2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
49 static inline struct kmem_cache *
50 scsi_select_sense_cache(bool unchecked_isa_dma)
52 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56 unsigned char *sense_buffer)
58 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
63 gfp_t gfp_mask, int numa_node)
65 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
69 int scsi_init_sense_cache(struct Scsi_Host *shost)
71 struct kmem_cache *cache;
74 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
78 mutex_lock(&scsi_sense_cache_mutex);
79 if (shost->unchecked_isa_dma) {
80 scsi_sense_isadma_cache =
81 kmem_cache_create("scsi_sense_cache(DMA)",
82 SCSI_SENSE_BUFFERSIZE, 0,
83 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
84 if (!scsi_sense_isadma_cache)
88 kmem_cache_create_usercopy("scsi_sense_cache",
89 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
90 0, SCSI_SENSE_BUFFERSIZE, NULL);
91 if (!scsi_sense_cache)
95 mutex_unlock(&scsi_sense_cache_mutex);
100 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
101 * not change behaviour from the previous unplug mechanism, experimentation
102 * may prove this needs changing.
104 #define SCSI_QUEUE_DELAY 3
107 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
109 struct Scsi_Host *host = cmd->device->host;
110 struct scsi_device *device = cmd->device;
111 struct scsi_target *starget = scsi_target(device);
114 * Set the appropriate busy bit for the device/host.
116 * If the host/device isn't busy, assume that something actually
117 * completed, and that we should be able to queue a command now.
119 * Note that the prior mid-layer assumption that any host could
120 * always queue at least one command is now broken. The mid-layer
121 * will implement a user specifiable stall (see
122 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
123 * if a command is requeued with no other commands outstanding
124 * either for the device or for the host.
127 case SCSI_MLQUEUE_HOST_BUSY:
128 atomic_set(&host->host_blocked, host->max_host_blocked);
130 case SCSI_MLQUEUE_DEVICE_BUSY:
131 case SCSI_MLQUEUE_EH_RETRY:
132 atomic_set(&device->device_blocked,
133 device->max_device_blocked);
135 case SCSI_MLQUEUE_TARGET_BUSY:
136 atomic_set(&starget->target_blocked,
137 starget->max_target_blocked);
142 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
144 if (cmd->request->rq_flags & RQF_DONTPREP) {
145 cmd->request->rq_flags &= ~RQF_DONTPREP;
146 scsi_mq_uninit_cmd(cmd);
150 blk_mq_requeue_request(cmd->request, true);
154 * __scsi_queue_insert - private queue insertion
155 * @cmd: The SCSI command being requeued
156 * @reason: The reason for the requeue
157 * @unbusy: Whether the queue should be unbusied
159 * This is a private queue insertion. The public interface
160 * scsi_queue_insert() always assumes the queue should be unbusied
161 * because it's always called before the completion. This function is
162 * for a requeue after completion, which should only occur in this
165 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
167 struct scsi_device *device = cmd->device;
169 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
170 "Inserting command %p into mlqueue\n", cmd));
172 scsi_set_blocked(cmd, reason);
175 * Decrement the counters, since these commands are no longer
176 * active on the host/device.
179 scsi_device_unbusy(device);
182 * Requeue this command. It will go before all other commands
183 * that are already in the queue. Schedule requeue work under
184 * lock such that the kblockd_schedule_work() call happens
185 * before blk_cleanup_queue() finishes.
189 blk_mq_requeue_request(cmd->request, true);
193 * Function: scsi_queue_insert()
195 * Purpose: Insert a command in the midlevel queue.
197 * Arguments: cmd - command that we are adding to queue.
198 * reason - why we are inserting command to queue.
200 * Lock status: Assumed that lock is not held upon entry.
204 * Notes: We do this for one of two cases. Either the host is busy
205 * and it cannot accept any more commands for the time being,
206 * or the device returned QUEUE_FULL and can accept no more
208 * Notes: This could be called either from an interrupt context or a
209 * normal process context.
211 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
213 __scsi_queue_insert(cmd, reason, true);
218 * __scsi_execute - insert request and wait for the result
221 * @data_direction: data direction
222 * @buffer: data buffer
223 * @bufflen: len of buffer
224 * @sense: optional sense buffer
225 * @sshdr: optional decoded sense header
226 * @timeout: request timeout in seconds
227 * @retries: number of times to retry request
228 * @flags: flags for ->cmd_flags
229 * @rq_flags: flags for ->rq_flags
230 * @resid: optional residual length
232 * Returns the scsi_cmnd result field if a command was executed, or a negative
233 * Linux error code if we didn't get that far.
235 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
236 int data_direction, void *buffer, unsigned bufflen,
237 unsigned char *sense, struct scsi_sense_hdr *sshdr,
238 int timeout, int retries, u64 flags, req_flags_t rq_flags,
242 struct scsi_request *rq;
243 int ret = DRIVER_ERROR << 24;
245 req = blk_get_request(sdev->request_queue,
246 data_direction == DMA_TO_DEVICE ?
247 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
252 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
253 buffer, bufflen, GFP_NOIO))
256 rq->cmd_len = COMMAND_SIZE(cmd[0]);
257 memcpy(rq->cmd, cmd, rq->cmd_len);
258 rq->retries = retries;
259 req->timeout = timeout;
260 req->cmd_flags |= flags;
261 req->rq_flags |= rq_flags | RQF_QUIET;
264 * head injection *required* here otherwise quiesce won't work
266 blk_execute_rq(req->q, NULL, req, 1);
269 * Some devices (USB mass-storage in particular) may transfer
270 * garbage data together with a residue indicating that the data
271 * is invalid. Prevent the garbage from being misinterpreted
272 * and prevent security leaks by zeroing out the excess data.
274 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
275 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
278 *resid = rq->resid_len;
279 if (sense && rq->sense_len)
280 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
282 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
285 blk_put_request(req);
289 EXPORT_SYMBOL(__scsi_execute);
292 * Function: scsi_init_cmd_errh()
294 * Purpose: Initialize cmd fields related to error handling.
296 * Arguments: cmd - command that is ready to be queued.
298 * Notes: This function has the job of initializing a number of
299 * fields related to error handling. Typically this will
300 * be called once for each command, as required.
302 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
304 scsi_set_resid(cmd, 0);
305 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
306 if (cmd->cmd_len == 0)
307 cmd->cmd_len = scsi_command_size(cmd->cmnd);
311 * Decrement the host_busy counter and wake up the error handler if necessary.
312 * Avoid as follows that the error handler is not woken up if shost->host_busy
313 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
314 * with an RCU read lock in this function to ensure that this function in its
315 * entirety either finishes before scsi_eh_scmd_add() increases the
316 * host_failed counter or that it notices the shost state change made by
317 * scsi_eh_scmd_add().
319 static void scsi_dec_host_busy(struct Scsi_Host *shost)
324 atomic_dec(&shost->host_busy);
325 if (unlikely(scsi_host_in_recovery(shost))) {
326 spin_lock_irqsave(shost->host_lock, flags);
327 if (shost->host_failed || shost->host_eh_scheduled)
328 scsi_eh_wakeup(shost);
329 spin_unlock_irqrestore(shost->host_lock, flags);
334 void scsi_device_unbusy(struct scsi_device *sdev)
336 struct Scsi_Host *shost = sdev->host;
337 struct scsi_target *starget = scsi_target(sdev);
339 scsi_dec_host_busy(shost);
341 if (starget->can_queue > 0)
342 atomic_dec(&starget->target_busy);
344 atomic_dec(&sdev->device_busy);
347 static void scsi_kick_queue(struct request_queue *q)
349 blk_mq_run_hw_queues(q, false);
353 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
354 * and call blk_run_queue for all the scsi_devices on the target -
355 * including current_sdev first.
357 * Called with *no* scsi locks held.
359 static void scsi_single_lun_run(struct scsi_device *current_sdev)
361 struct Scsi_Host *shost = current_sdev->host;
362 struct scsi_device *sdev, *tmp;
363 struct scsi_target *starget = scsi_target(current_sdev);
366 spin_lock_irqsave(shost->host_lock, flags);
367 starget->starget_sdev_user = NULL;
368 spin_unlock_irqrestore(shost->host_lock, flags);
371 * Call blk_run_queue for all LUNs on the target, starting with
372 * current_sdev. We race with others (to set starget_sdev_user),
373 * but in most cases, we will be first. Ideally, each LU on the
374 * target would get some limited time or requests on the target.
376 scsi_kick_queue(current_sdev->request_queue);
378 spin_lock_irqsave(shost->host_lock, flags);
379 if (starget->starget_sdev_user)
381 list_for_each_entry_safe(sdev, tmp, &starget->devices,
382 same_target_siblings) {
383 if (sdev == current_sdev)
385 if (scsi_device_get(sdev))
388 spin_unlock_irqrestore(shost->host_lock, flags);
389 scsi_kick_queue(sdev->request_queue);
390 spin_lock_irqsave(shost->host_lock, flags);
392 scsi_device_put(sdev);
395 spin_unlock_irqrestore(shost->host_lock, flags);
398 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
400 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
402 if (atomic_read(&sdev->device_blocked) > 0)
407 static inline bool scsi_target_is_busy(struct scsi_target *starget)
409 if (starget->can_queue > 0) {
410 if (atomic_read(&starget->target_busy) >= starget->can_queue)
412 if (atomic_read(&starget->target_blocked) > 0)
418 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
420 if (shost->can_queue > 0 &&
421 atomic_read(&shost->host_busy) >= shost->can_queue)
423 if (atomic_read(&shost->host_blocked) > 0)
425 if (shost->host_self_blocked)
430 static void scsi_starved_list_run(struct Scsi_Host *shost)
432 LIST_HEAD(starved_list);
433 struct scsi_device *sdev;
436 spin_lock_irqsave(shost->host_lock, flags);
437 list_splice_init(&shost->starved_list, &starved_list);
439 while (!list_empty(&starved_list)) {
440 struct request_queue *slq;
443 * As long as shost is accepting commands and we have
444 * starved queues, call blk_run_queue. scsi_request_fn
445 * drops the queue_lock and can add us back to the
448 * host_lock protects the starved_list and starved_entry.
449 * scsi_request_fn must get the host_lock before checking
450 * or modifying starved_list or starved_entry.
452 if (scsi_host_is_busy(shost))
455 sdev = list_entry(starved_list.next,
456 struct scsi_device, starved_entry);
457 list_del_init(&sdev->starved_entry);
458 if (scsi_target_is_busy(scsi_target(sdev))) {
459 list_move_tail(&sdev->starved_entry,
460 &shost->starved_list);
465 * Once we drop the host lock, a racing scsi_remove_device()
466 * call may remove the sdev from the starved list and destroy
467 * it and the queue. Mitigate by taking a reference to the
468 * queue and never touching the sdev again after we drop the
469 * host lock. Note: if __scsi_remove_device() invokes
470 * blk_cleanup_queue() before the queue is run from this
471 * function then blk_run_queue() will return immediately since
472 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
474 slq = sdev->request_queue;
475 if (!blk_get_queue(slq))
477 spin_unlock_irqrestore(shost->host_lock, flags);
479 scsi_kick_queue(slq);
482 spin_lock_irqsave(shost->host_lock, flags);
484 /* put any unprocessed entries back */
485 list_splice(&starved_list, &shost->starved_list);
486 spin_unlock_irqrestore(shost->host_lock, flags);
490 * Function: scsi_run_queue()
492 * Purpose: Select a proper request queue to serve next
494 * Arguments: q - last request's queue
498 * Notes: The previous command was completely finished, start
499 * a new one if possible.
501 static void scsi_run_queue(struct request_queue *q)
503 struct scsi_device *sdev = q->queuedata;
505 if (scsi_target(sdev)->single_lun)
506 scsi_single_lun_run(sdev);
507 if (!list_empty(&sdev->host->starved_list))
508 scsi_starved_list_run(sdev->host);
510 blk_mq_run_hw_queues(q, false);
513 void scsi_requeue_run_queue(struct work_struct *work)
515 struct scsi_device *sdev;
516 struct request_queue *q;
518 sdev = container_of(work, struct scsi_device, requeue_work);
519 q = sdev->request_queue;
523 void scsi_run_host_queues(struct Scsi_Host *shost)
525 struct scsi_device *sdev;
527 shost_for_each_device(sdev, shost)
528 scsi_run_queue(sdev->request_queue);
531 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
533 if (!blk_rq_is_passthrough(cmd->request)) {
534 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
536 if (drv->uninit_command)
537 drv->uninit_command(cmd);
541 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
543 if (cmd->sdb.table.nents)
544 sg_free_table_chained(&cmd->sdb.table, true);
545 if (scsi_prot_sg_count(cmd))
546 sg_free_table_chained(&cmd->prot_sdb->table, true);
549 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
551 scsi_mq_free_sgtables(cmd);
552 scsi_uninit_cmd(cmd);
553 scsi_del_cmd_from_list(cmd);
556 /* Returns false when no more bytes to process, true if there are more */
557 static bool scsi_end_request(struct request *req, blk_status_t error,
560 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
561 struct scsi_device *sdev = cmd->device;
562 struct request_queue *q = sdev->request_queue;
564 if (blk_update_request(req, error, bytes))
567 if (blk_queue_add_random(q))
568 add_disk_randomness(req->rq_disk);
570 if (!blk_rq_is_scsi(req)) {
571 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
572 cmd->flags &= ~SCMD_INITIALIZED;
576 * Calling rcu_barrier() is not necessary here because the
577 * SCSI error handler guarantees that the function called by
578 * call_rcu() has been called before scsi_end_request() is
581 destroy_rcu_head(&cmd->rcu);
584 * In the MQ case the command gets freed by __blk_mq_end_request,
585 * so we have to do all cleanup that depends on it earlier.
587 * We also can't kick the queues from irq context, so we
588 * will have to defer it to a workqueue.
590 scsi_mq_uninit_cmd(cmd);
593 * queue is still alive, so grab the ref for preventing it
594 * from being cleaned up during running queue.
596 percpu_ref_get(&q->q_usage_counter);
598 __blk_mq_end_request(req, error);
600 if (scsi_target(sdev)->single_lun ||
601 !list_empty(&sdev->host->starved_list))
602 kblockd_schedule_work(&sdev->requeue_work);
604 blk_mq_run_hw_queues(q, true);
606 percpu_ref_put(&q->q_usage_counter);
611 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
613 * @result: scsi error code
615 * Translate a SCSI result code into a blk_status_t value. May reset the host
616 * byte of @cmd->result.
618 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
620 switch (host_byte(result)) {
623 * Also check the other bytes than the status byte in result
624 * to handle the case when a SCSI LLD sets result to
625 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
627 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
629 return BLK_STS_IOERR;
630 case DID_TRANSPORT_FAILFAST:
631 return BLK_STS_TRANSPORT;
632 case DID_TARGET_FAILURE:
633 set_host_byte(cmd, DID_OK);
634 return BLK_STS_TARGET;
635 case DID_NEXUS_FAILURE:
636 set_host_byte(cmd, DID_OK);
637 return BLK_STS_NEXUS;
638 case DID_ALLOC_FAILURE:
639 set_host_byte(cmd, DID_OK);
640 return BLK_STS_NOSPC;
641 case DID_MEDIUM_ERROR:
642 set_host_byte(cmd, DID_OK);
643 return BLK_STS_MEDIUM;
645 return BLK_STS_IOERR;
649 /* Helper for scsi_io_completion() when "reprep" action required. */
650 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
651 struct request_queue *q)
653 /* A new command will be prepared and issued. */
654 scsi_mq_requeue_cmd(cmd);
657 /* Helper for scsi_io_completion() when special action required. */
658 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
660 struct request_queue *q = cmd->device->request_queue;
661 struct request *req = cmd->request;
663 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
664 ACTION_DELAYED_RETRY} action;
665 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
666 struct scsi_sense_hdr sshdr;
668 bool sense_current = true; /* false implies "deferred sense" */
669 blk_status_t blk_stat;
671 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
673 sense_current = !scsi_sense_is_deferred(&sshdr);
675 blk_stat = scsi_result_to_blk_status(cmd, result);
677 if (host_byte(result) == DID_RESET) {
678 /* Third party bus reset or reset for error recovery
679 * reasons. Just retry the command and see what
682 action = ACTION_RETRY;
683 } else if (sense_valid && sense_current) {
684 switch (sshdr.sense_key) {
686 if (cmd->device->removable) {
687 /* Detected disc change. Set a bit
688 * and quietly refuse further access.
690 cmd->device->changed = 1;
691 action = ACTION_FAIL;
693 /* Must have been a power glitch, or a
694 * bus reset. Could not have been a
695 * media change, so we just retry the
696 * command and see what happens.
698 action = ACTION_RETRY;
701 case ILLEGAL_REQUEST:
702 /* If we had an ILLEGAL REQUEST returned, then
703 * we may have performed an unsupported
704 * command. The only thing this should be
705 * would be a ten byte read where only a six
706 * byte read was supported. Also, on a system
707 * where READ CAPACITY failed, we may have
708 * read past the end of the disk.
710 if ((cmd->device->use_10_for_rw &&
711 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
712 (cmd->cmnd[0] == READ_10 ||
713 cmd->cmnd[0] == WRITE_10)) {
714 /* This will issue a new 6-byte command. */
715 cmd->device->use_10_for_rw = 0;
716 action = ACTION_REPREP;
717 } else if (sshdr.asc == 0x10) /* DIX */ {
718 action = ACTION_FAIL;
719 blk_stat = BLK_STS_PROTECTION;
720 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
721 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
722 action = ACTION_FAIL;
723 blk_stat = BLK_STS_TARGET;
725 action = ACTION_FAIL;
727 case ABORTED_COMMAND:
728 action = ACTION_FAIL;
729 if (sshdr.asc == 0x10) /* DIF */
730 blk_stat = BLK_STS_PROTECTION;
733 /* If the device is in the process of becoming
734 * ready, or has a temporary blockage, retry.
736 if (sshdr.asc == 0x04) {
737 switch (sshdr.ascq) {
738 case 0x01: /* becoming ready */
739 case 0x04: /* format in progress */
740 case 0x05: /* rebuild in progress */
741 case 0x06: /* recalculation in progress */
742 case 0x07: /* operation in progress */
743 case 0x08: /* Long write in progress */
744 case 0x09: /* self test in progress */
745 case 0x14: /* space allocation in progress */
746 case 0x1a: /* start stop unit in progress */
747 case 0x1b: /* sanitize in progress */
748 case 0x1d: /* configuration in progress */
749 case 0x24: /* depopulation in progress */
750 action = ACTION_DELAYED_RETRY;
753 action = ACTION_FAIL;
757 action = ACTION_FAIL;
759 case VOLUME_OVERFLOW:
760 /* See SSC3rXX or current. */
761 action = ACTION_FAIL;
764 action = ACTION_FAIL;
768 action = ACTION_FAIL;
770 if (action != ACTION_FAIL &&
771 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
772 action = ACTION_FAIL;
776 /* Give up and fail the remainder of the request */
777 if (!(req->rq_flags & RQF_QUIET)) {
778 static DEFINE_RATELIMIT_STATE(_rs,
779 DEFAULT_RATELIMIT_INTERVAL,
780 DEFAULT_RATELIMIT_BURST);
782 if (unlikely(scsi_logging_level))
784 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
785 SCSI_LOG_MLCOMPLETE_BITS);
788 * if logging is enabled the failure will be printed
789 * in scsi_log_completion(), so avoid duplicate messages
791 if (!level && __ratelimit(&_rs)) {
792 scsi_print_result(cmd, NULL, FAILED);
793 if (driver_byte(result) == DRIVER_SENSE)
794 scsi_print_sense(cmd);
795 scsi_print_command(cmd);
798 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
802 scsi_io_completion_reprep(cmd, q);
805 /* Retry the same command immediately */
806 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
808 case ACTION_DELAYED_RETRY:
809 /* Retry the same command after a delay */
810 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
816 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
817 * new result that may suppress further error checking. Also modifies
818 * *blk_statp in some cases.
820 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
821 blk_status_t *blk_statp)
824 bool sense_current = true; /* false implies "deferred sense" */
825 struct request *req = cmd->request;
826 struct scsi_sense_hdr sshdr;
828 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
830 sense_current = !scsi_sense_is_deferred(&sshdr);
832 if (blk_rq_is_passthrough(req)) {
835 * SG_IO wants current and deferred errors
837 scsi_req(req)->sense_len =
838 min(8 + cmd->sense_buffer[7],
839 SCSI_SENSE_BUFFERSIZE);
842 *blk_statp = scsi_result_to_blk_status(cmd, result);
843 } else if (blk_rq_bytes(req) == 0 && sense_current) {
845 * Flush commands do not transfers any data, and thus cannot use
846 * good_bytes != blk_rq_bytes(req) as the signal for an error.
847 * This sets *blk_statp explicitly for the problem case.
849 *blk_statp = scsi_result_to_blk_status(cmd, result);
852 * Recovered errors need reporting, but they're always treated as
853 * success, so fiddle the result code here. For passthrough requests
854 * we already took a copy of the original into sreq->result which
855 * is what gets returned to the user
857 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
858 bool do_print = true;
860 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
861 * skip print since caller wants ATA registers. Only occurs
862 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
864 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
866 else if (req->rq_flags & RQF_QUIET)
869 scsi_print_sense(cmd);
871 /* for passthrough, *blk_statp may be set */
872 *blk_statp = BLK_STS_OK;
875 * Another corner case: the SCSI status byte is non-zero but 'good'.
876 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
877 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
878 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
879 * intermediate statuses (both obsolete in SAM-4) as good.
881 if (status_byte(result) && scsi_status_is_good(result)) {
883 *blk_statp = BLK_STS_OK;
889 * Function: scsi_io_completion()
891 * Purpose: Completion processing for block device I/O requests.
893 * Arguments: cmd - command that is finished.
895 * Lock status: Assumed that no lock is held upon entry.
899 * Notes: We will finish off the specified number of sectors. If we
900 * are done, the command block will be released and the queue
901 * function will be goosed. If we are not done then we have to
902 * figure out what to do next:
904 * a) We can call scsi_requeue_command(). The request
905 * will be unprepared and put back on the queue. Then
906 * a new command will be created for it. This should
907 * be used if we made forward progress, or if we want
908 * to switch from READ(10) to READ(6) for example.
910 * b) We can call __scsi_queue_insert(). The request will
911 * be put back on the queue and retried using the same
912 * command as before, possibly after a delay.
914 * c) We can call scsi_end_request() with blk_stat other than
915 * BLK_STS_OK, to fail the remainder of the request.
917 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
919 int result = cmd->result;
920 struct request_queue *q = cmd->device->request_queue;
921 struct request *req = cmd->request;
922 blk_status_t blk_stat = BLK_STS_OK;
924 if (unlikely(result)) /* a nz result may or may not be an error */
925 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
927 if (unlikely(blk_rq_is_passthrough(req))) {
929 * scsi_result_to_blk_status may have reset the host_byte
931 scsi_req(req)->result = cmd->result;
935 * Next deal with any sectors which we were able to correctly
938 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
939 "%u sectors total, %d bytes done.\n",
940 blk_rq_sectors(req), good_bytes));
943 * Next deal with any sectors which we were able to correctly
944 * handle. Failed, zero length commands always need to drop down
945 * to retry code. Fast path should return in this block.
947 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
948 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
949 return; /* no bytes remaining */
952 /* Kill remainder if no retries. */
953 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
954 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
956 "Bytes remaining after failed, no-retry command");
961 * If there had been no error, but we have leftover bytes in the
962 * requeues just queue the command up again.
964 if (likely(result == 0))
965 scsi_io_completion_reprep(cmd, q);
967 scsi_io_completion_action(cmd, result);
970 static blk_status_t scsi_init_sgtable(struct request *req,
971 struct scsi_data_buffer *sdb)
976 * If sg table allocation fails, requeue request later.
978 if (unlikely(sg_alloc_table_chained(&sdb->table,
979 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
980 return BLK_STS_RESOURCE;
983 * Next, walk the list, and fill in the addresses and sizes of
986 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
987 BUG_ON(count > sdb->table.nents);
988 sdb->table.nents = count;
989 sdb->length = blk_rq_payload_bytes(req);
994 * Function: scsi_init_io()
996 * Purpose: SCSI I/O initialize function.
998 * Arguments: cmd - Command descriptor we wish to initialize
1000 * Returns: BLK_STS_OK on success
1001 * BLK_STS_RESOURCE if the failure is retryable
1002 * BLK_STS_IOERR if the failure is fatal
1004 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1006 struct request *rq = cmd->request;
1009 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1010 return BLK_STS_IOERR;
1012 ret = scsi_init_sgtable(rq, &cmd->sdb);
1016 if (blk_integrity_rq(rq)) {
1017 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1020 if (WARN_ON_ONCE(!prot_sdb)) {
1022 * This can happen if someone (e.g. multipath)
1023 * queues a command to a device on an adapter
1024 * that does not support DIX.
1026 ret = BLK_STS_IOERR;
1027 goto out_free_sgtables;
1030 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1032 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1033 prot_sdb->table.sgl)) {
1034 ret = BLK_STS_RESOURCE;
1035 goto out_free_sgtables;
1038 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1039 prot_sdb->table.sgl);
1040 BUG_ON(count > ivecs);
1041 BUG_ON(count > queue_max_integrity_segments(rq->q));
1043 cmd->prot_sdb = prot_sdb;
1044 cmd->prot_sdb->table.nents = count;
1049 scsi_mq_free_sgtables(cmd);
1052 EXPORT_SYMBOL(scsi_init_io);
1055 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1056 * @rq: Request associated with the SCSI command to be initialized.
1058 * This function initializes the members of struct scsi_cmnd that must be
1059 * initialized before request processing starts and that won't be
1060 * reinitialized if a SCSI command is requeued.
1062 * Called from inside blk_get_request() for pass-through requests and from
1063 * inside scsi_init_command() for filesystem requests.
1065 static void scsi_initialize_rq(struct request *rq)
1067 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1069 scsi_req_init(&cmd->req);
1070 init_rcu_head(&cmd->rcu);
1071 cmd->jiffies_at_alloc = jiffies;
1075 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1076 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1078 struct scsi_device *sdev = cmd->device;
1079 struct Scsi_Host *shost = sdev->host;
1080 unsigned long flags;
1082 if (shost->use_cmd_list) {
1083 spin_lock_irqsave(&sdev->list_lock, flags);
1084 list_add_tail(&cmd->list, &sdev->cmd_list);
1085 spin_unlock_irqrestore(&sdev->list_lock, flags);
1089 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1090 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1092 struct scsi_device *sdev = cmd->device;
1093 struct Scsi_Host *shost = sdev->host;
1094 unsigned long flags;
1096 if (shost->use_cmd_list) {
1097 spin_lock_irqsave(&sdev->list_lock, flags);
1098 BUG_ON(list_empty(&cmd->list));
1099 list_del_init(&cmd->list);
1100 spin_unlock_irqrestore(&sdev->list_lock, flags);
1104 /* Called after a request has been started. */
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;
1114 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1115 flags |= SCMD_INITIALIZED;
1116 scsi_initialize_rq(rq);
1119 jiffies_at_alloc = cmd->jiffies_at_alloc;
1120 retries = cmd->retries;
1121 /* zero out the cmd, except for the embedded scsi_request */
1122 memset((char *)cmd + sizeof(cmd->req), 0,
1123 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1126 cmd->sense_buffer = buf;
1127 cmd->prot_sdb = prot;
1129 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1130 cmd->jiffies_at_alloc = jiffies_at_alloc;
1131 cmd->retries = retries;
1133 scsi_add_cmd_to_list(cmd);
1136 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1137 struct request *req)
1139 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1142 * Passthrough requests may transfer data, in which case they must
1143 * a bio attached to them. Or they might contain a SCSI command
1144 * that does not transfer data, in which case they may optionally
1145 * submit a request without an attached bio.
1148 blk_status_t ret = scsi_init_io(cmd);
1149 if (unlikely(ret != BLK_STS_OK))
1152 BUG_ON(blk_rq_bytes(req));
1154 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1157 cmd->cmd_len = scsi_req(req)->cmd_len;
1158 cmd->cmnd = scsi_req(req)->cmd;
1159 cmd->transfersize = blk_rq_bytes(req);
1160 cmd->allowed = scsi_req(req)->retries;
1165 * Setup a normal block command. These are simple request from filesystems
1166 * that still need to be translated to SCSI CDBs from the ULD.
1168 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1169 struct request *req)
1171 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1173 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1174 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1175 if (ret != BLK_STS_OK)
1179 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1180 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1181 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1184 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1185 struct request *req)
1187 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1189 if (!blk_rq_bytes(req))
1190 cmd->sc_data_direction = DMA_NONE;
1191 else if (rq_data_dir(req) == WRITE)
1192 cmd->sc_data_direction = DMA_TO_DEVICE;
1194 cmd->sc_data_direction = DMA_FROM_DEVICE;
1196 if (blk_rq_is_scsi(req))
1197 return scsi_setup_scsi_cmnd(sdev, req);
1199 return scsi_setup_fs_cmnd(sdev, req);
1203 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1205 switch (sdev->sdev_state) {
1207 case SDEV_TRANSPORT_OFFLINE:
1209 * If the device is offline we refuse to process any
1210 * commands. The device must be brought online
1211 * before trying any recovery commands.
1213 sdev_printk(KERN_ERR, sdev,
1214 "rejecting I/O to offline device\n");
1215 return BLK_STS_IOERR;
1218 * If the device is fully deleted, we refuse to
1219 * process any commands as well.
1221 sdev_printk(KERN_ERR, sdev,
1222 "rejecting I/O to dead device\n");
1223 return BLK_STS_IOERR;
1225 case SDEV_CREATED_BLOCK:
1226 return BLK_STS_RESOURCE;
1229 * If the devices is blocked we defer normal commands.
1231 if (req && !(req->rq_flags & RQF_PREEMPT))
1232 return BLK_STS_RESOURCE;
1236 * For any other not fully online state we only allow
1237 * special commands. In particular any user initiated
1238 * command is not allowed.
1240 if (req && !(req->rq_flags & RQF_PREEMPT))
1241 return BLK_STS_IOERR;
1247 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1250 * Called with the queue_lock held.
1252 static inline int scsi_dev_queue_ready(struct request_queue *q,
1253 struct scsi_device *sdev)
1257 busy = atomic_inc_return(&sdev->device_busy) - 1;
1258 if (atomic_read(&sdev->device_blocked)) {
1263 * unblock after device_blocked iterates to zero
1265 if (atomic_dec_return(&sdev->device_blocked) > 0)
1267 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1268 "unblocking device at zero depth\n"));
1271 if (busy >= sdev->queue_depth)
1276 atomic_dec(&sdev->device_busy);
1281 * scsi_target_queue_ready: checks if there we can send commands to target
1282 * @sdev: scsi device on starget to check.
1284 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1285 struct scsi_device *sdev)
1287 struct scsi_target *starget = scsi_target(sdev);
1290 if (starget->single_lun) {
1291 spin_lock_irq(shost->host_lock);
1292 if (starget->starget_sdev_user &&
1293 starget->starget_sdev_user != sdev) {
1294 spin_unlock_irq(shost->host_lock);
1297 starget->starget_sdev_user = sdev;
1298 spin_unlock_irq(shost->host_lock);
1301 if (starget->can_queue <= 0)
1304 busy = atomic_inc_return(&starget->target_busy) - 1;
1305 if (atomic_read(&starget->target_blocked) > 0) {
1310 * unblock after target_blocked iterates to zero
1312 if (atomic_dec_return(&starget->target_blocked) > 0)
1315 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1316 "unblocking target at zero depth\n"));
1319 if (busy >= starget->can_queue)
1325 spin_lock_irq(shost->host_lock);
1326 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1327 spin_unlock_irq(shost->host_lock);
1329 if (starget->can_queue > 0)
1330 atomic_dec(&starget->target_busy);
1335 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1336 * return 0. We must end up running the queue again whenever 0 is
1337 * returned, else IO can hang.
1339 static inline int scsi_host_queue_ready(struct request_queue *q,
1340 struct Scsi_Host *shost,
1341 struct scsi_device *sdev)
1345 if (scsi_host_in_recovery(shost))
1348 busy = atomic_inc_return(&shost->host_busy) - 1;
1349 if (atomic_read(&shost->host_blocked) > 0) {
1354 * unblock after host_blocked iterates to zero
1356 if (atomic_dec_return(&shost->host_blocked) > 0)
1360 shost_printk(KERN_INFO, shost,
1361 "unblocking host at zero depth\n"));
1364 if (shost->can_queue > 0 && busy >= shost->can_queue)
1366 if (shost->host_self_blocked)
1369 /* We're OK to process the command, so we can't be starved */
1370 if (!list_empty(&sdev->starved_entry)) {
1371 spin_lock_irq(shost->host_lock);
1372 if (!list_empty(&sdev->starved_entry))
1373 list_del_init(&sdev->starved_entry);
1374 spin_unlock_irq(shost->host_lock);
1380 spin_lock_irq(shost->host_lock);
1381 if (list_empty(&sdev->starved_entry))
1382 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1383 spin_unlock_irq(shost->host_lock);
1385 scsi_dec_host_busy(shost);
1390 * Busy state exporting function for request stacking drivers.
1392 * For efficiency, no lock is taken to check the busy state of
1393 * shost/starget/sdev, since the returned value is not guaranteed and
1394 * may be changed after request stacking drivers call the function,
1395 * regardless of taking lock or not.
1397 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1398 * needs to return 'not busy'. Otherwise, request stacking drivers
1399 * may hold requests forever.
1401 static bool scsi_mq_lld_busy(struct request_queue *q)
1403 struct scsi_device *sdev = q->queuedata;
1404 struct Scsi_Host *shost;
1406 if (blk_queue_dying(q))
1412 * Ignore host/starget busy state.
1413 * Since block layer does not have a concept of fairness across
1414 * multiple queues, congestion of host/starget needs to be handled
1417 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1423 static void scsi_softirq_done(struct request *rq)
1425 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1426 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1429 INIT_LIST_HEAD(&cmd->eh_entry);
1431 atomic_inc(&cmd->device->iodone_cnt);
1433 atomic_inc(&cmd->device->ioerr_cnt);
1435 disposition = scsi_decide_disposition(cmd);
1436 if (disposition != SUCCESS &&
1437 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1438 sdev_printk(KERN_ERR, cmd->device,
1439 "timing out command, waited %lus\n",
1441 disposition = SUCCESS;
1444 scsi_log_completion(cmd, disposition);
1446 switch (disposition) {
1448 scsi_finish_command(cmd);
1451 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1453 case ADD_TO_MLQUEUE:
1454 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1457 scsi_eh_scmd_add(cmd);
1463 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1464 * @cmd: command block we are dispatching.
1466 * Return: nonzero return request was rejected and device's queue needs to be
1469 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1471 struct Scsi_Host *host = cmd->device->host;
1474 atomic_inc(&cmd->device->iorequest_cnt);
1476 /* check if the device is still usable */
1477 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1478 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1479 * returns an immediate error upwards, and signals
1480 * that the device is no longer present */
1481 cmd->result = DID_NO_CONNECT << 16;
1485 /* Check to see if the scsi lld made this device blocked. */
1486 if (unlikely(scsi_device_blocked(cmd->device))) {
1488 * in blocked state, the command is just put back on
1489 * the device queue. The suspend state has already
1490 * blocked the queue so future requests should not
1491 * occur until the device transitions out of the
1494 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1495 "queuecommand : device blocked\n"));
1496 return SCSI_MLQUEUE_DEVICE_BUSY;
1499 /* Store the LUN value in cmnd, if needed. */
1500 if (cmd->device->lun_in_cdb)
1501 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1502 (cmd->device->lun << 5 & 0xe0);
1507 * Before we queue this command, check if the command
1508 * length exceeds what the host adapter can handle.
1510 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1511 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1512 "queuecommand : command too long. "
1513 "cdb_size=%d host->max_cmd_len=%d\n",
1514 cmd->cmd_len, cmd->device->host->max_cmd_len));
1515 cmd->result = (DID_ABORT << 16);
1519 if (unlikely(host->shost_state == SHOST_DEL)) {
1520 cmd->result = (DID_NO_CONNECT << 16);
1525 trace_scsi_dispatch_cmd_start(cmd);
1526 rtn = host->hostt->queuecommand(host, cmd);
1528 trace_scsi_dispatch_cmd_error(cmd, rtn);
1529 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1530 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1531 rtn = SCSI_MLQUEUE_HOST_BUSY;
1533 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1534 "queuecommand : request rejected\n"));
1539 cmd->scsi_done(cmd);
1543 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1544 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1546 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1547 sizeof(struct scatterlist);
1550 static blk_status_t scsi_mq_prep_fn(struct request *req)
1552 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1553 struct scsi_device *sdev = req->q->queuedata;
1554 struct Scsi_Host *shost = sdev->host;
1555 struct scatterlist *sg;
1557 scsi_init_command(sdev, cmd);
1560 cmd->tag = req->tag;
1561 cmd->prot_op = SCSI_PROT_NORMAL;
1563 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1564 cmd->sdb.table.sgl = sg;
1566 if (scsi_host_get_prot(shost)) {
1567 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1569 cmd->prot_sdb->table.sgl =
1570 (struct scatterlist *)(cmd->prot_sdb + 1);
1573 blk_mq_start_request(req);
1575 return scsi_setup_cmnd(sdev, req);
1578 static void scsi_mq_done(struct scsi_cmnd *cmd)
1580 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1582 trace_scsi_dispatch_cmd_done(cmd);
1585 * If the block layer didn't complete the request due to a timeout
1586 * injection, scsi must clear its internal completed state so that the
1587 * timeout handler will see it needs to escalate its own error
1590 if (unlikely(!blk_mq_complete_request(cmd->request)))
1591 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1594 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1596 struct request_queue *q = hctx->queue;
1597 struct scsi_device *sdev = q->queuedata;
1599 atomic_dec(&sdev->device_busy);
1602 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1604 struct request_queue *q = hctx->queue;
1605 struct scsi_device *sdev = q->queuedata;
1607 if (scsi_dev_queue_ready(q, sdev))
1610 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1611 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1615 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1616 const struct blk_mq_queue_data *bd)
1618 struct request *req = bd->rq;
1619 struct request_queue *q = req->q;
1620 struct scsi_device *sdev = q->queuedata;
1621 struct Scsi_Host *shost = sdev->host;
1622 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1627 * If the device is not in running state we will reject some or all
1630 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1631 ret = scsi_prep_state_check(sdev, req);
1632 if (ret != BLK_STS_OK)
1633 goto out_put_budget;
1636 ret = BLK_STS_RESOURCE;
1637 if (!scsi_target_queue_ready(shost, sdev))
1638 goto out_put_budget;
1639 if (!scsi_host_queue_ready(q, shost, sdev))
1640 goto out_dec_target_busy;
1642 if (!(req->rq_flags & RQF_DONTPREP)) {
1643 ret = scsi_mq_prep_fn(req);
1644 if (ret != BLK_STS_OK)
1645 goto out_dec_host_busy;
1646 req->rq_flags |= RQF_DONTPREP;
1648 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1649 blk_mq_start_request(req);
1652 if (sdev->simple_tags)
1653 cmd->flags |= SCMD_TAGGED;
1655 cmd->flags &= ~SCMD_TAGGED;
1657 scsi_init_cmd_errh(cmd);
1658 cmd->scsi_done = scsi_mq_done;
1660 reason = scsi_dispatch_cmd(cmd);
1662 scsi_set_blocked(cmd, reason);
1663 ret = BLK_STS_RESOURCE;
1664 goto out_dec_host_busy;
1670 scsi_dec_host_busy(shost);
1671 out_dec_target_busy:
1672 if (scsi_target(sdev)->can_queue > 0)
1673 atomic_dec(&scsi_target(sdev)->target_busy);
1675 scsi_mq_put_budget(hctx);
1679 case BLK_STS_RESOURCE:
1680 if (atomic_read(&sdev->device_busy) ||
1681 scsi_device_blocked(sdev))
1682 ret = BLK_STS_DEV_RESOURCE;
1685 if (unlikely(!scsi_device_online(sdev)))
1686 scsi_req(req)->result = DID_NO_CONNECT << 16;
1688 scsi_req(req)->result = DID_ERROR << 16;
1690 * Make sure to release all allocated resources when
1691 * we hit an error, as we will never see this command
1694 if (req->rq_flags & RQF_DONTPREP)
1695 scsi_mq_uninit_cmd(cmd);
1701 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1705 return BLK_EH_RESET_TIMER;
1706 return scsi_times_out(req);
1709 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1710 unsigned int hctx_idx, unsigned int numa_node)
1712 struct Scsi_Host *shost = set->driver_data;
1713 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1714 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1715 struct scatterlist *sg;
1717 if (unchecked_isa_dma)
1718 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1719 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1720 GFP_KERNEL, numa_node);
1721 if (!cmd->sense_buffer)
1723 cmd->req.sense = cmd->sense_buffer;
1725 if (scsi_host_get_prot(shost)) {
1726 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1727 shost->hostt->cmd_size;
1728 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
1734 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1735 unsigned int hctx_idx)
1737 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1739 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1743 static int scsi_map_queues(struct blk_mq_tag_set *set)
1745 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1747 if (shost->hostt->map_queues)
1748 return shost->hostt->map_queues(shost);
1749 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1752 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1754 struct device *dev = shost->dma_dev;
1757 * this limit is imposed by hardware restrictions
1759 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1762 if (scsi_host_prot_dma(shost)) {
1763 shost->sg_prot_tablesize =
1764 min_not_zero(shost->sg_prot_tablesize,
1765 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1766 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1767 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1770 blk_queue_max_hw_sectors(q, shost->max_sectors);
1771 if (shost->unchecked_isa_dma)
1772 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1773 blk_queue_segment_boundary(q, shost->dma_boundary);
1774 dma_set_seg_boundary(dev, shost->dma_boundary);
1776 blk_queue_max_segment_size(q, shost->max_segment_size);
1777 dma_set_max_seg_size(dev, shost->max_segment_size);
1780 * Set a reasonable default alignment: The larger of 32-byte (dword),
1781 * which is a common minimum for HBAs, and the minimum DMA alignment,
1782 * which is set by the platform.
1784 * Devices that require a bigger alignment can increase it later.
1786 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1788 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1790 static const struct blk_mq_ops scsi_mq_ops = {
1791 .get_budget = scsi_mq_get_budget,
1792 .put_budget = scsi_mq_put_budget,
1793 .queue_rq = scsi_queue_rq,
1794 .complete = scsi_softirq_done,
1795 .timeout = scsi_timeout,
1796 #ifdef CONFIG_BLK_DEBUG_FS
1797 .show_rq = scsi_show_rq,
1799 .init_request = scsi_mq_init_request,
1800 .exit_request = scsi_mq_exit_request,
1801 .initialize_rq_fn = scsi_initialize_rq,
1802 .busy = scsi_mq_lld_busy,
1803 .map_queues = scsi_map_queues,
1806 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1808 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1809 if (IS_ERR(sdev->request_queue))
1812 sdev->request_queue->queuedata = sdev;
1813 __scsi_init_queue(sdev->host, sdev->request_queue);
1814 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1815 return sdev->request_queue;
1818 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1820 unsigned int cmd_size, sgl_size;
1822 sgl_size = scsi_mq_sgl_size(shost);
1823 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1824 if (scsi_host_get_prot(shost))
1825 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
1827 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1828 shost->tag_set.ops = &scsi_mq_ops;
1829 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1830 shost->tag_set.queue_depth = shost->can_queue;
1831 shost->tag_set.cmd_size = cmd_size;
1832 shost->tag_set.numa_node = NUMA_NO_NODE;
1833 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1834 shost->tag_set.flags |=
1835 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1836 shost->tag_set.driver_data = shost;
1838 return blk_mq_alloc_tag_set(&shost->tag_set);
1841 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1843 blk_mq_free_tag_set(&shost->tag_set);
1847 * scsi_device_from_queue - return sdev associated with a request_queue
1848 * @q: The request queue to return the sdev from
1850 * Return the sdev associated with a request queue or NULL if the
1851 * request_queue does not reference a SCSI device.
1853 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1855 struct scsi_device *sdev = NULL;
1857 if (q->mq_ops == &scsi_mq_ops)
1858 sdev = q->queuedata;
1859 if (!sdev || !get_device(&sdev->sdev_gendev))
1864 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1867 * Function: scsi_block_requests()
1869 * Purpose: Utility function used by low-level drivers to prevent further
1870 * commands from being queued to the device.
1872 * Arguments: shost - Host in question
1876 * Lock status: No locks are assumed held.
1878 * Notes: There is no timer nor any other means by which the requests
1879 * get unblocked other than the low-level driver calling
1880 * scsi_unblock_requests().
1882 void scsi_block_requests(struct Scsi_Host *shost)
1884 shost->host_self_blocked = 1;
1886 EXPORT_SYMBOL(scsi_block_requests);
1889 * Function: scsi_unblock_requests()
1891 * Purpose: Utility function used by low-level drivers to allow further
1892 * commands from being queued to the device.
1894 * Arguments: shost - Host in question
1898 * Lock status: No locks are assumed held.
1900 * Notes: There is no timer nor any other means by which the requests
1901 * get unblocked other than the low-level driver calling
1902 * scsi_unblock_requests().
1904 * This is done as an API function so that changes to the
1905 * internals of the scsi mid-layer won't require wholesale
1906 * changes to drivers that use this feature.
1908 void scsi_unblock_requests(struct Scsi_Host *shost)
1910 shost->host_self_blocked = 0;
1911 scsi_run_host_queues(shost);
1913 EXPORT_SYMBOL(scsi_unblock_requests);
1915 int __init scsi_init_queue(void)
1917 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1918 sizeof(struct scsi_data_buffer),
1920 if (!scsi_sdb_cache) {
1921 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1928 void scsi_exit_queue(void)
1930 kmem_cache_destroy(scsi_sense_cache);
1931 kmem_cache_destroy(scsi_sense_isadma_cache);
1932 kmem_cache_destroy(scsi_sdb_cache);
1936 * scsi_mode_select - issue a mode select
1937 * @sdev: SCSI device to be queried
1938 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1939 * @sp: Save page bit (0 == don't save, 1 == save)
1940 * @modepage: mode page being requested
1941 * @buffer: request buffer (may not be smaller than eight bytes)
1942 * @len: length of request buffer.
1943 * @timeout: command timeout
1944 * @retries: number of retries before failing
1945 * @data: returns a structure abstracting the mode header data
1946 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1947 * must be SCSI_SENSE_BUFFERSIZE big.
1949 * Returns zero if successful; negative error number or scsi
1954 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1955 unsigned char *buffer, int len, int timeout, int retries,
1956 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1958 unsigned char cmd[10];
1959 unsigned char *real_buffer;
1962 memset(cmd, 0, sizeof(cmd));
1963 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1965 if (sdev->use_10_for_ms) {
1968 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1971 memcpy(real_buffer + 8, buffer, len);
1975 real_buffer[2] = data->medium_type;
1976 real_buffer[3] = data->device_specific;
1977 real_buffer[4] = data->longlba ? 0x01 : 0;
1979 real_buffer[6] = data->block_descriptor_length >> 8;
1980 real_buffer[7] = data->block_descriptor_length;
1982 cmd[0] = MODE_SELECT_10;
1986 if (len > 255 || data->block_descriptor_length > 255 ||
1990 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1993 memcpy(real_buffer + 4, buffer, len);
1996 real_buffer[1] = data->medium_type;
1997 real_buffer[2] = data->device_specific;
1998 real_buffer[3] = data->block_descriptor_length;
2001 cmd[0] = MODE_SELECT;
2005 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2006 sshdr, timeout, retries, NULL);
2010 EXPORT_SYMBOL_GPL(scsi_mode_select);
2013 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2014 * @sdev: SCSI device to be queried
2015 * @dbd: set if mode sense will allow block descriptors to be returned
2016 * @modepage: mode page being requested
2017 * @buffer: request buffer (may not be smaller than eight bytes)
2018 * @len: length of request buffer.
2019 * @timeout: command timeout
2020 * @retries: number of retries before failing
2021 * @data: returns a structure abstracting the mode header data
2022 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2023 * must be SCSI_SENSE_BUFFERSIZE big.
2025 * Returns zero if unsuccessful, or the header offset (either 4
2026 * or 8 depending on whether a six or ten byte command was
2027 * issued) if successful.
2030 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2031 unsigned char *buffer, int len, int timeout, int retries,
2032 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2034 unsigned char cmd[12];
2037 int result, retry_count = retries;
2038 struct scsi_sense_hdr my_sshdr;
2040 memset(data, 0, sizeof(*data));
2041 memset(&cmd[0], 0, 12);
2042 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2045 /* caller might not be interested in sense, but we need it */
2050 use_10_for_ms = sdev->use_10_for_ms;
2052 if (use_10_for_ms) {
2056 cmd[0] = MODE_SENSE_10;
2063 cmd[0] = MODE_SENSE;
2068 memset(buffer, 0, len);
2070 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2071 sshdr, timeout, retries, NULL);
2073 /* This code looks awful: what it's doing is making sure an
2074 * ILLEGAL REQUEST sense return identifies the actual command
2075 * byte as the problem. MODE_SENSE commands can return
2076 * ILLEGAL REQUEST if the code page isn't supported */
2078 if (use_10_for_ms && !scsi_status_is_good(result) &&
2079 driver_byte(result) == DRIVER_SENSE) {
2080 if (scsi_sense_valid(sshdr)) {
2081 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2082 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2084 * Invalid command operation code
2086 sdev->use_10_for_ms = 0;
2092 if(scsi_status_is_good(result)) {
2093 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2094 (modepage == 6 || modepage == 8))) {
2095 /* Initio breakage? */
2098 data->medium_type = 0;
2099 data->device_specific = 0;
2101 data->block_descriptor_length = 0;
2102 } else if(use_10_for_ms) {
2103 data->length = buffer[0]*256 + buffer[1] + 2;
2104 data->medium_type = buffer[2];
2105 data->device_specific = buffer[3];
2106 data->longlba = buffer[4] & 0x01;
2107 data->block_descriptor_length = buffer[6]*256
2110 data->length = buffer[0] + 1;
2111 data->medium_type = buffer[1];
2112 data->device_specific = buffer[2];
2113 data->block_descriptor_length = buffer[3];
2115 data->header_length = header_length;
2116 } else if ((status_byte(result) == CHECK_CONDITION) &&
2117 scsi_sense_valid(sshdr) &&
2118 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2125 EXPORT_SYMBOL(scsi_mode_sense);
2128 * scsi_test_unit_ready - test if unit is ready
2129 * @sdev: scsi device to change the state of.
2130 * @timeout: command timeout
2131 * @retries: number of retries before failing
2132 * @sshdr: outpout pointer for decoded sense information.
2134 * Returns zero if unsuccessful or an error if TUR failed. For
2135 * removable media, UNIT_ATTENTION sets ->changed flag.
2138 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2139 struct scsi_sense_hdr *sshdr)
2142 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2146 /* try to eat the UNIT_ATTENTION if there are enough retries */
2148 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2150 if (sdev->removable && scsi_sense_valid(sshdr) &&
2151 sshdr->sense_key == UNIT_ATTENTION)
2153 } while (scsi_sense_valid(sshdr) &&
2154 sshdr->sense_key == UNIT_ATTENTION && --retries);
2158 EXPORT_SYMBOL(scsi_test_unit_ready);
2161 * scsi_device_set_state - Take the given device through the device state model.
2162 * @sdev: scsi device to change the state of.
2163 * @state: state to change to.
2165 * Returns zero if successful or an error if the requested
2166 * transition is illegal.
2169 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2171 enum scsi_device_state oldstate = sdev->sdev_state;
2173 if (state == oldstate)
2179 case SDEV_CREATED_BLOCK:
2190 case SDEV_TRANSPORT_OFFLINE:
2203 case SDEV_TRANSPORT_OFFLINE:
2211 case SDEV_TRANSPORT_OFFLINE:
2226 case SDEV_CREATED_BLOCK:
2234 case SDEV_CREATED_BLOCK:
2249 case SDEV_TRANSPORT_OFFLINE:
2261 case SDEV_TRANSPORT_OFFLINE:
2264 case SDEV_CREATED_BLOCK:
2272 sdev->sdev_state = state;
2276 SCSI_LOG_ERROR_RECOVERY(1,
2277 sdev_printk(KERN_ERR, sdev,
2278 "Illegal state transition %s->%s",
2279 scsi_device_state_name(oldstate),
2280 scsi_device_state_name(state))
2284 EXPORT_SYMBOL(scsi_device_set_state);
2287 * sdev_evt_emit - emit a single SCSI device uevent
2288 * @sdev: associated SCSI device
2289 * @evt: event to emit
2291 * Send a single uevent (scsi_event) to the associated scsi_device.
2293 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2298 switch (evt->evt_type) {
2299 case SDEV_EVT_MEDIA_CHANGE:
2300 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2302 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2303 scsi_rescan_device(&sdev->sdev_gendev);
2304 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2306 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2307 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2309 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2310 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2312 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2313 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2315 case SDEV_EVT_LUN_CHANGE_REPORTED:
2316 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2318 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2319 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2321 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2322 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2331 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2335 * sdev_evt_thread - send a uevent for each scsi event
2336 * @work: work struct for scsi_device
2338 * Dispatch queued events to their associated scsi_device kobjects
2341 void scsi_evt_thread(struct work_struct *work)
2343 struct scsi_device *sdev;
2344 enum scsi_device_event evt_type;
2345 LIST_HEAD(event_list);
2347 sdev = container_of(work, struct scsi_device, event_work);
2349 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2350 if (test_and_clear_bit(evt_type, sdev->pending_events))
2351 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2354 struct scsi_event *evt;
2355 struct list_head *this, *tmp;
2356 unsigned long flags;
2358 spin_lock_irqsave(&sdev->list_lock, flags);
2359 list_splice_init(&sdev->event_list, &event_list);
2360 spin_unlock_irqrestore(&sdev->list_lock, flags);
2362 if (list_empty(&event_list))
2365 list_for_each_safe(this, tmp, &event_list) {
2366 evt = list_entry(this, struct scsi_event, node);
2367 list_del(&evt->node);
2368 scsi_evt_emit(sdev, evt);
2375 * sdev_evt_send - send asserted event to uevent thread
2376 * @sdev: scsi_device event occurred on
2377 * @evt: event to send
2379 * Assert scsi device event asynchronously.
2381 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2383 unsigned long flags;
2386 /* FIXME: currently this check eliminates all media change events
2387 * for polled devices. Need to update to discriminate between AN
2388 * and polled events */
2389 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2395 spin_lock_irqsave(&sdev->list_lock, flags);
2396 list_add_tail(&evt->node, &sdev->event_list);
2397 schedule_work(&sdev->event_work);
2398 spin_unlock_irqrestore(&sdev->list_lock, flags);
2400 EXPORT_SYMBOL_GPL(sdev_evt_send);
2403 * sdev_evt_alloc - allocate a new scsi event
2404 * @evt_type: type of event to allocate
2405 * @gfpflags: GFP flags for allocation
2407 * Allocates and returns a new scsi_event.
2409 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2412 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2416 evt->evt_type = evt_type;
2417 INIT_LIST_HEAD(&evt->node);
2419 /* evt_type-specific initialization, if any */
2421 case SDEV_EVT_MEDIA_CHANGE:
2422 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2423 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2424 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2425 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2426 case SDEV_EVT_LUN_CHANGE_REPORTED:
2427 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2428 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2436 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2439 * sdev_evt_send_simple - send asserted event to uevent thread
2440 * @sdev: scsi_device event occurred on
2441 * @evt_type: type of event to send
2442 * @gfpflags: GFP flags for allocation
2444 * Assert scsi device event asynchronously, given an event type.
2446 void sdev_evt_send_simple(struct scsi_device *sdev,
2447 enum scsi_device_event evt_type, gfp_t gfpflags)
2449 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2451 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2456 sdev_evt_send(sdev, evt);
2458 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2461 * scsi_device_quiesce - Block user issued commands.
2462 * @sdev: scsi device to quiesce.
2464 * This works by trying to transition to the SDEV_QUIESCE state
2465 * (which must be a legal transition). When the device is in this
2466 * state, only special requests will be accepted, all others will
2467 * be deferred. Since special requests may also be requeued requests,
2468 * a successful return doesn't guarantee the device will be
2469 * totally quiescent.
2471 * Must be called with user context, may sleep.
2473 * Returns zero if unsuccessful or an error if not.
2476 scsi_device_quiesce(struct scsi_device *sdev)
2478 struct request_queue *q = sdev->request_queue;
2482 * It is allowed to call scsi_device_quiesce() multiple times from
2483 * the same context but concurrent scsi_device_quiesce() calls are
2486 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2488 if (sdev->quiesced_by == current)
2493 blk_mq_freeze_queue(q);
2495 * Ensure that the effect of blk_set_pm_only() will be visible
2496 * for percpu_ref_tryget() callers that occur after the queue
2497 * unfreeze even if the queue was already frozen before this function
2498 * was called. See also https://lwn.net/Articles/573497/.
2501 blk_mq_unfreeze_queue(q);
2503 mutex_lock(&sdev->state_mutex);
2504 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2506 sdev->quiesced_by = current;
2508 blk_clear_pm_only(q);
2509 mutex_unlock(&sdev->state_mutex);
2513 EXPORT_SYMBOL(scsi_device_quiesce);
2516 * scsi_device_resume - Restart user issued commands to a quiesced device.
2517 * @sdev: scsi device to resume.
2519 * Moves the device from quiesced back to running and restarts the
2522 * Must be called with user context, may sleep.
2524 void scsi_device_resume(struct scsi_device *sdev)
2526 /* check if the device state was mutated prior to resume, and if
2527 * so assume the state is being managed elsewhere (for example
2528 * device deleted during suspend)
2530 mutex_lock(&sdev->state_mutex);
2531 if (sdev->quiesced_by) {
2532 sdev->quiesced_by = NULL;
2533 blk_clear_pm_only(sdev->request_queue);
2535 if (sdev->sdev_state == SDEV_QUIESCE)
2536 scsi_device_set_state(sdev, SDEV_RUNNING);
2537 mutex_unlock(&sdev->state_mutex);
2539 EXPORT_SYMBOL(scsi_device_resume);
2542 device_quiesce_fn(struct scsi_device *sdev, void *data)
2544 scsi_device_quiesce(sdev);
2548 scsi_target_quiesce(struct scsi_target *starget)
2550 starget_for_each_device(starget, NULL, device_quiesce_fn);
2552 EXPORT_SYMBOL(scsi_target_quiesce);
2555 device_resume_fn(struct scsi_device *sdev, void *data)
2557 scsi_device_resume(sdev);
2561 scsi_target_resume(struct scsi_target *starget)
2563 starget_for_each_device(starget, NULL, device_resume_fn);
2565 EXPORT_SYMBOL(scsi_target_resume);
2568 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2569 * @sdev: device to block
2571 * Pause SCSI command processing on the specified device. Does not sleep.
2573 * Returns zero if successful or a negative error code upon failure.
2576 * This routine transitions the device to the SDEV_BLOCK state (which must be
2577 * a legal transition). When the device is in this state, command processing
2578 * is paused until the device leaves the SDEV_BLOCK state. See also
2579 * scsi_internal_device_unblock_nowait().
2581 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2583 struct request_queue *q = sdev->request_queue;
2586 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2588 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2595 * The device has transitioned to SDEV_BLOCK. Stop the
2596 * block layer from calling the midlayer with this device's
2599 blk_mq_quiesce_queue_nowait(q);
2602 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2605 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2606 * @sdev: device to block
2608 * Pause SCSI command processing on the specified device and wait until all
2609 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2611 * Returns zero if successful or a negative error code upon failure.
2614 * This routine transitions the device to the SDEV_BLOCK state (which must be
2615 * a legal transition). When the device is in this state, command processing
2616 * is paused until the device leaves the SDEV_BLOCK state. See also
2617 * scsi_internal_device_unblock().
2619 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2620 * scsi_internal_device_block() has blocked a SCSI device and also
2621 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2623 static int scsi_internal_device_block(struct scsi_device *sdev)
2625 struct request_queue *q = sdev->request_queue;
2628 mutex_lock(&sdev->state_mutex);
2629 err = scsi_internal_device_block_nowait(sdev);
2631 blk_mq_quiesce_queue(q);
2632 mutex_unlock(&sdev->state_mutex);
2637 void scsi_start_queue(struct scsi_device *sdev)
2639 struct request_queue *q = sdev->request_queue;
2641 blk_mq_unquiesce_queue(q);
2645 * scsi_internal_device_unblock_nowait - resume a device after a block request
2646 * @sdev: device to resume
2647 * @new_state: state to set the device to after unblocking
2649 * Restart the device queue for a previously suspended SCSI device. Does not
2652 * Returns zero if successful or a negative error code upon failure.
2655 * This routine transitions the device to the SDEV_RUNNING state or to one of
2656 * the offline states (which must be a legal transition) allowing the midlayer
2657 * to goose the queue for this device.
2659 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2660 enum scsi_device_state new_state)
2663 * Try to transition the scsi device to SDEV_RUNNING or one of the
2664 * offlined states and goose the device queue if successful.
2666 switch (sdev->sdev_state) {
2668 case SDEV_TRANSPORT_OFFLINE:
2669 sdev->sdev_state = new_state;
2671 case SDEV_CREATED_BLOCK:
2672 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2673 new_state == SDEV_OFFLINE)
2674 sdev->sdev_state = new_state;
2676 sdev->sdev_state = SDEV_CREATED;
2684 scsi_start_queue(sdev);
2688 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2691 * scsi_internal_device_unblock - resume a device after a block request
2692 * @sdev: device to resume
2693 * @new_state: state to set the device to after unblocking
2695 * Restart the device queue for a previously suspended SCSI device. May sleep.
2697 * Returns zero if successful or a negative error code upon failure.
2700 * This routine transitions the device to the SDEV_RUNNING state or to one of
2701 * the offline states (which must be a legal transition) allowing the midlayer
2702 * to goose the queue for this device.
2704 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2705 enum scsi_device_state new_state)
2709 mutex_lock(&sdev->state_mutex);
2710 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2711 mutex_unlock(&sdev->state_mutex);
2717 device_block(struct scsi_device *sdev, void *data)
2719 scsi_internal_device_block(sdev);
2723 target_block(struct device *dev, void *data)
2725 if (scsi_is_target_device(dev))
2726 starget_for_each_device(to_scsi_target(dev), NULL,
2732 scsi_target_block(struct device *dev)
2734 if (scsi_is_target_device(dev))
2735 starget_for_each_device(to_scsi_target(dev), NULL,
2738 device_for_each_child(dev, NULL, target_block);
2740 EXPORT_SYMBOL_GPL(scsi_target_block);
2743 device_unblock(struct scsi_device *sdev, void *data)
2745 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2749 target_unblock(struct device *dev, void *data)
2751 if (scsi_is_target_device(dev))
2752 starget_for_each_device(to_scsi_target(dev), data,
2758 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2760 if (scsi_is_target_device(dev))
2761 starget_for_each_device(to_scsi_target(dev), &new_state,
2764 device_for_each_child(dev, &new_state, target_unblock);
2766 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2769 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2770 * @sgl: scatter-gather list
2771 * @sg_count: number of segments in sg
2772 * @offset: offset in bytes into sg, on return offset into the mapped area
2773 * @len: bytes to map, on return number of bytes mapped
2775 * Returns virtual address of the start of the mapped page
2777 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2778 size_t *offset, size_t *len)
2781 size_t sg_len = 0, len_complete = 0;
2782 struct scatterlist *sg;
2785 WARN_ON(!irqs_disabled());
2787 for_each_sg(sgl, sg, sg_count, i) {
2788 len_complete = sg_len; /* Complete sg-entries */
2789 sg_len += sg->length;
2790 if (sg_len > *offset)
2794 if (unlikely(i == sg_count)) {
2795 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2797 __func__, sg_len, *offset, sg_count);
2802 /* Offset starting from the beginning of first page in this sg-entry */
2803 *offset = *offset - len_complete + sg->offset;
2805 /* Assumption: contiguous pages can be accessed as "page + i" */
2806 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2807 *offset &= ~PAGE_MASK;
2809 /* Bytes in this sg-entry from *offset to the end of the page */
2810 sg_len = PAGE_SIZE - *offset;
2814 return kmap_atomic(page);
2816 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2819 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2820 * @virt: virtual address to be unmapped
2822 void scsi_kunmap_atomic_sg(void *virt)
2824 kunmap_atomic(virt);
2826 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2828 void sdev_disable_disk_events(struct scsi_device *sdev)
2830 atomic_inc(&sdev->disk_events_disable_depth);
2832 EXPORT_SYMBOL(sdev_disable_disk_events);
2834 void sdev_enable_disk_events(struct scsi_device *sdev)
2836 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2838 atomic_dec(&sdev->disk_events_disable_depth);
2840 EXPORT_SYMBOL(sdev_enable_disk_events);
2843 * scsi_vpd_lun_id - return a unique device identification
2844 * @sdev: SCSI device
2845 * @id: buffer for the identification
2846 * @id_len: length of the buffer
2848 * Copies a unique device identification into @id based
2849 * on the information in the VPD page 0x83 of the device.
2850 * The string will be formatted as a SCSI name string.
2852 * Returns the length of the identification or error on failure.
2853 * If the identifier is longer than the supplied buffer the actual
2854 * identifier length is returned and the buffer is not zero-padded.
2856 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2858 u8 cur_id_type = 0xff;
2860 const unsigned char *d, *cur_id_str;
2861 const struct scsi_vpd *vpd_pg83;
2862 int id_size = -EINVAL;
2865 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2872 * Look for the correct descriptor.
2873 * Order of preference for lun descriptor:
2874 * - SCSI name string
2875 * - NAA IEEE Registered Extended
2876 * - EUI-64 based 16-byte
2877 * - EUI-64 based 12-byte
2878 * - NAA IEEE Registered
2879 * - NAA IEEE Extended
2881 * as longer descriptors reduce the likelyhood
2882 * of identification clashes.
2885 /* The id string must be at least 20 bytes + terminating NULL byte */
2891 memset(id, 0, id_len);
2892 d = vpd_pg83->data + 4;
2893 while (d < vpd_pg83->data + vpd_pg83->len) {
2894 /* Skip designators not referring to the LUN */
2895 if ((d[1] & 0x30) != 0x00)
2898 switch (d[1] & 0xf) {
2901 if (cur_id_size > d[3])
2903 /* Prefer anything */
2904 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2907 if (cur_id_size + 4 > id_len)
2908 cur_id_size = id_len - 4;
2910 cur_id_type = d[1] & 0xf;
2911 id_size = snprintf(id, id_len, "t10.%*pE",
2912 cur_id_size, cur_id_str);
2916 if (cur_id_size > d[3])
2918 /* Prefer NAA IEEE Registered Extended */
2919 if (cur_id_type == 0x3 &&
2920 cur_id_size == d[3])
2924 cur_id_type = d[1] & 0xf;
2925 switch (cur_id_size) {
2927 id_size = snprintf(id, id_len,
2932 id_size = snprintf(id, id_len,
2937 id_size = snprintf(id, id_len,
2948 if (cur_id_size > d[3])
2952 cur_id_type = d[1] & 0xf;
2953 switch (cur_id_size) {
2955 id_size = snprintf(id, id_len,
2960 id_size = snprintf(id, id_len,
2970 /* SCSI name string */
2971 if (cur_id_size + 4 > d[3])
2973 /* Prefer others for truncated descriptor */
2974 if (cur_id_size && d[3] > id_len)
2976 cur_id_size = id_size = d[3];
2978 cur_id_type = d[1] & 0xf;
2979 if (cur_id_size >= id_len)
2980 cur_id_size = id_len - 1;
2981 memcpy(id, cur_id_str, cur_id_size);
2982 /* Decrease priority for truncated descriptor */
2983 if (cur_id_size != id_size)
2996 EXPORT_SYMBOL(scsi_vpd_lun_id);
2999 * scsi_vpd_tpg_id - return a target port group identifier
3000 * @sdev: SCSI device
3002 * Returns the Target Port Group identifier from the information
3003 * froom VPD page 0x83 of the device.
3005 * Returns the identifier or error on failure.
3007 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3009 const unsigned char *d;
3010 const struct scsi_vpd *vpd_pg83;
3011 int group_id = -EAGAIN, rel_port = -1;
3014 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3020 d = vpd_pg83->data + 4;
3021 while (d < vpd_pg83->data + vpd_pg83->len) {
3022 switch (d[1] & 0xf) {
3024 /* Relative target port */
3025 rel_port = get_unaligned_be16(&d[6]);
3028 /* Target port group */
3029 group_id = get_unaligned_be16(&d[6]);
3038 if (group_id >= 0 && rel_id && rel_port != -1)
3043 EXPORT_SYMBOL(scsi_vpd_tpg_id);