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);
555 blk_mq_run_hw_queues(sdev->request_queue, true);
558 /* Returns false when no more bytes to process, true if there are more */
559 static bool scsi_end_request(struct request *req, blk_status_t error,
562 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
563 struct scsi_device *sdev = cmd->device;
564 struct request_queue *q = sdev->request_queue;
566 if (blk_update_request(req, error, bytes))
569 if (blk_queue_add_random(q))
570 add_disk_randomness(req->rq_disk);
572 if (!blk_rq_is_scsi(req)) {
573 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
574 cmd->flags &= ~SCMD_INITIALIZED;
578 * Calling rcu_barrier() is not necessary here because the
579 * SCSI error handler guarantees that the function called by
580 * call_rcu() has been called before scsi_end_request() is
583 destroy_rcu_head(&cmd->rcu);
586 * In the MQ case the command gets freed by __blk_mq_end_request,
587 * so we have to do all cleanup that depends on it earlier.
589 * We also can't kick the queues from irq context, so we
590 * will have to defer it to a workqueue.
592 scsi_mq_uninit_cmd(cmd);
595 * queue is still alive, so grab the ref for preventing it
596 * from being cleaned up during running queue.
598 percpu_ref_get(&q->q_usage_counter);
600 __blk_mq_end_request(req, error);
602 scsi_run_queue_async(sdev);
604 percpu_ref_put(&q->q_usage_counter);
609 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
611 * @result: scsi error code
613 * Translate a SCSI result code into a blk_status_t value. May reset the host
614 * byte of @cmd->result.
616 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
618 switch (host_byte(result)) {
621 * Also check the other bytes than the status byte in result
622 * to handle the case when a SCSI LLD sets result to
623 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
625 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
627 return BLK_STS_IOERR;
628 case DID_TRANSPORT_FAILFAST:
629 return BLK_STS_TRANSPORT;
630 case DID_TARGET_FAILURE:
631 set_host_byte(cmd, DID_OK);
632 return BLK_STS_TARGET;
633 case DID_NEXUS_FAILURE:
634 set_host_byte(cmd, DID_OK);
635 return BLK_STS_NEXUS;
636 case DID_ALLOC_FAILURE:
637 set_host_byte(cmd, DID_OK);
638 return BLK_STS_NOSPC;
639 case DID_MEDIUM_ERROR:
640 set_host_byte(cmd, DID_OK);
641 return BLK_STS_MEDIUM;
643 return BLK_STS_IOERR;
647 /* Helper for scsi_io_completion() when "reprep" action required. */
648 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
649 struct request_queue *q)
651 /* A new command will be prepared and issued. */
652 scsi_mq_requeue_cmd(cmd);
655 /* Helper for scsi_io_completion() when special action required. */
656 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
658 struct request_queue *q = cmd->device->request_queue;
659 struct request *req = cmd->request;
661 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
662 ACTION_DELAYED_RETRY} action;
663 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
664 struct scsi_sense_hdr sshdr;
666 bool sense_current = true; /* false implies "deferred sense" */
667 blk_status_t blk_stat;
669 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
671 sense_current = !scsi_sense_is_deferred(&sshdr);
673 blk_stat = scsi_result_to_blk_status(cmd, result);
675 if (host_byte(result) == DID_RESET) {
676 /* Third party bus reset or reset for error recovery
677 * reasons. Just retry the command and see what
680 action = ACTION_RETRY;
681 } else if (sense_valid && sense_current) {
682 switch (sshdr.sense_key) {
684 if (cmd->device->removable) {
685 /* Detected disc change. Set a bit
686 * and quietly refuse further access.
688 cmd->device->changed = 1;
689 action = ACTION_FAIL;
691 /* Must have been a power glitch, or a
692 * bus reset. Could not have been a
693 * media change, so we just retry the
694 * command and see what happens.
696 action = ACTION_RETRY;
699 case ILLEGAL_REQUEST:
700 /* If we had an ILLEGAL REQUEST returned, then
701 * we may have performed an unsupported
702 * command. The only thing this should be
703 * would be a ten byte read where only a six
704 * byte read was supported. Also, on a system
705 * where READ CAPACITY failed, we may have
706 * read past the end of the disk.
708 if ((cmd->device->use_10_for_rw &&
709 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
710 (cmd->cmnd[0] == READ_10 ||
711 cmd->cmnd[0] == WRITE_10)) {
712 /* This will issue a new 6-byte command. */
713 cmd->device->use_10_for_rw = 0;
714 action = ACTION_REPREP;
715 } else if (sshdr.asc == 0x10) /* DIX */ {
716 action = ACTION_FAIL;
717 blk_stat = BLK_STS_PROTECTION;
718 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
719 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
720 action = ACTION_FAIL;
721 blk_stat = BLK_STS_TARGET;
723 action = ACTION_FAIL;
725 case ABORTED_COMMAND:
726 action = ACTION_FAIL;
727 if (sshdr.asc == 0x10) /* DIF */
728 blk_stat = BLK_STS_PROTECTION;
731 /* If the device is in the process of becoming
732 * ready, or has a temporary blockage, retry.
734 if (sshdr.asc == 0x04) {
735 switch (sshdr.ascq) {
736 case 0x01: /* becoming ready */
737 case 0x04: /* format in progress */
738 case 0x05: /* rebuild in progress */
739 case 0x06: /* recalculation in progress */
740 case 0x07: /* operation in progress */
741 case 0x08: /* Long write in progress */
742 case 0x09: /* self test in progress */
743 case 0x14: /* space allocation in progress */
744 case 0x1a: /* start stop unit in progress */
745 case 0x1b: /* sanitize in progress */
746 case 0x1d: /* configuration in progress */
747 case 0x24: /* depopulation in progress */
748 action = ACTION_DELAYED_RETRY;
751 action = ACTION_FAIL;
755 action = ACTION_FAIL;
757 case VOLUME_OVERFLOW:
758 /* See SSC3rXX or current. */
759 action = ACTION_FAIL;
762 action = ACTION_FAIL;
766 action = ACTION_FAIL;
768 if (action != ACTION_FAIL &&
769 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
770 action = ACTION_FAIL;
774 /* Give up and fail the remainder of the request */
775 if (!(req->rq_flags & RQF_QUIET)) {
776 static DEFINE_RATELIMIT_STATE(_rs,
777 DEFAULT_RATELIMIT_INTERVAL,
778 DEFAULT_RATELIMIT_BURST);
780 if (unlikely(scsi_logging_level))
782 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
783 SCSI_LOG_MLCOMPLETE_BITS);
786 * if logging is enabled the failure will be printed
787 * in scsi_log_completion(), so avoid duplicate messages
789 if (!level && __ratelimit(&_rs)) {
790 scsi_print_result(cmd, NULL, FAILED);
791 if (driver_byte(result) == DRIVER_SENSE)
792 scsi_print_sense(cmd);
793 scsi_print_command(cmd);
796 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
800 scsi_io_completion_reprep(cmd, q);
803 /* Retry the same command immediately */
804 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
806 case ACTION_DELAYED_RETRY:
807 /* Retry the same command after a delay */
808 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
814 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
815 * new result that may suppress further error checking. Also modifies
816 * *blk_statp in some cases.
818 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
819 blk_status_t *blk_statp)
822 bool sense_current = true; /* false implies "deferred sense" */
823 struct request *req = cmd->request;
824 struct scsi_sense_hdr sshdr;
826 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
828 sense_current = !scsi_sense_is_deferred(&sshdr);
830 if (blk_rq_is_passthrough(req)) {
833 * SG_IO wants current and deferred errors
835 scsi_req(req)->sense_len =
836 min(8 + cmd->sense_buffer[7],
837 SCSI_SENSE_BUFFERSIZE);
840 *blk_statp = scsi_result_to_blk_status(cmd, result);
841 } else if (blk_rq_bytes(req) == 0 && sense_current) {
843 * Flush commands do not transfers any data, and thus cannot use
844 * good_bytes != blk_rq_bytes(req) as the signal for an error.
845 * This sets *blk_statp explicitly for the problem case.
847 *blk_statp = scsi_result_to_blk_status(cmd, result);
850 * Recovered errors need reporting, but they're always treated as
851 * success, so fiddle the result code here. For passthrough requests
852 * we already took a copy of the original into sreq->result which
853 * is what gets returned to the user
855 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
856 bool do_print = true;
858 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
859 * skip print since caller wants ATA registers. Only occurs
860 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
862 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
864 else if (req->rq_flags & RQF_QUIET)
867 scsi_print_sense(cmd);
869 /* for passthrough, *blk_statp may be set */
870 *blk_statp = BLK_STS_OK;
873 * Another corner case: the SCSI status byte is non-zero but 'good'.
874 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
875 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
876 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
877 * intermediate statuses (both obsolete in SAM-4) as good.
879 if (status_byte(result) && scsi_status_is_good(result)) {
881 *blk_statp = BLK_STS_OK;
887 * scsi_io_completion - Completion processing for SCSI commands.
888 * @cmd: command that is finished.
889 * @good_bytes: number of processed bytes.
891 * We will finish off the specified number of sectors. If we are done, the
892 * command block will be released and the queue function will be goosed. If we
893 * are not done then we have to figure out what to do next:
895 * a) We can call scsi_io_completion_reprep(). The request will be
896 * unprepared and put back on the queue. Then a new command will
897 * be created for it. This should be used if we made forward
898 * progress, or if we want to switch from READ(10) to READ(6) for
901 * b) We can call scsi_io_completion_action(). The request will be
902 * put back on the queue and retried using the same command as
903 * before, possibly after a delay.
905 * c) We can call scsi_end_request() with blk_stat other than
906 * BLK_STS_OK, to fail the remainder of the request.
908 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
910 int result = cmd->result;
911 struct request_queue *q = cmd->device->request_queue;
912 struct request *req = cmd->request;
913 blk_status_t blk_stat = BLK_STS_OK;
915 if (unlikely(result)) /* a nz result may or may not be an error */
916 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
918 if (unlikely(blk_rq_is_passthrough(req))) {
920 * scsi_result_to_blk_status may have reset the host_byte
922 scsi_req(req)->result = cmd->result;
926 * Next deal with any sectors which we were able to correctly
929 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
930 "%u sectors total, %d bytes done.\n",
931 blk_rq_sectors(req), good_bytes));
934 * Failed, zero length commands always need to drop down
935 * to retry code. Fast path should return in this block.
937 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
938 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
939 return; /* no bytes remaining */
942 /* Kill remainder if no retries. */
943 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
944 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
946 "Bytes remaining after failed, no-retry command");
951 * If there had been no error, but we have leftover bytes in the
952 * requeues just queue the command up again.
954 if (likely(result == 0))
955 scsi_io_completion_reprep(cmd, q);
957 scsi_io_completion_action(cmd, result);
960 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
963 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
964 !op_is_write(req_op(rq)) &&
965 sdev->host->hostt->dma_need_drain(rq);
969 * scsi_init_io - SCSI I/O initialization function.
970 * @cmd: command descriptor we wish to initialize
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_init_io(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_init_io);
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;
1115 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1116 flags |= SCMD_INITIALIZED;
1117 scsi_initialize_rq(rq);
1120 jiffies_at_alloc = cmd->jiffies_at_alloc;
1121 retries = cmd->retries;
1122 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1124 * Zero out the cmd, except for the embedded scsi_request. Only clear
1125 * the driver-private command data if the LLD does not supply a
1126 * function to initialize that data.
1128 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1129 if (!dev->host->hostt->init_cmd_priv)
1130 to_clear += dev->host->hostt->cmd_size;
1131 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1134 cmd->sense_buffer = buf;
1135 cmd->prot_sdb = prot;
1137 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1138 cmd->jiffies_at_alloc = jiffies_at_alloc;
1139 cmd->retries = retries;
1141 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1145 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1146 struct request *req)
1148 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1151 * Passthrough requests may transfer data, in which case they must
1152 * a bio attached to them. Or they might contain a SCSI command
1153 * that does not transfer data, in which case they may optionally
1154 * submit a request without an attached bio.
1157 blk_status_t ret = scsi_init_io(cmd);
1158 if (unlikely(ret != BLK_STS_OK))
1161 BUG_ON(blk_rq_bytes(req));
1163 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1166 cmd->cmd_len = scsi_req(req)->cmd_len;
1167 cmd->cmnd = scsi_req(req)->cmd;
1168 cmd->transfersize = blk_rq_bytes(req);
1169 cmd->allowed = scsi_req(req)->retries;
1174 * Setup a normal block command. These are simple request from filesystems
1175 * that still need to be translated to SCSI CDBs from the ULD.
1177 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1178 struct request *req)
1180 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1182 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1183 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1184 if (ret != BLK_STS_OK)
1188 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1189 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1190 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1193 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1194 struct request *req)
1196 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1199 if (!blk_rq_bytes(req))
1200 cmd->sc_data_direction = DMA_NONE;
1201 else if (rq_data_dir(req) == WRITE)
1202 cmd->sc_data_direction = DMA_TO_DEVICE;
1204 cmd->sc_data_direction = DMA_FROM_DEVICE;
1206 if (blk_rq_is_scsi(req))
1207 ret = scsi_setup_scsi_cmnd(sdev, req);
1209 ret = scsi_setup_fs_cmnd(sdev, req);
1211 if (ret != BLK_STS_OK)
1212 scsi_free_sgtables(cmd);
1218 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1220 switch (sdev->sdev_state) {
1222 case SDEV_TRANSPORT_OFFLINE:
1224 * If the device is offline we refuse to process any
1225 * commands. The device must be brought online
1226 * before trying any recovery commands.
1228 if (!sdev->offline_already) {
1229 sdev->offline_already = true;
1230 sdev_printk(KERN_ERR, sdev,
1231 "rejecting I/O to offline device\n");
1233 return BLK_STS_IOERR;
1236 * If the device is fully deleted, we refuse to
1237 * process any commands as well.
1239 sdev_printk(KERN_ERR, sdev,
1240 "rejecting I/O to dead device\n");
1241 return BLK_STS_IOERR;
1243 case SDEV_CREATED_BLOCK:
1244 return BLK_STS_RESOURCE;
1247 * If the devices is blocked we defer normal commands.
1249 if (req && !(req->rq_flags & RQF_PREEMPT))
1250 return BLK_STS_RESOURCE;
1254 * For any other not fully online state we only allow
1255 * special commands. In particular any user initiated
1256 * command is not allowed.
1258 if (req && !(req->rq_flags & RQF_PREEMPT))
1259 return BLK_STS_IOERR;
1265 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1268 * Called with the queue_lock held.
1270 static inline int scsi_dev_queue_ready(struct request_queue *q,
1271 struct scsi_device *sdev)
1275 busy = atomic_inc_return(&sdev->device_busy) - 1;
1276 if (atomic_read(&sdev->device_blocked)) {
1281 * unblock after device_blocked iterates to zero
1283 if (atomic_dec_return(&sdev->device_blocked) > 0)
1285 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1286 "unblocking device at zero depth\n"));
1289 if (busy >= sdev->queue_depth)
1294 atomic_dec(&sdev->device_busy);
1299 * scsi_target_queue_ready: checks if there we can send commands to target
1300 * @sdev: scsi device on starget to check.
1302 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1303 struct scsi_device *sdev)
1305 struct scsi_target *starget = scsi_target(sdev);
1308 if (starget->single_lun) {
1309 spin_lock_irq(shost->host_lock);
1310 if (starget->starget_sdev_user &&
1311 starget->starget_sdev_user != sdev) {
1312 spin_unlock_irq(shost->host_lock);
1315 starget->starget_sdev_user = sdev;
1316 spin_unlock_irq(shost->host_lock);
1319 if (starget->can_queue <= 0)
1322 busy = atomic_inc_return(&starget->target_busy) - 1;
1323 if (atomic_read(&starget->target_blocked) > 0) {
1328 * unblock after target_blocked iterates to zero
1330 if (atomic_dec_return(&starget->target_blocked) > 0)
1333 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1334 "unblocking target at zero depth\n"));
1337 if (busy >= starget->can_queue)
1343 spin_lock_irq(shost->host_lock);
1344 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1345 spin_unlock_irq(shost->host_lock);
1347 if (starget->can_queue > 0)
1348 atomic_dec(&starget->target_busy);
1353 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1354 * return 0. We must end up running the queue again whenever 0 is
1355 * returned, else IO can hang.
1357 static inline int scsi_host_queue_ready(struct request_queue *q,
1358 struct Scsi_Host *shost,
1359 struct scsi_device *sdev,
1360 struct scsi_cmnd *cmd)
1362 if (scsi_host_in_recovery(shost))
1365 if (atomic_read(&shost->host_blocked) > 0) {
1366 if (scsi_host_busy(shost) > 0)
1370 * unblock after host_blocked iterates to zero
1372 if (atomic_dec_return(&shost->host_blocked) > 0)
1376 shost_printk(KERN_INFO, shost,
1377 "unblocking host at zero depth\n"));
1380 if (shost->host_self_blocked)
1383 /* We're OK to process the command, so we can't be starved */
1384 if (!list_empty(&sdev->starved_entry)) {
1385 spin_lock_irq(shost->host_lock);
1386 if (!list_empty(&sdev->starved_entry))
1387 list_del_init(&sdev->starved_entry);
1388 spin_unlock_irq(shost->host_lock);
1391 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1396 spin_lock_irq(shost->host_lock);
1397 if (list_empty(&sdev->starved_entry))
1398 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1399 spin_unlock_irq(shost->host_lock);
1401 scsi_dec_host_busy(shost, cmd);
1406 * Busy state exporting function for request stacking drivers.
1408 * For efficiency, no lock is taken to check the busy state of
1409 * shost/starget/sdev, since the returned value is not guaranteed and
1410 * may be changed after request stacking drivers call the function,
1411 * regardless of taking lock or not.
1413 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1414 * needs to return 'not busy'. Otherwise, request stacking drivers
1415 * may hold requests forever.
1417 static bool scsi_mq_lld_busy(struct request_queue *q)
1419 struct scsi_device *sdev = q->queuedata;
1420 struct Scsi_Host *shost;
1422 if (blk_queue_dying(q))
1428 * Ignore host/starget busy state.
1429 * Since block layer does not have a concept of fairness across
1430 * multiple queues, congestion of host/starget needs to be handled
1433 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1439 static void scsi_softirq_done(struct request *rq)
1441 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1442 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1445 INIT_LIST_HEAD(&cmd->eh_entry);
1447 atomic_inc(&cmd->device->iodone_cnt);
1449 atomic_inc(&cmd->device->ioerr_cnt);
1451 disposition = scsi_decide_disposition(cmd);
1452 if (disposition != SUCCESS &&
1453 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1454 scmd_printk(KERN_ERR, cmd,
1455 "timing out command, waited %lus\n",
1457 disposition = SUCCESS;
1460 scsi_log_completion(cmd, disposition);
1462 switch (disposition) {
1464 scsi_finish_command(cmd);
1467 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1469 case ADD_TO_MLQUEUE:
1470 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1473 scsi_eh_scmd_add(cmd);
1479 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1480 * @cmd: command block we are dispatching.
1482 * Return: nonzero return request was rejected and device's queue needs to be
1485 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1487 struct Scsi_Host *host = cmd->device->host;
1490 atomic_inc(&cmd->device->iorequest_cnt);
1492 /* check if the device is still usable */
1493 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1494 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1495 * returns an immediate error upwards, and signals
1496 * that the device is no longer present */
1497 cmd->result = DID_NO_CONNECT << 16;
1501 /* Check to see if the scsi lld made this device blocked. */
1502 if (unlikely(scsi_device_blocked(cmd->device))) {
1504 * in blocked state, the command is just put back on
1505 * the device queue. The suspend state has already
1506 * blocked the queue so future requests should not
1507 * occur until the device transitions out of the
1510 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1511 "queuecommand : device blocked\n"));
1512 return SCSI_MLQUEUE_DEVICE_BUSY;
1515 /* Store the LUN value in cmnd, if needed. */
1516 if (cmd->device->lun_in_cdb)
1517 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1518 (cmd->device->lun << 5 & 0xe0);
1523 * Before we queue this command, check if the command
1524 * length exceeds what the host adapter can handle.
1526 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1527 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1528 "queuecommand : command too long. "
1529 "cdb_size=%d host->max_cmd_len=%d\n",
1530 cmd->cmd_len, cmd->device->host->max_cmd_len));
1531 cmd->result = (DID_ABORT << 16);
1535 if (unlikely(host->shost_state == SHOST_DEL)) {
1536 cmd->result = (DID_NO_CONNECT << 16);
1541 trace_scsi_dispatch_cmd_start(cmd);
1542 rtn = host->hostt->queuecommand(host, cmd);
1544 trace_scsi_dispatch_cmd_error(cmd, rtn);
1545 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1546 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1547 rtn = SCSI_MLQUEUE_HOST_BUSY;
1549 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1550 "queuecommand : request rejected\n"));
1555 cmd->scsi_done(cmd);
1559 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1560 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1562 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1563 sizeof(struct scatterlist);
1566 static blk_status_t scsi_mq_prep_fn(struct request *req)
1568 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1569 struct scsi_device *sdev = req->q->queuedata;
1570 struct Scsi_Host *shost = sdev->host;
1571 struct scatterlist *sg;
1573 scsi_init_command(sdev, cmd);
1576 cmd->tag = req->tag;
1577 cmd->prot_op = SCSI_PROT_NORMAL;
1579 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 cmd->sdb.table.sgl = sg;
1582 if (scsi_host_get_prot(shost)) {
1583 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1585 cmd->prot_sdb->table.sgl =
1586 (struct scatterlist *)(cmd->prot_sdb + 1);
1589 blk_mq_start_request(req);
1591 return scsi_setup_cmnd(sdev, req);
1594 static void scsi_mq_done(struct scsi_cmnd *cmd)
1596 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1598 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1600 trace_scsi_dispatch_cmd_done(cmd);
1601 blk_mq_complete_request(cmd->request);
1604 static void scsi_mq_put_budget(struct request_queue *q)
1606 struct scsi_device *sdev = q->queuedata;
1608 atomic_dec(&sdev->device_busy);
1611 static bool scsi_mq_get_budget(struct request_queue *q)
1613 struct scsi_device *sdev = q->queuedata;
1615 return scsi_dev_queue_ready(q, sdev);
1618 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1619 const struct blk_mq_queue_data *bd)
1621 struct request *req = bd->rq;
1622 struct request_queue *q = req->q;
1623 struct scsi_device *sdev = q->queuedata;
1624 struct Scsi_Host *shost = sdev->host;
1625 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1630 * If the device is not in running state we will reject some or all
1633 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1634 ret = scsi_prep_state_check(sdev, req);
1635 if (ret != BLK_STS_OK)
1636 goto out_put_budget;
1639 ret = BLK_STS_RESOURCE;
1640 if (!scsi_target_queue_ready(shost, sdev))
1641 goto out_put_budget;
1642 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1643 goto out_dec_target_busy;
1645 if (!(req->rq_flags & RQF_DONTPREP)) {
1646 ret = scsi_mq_prep_fn(req);
1647 if (ret != BLK_STS_OK)
1648 goto out_dec_host_busy;
1649 req->rq_flags |= RQF_DONTPREP;
1651 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1652 blk_mq_start_request(req);
1655 cmd->flags &= SCMD_PRESERVED_FLAGS;
1656 if (sdev->simple_tags)
1657 cmd->flags |= SCMD_TAGGED;
1659 cmd->flags |= SCMD_LAST;
1661 scsi_init_cmd_errh(cmd);
1662 cmd->scsi_done = scsi_mq_done;
1664 reason = scsi_dispatch_cmd(cmd);
1666 scsi_set_blocked(cmd, reason);
1667 ret = BLK_STS_RESOURCE;
1668 goto out_dec_host_busy;
1674 scsi_dec_host_busy(shost, cmd);
1675 out_dec_target_busy:
1676 if (scsi_target(sdev)->can_queue > 0)
1677 atomic_dec(&scsi_target(sdev)->target_busy);
1679 scsi_mq_put_budget(q);
1683 case BLK_STS_RESOURCE:
1684 case BLK_STS_ZONE_RESOURCE:
1685 if (atomic_read(&sdev->device_busy) ||
1686 scsi_device_blocked(sdev))
1687 ret = BLK_STS_DEV_RESOURCE;
1690 if (unlikely(!scsi_device_online(sdev)))
1691 scsi_req(req)->result = DID_NO_CONNECT << 16;
1693 scsi_req(req)->result = DID_ERROR << 16;
1695 * Make sure to release all allocated resources when
1696 * we hit an error, as we will never see this command
1699 if (req->rq_flags & RQF_DONTPREP)
1700 scsi_mq_uninit_cmd(cmd);
1701 scsi_run_queue_async(sdev);
1707 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1711 return BLK_EH_RESET_TIMER;
1712 return scsi_times_out(req);
1715 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1716 unsigned int hctx_idx, unsigned int numa_node)
1718 struct Scsi_Host *shost = set->driver_data;
1719 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1720 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1721 struct scatterlist *sg;
1724 if (unchecked_isa_dma)
1725 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1726 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1727 GFP_KERNEL, numa_node);
1728 if (!cmd->sense_buffer)
1730 cmd->req.sense = cmd->sense_buffer;
1732 if (scsi_host_get_prot(shost)) {
1733 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1734 shost->hostt->cmd_size;
1735 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1738 if (shost->hostt->init_cmd_priv) {
1739 ret = shost->hostt->init_cmd_priv(shost, cmd);
1741 scsi_free_sense_buffer(unchecked_isa_dma,
1748 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1749 unsigned int hctx_idx)
1751 struct Scsi_Host *shost = set->driver_data;
1752 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1754 if (shost->hostt->exit_cmd_priv)
1755 shost->hostt->exit_cmd_priv(shost, cmd);
1756 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1760 static int scsi_map_queues(struct blk_mq_tag_set *set)
1762 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1764 if (shost->hostt->map_queues)
1765 return shost->hostt->map_queues(shost);
1766 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1769 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1771 struct device *dev = shost->dma_dev;
1774 * this limit is imposed by hardware restrictions
1776 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1779 if (scsi_host_prot_dma(shost)) {
1780 shost->sg_prot_tablesize =
1781 min_not_zero(shost->sg_prot_tablesize,
1782 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1783 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1784 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1787 if (dev->dma_mask) {
1788 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1789 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1791 blk_queue_max_hw_sectors(q, shost->max_sectors);
1792 if (shost->unchecked_isa_dma)
1793 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1794 blk_queue_segment_boundary(q, shost->dma_boundary);
1795 dma_set_seg_boundary(dev, shost->dma_boundary);
1797 blk_queue_max_segment_size(q, shost->max_segment_size);
1798 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1799 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1802 * Set a reasonable default alignment: The larger of 32-byte (dword),
1803 * which is a common minimum for HBAs, and the minimum DMA alignment,
1804 * which is set by the platform.
1806 * Devices that require a bigger alignment can increase it later.
1808 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1810 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1812 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1813 .get_budget = scsi_mq_get_budget,
1814 .put_budget = scsi_mq_put_budget,
1815 .queue_rq = scsi_queue_rq,
1816 .complete = scsi_softirq_done,
1817 .timeout = scsi_timeout,
1818 #ifdef CONFIG_BLK_DEBUG_FS
1819 .show_rq = scsi_show_rq,
1821 .init_request = scsi_mq_init_request,
1822 .exit_request = scsi_mq_exit_request,
1823 .initialize_rq_fn = scsi_initialize_rq,
1824 .cleanup_rq = scsi_cleanup_rq,
1825 .busy = scsi_mq_lld_busy,
1826 .map_queues = scsi_map_queues,
1830 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1832 struct request_queue *q = hctx->queue;
1833 struct scsi_device *sdev = q->queuedata;
1834 struct Scsi_Host *shost = sdev->host;
1836 shost->hostt->commit_rqs(shost, hctx->queue_num);
1839 static const struct blk_mq_ops scsi_mq_ops = {
1840 .get_budget = scsi_mq_get_budget,
1841 .put_budget = scsi_mq_put_budget,
1842 .queue_rq = scsi_queue_rq,
1843 .commit_rqs = scsi_commit_rqs,
1844 .complete = scsi_softirq_done,
1845 .timeout = scsi_timeout,
1846 #ifdef CONFIG_BLK_DEBUG_FS
1847 .show_rq = scsi_show_rq,
1849 .init_request = scsi_mq_init_request,
1850 .exit_request = scsi_mq_exit_request,
1851 .initialize_rq_fn = scsi_initialize_rq,
1852 .cleanup_rq = scsi_cleanup_rq,
1853 .busy = scsi_mq_lld_busy,
1854 .map_queues = scsi_map_queues,
1857 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1859 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1860 if (IS_ERR(sdev->request_queue))
1863 sdev->request_queue->queuedata = sdev;
1864 __scsi_init_queue(sdev->host, sdev->request_queue);
1865 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1866 return sdev->request_queue;
1869 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1871 unsigned int cmd_size, sgl_size;
1872 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1874 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1875 scsi_mq_inline_sgl_size(shost));
1876 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1877 if (scsi_host_get_prot(shost))
1878 cmd_size += sizeof(struct scsi_data_buffer) +
1879 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1881 memset(tag_set, 0, sizeof(*tag_set));
1882 if (shost->hostt->commit_rqs)
1883 tag_set->ops = &scsi_mq_ops;
1885 tag_set->ops = &scsi_mq_ops_no_commit;
1886 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1887 tag_set->queue_depth = shost->can_queue;
1888 tag_set->cmd_size = cmd_size;
1889 tag_set->numa_node = NUMA_NO_NODE;
1890 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1892 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1893 tag_set->driver_data = shost;
1895 return blk_mq_alloc_tag_set(tag_set);
1898 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1900 blk_mq_free_tag_set(&shost->tag_set);
1904 * scsi_device_from_queue - return sdev associated with a request_queue
1905 * @q: The request queue to return the sdev from
1907 * Return the sdev associated with a request queue or NULL if the
1908 * request_queue does not reference a SCSI device.
1910 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1912 struct scsi_device *sdev = NULL;
1914 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1915 q->mq_ops == &scsi_mq_ops)
1916 sdev = q->queuedata;
1917 if (!sdev || !get_device(&sdev->sdev_gendev))
1922 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1925 * scsi_block_requests - Utility function used by low-level drivers to prevent
1926 * further commands from being queued to the device.
1927 * @shost: host in question
1929 * There is no timer nor any other means by which the requests get unblocked
1930 * other than the low-level driver calling scsi_unblock_requests().
1932 void scsi_block_requests(struct Scsi_Host *shost)
1934 shost->host_self_blocked = 1;
1936 EXPORT_SYMBOL(scsi_block_requests);
1939 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1940 * further commands to be queued to the device.
1941 * @shost: host in question
1943 * There is no timer nor any other means by which the requests get unblocked
1944 * other than the low-level driver calling scsi_unblock_requests(). This is done
1945 * as an API function so that changes to the internals of the scsi mid-layer
1946 * won't require wholesale changes to drivers that use this feature.
1948 void scsi_unblock_requests(struct Scsi_Host *shost)
1950 shost->host_self_blocked = 0;
1951 scsi_run_host_queues(shost);
1953 EXPORT_SYMBOL(scsi_unblock_requests);
1955 void scsi_exit_queue(void)
1957 kmem_cache_destroy(scsi_sense_cache);
1958 kmem_cache_destroy(scsi_sense_isadma_cache);
1962 * scsi_mode_select - issue a mode select
1963 * @sdev: SCSI device to be queried
1964 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1965 * @sp: Save page bit (0 == don't save, 1 == save)
1966 * @modepage: mode page being requested
1967 * @buffer: request buffer (may not be smaller than eight bytes)
1968 * @len: length of request buffer.
1969 * @timeout: command timeout
1970 * @retries: number of retries before failing
1971 * @data: returns a structure abstracting the mode header data
1972 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1973 * must be SCSI_SENSE_BUFFERSIZE big.
1975 * Returns zero if successful; negative error number or scsi
1980 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1981 unsigned char *buffer, int len, int timeout, int retries,
1982 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1984 unsigned char cmd[10];
1985 unsigned char *real_buffer;
1988 memset(cmd, 0, sizeof(cmd));
1989 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1991 if (sdev->use_10_for_ms) {
1994 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1997 memcpy(real_buffer + 8, buffer, len);
2001 real_buffer[2] = data->medium_type;
2002 real_buffer[3] = data->device_specific;
2003 real_buffer[4] = data->longlba ? 0x01 : 0;
2005 real_buffer[6] = data->block_descriptor_length >> 8;
2006 real_buffer[7] = data->block_descriptor_length;
2008 cmd[0] = MODE_SELECT_10;
2012 if (len > 255 || data->block_descriptor_length > 255 ||
2016 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2019 memcpy(real_buffer + 4, buffer, len);
2022 real_buffer[1] = data->medium_type;
2023 real_buffer[2] = data->device_specific;
2024 real_buffer[3] = data->block_descriptor_length;
2026 cmd[0] = MODE_SELECT;
2030 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2031 sshdr, timeout, retries, NULL);
2035 EXPORT_SYMBOL_GPL(scsi_mode_select);
2038 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2039 * @sdev: SCSI device to be queried
2040 * @dbd: set if mode sense will allow block descriptors to be returned
2041 * @modepage: mode page being requested
2042 * @buffer: request buffer (may not be smaller than eight bytes)
2043 * @len: length of request buffer.
2044 * @timeout: command timeout
2045 * @retries: number of retries before failing
2046 * @data: returns a structure abstracting the mode header data
2047 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2048 * must be SCSI_SENSE_BUFFERSIZE big.
2050 * Returns zero if unsuccessful, or the header offset (either 4
2051 * or 8 depending on whether a six or ten byte command was
2052 * issued) if successful.
2055 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2056 unsigned char *buffer, int len, int timeout, int retries,
2057 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2059 unsigned char cmd[12];
2062 int result, retry_count = retries;
2063 struct scsi_sense_hdr my_sshdr;
2065 memset(data, 0, sizeof(*data));
2066 memset(&cmd[0], 0, 12);
2068 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2069 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2072 /* caller might not be interested in sense, but we need it */
2077 use_10_for_ms = sdev->use_10_for_ms;
2079 if (use_10_for_ms) {
2083 cmd[0] = MODE_SENSE_10;
2090 cmd[0] = MODE_SENSE;
2095 memset(buffer, 0, len);
2097 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2098 sshdr, timeout, retries, NULL);
2100 /* This code looks awful: what it's doing is making sure an
2101 * ILLEGAL REQUEST sense return identifies the actual command
2102 * byte as the problem. MODE_SENSE commands can return
2103 * ILLEGAL REQUEST if the code page isn't supported */
2105 if (use_10_for_ms && !scsi_status_is_good(result) &&
2106 driver_byte(result) == DRIVER_SENSE) {
2107 if (scsi_sense_valid(sshdr)) {
2108 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2109 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2111 * Invalid command operation code
2113 sdev->use_10_for_ms = 0;
2119 if (scsi_status_is_good(result)) {
2120 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2121 (modepage == 6 || modepage == 8))) {
2122 /* Initio breakage? */
2125 data->medium_type = 0;
2126 data->device_specific = 0;
2128 data->block_descriptor_length = 0;
2129 } else if (use_10_for_ms) {
2130 data->length = buffer[0]*256 + buffer[1] + 2;
2131 data->medium_type = buffer[2];
2132 data->device_specific = buffer[3];
2133 data->longlba = buffer[4] & 0x01;
2134 data->block_descriptor_length = buffer[6]*256
2137 data->length = buffer[0] + 1;
2138 data->medium_type = buffer[1];
2139 data->device_specific = buffer[2];
2140 data->block_descriptor_length = buffer[3];
2142 data->header_length = header_length;
2143 } else if ((status_byte(result) == CHECK_CONDITION) &&
2144 scsi_sense_valid(sshdr) &&
2145 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2152 EXPORT_SYMBOL(scsi_mode_sense);
2155 * scsi_test_unit_ready - test if unit is ready
2156 * @sdev: scsi device to change the state of.
2157 * @timeout: command timeout
2158 * @retries: number of retries before failing
2159 * @sshdr: outpout pointer for decoded sense information.
2161 * Returns zero if unsuccessful or an error if TUR failed. For
2162 * removable media, UNIT_ATTENTION sets ->changed flag.
2165 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2166 struct scsi_sense_hdr *sshdr)
2169 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2173 /* try to eat the UNIT_ATTENTION if there are enough retries */
2175 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2177 if (sdev->removable && scsi_sense_valid(sshdr) &&
2178 sshdr->sense_key == UNIT_ATTENTION)
2180 } while (scsi_sense_valid(sshdr) &&
2181 sshdr->sense_key == UNIT_ATTENTION && --retries);
2185 EXPORT_SYMBOL(scsi_test_unit_ready);
2188 * scsi_device_set_state - Take the given device through the device state model.
2189 * @sdev: scsi device to change the state of.
2190 * @state: state to change to.
2192 * Returns zero if successful or an error if the requested
2193 * transition is illegal.
2196 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2198 enum scsi_device_state oldstate = sdev->sdev_state;
2200 if (state == oldstate)
2206 case SDEV_CREATED_BLOCK:
2217 case SDEV_TRANSPORT_OFFLINE:
2230 case SDEV_TRANSPORT_OFFLINE:
2238 case SDEV_TRANSPORT_OFFLINE:
2253 case SDEV_CREATED_BLOCK:
2262 case SDEV_CREATED_BLOCK:
2277 case SDEV_TRANSPORT_OFFLINE:
2289 case SDEV_TRANSPORT_OFFLINE:
2292 case SDEV_CREATED_BLOCK:
2300 sdev->offline_already = false;
2301 sdev->sdev_state = state;
2305 SCSI_LOG_ERROR_RECOVERY(1,
2306 sdev_printk(KERN_ERR, sdev,
2307 "Illegal state transition %s->%s",
2308 scsi_device_state_name(oldstate),
2309 scsi_device_state_name(state))
2313 EXPORT_SYMBOL(scsi_device_set_state);
2316 * sdev_evt_emit - emit a single SCSI device uevent
2317 * @sdev: associated SCSI device
2318 * @evt: event to emit
2320 * Send a single uevent (scsi_event) to the associated scsi_device.
2322 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2327 switch (evt->evt_type) {
2328 case SDEV_EVT_MEDIA_CHANGE:
2329 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2331 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2332 scsi_rescan_device(&sdev->sdev_gendev);
2333 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2335 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2336 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2338 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2339 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2341 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2342 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2344 case SDEV_EVT_LUN_CHANGE_REPORTED:
2345 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2347 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2348 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2350 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2351 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2360 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2364 * sdev_evt_thread - send a uevent for each scsi event
2365 * @work: work struct for scsi_device
2367 * Dispatch queued events to their associated scsi_device kobjects
2370 void scsi_evt_thread(struct work_struct *work)
2372 struct scsi_device *sdev;
2373 enum scsi_device_event evt_type;
2374 LIST_HEAD(event_list);
2376 sdev = container_of(work, struct scsi_device, event_work);
2378 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2379 if (test_and_clear_bit(evt_type, sdev->pending_events))
2380 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2383 struct scsi_event *evt;
2384 struct list_head *this, *tmp;
2385 unsigned long flags;
2387 spin_lock_irqsave(&sdev->list_lock, flags);
2388 list_splice_init(&sdev->event_list, &event_list);
2389 spin_unlock_irqrestore(&sdev->list_lock, flags);
2391 if (list_empty(&event_list))
2394 list_for_each_safe(this, tmp, &event_list) {
2395 evt = list_entry(this, struct scsi_event, node);
2396 list_del(&evt->node);
2397 scsi_evt_emit(sdev, evt);
2404 * sdev_evt_send - send asserted event to uevent thread
2405 * @sdev: scsi_device event occurred on
2406 * @evt: event to send
2408 * Assert scsi device event asynchronously.
2410 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2412 unsigned long flags;
2415 /* FIXME: currently this check eliminates all media change events
2416 * for polled devices. Need to update to discriminate between AN
2417 * and polled events */
2418 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2424 spin_lock_irqsave(&sdev->list_lock, flags);
2425 list_add_tail(&evt->node, &sdev->event_list);
2426 schedule_work(&sdev->event_work);
2427 spin_unlock_irqrestore(&sdev->list_lock, flags);
2429 EXPORT_SYMBOL_GPL(sdev_evt_send);
2432 * sdev_evt_alloc - allocate a new scsi event
2433 * @evt_type: type of event to allocate
2434 * @gfpflags: GFP flags for allocation
2436 * Allocates and returns a new scsi_event.
2438 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2441 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2445 evt->evt_type = evt_type;
2446 INIT_LIST_HEAD(&evt->node);
2448 /* evt_type-specific initialization, if any */
2450 case SDEV_EVT_MEDIA_CHANGE:
2451 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2452 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2453 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2454 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2455 case SDEV_EVT_LUN_CHANGE_REPORTED:
2456 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2457 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2465 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2468 * sdev_evt_send_simple - send asserted event to uevent thread
2469 * @sdev: scsi_device event occurred on
2470 * @evt_type: type of event to send
2471 * @gfpflags: GFP flags for allocation
2473 * Assert scsi device event asynchronously, given an event type.
2475 void sdev_evt_send_simple(struct scsi_device *sdev,
2476 enum scsi_device_event evt_type, gfp_t gfpflags)
2478 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2480 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2485 sdev_evt_send(sdev, evt);
2487 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2490 * scsi_device_quiesce - Block user issued commands.
2491 * @sdev: scsi device to quiesce.
2493 * This works by trying to transition to the SDEV_QUIESCE state
2494 * (which must be a legal transition). When the device is in this
2495 * state, only special requests will be accepted, all others will
2496 * be deferred. Since special requests may also be requeued requests,
2497 * a successful return doesn't guarantee the device will be
2498 * totally quiescent.
2500 * Must be called with user context, may sleep.
2502 * Returns zero if unsuccessful or an error if not.
2505 scsi_device_quiesce(struct scsi_device *sdev)
2507 struct request_queue *q = sdev->request_queue;
2511 * It is allowed to call scsi_device_quiesce() multiple times from
2512 * the same context but concurrent scsi_device_quiesce() calls are
2515 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2517 if (sdev->quiesced_by == current)
2522 blk_mq_freeze_queue(q);
2524 * Ensure that the effect of blk_set_pm_only() will be visible
2525 * for percpu_ref_tryget() callers that occur after the queue
2526 * unfreeze even if the queue was already frozen before this function
2527 * was called. See also https://lwn.net/Articles/573497/.
2530 blk_mq_unfreeze_queue(q);
2532 mutex_lock(&sdev->state_mutex);
2533 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2535 sdev->quiesced_by = current;
2537 blk_clear_pm_only(q);
2538 mutex_unlock(&sdev->state_mutex);
2542 EXPORT_SYMBOL(scsi_device_quiesce);
2545 * scsi_device_resume - Restart user issued commands to a quiesced device.
2546 * @sdev: scsi device to resume.
2548 * Moves the device from quiesced back to running and restarts the
2551 * Must be called with user context, may sleep.
2553 void scsi_device_resume(struct scsi_device *sdev)
2555 /* check if the device state was mutated prior to resume, and if
2556 * so assume the state is being managed elsewhere (for example
2557 * device deleted during suspend)
2559 mutex_lock(&sdev->state_mutex);
2560 if (sdev->quiesced_by) {
2561 sdev->quiesced_by = NULL;
2562 blk_clear_pm_only(sdev->request_queue);
2564 if (sdev->sdev_state == SDEV_QUIESCE)
2565 scsi_device_set_state(sdev, SDEV_RUNNING);
2566 mutex_unlock(&sdev->state_mutex);
2568 EXPORT_SYMBOL(scsi_device_resume);
2571 device_quiesce_fn(struct scsi_device *sdev, void *data)
2573 scsi_device_quiesce(sdev);
2577 scsi_target_quiesce(struct scsi_target *starget)
2579 starget_for_each_device(starget, NULL, device_quiesce_fn);
2581 EXPORT_SYMBOL(scsi_target_quiesce);
2584 device_resume_fn(struct scsi_device *sdev, void *data)
2586 scsi_device_resume(sdev);
2590 scsi_target_resume(struct scsi_target *starget)
2592 starget_for_each_device(starget, NULL, device_resume_fn);
2594 EXPORT_SYMBOL(scsi_target_resume);
2597 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2598 * @sdev: device to block
2600 * Pause SCSI command processing on the specified device. Does not sleep.
2602 * Returns zero if successful or a negative error code upon failure.
2605 * This routine transitions the device to the SDEV_BLOCK state (which must be
2606 * a legal transition). When the device is in this state, command processing
2607 * is paused until the device leaves the SDEV_BLOCK state. See also
2608 * scsi_internal_device_unblock_nowait().
2610 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2612 struct request_queue *q = sdev->request_queue;
2615 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2617 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2624 * The device has transitioned to SDEV_BLOCK. Stop the
2625 * block layer from calling the midlayer with this device's
2628 blk_mq_quiesce_queue_nowait(q);
2631 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2634 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2635 * @sdev: device to block
2637 * Pause SCSI command processing on the specified device and wait until all
2638 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2640 * Returns zero if successful or a negative error code upon failure.
2643 * This routine transitions the device to the SDEV_BLOCK state (which must be
2644 * a legal transition). When the device is in this state, command processing
2645 * is paused until the device leaves the SDEV_BLOCK state. See also
2646 * scsi_internal_device_unblock().
2648 static int scsi_internal_device_block(struct scsi_device *sdev)
2650 struct request_queue *q = sdev->request_queue;
2653 mutex_lock(&sdev->state_mutex);
2654 err = scsi_internal_device_block_nowait(sdev);
2656 blk_mq_quiesce_queue(q);
2657 mutex_unlock(&sdev->state_mutex);
2662 void scsi_start_queue(struct scsi_device *sdev)
2664 struct request_queue *q = sdev->request_queue;
2666 blk_mq_unquiesce_queue(q);
2670 * scsi_internal_device_unblock_nowait - resume a device after a block request
2671 * @sdev: device to resume
2672 * @new_state: state to set the device to after unblocking
2674 * Restart the device queue for a previously suspended SCSI device. Does not
2677 * Returns zero if successful or a negative error code upon failure.
2680 * This routine transitions the device to the SDEV_RUNNING state or to one of
2681 * the offline states (which must be a legal transition) allowing the midlayer
2682 * to goose the queue for this device.
2684 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2685 enum scsi_device_state new_state)
2687 switch (new_state) {
2689 case SDEV_TRANSPORT_OFFLINE:
2696 * Try to transition the scsi device to SDEV_RUNNING or one of the
2697 * offlined states and goose the device queue if successful.
2699 switch (sdev->sdev_state) {
2701 case SDEV_TRANSPORT_OFFLINE:
2702 sdev->sdev_state = new_state;
2704 case SDEV_CREATED_BLOCK:
2705 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2706 new_state == SDEV_OFFLINE)
2707 sdev->sdev_state = new_state;
2709 sdev->sdev_state = SDEV_CREATED;
2717 scsi_start_queue(sdev);
2721 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2724 * scsi_internal_device_unblock - resume a device after a block request
2725 * @sdev: device to resume
2726 * @new_state: state to set the device to after unblocking
2728 * Restart the device queue for a previously suspended SCSI device. May sleep.
2730 * Returns zero if successful or a negative error code upon failure.
2733 * This routine transitions the device to the SDEV_RUNNING state or to one of
2734 * the offline states (which must be a legal transition) allowing the midlayer
2735 * to goose the queue for this device.
2737 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2738 enum scsi_device_state new_state)
2742 mutex_lock(&sdev->state_mutex);
2743 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2744 mutex_unlock(&sdev->state_mutex);
2750 device_block(struct scsi_device *sdev, void *data)
2754 ret = scsi_internal_device_block(sdev);
2756 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2757 dev_name(&sdev->sdev_gendev), ret);
2761 target_block(struct device *dev, void *data)
2763 if (scsi_is_target_device(dev))
2764 starget_for_each_device(to_scsi_target(dev), NULL,
2770 scsi_target_block(struct device *dev)
2772 if (scsi_is_target_device(dev))
2773 starget_for_each_device(to_scsi_target(dev), NULL,
2776 device_for_each_child(dev, NULL, target_block);
2778 EXPORT_SYMBOL_GPL(scsi_target_block);
2781 device_unblock(struct scsi_device *sdev, void *data)
2783 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2787 target_unblock(struct device *dev, void *data)
2789 if (scsi_is_target_device(dev))
2790 starget_for_each_device(to_scsi_target(dev), data,
2796 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2798 if (scsi_is_target_device(dev))
2799 starget_for_each_device(to_scsi_target(dev), &new_state,
2802 device_for_each_child(dev, &new_state, target_unblock);
2804 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2807 scsi_host_block(struct Scsi_Host *shost)
2809 struct scsi_device *sdev;
2813 * Call scsi_internal_device_block_nowait so we can avoid
2814 * calling synchronize_rcu() for each LUN.
2816 shost_for_each_device(sdev, shost) {
2817 mutex_lock(&sdev->state_mutex);
2818 ret = scsi_internal_device_block_nowait(sdev);
2819 mutex_unlock(&sdev->state_mutex);
2821 scsi_device_put(sdev);
2827 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2828 * calling synchronize_rcu() once is enough.
2830 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2837 EXPORT_SYMBOL_GPL(scsi_host_block);
2840 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2842 struct scsi_device *sdev;
2845 shost_for_each_device(sdev, shost) {
2846 ret = scsi_internal_device_unblock(sdev, new_state);
2848 scsi_device_put(sdev);
2854 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2857 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2858 * @sgl: scatter-gather list
2859 * @sg_count: number of segments in sg
2860 * @offset: offset in bytes into sg, on return offset into the mapped area
2861 * @len: bytes to map, on return number of bytes mapped
2863 * Returns virtual address of the start of the mapped page
2865 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2866 size_t *offset, size_t *len)
2869 size_t sg_len = 0, len_complete = 0;
2870 struct scatterlist *sg;
2873 WARN_ON(!irqs_disabled());
2875 for_each_sg(sgl, sg, sg_count, i) {
2876 len_complete = sg_len; /* Complete sg-entries */
2877 sg_len += sg->length;
2878 if (sg_len > *offset)
2882 if (unlikely(i == sg_count)) {
2883 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2885 __func__, sg_len, *offset, sg_count);
2890 /* Offset starting from the beginning of first page in this sg-entry */
2891 *offset = *offset - len_complete + sg->offset;
2893 /* Assumption: contiguous pages can be accessed as "page + i" */
2894 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2895 *offset &= ~PAGE_MASK;
2897 /* Bytes in this sg-entry from *offset to the end of the page */
2898 sg_len = PAGE_SIZE - *offset;
2902 return kmap_atomic(page);
2904 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2907 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2908 * @virt: virtual address to be unmapped
2910 void scsi_kunmap_atomic_sg(void *virt)
2912 kunmap_atomic(virt);
2914 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2916 void sdev_disable_disk_events(struct scsi_device *sdev)
2918 atomic_inc(&sdev->disk_events_disable_depth);
2920 EXPORT_SYMBOL(sdev_disable_disk_events);
2922 void sdev_enable_disk_events(struct scsi_device *sdev)
2924 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2926 atomic_dec(&sdev->disk_events_disable_depth);
2928 EXPORT_SYMBOL(sdev_enable_disk_events);
2931 * scsi_vpd_lun_id - return a unique device identification
2932 * @sdev: SCSI device
2933 * @id: buffer for the identification
2934 * @id_len: length of the buffer
2936 * Copies a unique device identification into @id based
2937 * on the information in the VPD page 0x83 of the device.
2938 * The string will be formatted as a SCSI name string.
2940 * Returns the length of the identification or error on failure.
2941 * If the identifier is longer than the supplied buffer the actual
2942 * identifier length is returned and the buffer is not zero-padded.
2944 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2946 u8 cur_id_type = 0xff;
2948 const unsigned char *d, *cur_id_str;
2949 const struct scsi_vpd *vpd_pg83;
2950 int id_size = -EINVAL;
2953 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2960 * Look for the correct descriptor.
2961 * Order of preference for lun descriptor:
2962 * - SCSI name string
2963 * - NAA IEEE Registered Extended
2964 * - EUI-64 based 16-byte
2965 * - EUI-64 based 12-byte
2966 * - NAA IEEE Registered
2967 * - NAA IEEE Extended
2969 * as longer descriptors reduce the likelyhood
2970 * of identification clashes.
2973 /* The id string must be at least 20 bytes + terminating NULL byte */
2979 memset(id, 0, id_len);
2980 d = vpd_pg83->data + 4;
2981 while (d < vpd_pg83->data + vpd_pg83->len) {
2982 /* Skip designators not referring to the LUN */
2983 if ((d[1] & 0x30) != 0x00)
2986 switch (d[1] & 0xf) {
2989 if (cur_id_size > d[3])
2991 /* Prefer anything */
2992 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2995 if (cur_id_size + 4 > id_len)
2996 cur_id_size = id_len - 4;
2998 cur_id_type = d[1] & 0xf;
2999 id_size = snprintf(id, id_len, "t10.%*pE",
3000 cur_id_size, cur_id_str);
3004 if (cur_id_size > d[3])
3006 /* Prefer NAA IEEE Registered Extended */
3007 if (cur_id_type == 0x3 &&
3008 cur_id_size == d[3])
3012 cur_id_type = d[1] & 0xf;
3013 switch (cur_id_size) {
3015 id_size = snprintf(id, id_len,
3020 id_size = snprintf(id, id_len,
3025 id_size = snprintf(id, id_len,
3036 if (cur_id_size > d[3])
3040 cur_id_type = d[1] & 0xf;
3041 switch (cur_id_size) {
3043 id_size = snprintf(id, id_len,
3048 id_size = snprintf(id, id_len,
3058 /* SCSI name string */
3059 if (cur_id_size + 4 > d[3])
3061 /* Prefer others for truncated descriptor */
3062 if (cur_id_size && d[3] > id_len)
3064 cur_id_size = id_size = d[3];
3066 cur_id_type = d[1] & 0xf;
3067 if (cur_id_size >= id_len)
3068 cur_id_size = id_len - 1;
3069 memcpy(id, cur_id_str, cur_id_size);
3070 /* Decrease priority for truncated descriptor */
3071 if (cur_id_size != id_size)
3084 EXPORT_SYMBOL(scsi_vpd_lun_id);
3087 * scsi_vpd_tpg_id - return a target port group identifier
3088 * @sdev: SCSI device
3090 * Returns the Target Port Group identifier from the information
3091 * froom VPD page 0x83 of the device.
3093 * Returns the identifier or error on failure.
3095 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3097 const unsigned char *d;
3098 const struct scsi_vpd *vpd_pg83;
3099 int group_id = -EAGAIN, rel_port = -1;
3102 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3108 d = vpd_pg83->data + 4;
3109 while (d < vpd_pg83->data + vpd_pg83->len) {
3110 switch (d[1] & 0xf) {
3112 /* Relative target port */
3113 rel_port = get_unaligned_be16(&d[6]);
3116 /* Target port group */
3117 group_id = get_unaligned_be16(&d[6]);
3126 if (group_id >= 0 && rel_id && rel_port != -1)
3131 EXPORT_SYMBOL(scsi_vpd_tpg_id);