len);
2148 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2149 sshdr, timeout, retries, NULL);
2151 /* This code looks awful: what it's doing is making sure an
2152 * ILLEGAL REQUEST sense return identifies the actual command
2153 * byte as the problem. MODE_SENSE commands can return
2154 * ILLEGAL REQUEST if the code page isn't supported */
2156 if (use_10_for_ms && !scsi_status_is_good(result) &&
2157 driver_byte(result) == DRIVER_SENSE) {
2158 if (scsi_sense_valid(sshdr)) {
2159 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2160 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2162 * Invalid command operation code
2164 sdev->use_10_for_ms = 0;
2170 if (scsi_status_is_good(result)) {
2171 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2172 (modepage == 6 || modepage == 8))) {
2173 /* Initio breakage? */
2176 data->medium_type = 0;
2177 data->device_specific = 0;
2179 data->block_descriptor_length = 0;
2180 } else if (use_10_for_ms) {
2181 data->length = buffer[0]*256 + buffer[1] + 2;
2182 data->medium_type = buffer[2];
2183 data->device_specific = buffer[3];
2184 data->longlba = buffer[4] & 0x01;
2185 data->block_descriptor_length = buffer[6]*256
2188 data->length = buffer[0] + 1;
2189 data->medium_type = buffer[1];
2190 data->device_specific = buffer[2];
2191 data->block_descriptor_length = buffer[3];
2193 data->header_length = header_length;
2194 } else if ((status_byte(result) == CHECK_CONDITION) &&
2195 scsi_sense_valid(sshdr) &&
2196 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2203 EXPORT_SYMBOL(scsi_mode_sense);
2206 * scsi_test_unit_ready - test if unit is ready
2207 * @sdev: scsi device to change the state of.
2208 * @timeout: command timeout
2209 * @retries: number of retries before failing
2210 * @sshdr: outpout pointer for decoded sense information.
2212 * Returns zero if unsuccessful or an error if TUR failed. For
2213 * removable media, UNIT_ATTENTION sets ->changed flag.
2216 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2217 struct scsi_sense_hdr *sshdr)
2220 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2224 /* try to eat the UNIT_ATTENTION if there are enough retries */
2226 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2228 if (sdev->removable && scsi_sense_valid(sshdr) &&
2229 sshdr->sense_key == UNIT_ATTENTION)
2231 } while (scsi_sense_valid(sshdr) &&
2232 sshdr->sense_key == UNIT_ATTENTION && --retries);
2236 EXPORT_SYMBOL(scsi_test_unit_ready);
2239 * scsi_device_set_state - Take the given device through the device state model.
2240 * @sdev: scsi device to change the state of.
2241 * @state: state to change to.
2243 * Returns zero if successful or an error if the requested
2244 * transition is illegal.
2247 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2249 enum scsi_device_state oldstate = sdev->sdev_state;
2251 if (state == oldstate)
2257 case SDEV_CREATED_BLOCK:
2268 case SDEV_TRANSPORT_OFFLINE:
2281 case SDEV_TRANSPORT_OFFLINE:
2289 case SDEV_TRANSPORT_OFFLINE:
2304 case SDEV_CREATED_BLOCK:
2313 case SDEV_CREATED_BLOCK:
2328 case SDEV_TRANSPORT_OFFLINE:
2340 case SDEV_TRANSPORT_OFFLINE:
2343 case SDEV_CREATED_BLOCK:
2351 sdev->offline_already = false;
2352 sdev->sdev_state = state;
2356 SCSI_LOG_ERROR_RECOVERY(1,
2357 sdev_printk(KERN_ERR, sdev,
2358 "Illegal state transition %s->%s",
2359 scsi_device_state_name(oldstate),
2360 scsi_device_state_name(state))
2364 EXPORT_SYMBOL(scsi_device_set_state);
2367 * sdev_evt_emit - emit a single SCSI device uevent
2368 * @sdev: associated SCSI device
2369 * @evt: event to emit
2371 * Send a single uevent (scsi_event) to the associated scsi_device.
2373 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2378 switch (evt->evt_type) {
2379 case SDEV_EVT_MEDIA_CHANGE:
2380 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2382 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2383 scsi_rescan_device(&sdev->sdev_gendev);
2384 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2386 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2387 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2389 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2390 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2392 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2393 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2395 case SDEV_EVT_LUN_CHANGE_REPORTED:
2396 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2398 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2399 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2401 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2402 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2411 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2415 * sdev_evt_thread - send a uevent for each scsi event
2416 * @work: work struct for scsi_device
2418 * Dispatch queued events to their associated scsi_device kobjects
2421 void scsi_evt_thread(struct work_struct *work)
2423 struct scsi_device *sdev;
2424 enum scsi_device_event evt_type;
2425 LIST_HEAD(event_list);
2427 sdev = container_of(work, struct scsi_device, event_work);
2429 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2430 if (test_and_clear_bit(evt_type, sdev->pending_events))
2431 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2434 struct scsi_event *evt;
2435 struct list_head *this, *tmp;
2436 unsigned long flags;
2438 spin_lock_irqsave(&sdev->list_lock, flags);
2439 list_splice_init(&sdev->event_list, &event_list);
2440 spin_unlock_irqrestore(&sdev->list_lock, flags);
2442 if (list_empty(&event_list))
2445 list_for_each_safe(this, tmp, &event_list) {
2446 evt = list_entry(this, struct scsi_event, node);
2447 list_del(&evt->node);
2448 scsi_evt_emit(sdev, evt);
2455 * sdev_evt_send - send asserted event to uevent thread
2456 * @sdev: scsi_device event occurred on
2457 * @evt: event to send
2459 * Assert scsi device event asynchronously.
2461 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2463 unsigned long flags;
2466 /* FIXME: currently this check eliminates all media change events
2467 * for polled devices. Need to update to discriminate between AN
2468 * and polled events */
2469 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2475 spin_lock_irqsave(&sdev->list_lock, flags);
2476 list_add_tail(&evt->node, &sdev->event_list);
2477 schedule_work(&sdev->event_work);
2478 spin_unlock_irqrestore(&sdev->list_lock, flags);
2480 EXPORT_SYMBOL_GPL(sdev_evt_send);
2483 * sdev_evt_alloc - allocate a new scsi event
2484 * @evt_type: type of event to allocate
2485 * @gfpflags: GFP flags for allocation
2487 * Allocates and returns a new scsi_event.
2489 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2492 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2496 evt->evt_type = evt_type;
2497 INIT_LIST_HEAD(&evt->node);
2499 /* evt_type-specific initialization, if any */
2501 case SDEV_EVT_MEDIA_CHANGE:
2502 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2503 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2504 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2505 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2506 case SDEV_EVT_LUN_CHANGE_REPORTED:
2507 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2508 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2516 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2519 * sdev_evt_send_simple - send asserted event to uevent thread
2520 * @sdev: scsi_device event occurred on
2521 * @evt_type: type of event to send
2522 * @gfpflags: GFP flags for allocation
2524 * Assert scsi device event asynchronously, given an event type.
2526 void sdev_evt_send_simple(struct scsi_device *sdev,
2527 enum scsi_device_event evt_type, gfp_t gfpflags)
2529 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2531 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2536 sdev_evt_send(sdev, evt);
2538 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2541 * scsi_device_quiesce - Block user issued commands.
2542 * @sdev: scsi device to quiesce.
2544 * This works by trying to transition to the SDEV_QUIESCE state
2545 * (which must be a legal transition). When the device is in this
2546 * state, only special requests will be accepted, all others will
2547 * be deferred. Since special requests may also be requeued requests,
2548 * a successful return doesn't guarantee the device will be
2549 * totally quiescent.
2551 * Must be called with user context, may sleep.
2553 * Returns zero if unsuccessful or an error if not.
2556 scsi_device_quiesce(struct scsi_device *sdev)
2558 struct request_queue *q = sdev->request_queue;
2562 * It is allowed to call scsi_device_quiesce() multiple times from
2563 * the same context but concurrent scsi_device_quiesce() calls are
2566 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2568 if (sdev->quiesced_by == current)
2573 blk_mq_freeze_queue(q);
2575 * Ensure that the effect of blk_set_pm_only() will be visible
2576 * for percpu_ref_tryget() callers that occur after the queue
2577 * unfreeze even if the queue was already frozen before this function
2578 * was called. See also https://lwn.net/Articles/573497/.
2581 blk_mq_unfreeze_queue(q);
2583 mutex_lock(&sdev->state_mutex);
2584 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2586 sdev->quiesced_by = current;
2588 blk_clear_pm_only(q);
2589 mutex_unlock(&sdev->state_mutex);
2593 EXPORT_SYMBOL(scsi_device_quiesce);
2596 * scsi_device_resume - Restart user issued commands to a quiesced device.
2597 * @sdev: scsi device to resume.
2599 * Moves the device from quiesced back to running and restarts the
2602 * Must be called with user context, may sleep.
2604 void scsi_device_resume(struct scsi_device *sdev)
2606 /* check if the device state was mutated prior to resume, and if
2607 * so assume the state is being managed elsewhere (for example
2608 * device deleted during suspend)
2610 mutex_lock(&sdev->state_mutex);
2611 if (sdev->quiesced_by) {
2612 sdev->quiesced_by = NULL;
2613 blk_clear_pm_only(sdev->request_queue);
2615 if (sdev->sdev_state == SDEV_QUIESCE)
2616 scsi_device_set_state(sdev, SDEV_RUNNING);
2617 mutex_unlock(&sdev->state_mutex);
2619 EXPORT_SYMBOL(scsi_device_resume);
2622 device_quiesce_fn(struct scsi_device *sdev, void *data)
2624 scsi_device_quiesce(sdev);
2628 scsi_target_quiesce(struct scsi_target *starget)
2630 starget_for_each_device(starget, NULL, device_quiesce_fn);
2632 EXPORT_SYMBOL(scsi_target_quiesce);
2635 device_resume_fn(struct scsi_device *sdev, void *data)
2637 scsi_device_resume(sdev);
2641 scsi_target_resume(struct scsi_target *starget)
2643 starget_for_each_device(starget, NULL, device_resume_fn);
2645 EXPORT_SYMBOL(scsi_target_resume);
2648 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2649 * @sdev: device to block
2651 * Pause SCSI command processing on the specified device. Does not sleep.
2653 * Returns zero if successful or a negative error code upon failure.
2656 * This routine transitions the device to the SDEV_BLOCK state (which must be
2657 * a legal transition). When the device is in this state, command processing
2658 * is paused until the device leaves the SDEV_BLOCK state. See also
2659 * scsi_internal_device_unblock_nowait().
2661 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2663 struct request_queue *q = sdev->request_queue;
2666 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2668 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2675 * The device has transitioned to SDEV_BLOCK. Stop the
2676 * block layer from calling the midlayer with this device's
2679 blk_mq_quiesce_queue_nowait(q);
2682 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2685 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2686 * @sdev: device to block
2688 * Pause SCSI command processing on the specified device and wait until all
2689 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2691 * Returns zero if successful or a negative error code upon failure.
2694 * This routine transitions the device to the SDEV_BLOCK state (which must be
2695 * a legal transition). When the device is in this state, command processing
2696 * is paused until the device leaves the SDEV_BLOCK state. See also
2697 * scsi_internal_device_unblock().
2699 static int scsi_internal_device_block(struct scsi_device *sdev)
2701 struct request_queue *q = sdev->request_queue;
2704 mutex_lock(&sdev->state_mutex);
2705 err = scsi_internal_device_block_nowait(sdev);
2707 blk_mq_quiesce_queue(q);
2708 mutex_unlock(&sdev->state_mutex);
2713 void scsi_start_queue(struct scsi_device *sdev)
2715 struct request_queue *q = sdev->request_queue;
2717 blk_mq_unquiesce_queue(q);
2721 * scsi_internal_device_unblock_nowait - resume a device after a block request
2722 * @sdev: device to resume
2723 * @new_state: state to set the device to after unblocking
2725 * Restart the device queue for a previously suspended SCSI device. Does not
2728 * Returns zero if successful or a negative error code upon failure.
2731 * This routine transitions the device to the SDEV_RUNNING state or to one of
2732 * the offline states (which must be a legal transition) allowing the midlayer
2733 * to goose the queue for this device.
2735 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2736 enum scsi_device_state new_state)
2738 switch (new_state) {
2740 case SDEV_TRANSPORT_OFFLINE:
2747 * Try to transition the scsi device to SDEV_RUNNING or one of the
2748 * offlined states and goose the device queue if successful.
2750 switch (sdev->sdev_state) {
2752 case SDEV_TRANSPORT_OFFLINE:
2753 sdev->sdev_state = new_state;
2755 case SDEV_CREATED_BLOCK:
2756 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2757 new_state == SDEV_OFFLINE)
2758 sdev->sdev_state = new_state;
2760 sdev->sdev_state = SDEV_CREATED;
2768 scsi_start_queue(sdev);
2772 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2775 * scsi_internal_device_unblock - resume a device after a block request
2776 * @sdev: device to resume
2777 * @new_state: state to set the device to after unblocking
2779 * Restart the device queue for a previously suspended SCSI device. May sleep.
2781 * Returns zero if successful or a negative error code upon failure.
2784 * This routine transitions the device to the SDEV_RUNNING state or to one of
2785 * the offline states (which must be a legal transition) allowing the midlayer
2786 * to goose the queue for this device.
2788 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2789 enum scsi_device_state new_state)
2793 mutex_lock(&sdev->state_mutex);
2794 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2795 mutex_unlock(&sdev->state_mutex);
2801 device_block(struct scsi_device *sdev, void *data)
2805 ret = scsi_internal_device_block(sdev);
2807 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2808 dev_name(&sdev->sdev_gendev), ret);
2812 target_block(struct device *dev, void *data)
2814 if (scsi_is_target_device(dev))
2815 starget_for_each_device(to_scsi_target(dev), NULL,
2821 scsi_target_block(struct device *dev)
2823 if (scsi_is_target_device(dev))
2824 starget_for_each_device(to_scsi_target(dev), NULL,
2827 device_for_each_child(dev, NULL, target_block);
2829 EXPORT_SYMBOL_GPL(scsi_target_block);
2832 device_unblock(struct scsi_device *sdev, void *data)
2834 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2838 target_unblock(struct device *dev, void *data)
2840 if (scsi_is_target_device(dev))
2841 starget_for_each_device(to_scsi_target(dev), data,
2847 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2849 if (scsi_is_target_device(dev))
2850 starget_for_each_device(to_scsi_target(dev), &new_state,
2853 device_for_each_child(dev, &new_state, target_unblock);
2855 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2858 scsi_host_block(struct Scsi_Host *shost)
2860 struct scsi_device *sdev;
2864 * Call scsi_internal_device_block_nowait so we can avoid
2865 * calling synchronize_rcu() for each LUN.
2867 shost_for_each_device(sdev, shost) {
2868 mutex_lock(&sdev->state_mutex);
2869 ret = scsi_internal_device_block_nowait(sdev);
2870 mutex_unlock(&sdev->state_mutex);
2872 scsi_device_put(sdev);
2878 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2879 * calling synchronize_rcu() once is enough.
2881 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2888 EXPORT_SYMBOL_GPL(scsi_host_block);
2891 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2893 struct scsi_device *sdev;
2896 shost_for_each_device(sdev, shost) {
2897 ret = scsi_internal_device_unblock(sdev, new_state);
2899 scsi_device_put(sdev);
2905 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2908 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2909 * @sgl: scatter-gather list
2910 * @sg_count: number of segments in sg
2911 * @offset: offset in bytes into sg, on return offset into the mapped area
2912 * @len: bytes to map, on return number of bytes mapped
2914 * Returns virtual address of the start of the mapped page
2916 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2917 size_t *offset, size_t *len)
2920 size_t sg_len = 0, len_complete = 0;
2921 struct scatterlist *sg;
2924 WARN_ON(!irqs_disabled());
2926 for_each_sg(sgl, sg, sg_count, i) {
2927 len_complete = sg_len; /* Complete sg-entries */
2928 sg_len += sg->length;
2929 if (sg_len > *offset)
2933 if (unlikely(i == sg_count)) {
2934 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2936 __func__, sg_len, *offset, sg_count);
2941 /* Offset starting from the beginning of first page in this sg-entry */
2942 *offset = *offset - len_complete + sg->offset;
2944 /* Assumption: contiguous pages can be accessed as "page + i" */
2945 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2946 *offset &= ~PAGE_MASK;
2948 /* Bytes in this sg-entry from *offset to the end of the page */
2949 sg_len = PAGE_SIZE - *offset;
2953 return kmap_atomic(page);
2955 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2958 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2959 * @virt: virtual address to be unmapped
2961 void scsi_kunmap_atomic_sg(void *virt)
2963 kunmap_atomic(virt);
2965 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2967 void sdev_disable_disk_events(struct scsi_device *sdev)
2969 atomic_inc(&sdev->disk_events_disable_depth);
2971 EXPORT_SYMBOL(sdev_disable_disk_events);
2973 void sdev_enable_disk_events(struct scsi_device *sdev)
2975 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2977 atomic_dec(&sdev->disk_events_disable_depth);
2979 EXPORT_SYMBOL(sdev_enable_disk_events);
2982 * scsi_vpd_lun_id - return a unique device identification
2983 * @sdev: SCSI device
2984 * @id: buffer for the identification
2985 * @id_len: length of the buffer
2987 * Copies a unique device identification into @id based
2988 * on the information in the VPD page 0x83 of the device.
2989 * The string will be formatted as a SCSI name string.
2991 * Returns the length of the identification or error on failure.
2992 * If the identifier is longer than the supplied buffer the actual
2993 * identifier length is returned and the buffer is not zero-padded.
2995 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2997 u8 cur_id_type = 0xff;
2999 const unsigned char *d, *cur_id_str;
3000 const struct scsi_vpd *vpd_pg83;
3001 int id_size = -EINVAL;
3004 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3011 * Look for the correct descriptor.
3012 * Order of preference for lun descriptor:
3013 * - SCSI name string
3014 * - NAA IEEE Registered Extended
3015 * - EUI-64 based 16-byte
3016 * - EUI-64 based 12-byte
3017 * - NAA IEEE Registered
3018 * - NAA IEEE Extended
3020 * as longer descriptors reduce the likelyhood
3021 * of identification clashes.
3024 /* The id string must be at least 20 bytes + terminating NULL byte */
3030 memset(id, 0, id_len);
3031 d = vpd_pg83->data + 4;
3032 while (d < vpd_pg83->data + vpd_pg83->len) {
3033 /* Skip designators not referring to the LUN */
3034 if ((d[1] & 0x30) != 0x00)
3037 switch (d[1] & 0xf) {
3040 if (cur_id_size > d[3])
3042 /* Prefer anything */
3043 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3046 if (cur_id_size + 4 > id_len)
3047 cur_id_size = id_len - 4;
3049 cur_id_type = d[1] & 0xf;
3050 id_size = snprintf(id, id_len, "t10.%*pE",
3051 cur_id_size, cur_id_str);
3055 if (cur_id_size > d[3])
3057 /* Prefer NAA IEEE Registered Extended */
3058 if (cur_id_type == 0x3 &&
3059 cur_id_size == d[3])
3063 cur_id_type = d[1] & 0xf;
3064 switch (cur_id_size) {
3066 id_size = snprintf(id, id_len,
3071 id_size = snprintf(id, id_len,
3076 id_size = snprintf(id, id_len,
3087 if (cur_id_size > d[3])
3091 cur_id_type = d[1] & 0xf;
3092 switch (cur_id_size) {
3094 id_size = snprintf(id, id_len,
3099 id_size = snprintf(id, id_len,
3109 /* SCSI name string */
3110 if (cur_id_size + 4 > d[3])
3112 /* Prefer others for truncated descriptor */
3113 if (cur_id_size && d[3] > id_len)
3115 cur_id_size = id_size = d[3];
3117 cur_id_type = d[1] & 0xf;
3118 if (cur_id_size >= id_len)
3119 cur_id_size = id_len - 1;
3120 memcpy(id, cur_id_str, cur_id_size);
3121 /* Decrease priority for truncated descriptor */
3122 if (cur_id_size != id_size)
3135 EXPORT_SYMBOL(scsi_vpd_lun_id);
3138 * scsi_vpd_tpg_id - return a target port group identifier
3139 * @sdev: SCSI device
3141 * Returns the Target Port Group identifier from the information
3142 * froom VPD page 0x83 of the device.
3144 * Returns the identifier or error on failure.
3146 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3148 const unsigned char *d;
3149 const struct scsi_vpd *vpd_pg83;
3150 int group_id = -EAGAIN, rel_port = -1;
3153 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3159 d = vpd_pg83->data + 4;
3160 while (d < vpd_pg83->data + vpd_pg83->len) {
3161 switch (d[1] & 0xf) {
3163 /* Relative target port */
3164 rel_port = get_unaligned_be16(&d[6]);
3167 /* Target port group */
3168 group_id = get_unaligned_be16(&d[6]);
3177 if (group_id >= 0 && rel_id && rel_port != -1)
3182 EXPORT_SYMBOL(scsi_vpd_tpg_id);