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"
43 * Size of integrity metadata is usually small, 1 inline sg should
46 #define SCSI_INLINE_PROT_SG_CNT 1
48 #define SCSI_INLINE_SG_CNT 2
50 static struct kmem_cache *scsi_sdb_cache;
51 static struct kmem_cache *scsi_sense_cache;
52 static struct kmem_cache *scsi_sense_isadma_cache;
53 static DEFINE_MUTEX(scsi_sense_cache_mutex);
55 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
57 static inline struct kmem_cache *
58 scsi_select_sense_cache(bool unchecked_isa_dma)
60 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
63 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
64 unsigned char *sense_buffer)
66 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
70 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
71 gfp_t gfp_mask, int numa_node)
73 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
77 int scsi_init_sense_cache(struct Scsi_Host *shost)
79 struct kmem_cache *cache;
82 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
86 mutex_lock(&scsi_sense_cache_mutex);
87 if (shost->unchecked_isa_dma) {
88 scsi_sense_isadma_cache =
89 kmem_cache_create("scsi_sense_cache(DMA)",
90 SCSI_SENSE_BUFFERSIZE, 0,
91 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
92 if (!scsi_sense_isadma_cache)
96 kmem_cache_create_usercopy("scsi_sense_cache",
97 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
98 0, SCSI_SENSE_BUFFERSIZE, NULL);
99 if (!scsi_sense_cache)
103 mutex_unlock(&scsi_sense_cache_mutex);
108 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
109 * not change behaviour from the previous unplug mechanism, experimentation
110 * may prove this needs changing.
112 #define SCSI_QUEUE_DELAY 3
115 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
117 struct Scsi_Host *host = cmd->device->host;
118 struct scsi_device *device = cmd->device;
119 struct scsi_target *starget = scsi_target(device);
122 * Set the appropriate busy bit for the device/host.
124 * If the host/device isn't busy, assume that something actually
125 * completed, and that we should be able to queue a command now.
127 * Note that the prior mid-layer assumption that any host could
128 * always queue at least one command is now broken. The mid-layer
129 * will implement a user specifiable stall (see
130 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
131 * if a command is requeued with no other commands outstanding
132 * either for the device or for the host.
135 case SCSI_MLQUEUE_HOST_BUSY:
136 atomic_set(&host->host_blocked, host->max_host_blocked);
138 case SCSI_MLQUEUE_DEVICE_BUSY:
139 case SCSI_MLQUEUE_EH_RETRY:
140 atomic_set(&device->device_blocked,
141 device->max_device_blocked);
143 case SCSI_MLQUEUE_TARGET_BUSY:
144 atomic_set(&starget->target_blocked,
145 starget->max_target_blocked);
150 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
152 if (cmd->request->rq_flags & RQF_DONTPREP) {
153 cmd->request->rq_flags &= ~RQF_DONTPREP;
154 scsi_mq_uninit_cmd(cmd);
158 blk_mq_requeue_request(cmd->request, true);
162 * __scsi_queue_insert - private queue insertion
163 * @cmd: The SCSI command being requeued
164 * @reason: The reason for the requeue
165 * @unbusy: Whether the queue should be unbusied
167 * This is a private queue insertion. The public interface
168 * scsi_queue_insert() always assumes the queue should be unbusied
169 * because it's always called before the completion. This function is
170 * for a requeue after completion, which should only occur in this
173 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
175 struct scsi_device *device = cmd->device;
177 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
178 "Inserting command %p into mlqueue\n", cmd));
180 scsi_set_blocked(cmd, reason);
183 * Decrement the counters, since these commands are no longer
184 * active on the host/device.
187 scsi_device_unbusy(device);
190 * Requeue this command. It will go before all other commands
191 * that are already in the queue. Schedule requeue work under
192 * lock such that the kblockd_schedule_work() call happens
193 * before blk_cleanup_queue() finishes.
197 blk_mq_requeue_request(cmd->request, true);
201 * Function: scsi_queue_insert()
203 * Purpose: Insert a command in the midlevel queue.
205 * Arguments: cmd - command that we are adding to queue.
206 * reason - why we are inserting command to queue.
208 * Lock status: Assumed that lock is not held upon entry.
212 * Notes: We do this for one of two cases. Either the host is busy
213 * and it cannot accept any more commands for the time being,
214 * or the device returned QUEUE_FULL and can accept no more
216 * Notes: This could be called either from an interrupt context or a
217 * normal process context.
219 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
221 __scsi_queue_insert(cmd, reason, true);
226 * __scsi_execute - insert request and wait for the result
229 * @data_direction: data direction
230 * @buffer: data buffer
231 * @bufflen: len of buffer
232 * @sense: optional sense buffer
233 * @sshdr: optional decoded sense header
234 * @timeout: request timeout in seconds
235 * @retries: number of times to retry request
236 * @flags: flags for ->cmd_flags
237 * @rq_flags: flags for ->rq_flags
238 * @resid: optional residual length
240 * Returns the scsi_cmnd result field if a command was executed, or a negative
241 * Linux error code if we didn't get that far.
243 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
244 int data_direction, void *buffer, unsigned bufflen,
245 unsigned char *sense, struct scsi_sense_hdr *sshdr,
246 int timeout, int retries, u64 flags, req_flags_t rq_flags,
250 struct scsi_request *rq;
251 int ret = DRIVER_ERROR << 24;
253 req = blk_get_request(sdev->request_queue,
254 data_direction == DMA_TO_DEVICE ?
255 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
260 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
261 buffer, bufflen, GFP_NOIO))
264 rq->cmd_len = COMMAND_SIZE(cmd[0]);
265 memcpy(rq->cmd, cmd, rq->cmd_len);
266 rq->retries = retries;
267 req->timeout = timeout;
268 req->cmd_flags |= flags;
269 req->rq_flags |= rq_flags | RQF_QUIET;
272 * head injection *required* here otherwise quiesce won't work
274 blk_execute_rq(req->q, NULL, req, 1);
277 * Some devices (USB mass-storage in particular) may transfer
278 * garbage data together with a residue indicating that the data
279 * is invalid. Prevent the garbage from being misinterpreted
280 * and prevent security leaks by zeroing out the excess data.
282 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
283 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
286 *resid = rq->resid_len;
287 if (sense && rq->sense_len)
288 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
290 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
293 blk_put_request(req);
297 EXPORT_SYMBOL(__scsi_execute);
300 * Function: scsi_init_cmd_errh()
302 * Purpose: Initialize cmd fields related to error handling.
304 * Arguments: cmd - command that is ready to be queued.
306 * Notes: This function has the job of initializing a number of
307 * fields related to error handling. Typically this will
308 * be called once for each command, as required.
310 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
312 scsi_set_resid(cmd, 0);
313 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
314 if (cmd->cmd_len == 0)
315 cmd->cmd_len = scsi_command_size(cmd->cmnd);
319 * Decrement the host_busy counter and wake up the error handler if necessary.
320 * Avoid as follows that the error handler is not woken up if shost->host_busy
321 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
322 * with an RCU read lock in this function to ensure that this function in its
323 * entirety either finishes before scsi_eh_scmd_add() increases the
324 * host_failed counter or that it notices the shost state change made by
325 * scsi_eh_scmd_add().
327 static void scsi_dec_host_busy(struct Scsi_Host *shost)
332 atomic_dec(&shost->host_busy);
333 if (unlikely(scsi_host_in_recovery(shost))) {
334 spin_lock_irqsave(shost->host_lock, flags);
335 if (shost->host_failed || shost->host_eh_scheduled)
336 scsi_eh_wakeup(shost);
337 spin_unlock_irqrestore(shost->host_lock, flags);
342 void scsi_device_unbusy(struct scsi_device *sdev)
344 struct Scsi_Host *shost = sdev->host;
345 struct scsi_target *starget = scsi_target(sdev);
347 scsi_dec_host_busy(shost);
349 if (starget->can_queue > 0)
350 atomic_dec(&starget->target_busy);
352 atomic_dec(&sdev->device_busy);
355 static void scsi_kick_queue(struct request_queue *q)
357 blk_mq_run_hw_queues(q, false);
361 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
362 * and call blk_run_queue for all the scsi_devices on the target -
363 * including current_sdev first.
365 * Called with *no* scsi locks held.
367 static void scsi_single_lun_run(struct scsi_device *current_sdev)
369 struct Scsi_Host *shost = current_sdev->host;
370 struct scsi_device *sdev, *tmp;
371 struct scsi_target *starget = scsi_target(current_sdev);
374 spin_lock_irqsave(shost->host_lock, flags);
375 starget->starget_sdev_user = NULL;
376 spin_unlock_irqrestore(shost->host_lock, flags);
379 * Call blk_run_queue for all LUNs on the target, starting with
380 * current_sdev. We race with others (to set starget_sdev_user),
381 * but in most cases, we will be first. Ideally, each LU on the
382 * target would get some limited time or requests on the target.
384 scsi_kick_queue(current_sdev->request_queue);
386 spin_lock_irqsave(shost->host_lock, flags);
387 if (starget->starget_sdev_user)
389 list_for_each_entry_safe(sdev, tmp, &starget->devices,
390 same_target_siblings) {
391 if (sdev == current_sdev)
393 if (scsi_device_get(sdev))
396 spin_unlock_irqrestore(shost->host_lock, flags);
397 scsi_kick_queue(sdev->request_queue);
398 spin_lock_irqsave(shost->host_lock, flags);
400 scsi_device_put(sdev);
403 spin_unlock_irqrestore(shost->host_lock, flags);
406 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
408 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
410 if (atomic_read(&sdev->device_blocked) > 0)
415 static inline bool scsi_target_is_busy(struct scsi_target *starget)
417 if (starget->can_queue > 0) {
418 if (atomic_read(&starget->target_busy) >= starget->can_queue)
420 if (atomic_read(&starget->target_blocked) > 0)
426 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
428 if (shost->can_queue > 0 &&
429 atomic_read(&shost->host_busy) >= shost->can_queue)
431 if (atomic_read(&shost->host_blocked) > 0)
433 if (shost->host_self_blocked)
438 static void scsi_starved_list_run(struct Scsi_Host *shost)
440 LIST_HEAD(starved_list);
441 struct scsi_device *sdev;
444 spin_lock_irqsave(shost->host_lock, flags);
445 list_splice_init(&shost->starved_list, &starved_list);
447 while (!list_empty(&starved_list)) {
448 struct request_queue *slq;
451 * As long as shost is accepting commands and we have
452 * starved queues, call blk_run_queue. scsi_request_fn
453 * drops the queue_lock and can add us back to the
456 * host_lock protects the starved_list and starved_entry.
457 * scsi_request_fn must get the host_lock before checking
458 * or modifying starved_list or starved_entry.
460 if (scsi_host_is_busy(shost))
463 sdev = list_entry(starved_list.next,
464 struct scsi_device, starved_entry);
465 list_del_init(&sdev->starved_entry);
466 if (scsi_target_is_busy(scsi_target(sdev))) {
467 list_move_tail(&sdev->starved_entry,
468 &shost->starved_list);
473 * Once we drop the host lock, a racing scsi_remove_device()
474 * call may remove the sdev from the starved list and destroy
475 * it and the queue. Mitigate by taking a reference to the
476 * queue and never touching the sdev again after we drop the
477 * host lock. Note: if __scsi_remove_device() invokes
478 * blk_cleanup_queue() before the queue is run from this
479 * function then blk_run_queue() will return immediately since
480 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
482 slq = sdev->request_queue;
483 if (!blk_get_queue(slq))
485 spin_unlock_irqrestore(shost->host_lock, flags);
487 scsi_kick_queue(slq);
490 spin_lock_irqsave(shost->host_lock, flags);
492 /* put any unprocessed entries back */
493 list_splice(&starved_list, &shost->starved_list);
494 spin_unlock_irqrestore(shost->host_lock, flags);
498 * Function: scsi_run_queue()
500 * Purpose: Select a proper request queue to serve next
502 * Arguments: q - last request's queue
506 * Notes: The previous command was completely finished, start
507 * a new one if possible.
509 static void scsi_run_queue(struct request_queue *q)
511 struct scsi_device *sdev = q->queuedata;
513 if (scsi_target(sdev)->single_lun)
514 scsi_single_lun_run(sdev);
515 if (!list_empty(&sdev->host->starved_list))
516 scsi_starved_list_run(sdev->host);
518 blk_mq_run_hw_queues(q, false);
521 void scsi_requeue_run_queue(struct work_struct *work)
523 struct scsi_device *sdev;
524 struct request_queue *q;
526 sdev = container_of(work, struct scsi_device, requeue_work);
527 q = sdev->request_queue;
531 void scsi_run_host_queues(struct Scsi_Host *shost)
533 struct scsi_device *sdev;
535 shost_for_each_device(sdev, shost)
536 scsi_run_queue(sdev->request_queue);
539 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
541 if (!blk_rq_is_passthrough(cmd->request)) {
542 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
544 if (drv->uninit_command)
545 drv->uninit_command(cmd);
549 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
551 if (cmd->sdb.table.nents)
552 sg_free_table_chained(&cmd->sdb.table,
554 if (scsi_prot_sg_count(cmd))
555 sg_free_table_chained(&cmd->prot_sdb->table,
556 SCSI_INLINE_PROT_SG_CNT);
559 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
561 scsi_mq_free_sgtables(cmd);
562 scsi_uninit_cmd(cmd);
563 scsi_del_cmd_from_list(cmd);
566 /* Returns false when no more bytes to process, true if there are more */
567 static bool scsi_end_request(struct request *req, blk_status_t error,
570 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
571 struct scsi_device *sdev = cmd->device;
572 struct request_queue *q = sdev->request_queue;
574 if (blk_update_request(req, error, bytes))
577 if (blk_queue_add_random(q))
578 add_disk_randomness(req->rq_disk);
580 if (!blk_rq_is_scsi(req)) {
581 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
582 cmd->flags &= ~SCMD_INITIALIZED;
586 * Calling rcu_barrier() is not necessary here because the
587 * SCSI error handler guarantees that the function called by
588 * call_rcu() has been called before scsi_end_request() is
591 destroy_rcu_head(&cmd->rcu);
594 * In the MQ case the command gets freed by __blk_mq_end_request,
595 * so we have to do all cleanup that depends on it earlier.
597 * We also can't kick the queues from irq context, so we
598 * will have to defer it to a workqueue.
600 scsi_mq_uninit_cmd(cmd);
603 * queue is still alive, so grab the ref for preventing it
604 * from being cleaned up during running queue.
606 percpu_ref_get(&q->q_usage_counter);
608 __blk_mq_end_request(req, error);
610 if (scsi_target(sdev)->single_lun ||
611 !list_empty(&sdev->host->starved_list))
612 kblockd_schedule_work(&sdev->requeue_work);
614 blk_mq_run_hw_queues(q, true);
616 percpu_ref_put(&q->q_usage_counter);
621 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
623 * @result: scsi error code
625 * Translate a SCSI result code into a blk_status_t value. May reset the host
626 * byte of @cmd->result.
628 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
630 switch (host_byte(result)) {
633 * Also check the other bytes than the status byte in result
634 * to handle the case when a SCSI LLD sets result to
635 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
637 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
639 return BLK_STS_IOERR;
640 case DID_TRANSPORT_FAILFAST:
641 return BLK_STS_TRANSPORT;
642 case DID_TARGET_FAILURE:
643 set_host_byte(cmd, DID_OK);
644 return BLK_STS_TARGET;
645 case DID_NEXUS_FAILURE:
646 set_host_byte(cmd, DID_OK);
647 return BLK_STS_NEXUS;
648 case DID_ALLOC_FAILURE:
649 set_host_byte(cmd, DID_OK);
650 return BLK_STS_NOSPC;
651 case DID_MEDIUM_ERROR:
652 set_host_byte(cmd, DID_OK);
653 return BLK_STS_MEDIUM;
655 return BLK_STS_IOERR;
659 /* Helper for scsi_io_completion() when "reprep" action required. */
660 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
661 struct request_queue *q)
663 /* A new command will be prepared and issued. */
664 scsi_mq_requeue_cmd(cmd);
667 /* Helper for scsi_io_completion() when special action required. */
668 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
670 struct request_queue *q = cmd->device->request_queue;
671 struct request *req = cmd->request;
673 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
674 ACTION_DELAYED_RETRY} action;
675 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
676 struct scsi_sense_hdr sshdr;
678 bool sense_current = true; /* false implies "deferred sense" */
679 blk_status_t blk_stat;
681 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
683 sense_current = !scsi_sense_is_deferred(&sshdr);
685 blk_stat = scsi_result_to_blk_status(cmd, result);
687 if (host_byte(result) == DID_RESET) {
688 /* Third party bus reset or reset for error recovery
689 * reasons. Just retry the command and see what
692 action = ACTION_RETRY;
693 } else if (sense_valid && sense_current) {
694 switch (sshdr.sense_key) {
696 if (cmd->device->removable) {
697 /* Detected disc change. Set a bit
698 * and quietly refuse further access.
700 cmd->device->changed = 1;
701 action = ACTION_FAIL;
703 /* Must have been a power glitch, or a
704 * bus reset. Could not have been a
705 * media change, so we just retry the
706 * command and see what happens.
708 action = ACTION_RETRY;
711 case ILLEGAL_REQUEST:
712 /* If we had an ILLEGAL REQUEST returned, then
713 * we may have performed an unsupported
714 * command. The only thing this should be
715 * would be a ten byte read where only a six
716 * byte read was supported. Also, on a system
717 * where READ CAPACITY failed, we may have
718 * read past the end of the disk.
720 if ((cmd->device->use_10_for_rw &&
721 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
722 (cmd->cmnd[0] == READ_10 ||
723 cmd->cmnd[0] == WRITE_10)) {
724 /* This will issue a new 6-byte command. */
725 cmd->device->use_10_for_rw = 0;
726 action = ACTION_REPREP;
727 } else if (sshdr.asc == 0x10) /* DIX */ {
728 action = ACTION_FAIL;
729 blk_stat = BLK_STS_PROTECTION;
730 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
731 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
732 action = ACTION_FAIL;
733 blk_stat = BLK_STS_TARGET;
735 action = ACTION_FAIL;
737 case ABORTED_COMMAND:
738 action = ACTION_FAIL;
739 if (sshdr.asc == 0x10) /* DIF */
740 blk_stat = BLK_STS_PROTECTION;
743 /* If the device is in the process of becoming
744 * ready, or has a temporary blockage, retry.
746 if (sshdr.asc == 0x04) {
747 switch (sshdr.ascq) {
748 case 0x01: /* becoming ready */
749 case 0x04: /* format in progress */
750 case 0x05: /* rebuild in progress */
751 case 0x06: /* recalculation in progress */
752 case 0x07: /* operation in progress */
753 case 0x08: /* Long write in progress */
754 case 0x09: /* self test in progress */
755 case 0x14: /* space allocation in progress */
756 case 0x1a: /* start stop unit in progress */
757 case 0x1b: /* sanitize in progress */
758 case 0x1d: /* configuration in progress */
759 case 0x24: /* depopulation in progress */
760 action = ACTION_DELAYED_RETRY;
763 action = ACTION_FAIL;
767 action = ACTION_FAIL;
769 case VOLUME_OVERFLOW:
770 /* See SSC3rXX or current. */
771 action = ACTION_FAIL;
774 action = ACTION_FAIL;
778 action = ACTION_FAIL;
780 if (action != ACTION_FAIL &&
781 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
782 action = ACTION_FAIL;
786 /* Give up and fail the remainder of the request */
787 if (!(req->rq_flags & RQF_QUIET)) {
788 static DEFINE_RATELIMIT_STATE(_rs,
789 DEFAULT_RATELIMIT_INTERVAL,
790 DEFAULT_RATELIMIT_BURST);
792 if (unlikely(scsi_logging_level))
794 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
795 SCSI_LOG_MLCOMPLETE_BITS);
798 * if logging is enabled the failure will be printed
799 * in scsi_log_completion(), so avoid duplicate messages
801 if (!level && __ratelimit(&_rs)) {
802 scsi_print_result(cmd, NULL, FAILED);
803 if (driver_byte(result) == DRIVER_SENSE)
804 scsi_print_sense(cmd);
805 scsi_print_command(cmd);
808 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
812 scsi_io_completion_reprep(cmd, q);
815 /* Retry the same command immediately */
816 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
818 case ACTION_DELAYED_RETRY:
819 /* Retry the same command after a delay */
820 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
826 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
827 * new result that may suppress further error checking. Also modifies
828 * *blk_statp in some cases.
830 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
831 blk_status_t *blk_statp)
834 bool sense_current = true; /* false implies "deferred sense" */
835 struct request *req = cmd->request;
836 struct scsi_sense_hdr sshdr;
838 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
840 sense_current = !scsi_sense_is_deferred(&sshdr);
842 if (blk_rq_is_passthrough(req)) {
845 * SG_IO wants current and deferred errors
847 scsi_req(req)->sense_len =
848 min(8 + cmd->sense_buffer[7],
849 SCSI_SENSE_BUFFERSIZE);
852 *blk_statp = scsi_result_to_blk_status(cmd, result);
853 } else if (blk_rq_bytes(req) == 0 && sense_current) {
855 * Flush commands do not transfers any data, and thus cannot use
856 * good_bytes != blk_rq_bytes(req) as the signal for an error.
857 * This sets *blk_statp explicitly for the problem case.
859 *blk_statp = scsi_result_to_blk_status(cmd, result);
862 * Recovered errors need reporting, but they're always treated as
863 * success, so fiddle the result code here. For passthrough requests
864 * we already took a copy of the original into sreq->result which
865 * is what gets returned to the user
867 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
868 bool do_print = true;
870 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
871 * skip print since caller wants ATA registers. Only occurs
872 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
874 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
876 else if (req->rq_flags & RQF_QUIET)
879 scsi_print_sense(cmd);
881 /* for passthrough, *blk_statp may be set */
882 *blk_statp = BLK_STS_OK;
885 * Another corner case: the SCSI status byte is non-zero but 'good'.
886 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
887 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
888 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
889 * intermediate statuses (both obsolete in SAM-4) as good.
891 if (status_byte(result) && scsi_status_is_good(result)) {
893 *blk_statp = BLK_STS_OK;
899 * Function: scsi_io_completion()
901 * Purpose: Completion processing for block device I/O requests.
903 * Arguments: cmd - command that is finished.
905 * Lock status: Assumed that no lock is held upon entry.
909 * Notes: We will finish off the specified number of sectors. If we
910 * are done, the command block will be released and the queue
911 * function will be goosed. If we are not done then we have to
912 * figure out what to do next:
914 * a) We can call scsi_requeue_command(). The request
915 * will be unprepared and put back on the queue. Then
916 * a new command will be created for it. This should
917 * be used if we made forward progress, or if we want
918 * to switch from READ(10) to READ(6) for example.
920 * b) We can call __scsi_queue_insert(). The request will
921 * be put back on the queue and retried using the same
922 * command as before, possibly after a delay.
924 * c) We can call scsi_end_request() with blk_stat other than
925 * BLK_STS_OK, to fail the remainder of the request.
927 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
929 int result = cmd->result;
930 struct request_queue *q = cmd->device->request_queue;
931 struct request *req = cmd->request;
932 blk_status_t blk_stat = BLK_STS_OK;
934 if (unlikely(result)) /* a nz result may or may not be an error */
935 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
937 if (unlikely(blk_rq_is_passthrough(req))) {
939 * scsi_result_to_blk_status may have reset the host_byte
941 scsi_req(req)->result = cmd->result;
945 * Next deal with any sectors which we were able to correctly
948 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
949 "%u sectors total, %d bytes done.\n",
950 blk_rq_sectors(req), good_bytes));
953 * Next deal with any sectors which we were able to correctly
954 * handle. Failed, zero length commands always need to drop down
955 * to retry code. Fast path should return in this block.
957 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
958 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
959 return; /* no bytes remaining */
962 /* Kill remainder if no retries. */
963 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
964 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
966 "Bytes remaining after failed, no-retry command");
971 * If there had been no error, but we have leftover bytes in the
972 * requeues just queue the command up again.
974 if (likely(result == 0))
975 scsi_io_completion_reprep(cmd, q);
977 scsi_io_completion_action(cmd, result);
980 static blk_status_t scsi_init_sgtable(struct request *req,
981 struct scsi_data_buffer *sdb)
986 * If sg table allocation fails, requeue request later.
988 if (unlikely(sg_alloc_table_chained(&sdb->table,
989 blk_rq_nr_phys_segments(req), sdb->table.sgl,
990 SCSI_INLINE_SG_CNT)))
991 return BLK_STS_RESOURCE;
994 * Next, walk the list, and fill in the addresses and sizes of
997 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
998 BUG_ON(count > sdb->table.nents);
999 sdb->table.nents = count;
1000 sdb->length = blk_rq_payload_bytes(req);
1005 * Function: scsi_init_io()
1007 * Purpose: SCSI I/O initialize function.
1009 * Arguments: cmd - Command descriptor we wish to initialize
1011 * Returns: BLK_STS_OK on success
1012 * BLK_STS_RESOURCE if the failure is retryable
1013 * BLK_STS_IOERR if the failure is fatal
1015 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1017 struct request *rq = cmd->request;
1020 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1021 return BLK_STS_IOERR;
1023 ret = scsi_init_sgtable(rq, &cmd->sdb);
1027 if (blk_integrity_rq(rq)) {
1028 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1031 if (WARN_ON_ONCE(!prot_sdb)) {
1033 * This can happen if someone (e.g. multipath)
1034 * queues a command to a device on an adapter
1035 * that does not support DIX.
1037 ret = BLK_STS_IOERR;
1038 goto out_free_sgtables;
1041 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1043 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1044 prot_sdb->table.sgl,
1045 SCSI_INLINE_PROT_SG_CNT)) {
1046 ret = BLK_STS_RESOURCE;
1047 goto out_free_sgtables;
1050 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1051 prot_sdb->table.sgl);
1052 BUG_ON(count > ivecs);
1053 BUG_ON(count > queue_max_integrity_segments(rq->q));
1055 cmd->prot_sdb = prot_sdb;
1056 cmd->prot_sdb->table.nents = count;
1061 scsi_mq_free_sgtables(cmd);
1064 EXPORT_SYMBOL(scsi_init_io);
1067 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1068 * @rq: Request associated with the SCSI command to be initialized.
1070 * This function initializes the members of struct scsi_cmnd that must be
1071 * initialized before request processing starts and that won't be
1072 * reinitialized if a SCSI command is requeued.
1074 * Called from inside blk_get_request() for pass-through requests and from
1075 * inside scsi_init_command() for filesystem requests.
1077 static void scsi_initialize_rq(struct request *rq)
1079 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1081 scsi_req_init(&cmd->req);
1082 init_rcu_head(&cmd->rcu);
1083 cmd->jiffies_at_alloc = jiffies;
1087 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1088 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1090 struct scsi_device *sdev = cmd->device;
1091 struct Scsi_Host *shost = sdev->host;
1092 unsigned long flags;
1094 if (shost->use_cmd_list) {
1095 spin_lock_irqsave(&sdev->list_lock, flags);
1096 list_add_tail(&cmd->list, &sdev->cmd_list);
1097 spin_unlock_irqrestore(&sdev->list_lock, flags);
1101 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1102 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1104 struct scsi_device *sdev = cmd->device;
1105 struct Scsi_Host *shost = sdev->host;
1106 unsigned long flags;
1108 if (shost->use_cmd_list) {
1109 spin_lock_irqsave(&sdev->list_lock, flags);
1110 BUG_ON(list_empty(&cmd->list));
1111 list_del_init(&cmd->list);
1112 spin_unlock_irqrestore(&sdev->list_lock, flags);
1116 /* Called after a request has been started. */
1117 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1119 void *buf = cmd->sense_buffer;
1120 void *prot = cmd->prot_sdb;
1121 struct request *rq = blk_mq_rq_from_pdu(cmd);
1122 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1123 unsigned long jiffies_at_alloc;
1126 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1127 flags |= SCMD_INITIALIZED;
1128 scsi_initialize_rq(rq);
1131 jiffies_at_alloc = cmd->jiffies_at_alloc;
1132 retries = cmd->retries;
1133 /* zero out the cmd, except for the embedded scsi_request */
1134 memset((char *)cmd + sizeof(cmd->req), 0,
1135 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1138 cmd->sense_buffer = buf;
1139 cmd->prot_sdb = prot;
1141 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1142 cmd->jiffies_at_alloc = jiffies_at_alloc;
1143 cmd->retries = retries;
1145 scsi_add_cmd_to_list(cmd);
1148 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1149 struct request *req)
1151 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1154 * Passthrough requests may transfer data, in which case they must
1155 * a bio attached to them. Or they might contain a SCSI command
1156 * that does not transfer data, in which case they may optionally
1157 * submit a request without an attached bio.
1160 blk_status_t ret = scsi_init_io(cmd);
1161 if (unlikely(ret != BLK_STS_OK))
1164 BUG_ON(blk_rq_bytes(req));
1166 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1169 cmd->cmd_len = scsi_req(req)->cmd_len;
1170 cmd->cmnd = scsi_req(req)->cmd;
1171 cmd->transfersize = blk_rq_bytes(req);
1172 cmd->allowed = scsi_req(req)->retries;
1177 * Setup a normal block command. These are simple request from filesystems
1178 * that still need to be translated to SCSI CDBs from the ULD.
1180 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1181 struct request *req)
1183 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1185 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1186 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1187 if (ret != BLK_STS_OK)
1191 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1192 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1193 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1196 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1197 struct request *req)
1199 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1201 if (!blk_rq_bytes(req))
1202 cmd->sc_data_direction = DMA_NONE;
1203 else if (rq_data_dir(req) == WRITE)
1204 cmd->sc_data_direction = DMA_TO_DEVICE;
1206 cmd->sc_data_direction = DMA_FROM_DEVICE;
1208 if (blk_rq_is_scsi(req))
1209 return scsi_setup_scsi_cmnd(sdev, req);
1211 return scsi_setup_fs_cmnd(sdev, req);
1215 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1217 switch (sdev->sdev_state) {
1219 case SDEV_TRANSPORT_OFFLINE:
1221 * If the device is offline we refuse to process any
1222 * commands. The device must be brought online
1223 * before trying any recovery commands.
1225 sdev_printk(KERN_ERR, sdev,
1226 "rejecting I/O to offline device\n");
1227 return BLK_STS_IOERR;
1230 * If the device is fully deleted, we refuse to
1231 * process any commands as well.
1233 sdev_printk(KERN_ERR, sdev,
1234 "rejecting I/O to dead device\n");
1235 return BLK_STS_IOERR;
1237 case SDEV_CREATED_BLOCK:
1238 return BLK_STS_RESOURCE;
1241 * If the devices is blocked we defer normal commands.
1243 if (req && !(req->rq_flags & RQF_PREEMPT))
1244 return BLK_STS_RESOURCE;
1248 * For any other not fully online state we only allow
1249 * special commands. In particular any user initiated
1250 * command is not allowed.
1252 if (req && !(req->rq_flags & RQF_PREEMPT))
1253 return BLK_STS_IOERR;
1259 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1262 * Called with the queue_lock held.
1264 static inline int scsi_dev_queue_ready(struct request_queue *q,
1265 struct scsi_device *sdev)
1269 busy = atomic_inc_return(&sdev->device_busy) - 1;
1270 if (atomic_read(&sdev->device_blocked)) {
1275 * unblock after device_blocked iterates to zero
1277 if (atomic_dec_return(&sdev->device_blocked) > 0)
1279 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1280 "unblocking device at zero depth\n"));
1283 if (busy >= sdev->queue_depth)
1288 atomic_dec(&sdev->device_busy);
1293 * scsi_target_queue_ready: checks if there we can send commands to target
1294 * @sdev: scsi device on starget to check.
1296 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1297 struct scsi_device *sdev)
1299 struct scsi_target *starget = scsi_target(sdev);
1302 if (starget->single_lun) {
1303 spin_lock_irq(shost->host_lock);
1304 if (starget->starget_sdev_user &&
1305 starget->starget_sdev_user != sdev) {
1306 spin_unlock_irq(shost->host_lock);
1309 starget->starget_sdev_user = sdev;
1310 spin_unlock_irq(shost->host_lock);
1313 if (starget->can_queue <= 0)
1316 busy = atomic_inc_return(&starget->target_busy) - 1;
1317 if (atomic_read(&starget->target_blocked) > 0) {
1322 * unblock after target_blocked iterates to zero
1324 if (atomic_dec_return(&starget->target_blocked) > 0)
1327 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1328 "unblocking target at zero depth\n"));
1331 if (busy >= starget->can_queue)
1337 spin_lock_irq(shost->host_lock);
1338 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1339 spin_unlock_irq(shost->host_lock);
1341 if (starget->can_queue > 0)
1342 atomic_dec(&starget->target_busy);
1347 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1348 * return 0. We must end up running the queue again whenever 0 is
1349 * returned, else IO can hang.
1351 static inline int scsi_host_queue_ready(struct request_queue *q,
1352 struct Scsi_Host *shost,
1353 struct scsi_device *sdev)
1357 if (scsi_host_in_recovery(shost))
1360 busy = atomic_inc_return(&shost->host_busy) - 1;
1361 if (atomic_read(&shost->host_blocked) > 0) {
1366 * unblock after host_blocked iterates to zero
1368 if (atomic_dec_return(&shost->host_blocked) > 0)
1372 shost_printk(KERN_INFO, shost,
1373 "unblocking host at zero depth\n"));
1376 if (shost->can_queue > 0 && busy >= shost->can_queue)
1378 if (shost->host_self_blocked)
1381 /* We're OK to process the command, so we can't be starved */
1382 if (!list_empty(&sdev->starved_entry)) {
1383 spin_lock_irq(shost->host_lock);
1384 if (!list_empty(&sdev->starved_entry))
1385 list_del_init(&sdev->starved_entry);
1386 spin_unlock_irq(shost->host_lock);
1392 spin_lock_irq(shost->host_lock);
1393 if (list_empty(&sdev->starved_entry))
1394 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1395 spin_unlock_irq(shost->host_lock);
1397 scsi_dec_host_busy(shost);
1402 * Busy state exporting function for request stacking drivers.
1404 * For efficiency, no lock is taken to check the busy state of
1405 * shost/starget/sdev, since the returned value is not guaranteed and
1406 * may be changed after request stacking drivers call the function,
1407 * regardless of taking lock or not.
1409 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1410 * needs to return 'not busy'. Otherwise, request stacking drivers
1411 * may hold requests forever.
1413 static bool scsi_mq_lld_busy(struct request_queue *q)
1415 struct scsi_device *sdev = q->queuedata;
1416 struct Scsi_Host *shost;
1418 if (blk_queue_dying(q))
1424 * Ignore host/starget busy state.
1425 * Since block layer does not have a concept of fairness across
1426 * multiple queues, congestion of host/starget needs to be handled
1429 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1435 static void scsi_softirq_done(struct request *rq)
1437 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1438 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1441 INIT_LIST_HEAD(&cmd->eh_entry);
1443 atomic_inc(&cmd->device->iodone_cnt);
1445 atomic_inc(&cmd->device->ioerr_cnt);
1447 disposition = scsi_decide_disposition(cmd);
1448 if (disposition != SUCCESS &&
1449 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1450 sdev_printk(KERN_ERR, cmd->device,
1451 "timing out command, waited %lus\n",
1453 disposition = SUCCESS;
1456 scsi_log_completion(cmd, disposition);
1458 switch (disposition) {
1460 scsi_finish_command(cmd);
1463 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1465 case ADD_TO_MLQUEUE:
1466 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1469 scsi_eh_scmd_add(cmd);
1475 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1476 * @cmd: command block we are dispatching.
1478 * Return: nonzero return request was rejected and device's queue needs to be
1481 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1483 struct Scsi_Host *host = cmd->device->host;
1486 atomic_inc(&cmd->device->iorequest_cnt);
1488 /* check if the device is still usable */
1489 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1490 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1491 * returns an immediate error upwards, and signals
1492 * that the device is no longer present */
1493 cmd->result = DID_NO_CONNECT << 16;
1497 /* Check to see if the scsi lld made this device blocked. */
1498 if (unlikely(scsi_device_blocked(cmd->device))) {
1500 * in blocked state, the command is just put back on
1501 * the device queue. The suspend state has already
1502 * blocked the queue so future requests should not
1503 * occur until the device transitions out of the
1506 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1507 "queuecommand : device blocked\n"));
1508 return SCSI_MLQUEUE_DEVICE_BUSY;
1511 /* Store the LUN value in cmnd, if needed. */
1512 if (cmd->device->lun_in_cdb)
1513 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1514 (cmd->device->lun << 5 & 0xe0);
1519 * Before we queue this command, check if the command
1520 * length exceeds what the host adapter can handle.
1522 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1523 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1524 "queuecommand : command too long. "
1525 "cdb_size=%d host->max_cmd_len=%d\n",
1526 cmd->cmd_len, cmd->device->host->max_cmd_len));
1527 cmd->result = (DID_ABORT << 16);
1531 if (unlikely(host->shost_state == SHOST_DEL)) {
1532 cmd->result = (DID_NO_CONNECT << 16);
1537 trace_scsi_dispatch_cmd_start(cmd);
1538 rtn = host->hostt->queuecommand(host, cmd);
1540 trace_scsi_dispatch_cmd_error(cmd, rtn);
1541 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1542 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1543 rtn = SCSI_MLQUEUE_HOST_BUSY;
1545 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1546 "queuecommand : request rejected\n"));
1551 cmd->scsi_done(cmd);
1555 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1556 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1558 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1559 sizeof(struct scatterlist);
1562 static blk_status_t scsi_mq_prep_fn(struct request *req)
1564 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1565 struct scsi_device *sdev = req->q->queuedata;
1566 struct Scsi_Host *shost = sdev->host;
1567 struct scatterlist *sg;
1569 scsi_init_command(sdev, cmd);
1572 cmd->tag = req->tag;
1573 cmd->prot_op = SCSI_PROT_NORMAL;
1575 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1576 cmd->sdb.table.sgl = sg;
1578 if (scsi_host_get_prot(shost)) {
1579 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1581 cmd->prot_sdb->table.sgl =
1582 (struct scatterlist *)(cmd->prot_sdb + 1);
1585 blk_mq_start_request(req);
1587 return scsi_setup_cmnd(sdev, req);
1590 static void scsi_mq_done(struct scsi_cmnd *cmd)
1592 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1594 trace_scsi_dispatch_cmd_done(cmd);
1597 * If the block layer didn't complete the request due to a timeout
1598 * injection, scsi must clear its internal completed state so that the
1599 * timeout handler will see it needs to escalate its own error
1602 if (unlikely(!blk_mq_complete_request(cmd->request)))
1603 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1606 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1608 struct request_queue *q = hctx->queue;
1609 struct scsi_device *sdev = q->queuedata;
1611 atomic_dec(&sdev->device_busy);
1614 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1616 struct request_queue *q = hctx->queue;
1617 struct scsi_device *sdev = q->queuedata;
1619 if (scsi_dev_queue_ready(q, sdev))
1622 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1623 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1627 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1628 const struct blk_mq_queue_data *bd)
1630 struct request *req = bd->rq;
1631 struct request_queue *q = req->q;
1632 struct scsi_device *sdev = q->queuedata;
1633 struct Scsi_Host *shost = sdev->host;
1634 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1639 * If the device is not in running state we will reject some or all
1642 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1643 ret = scsi_prep_state_check(sdev, req);
1644 if (ret != BLK_STS_OK)
1645 goto out_put_budget;
1648 ret = BLK_STS_RESOURCE;
1649 if (!scsi_target_queue_ready(shost, sdev))
1650 goto out_put_budget;
1651 if (!scsi_host_queue_ready(q, shost, sdev))
1652 goto out_dec_target_busy;
1654 if (!(req->rq_flags & RQF_DONTPREP)) {
1655 ret = scsi_mq_prep_fn(req);
1656 if (ret != BLK_STS_OK)
1657 goto out_dec_host_busy;
1658 req->rq_flags |= RQF_DONTPREP;
1660 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1661 blk_mq_start_request(req);
1664 if (sdev->simple_tags)
1665 cmd->flags |= SCMD_TAGGED;
1667 cmd->flags &= ~SCMD_TAGGED;
1669 scsi_init_cmd_errh(cmd);
1670 cmd->scsi_done = scsi_mq_done;
1672 reason = scsi_dispatch_cmd(cmd);
1674 scsi_set_blocked(cmd, reason);
1675 ret = BLK_STS_RESOURCE;
1676 goto out_dec_host_busy;
1682 scsi_dec_host_busy(shost);
1683 out_dec_target_busy:
1684 if (scsi_target(sdev)->can_queue > 0)
1685 atomic_dec(&scsi_target(sdev)->target_busy);
1687 scsi_mq_put_budget(hctx);
1691 case BLK_STS_RESOURCE:
1692 if (atomic_read(&sdev->device_busy) ||
1693 scsi_device_blocked(sdev))
1694 ret = BLK_STS_DEV_RESOURCE;
1697 if (unlikely(!scsi_device_online(sdev)))
1698 scsi_req(req)->result = DID_NO_CONNECT << 16;
1700 scsi_req(req)->result = DID_ERROR << 16;
1702 * Make sure to release all allocated resources when
1703 * we hit an error, as we will never see this command
1706 if (req->rq_flags & RQF_DONTPREP)
1707 scsi_mq_uninit_cmd(cmd);
1713 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1717 return BLK_EH_RESET_TIMER;
1718 return scsi_times_out(req);
1721 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1722 unsigned int hctx_idx, unsigned int numa_node)
1724 struct Scsi_Host *shost = set->driver_data;
1725 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1726 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1727 struct scatterlist *sg;
1729 if (unchecked_isa_dma)
1730 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1731 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1732 GFP_KERNEL, numa_node);
1733 if (!cmd->sense_buffer)
1735 cmd->req.sense = cmd->sense_buffer;
1737 if (scsi_host_get_prot(shost)) {
1738 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1739 shost->hostt->cmd_size;
1740 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1746 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1747 unsigned int hctx_idx)
1749 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1751 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1755 static int scsi_map_queues(struct blk_mq_tag_set *set)
1757 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1759 if (shost->hostt->map_queues)
1760 return shost->hostt->map_queues(shost);
1761 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1764 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1766 struct device *dev = shost->dma_dev;
1769 * this limit is imposed by hardware restrictions
1771 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1774 if (scsi_host_prot_dma(shost)) {
1775 shost->sg_prot_tablesize =
1776 min_not_zero(shost->sg_prot_tablesize,
1777 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1778 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1779 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1782 blk_queue_max_hw_sectors(q, shost->max_sectors);
1783 if (shost->unchecked_isa_dma)
1784 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1785 blk_queue_segment_boundary(q, shost->dma_boundary);
1786 dma_set_seg_boundary(dev, shost->dma_boundary);
1788 blk_queue_max_segment_size(q, shost->max_segment_size);
1789 dma_set_max_seg_size(dev, shost->max_segment_size);
1792 * Set a reasonable default alignment: The larger of 32-byte (dword),
1793 * which is a common minimum for HBAs, and the minimum DMA alignment,
1794 * which is set by the platform.
1796 * Devices that require a bigger alignment can increase it later.
1798 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1800 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1802 static const struct blk_mq_ops scsi_mq_ops = {
1803 .get_budget = scsi_mq_get_budget,
1804 .put_budget = scsi_mq_put_budget,
1805 .queue_rq = scsi_queue_rq,
1806 .complete = scsi_softirq_done,
1807 .timeout = scsi_timeout,
1808 #ifdef CONFIG_BLK_DEBUG_FS
1809 .show_rq = scsi_show_rq,
1811 .init_request = scsi_mq_init_request,
1812 .exit_request = scsi_mq_exit_request,
1813 .initialize_rq_fn = scsi_initialize_rq,
1814 .busy = scsi_mq_lld_busy,
1815 .map_queues = scsi_map_queues,
1818 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1820 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1821 if (IS_ERR(sdev->request_queue))
1824 sdev->request_queue->queuedata = sdev;
1825 __scsi_init_queue(sdev->host, sdev->request_queue);
1826 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1827 return sdev->request_queue;
1830 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1832 unsigned int cmd_size, sgl_size;
1834 sgl_size = scsi_mq_inline_sgl_size(shost);
1835 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1836 if (scsi_host_get_prot(shost))
1837 cmd_size += sizeof(struct scsi_data_buffer) +
1838 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1840 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1841 shost->tag_set.ops = &scsi_mq_ops;
1842 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1843 shost->tag_set.queue_depth = shost->can_queue;
1844 shost->tag_set.cmd_size = cmd_size;
1845 shost->tag_set.numa_node = NUMA_NO_NODE;
1846 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1847 shost->tag_set.flags |=
1848 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1849 shost->tag_set.driver_data = shost;
1851 return blk_mq_alloc_tag_set(&shost->tag_set);
1854 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1856 blk_mq_free_tag_set(&shost->tag_set);
1860 * scsi_device_from_queue - return sdev associated with a request_queue
1861 * @q: The request queue to return the sdev from
1863 * Return the sdev associated with a request queue or NULL if the
1864 * request_queue does not reference a SCSI device.
1866 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1868 struct scsi_device *sdev = NULL;
1870 if (q->mq_ops == &scsi_mq_ops)
1871 sdev = q->queuedata;
1872 if (!sdev || !get_device(&sdev->sdev_gendev))
1877 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1880 * Function: scsi_block_requests()
1882 * Purpose: Utility function used by low-level drivers to prevent further
1883 * commands from being queued to the device.
1885 * Arguments: shost - Host in question
1889 * Lock status: No locks are assumed held.
1891 * Notes: There is no timer nor any other means by which the requests
1892 * get unblocked other than the low-level driver calling
1893 * scsi_unblock_requests().
1895 void scsi_block_requests(struct Scsi_Host *shost)
1897 shost->host_self_blocked = 1;
1899 EXPORT_SYMBOL(scsi_block_requests);
1902 * Function: scsi_unblock_requests()
1904 * Purpose: Utility function used by low-level drivers to allow further
1905 * commands from being queued to the device.
1907 * Arguments: shost - Host in question
1911 * Lock status: No locks are assumed held.
1913 * Notes: There is no timer nor any other means by which the requests
1914 * get unblocked other than the low-level driver calling
1915 * scsi_unblock_requests().
1917 * This is done as an API function so that changes to the
1918 * internals of the scsi mid-layer won't require wholesale
1919 * changes to drivers that use this feature.
1921 void scsi_unblock_requests(struct Scsi_Host *shost)
1923 shost->host_self_blocked = 0;
1924 scsi_run_host_queues(shost);
1926 EXPORT_SYMBOL(scsi_unblock_requests);
1928 int __init scsi_init_queue(void)
1930 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1931 sizeof(struct scsi_data_buffer),
1933 if (!scsi_sdb_cache) {
1934 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1941 void scsi_exit_queue(void)
1943 kmem_cache_destroy(scsi_sense_cache);
1944 kmem_cache_destroy(scsi_sense_isadma_cache);
1945 kmem_cache_destroy(scsi_sdb_cache);
1949 * scsi_mode_select - issue a mode select
1950 * @sdev: SCSI device to be queried
1951 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1952 * @sp: Save page bit (0 == don't save, 1 == save)
1953 * @modepage: mode page being requested
1954 * @buffer: request buffer (may not be smaller than eight bytes)
1955 * @len: length of request buffer.
1956 * @timeout: command timeout
1957 * @retries: number of retries before failing
1958 * @data: returns a structure abstracting the mode header data
1959 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1960 * must be SCSI_SENSE_BUFFERSIZE big.
1962 * Returns zero if successful; negative error number or scsi
1967 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1968 unsigned char *buffer, int len, int timeout, int retries,
1969 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1971 unsigned char cmd[10];
1972 unsigned char *real_buffer;
1975 memset(cmd, 0, sizeof(cmd));
1976 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1978 if (sdev->use_10_for_ms) {
1981 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1984 memcpy(real_buffer + 8, buffer, len);
1988 real_buffer[2] = data->medium_type;
1989 real_buffer[3] = data->device_specific;
1990 real_buffer[4] = data->longlba ? 0x01 : 0;
1992 real_buffer[6] = data->block_descriptor_length >> 8;
1993 real_buffer[7] = data->block_descriptor_length;
1995 cmd[0] = MODE_SELECT_10;
1999 if (len > 255 || data->block_descriptor_length > 255 ||
2003 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2006 memcpy(real_buffer + 4, buffer, len);
2009 real_buffer[1] = data->medium_type;
2010 real_buffer[2] = data->device_specific;
2011 real_buffer[3] = data->block_descriptor_length;
2014 cmd[0] = MODE_SELECT;
2018 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2019 sshdr, timeout, retries, NULL);
2023 EXPORT_SYMBOL_GPL(scsi_mode_select);
2026 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2027 * @sdev: SCSI device to be queried
2028 * @dbd: set if mode sense will allow block descriptors to be returned
2029 * @modepage: mode page being requested
2030 * @buffer: request buffer (may not be smaller than eight bytes)
2031 * @len: length of request buffer.
2032 * @timeout: command timeout
2033 * @retries: number of retries before failing
2034 * @data: returns a structure abstracting the mode header data
2035 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2036 * must be SCSI_SENSE_BUFFERSIZE big.
2038 * Returns zero if unsuccessful, or the header offset (either 4
2039 * or 8 depending on whether a six or ten byte command was
2040 * issued) if successful.
2043 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2044 unsigned char *buffer, int len, int timeout, int retries,
2045 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2047 unsigned char cmd[12];
2050 int result, retry_count = retries;
2051 struct scsi_sense_hdr my_sshdr;
2053 memset(data, 0, sizeof(*data));
2054 memset(&cmd[0], 0, 12);
2055 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2058 /* caller might not be interested in sense, but we need it */
2063 use_10_for_ms = sdev->use_10_for_ms;
2065 if (use_10_for_ms) {
2069 cmd[0] = MODE_SENSE_10;
2076 cmd[0] = MODE_SENSE;
2081 memset(buffer, 0, len);
2083 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2084 sshdr, timeout, retries, NULL);
2086 /* This code looks awful: what it's doing is making sure an
2087 * ILLEGAL REQUEST sense return identifies the actual command
2088 * byte as the problem. MODE_SENSE commands can return
2089 * ILLEGAL REQUEST if the code page isn't supported */
2091 if (use_10_for_ms && !scsi_status_is_good(result) &&
2092 driver_byte(result) == DRIVER_SENSE) {
2093 if (scsi_sense_valid(sshdr)) {
2094 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2095 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2097 * Invalid command operation code
2099 sdev->use_10_for_ms = 0;
2105 if(scsi_status_is_good(result)) {
2106 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2107 (modepage == 6 || modepage == 8))) {
2108 /* Initio breakage? */
2111 data->medium_type = 0;
2112 data->device_specific = 0;
2114 data->block_descriptor_length = 0;
2115 } else if(use_10_for_ms) {
2116 data->length = buffer[0]*256 + buffer[1] + 2;
2117 data->medium_type = buffer[2];
2118 data->device_specific = buffer[3];
2119 data->longlba = buffer[4] & 0x01;
2120 data->block_descriptor_length = buffer[6]*256
2123 data->length = buffer[0] + 1;
2124 data->medium_type = buffer[1];
2125 data->device_specific = buffer[2];
2126 data->block_descriptor_length = buffer[3];
2128 data->header_length = header_length;
2129 } else if ((status_byte(result) == CHECK_CONDITION) &&
2130 scsi_sense_valid(sshdr) &&
2131 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2138 EXPORT_SYMBOL(scsi_mode_sense);
2141 * scsi_test_unit_ready - test if unit is ready
2142 * @sdev: scsi device to change the state of.
2143 * @timeout: command timeout
2144 * @retries: number of retries before failing
2145 * @sshdr: outpout pointer for decoded sense information.
2147 * Returns zero if unsuccessful or an error if TUR failed. For
2148 * removable media, UNIT_ATTENTION sets ->changed flag.
2151 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2152 struct scsi_sense_hdr *sshdr)
2155 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2159 /* try to eat the UNIT_ATTENTION if there are enough retries */
2161 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2163 if (sdev->removable && scsi_sense_valid(sshdr) &&
2164 sshdr->sense_key == UNIT_ATTENTION)
2166 } while (scsi_sense_valid(sshdr) &&
2167 sshdr->sense_key == UNIT_ATTENTION && --retries);
2171 EXPORT_SYMBOL(scsi_test_unit_ready);
2174 * scsi_device_set_state - Take the given device through the device state model.
2175 * @sdev: scsi device to change the state of.
2176 * @state: state to change to.
2178 * Returns zero if successful or an error if the requested
2179 * transition is illegal.
2182 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2184 enum scsi_device_state oldstate = sdev->sdev_state;
2186 if (state == oldstate)
2192 case SDEV_CREATED_BLOCK:
2203 case SDEV_TRANSPORT_OFFLINE:
2216 case SDEV_TRANSPORT_OFFLINE:
2224 case SDEV_TRANSPORT_OFFLINE:
2239 case SDEV_CREATED_BLOCK:
2247 case SDEV_CREATED_BLOCK:
2262 case SDEV_TRANSPORT_OFFLINE:
2274 case SDEV_TRANSPORT_OFFLINE:
2277 case SDEV_CREATED_BLOCK:
2285 sdev->sdev_state = state;
2289 SCSI_LOG_ERROR_RECOVERY(1,
2290 sdev_printk(KERN_ERR, sdev,
2291 "Illegal state transition %s->%s",
2292 scsi_device_state_name(oldstate),
2293 scsi_device_state_name(state))
2297 EXPORT_SYMBOL(scsi_device_set_state);
2300 * sdev_evt_emit - emit a single SCSI device uevent
2301 * @sdev: associated SCSI device
2302 * @evt: event to emit
2304 * Send a single uevent (scsi_event) to the associated scsi_device.
2306 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2311 switch (evt->evt_type) {
2312 case SDEV_EVT_MEDIA_CHANGE:
2313 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2315 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2316 scsi_rescan_device(&sdev->sdev_gendev);
2317 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2319 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2320 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2322 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2323 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2325 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2326 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2328 case SDEV_EVT_LUN_CHANGE_REPORTED:
2329 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2331 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2332 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2334 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2335 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2344 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2348 * sdev_evt_thread - send a uevent for each scsi event
2349 * @work: work struct for scsi_device
2351 * Dispatch queued events to their associated scsi_device kobjects
2354 void scsi_evt_thread(struct work_struct *work)
2356 struct scsi_device *sdev;
2357 enum scsi_device_event evt_type;
2358 LIST_HEAD(event_list);
2360 sdev = container_of(work, struct scsi_device, event_work);
2362 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2363 if (test_and_clear_bit(evt_type, sdev->pending_events))
2364 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2367 struct scsi_event *evt;
2368 struct list_head *this, *tmp;
2369 unsigned long flags;
2371 spin_lock_irqsave(&sdev->list_lock, flags);
2372 list_splice_init(&sdev->event_list, &event_list);
2373 spin_unlock_irqrestore(&sdev->list_lock, flags);
2375 if (list_empty(&event_list))
2378 list_for_each_safe(this, tmp, &event_list) {
2379 evt = list_entry(this, struct scsi_event, node);
2380 list_del(&evt->node);
2381 scsi_evt_emit(sdev, evt);
2388 * sdev_evt_send - send asserted event to uevent thread
2389 * @sdev: scsi_device event occurred on
2390 * @evt: event to send
2392 * Assert scsi device event asynchronously.
2394 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2396 unsigned long flags;
2399 /* FIXME: currently this check eliminates all media change events
2400 * for polled devices. Need to update to discriminate between AN
2401 * and polled events */
2402 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2408 spin_lock_irqsave(&sdev->list_lock, flags);
2409 list_add_tail(&evt->node, &sdev->event_list);
2410 schedule_work(&sdev->event_work);
2411 spin_unlock_irqrestore(&sdev->list_lock, flags);
2413 EXPORT_SYMBOL_GPL(sdev_evt_send);
2416 * sdev_evt_alloc - allocate a new scsi event
2417 * @evt_type: type of event to allocate
2418 * @gfpflags: GFP flags for allocation
2420 * Allocates and returns a new scsi_event.
2422 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2425 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2429 evt->evt_type = evt_type;
2430 INIT_LIST_HEAD(&evt->node);
2432 /* evt_type-specific initialization, if any */
2434 case SDEV_EVT_MEDIA_CHANGE:
2435 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2436 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2437 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2438 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2439 case SDEV_EVT_LUN_CHANGE_REPORTED:
2440 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2441 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2449 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2452 * sdev_evt_send_simple - send asserted event to uevent thread
2453 * @sdev: scsi_device event occurred on
2454 * @evt_type: type of event to send
2455 * @gfpflags: GFP flags for allocation
2457 * Assert scsi device event asynchronously, given an event type.
2459 void sdev_evt_send_simple(struct scsi_device *sdev,
2460 enum scsi_device_event evt_type, gfp_t gfpflags)
2462 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2464 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2469 sdev_evt_send(sdev, evt);
2471 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2474 * scsi_device_quiesce - Block user issued commands.
2475 * @sdev: scsi device to quiesce.
2477 * This works by trying to transition to the SDEV_QUIESCE state
2478 * (which must be a legal transition). When the device is in this
2479 * state, only special requests will be accepted, all others will
2480 * be deferred. Since special requests may also be requeued requests,
2481 * a successful return doesn't guarantee the device will be
2482 * totally quiescent.
2484 * Must be called with user context, may sleep.
2486 * Returns zero if unsuccessful or an error if not.
2489 scsi_device_quiesce(struct scsi_device *sdev)
2491 struct request_queue *q = sdev->request_queue;
2495 * It is allowed to call scsi_device_quiesce() multiple times from
2496 * the same context but concurrent scsi_device_quiesce() calls are
2499 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2501 if (sdev->quiesced_by == current)
2506 blk_mq_freeze_queue(q);
2508 * Ensure that the effect of blk_set_pm_only() will be visible
2509 * for percpu_ref_tryget() callers that occur after the queue
2510 * unfreeze even if the queue was already frozen before this function
2511 * was called. See also https://lwn.net/Articles/573497/.
2514 blk_mq_unfreeze_queue(q);
2516 mutex_lock(&sdev->state_mutex);
2517 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2519 sdev->quiesced_by = current;
2521 blk_clear_pm_only(q);
2522 mutex_unlock(&sdev->state_mutex);
2526 EXPORT_SYMBOL(scsi_device_quiesce);
2529 * scsi_device_resume - Restart user issued commands to a quiesced device.
2530 * @sdev: scsi device to resume.
2532 * Moves the device from quiesced back to running and restarts the
2535 * Must be called with user context, may sleep.
2537 void scsi_device_resume(struct scsi_device *sdev)
2539 /* check if the device state was mutated prior to resume, and if
2540 * so assume the state is being managed elsewhere (for example
2541 * device deleted during suspend)
2543 mutex_lock(&sdev->state_mutex);
2544 if (sdev->quiesced_by) {
2545 sdev->quiesced_by = NULL;
2546 blk_clear_pm_only(sdev->request_queue);
2548 if (sdev->sdev_state == SDEV_QUIESCE)
2549 scsi_device_set_state(sdev, SDEV_RUNNING);
2550 mutex_unlock(&sdev->state_mutex);
2552 EXPORT_SYMBOL(scsi_device_resume);
2555 device_quiesce_fn(struct scsi_device *sdev, void *data)
2557 scsi_device_quiesce(sdev);
2561 scsi_target_quiesce(struct scsi_target *starget)
2563 starget_for_each_device(starget, NULL, device_quiesce_fn);
2565 EXPORT_SYMBOL(scsi_target_quiesce);
2568 device_resume_fn(struct scsi_device *sdev, void *data)
2570 scsi_device_resume(sdev);
2574 scsi_target_resume(struct scsi_target *starget)
2576 starget_for_each_device(starget, NULL, device_resume_fn);
2578 EXPORT_SYMBOL(scsi_target_resume);
2581 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2582 * @sdev: device to block
2584 * Pause SCSI command processing on the specified device. Does not sleep.
2586 * Returns zero if successful or a negative error code upon failure.
2589 * This routine transitions the device to the SDEV_BLOCK state (which must be
2590 * a legal transition). When the device is in this state, command processing
2591 * is paused until the device leaves the SDEV_BLOCK state. See also
2592 * scsi_internal_device_unblock_nowait().
2594 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2596 struct request_queue *q = sdev->request_queue;
2599 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2601 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2608 * The device has transitioned to SDEV_BLOCK. Stop the
2609 * block layer from calling the midlayer with this device's
2612 blk_mq_quiesce_queue_nowait(q);
2615 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2618 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2619 * @sdev: device to block
2621 * Pause SCSI command processing on the specified device and wait until all
2622 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2624 * Returns zero if successful or a negative error code upon failure.
2627 * This routine transitions the device to the SDEV_BLOCK state (which must be
2628 * a legal transition). When the device is in this state, command processing
2629 * is paused until the device leaves the SDEV_BLOCK state. See also
2630 * scsi_internal_device_unblock().
2632 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2633 * scsi_internal_device_block() has blocked a SCSI device and also
2634 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2636 static int scsi_internal_device_block(struct scsi_device *sdev)
2638 struct request_queue *q = sdev->request_queue;
2641 mutex_lock(&sdev->state_mutex);
2642 err = scsi_internal_device_block_nowait(sdev);
2644 blk_mq_quiesce_queue(q);
2645 mutex_unlock(&sdev->state_mutex);
2650 void scsi_start_queue(struct scsi_device *sdev)
2652 struct request_queue *q = sdev->request_queue;
2654 blk_mq_unquiesce_queue(q);
2658 * scsi_internal_device_unblock_nowait - resume a device after a block request
2659 * @sdev: device to resume
2660 * @new_state: state to set the device to after unblocking
2662 * Restart the device queue for a previously suspended SCSI device. Does not
2665 * Returns zero if successful or a negative error code upon failure.
2668 * This routine transitions the device to the SDEV_RUNNING state or to one of
2669 * the offline states (which must be a legal transition) allowing the midlayer
2670 * to goose the queue for this device.
2672 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2673 enum scsi_device_state new_state)
2676 * Try to transition the scsi device to SDEV_RUNNING or one of the
2677 * offlined states and goose the device queue if successful.
2679 switch (sdev->sdev_state) {
2681 case SDEV_TRANSPORT_OFFLINE:
2682 sdev->sdev_state = new_state;
2684 case SDEV_CREATED_BLOCK:
2685 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2686 new_state == SDEV_OFFLINE)
2687 sdev->sdev_state = new_state;
2689 sdev->sdev_state = SDEV_CREATED;
2697 scsi_start_queue(sdev);
2701 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2704 * scsi_internal_device_unblock - resume a device after a block request
2705 * @sdev: device to resume
2706 * @new_state: state to set the device to after unblocking
2708 * Restart the device queue for a previously suspended SCSI device. May sleep.
2710 * Returns zero if successful or a negative error code upon failure.
2713 * This routine transitions the device to the SDEV_RUNNING state or to one of
2714 * the offline states (which must be a legal transition) allowing the midlayer
2715 * to goose the queue for this device.
2717 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2718 enum scsi_device_state new_state)
2722 mutex_lock(&sdev->state_mutex);
2723 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2724 mutex_unlock(&sdev->state_mutex);
2730 device_block(struct scsi_device *sdev, void *data)
2732 scsi_internal_device_block(sdev);
2736 target_block(struct device *dev, void *data)
2738 if (scsi_is_target_device(dev))
2739 starget_for_each_device(to_scsi_target(dev), NULL,
2745 scsi_target_block(struct device *dev)
2747 if (scsi_is_target_device(dev))
2748 starget_for_each_device(to_scsi_target(dev), NULL,
2751 device_for_each_child(dev, NULL, target_block);
2753 EXPORT_SYMBOL_GPL(scsi_target_block);
2756 device_unblock(struct scsi_device *sdev, void *data)
2758 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2762 target_unblock(struct device *dev, void *data)
2764 if (scsi_is_target_device(dev))
2765 starget_for_each_device(to_scsi_target(dev), data,
2771 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2773 if (scsi_is_target_device(dev))
2774 starget_for_each_device(to_scsi_target(dev), &new_state,
2777 device_for_each_child(dev, &new_state, target_unblock);
2779 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2782 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2783 * @sgl: scatter-gather list
2784 * @sg_count: number of segments in sg
2785 * @offset: offset in bytes into sg, on return offset into the mapped area
2786 * @len: bytes to map, on return number of bytes mapped
2788 * Returns virtual address of the start of the mapped page
2790 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2791 size_t *offset, size_t *len)
2794 size_t sg_len = 0, len_complete = 0;
2795 struct scatterlist *sg;
2798 WARN_ON(!irqs_disabled());
2800 for_each_sg(sgl, sg, sg_count, i) {
2801 len_complete = sg_len; /* Complete sg-entries */
2802 sg_len += sg->length;
2803 if (sg_len > *offset)
2807 if (unlikely(i == sg_count)) {
2808 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2810 __func__, sg_len, *offset, sg_count);
2815 /* Offset starting from the beginning of first page in this sg-entry */
2816 *offset = *offset - len_complete + sg->offset;
2818 /* Assumption: contiguous pages can be accessed as "page + i" */
2819 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2820 *offset &= ~PAGE_MASK;
2822 /* Bytes in this sg-entry from *offset to the end of the page */
2823 sg_len = PAGE_SIZE - *offset;
2827 return kmap_atomic(page);
2829 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2832 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2833 * @virt: virtual address to be unmapped
2835 void scsi_kunmap_atomic_sg(void *virt)
2837 kunmap_atomic(virt);
2839 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2841 void sdev_disable_disk_events(struct scsi_device *sdev)
2843 atomic_inc(&sdev->disk_events_disable_depth);
2845 EXPORT_SYMBOL(sdev_disable_disk_events);
2847 void sdev_enable_disk_events(struct scsi_device *sdev)
2849 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2851 atomic_dec(&sdev->disk_events_disable_depth);
2853 EXPORT_SYMBOL(sdev_enable_disk_events);
2856 * scsi_vpd_lun_id - return a unique device identification
2857 * @sdev: SCSI device
2858 * @id: buffer for the identification
2859 * @id_len: length of the buffer
2861 * Copies a unique device identification into @id based
2862 * on the information in the VPD page 0x83 of the device.
2863 * The string will be formatted as a SCSI name string.
2865 * Returns the length of the identification or error on failure.
2866 * If the identifier is longer than the supplied buffer the actual
2867 * identifier length is returned and the buffer is not zero-padded.
2869 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2871 u8 cur_id_type = 0xff;
2873 const unsigned char *d, *cur_id_str;
2874 const struct scsi_vpd *vpd_pg83;
2875 int id_size = -EINVAL;
2878 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2885 * Look for the correct descriptor.
2886 * Order of preference for lun descriptor:
2887 * - SCSI name string
2888 * - NAA IEEE Registered Extended
2889 * - EUI-64 based 16-byte
2890 * - EUI-64 based 12-byte
2891 * - NAA IEEE Registered
2892 * - NAA IEEE Extended
2894 * as longer descriptors reduce the likelyhood
2895 * of identification clashes.
2898 /* The id string must be at least 20 bytes + terminating NULL byte */
2904 memset(id, 0, id_len);
2905 d = vpd_pg83->data + 4;
2906 while (d < vpd_pg83->data + vpd_pg83->len) {
2907 /* Skip designators not referring to the LUN */
2908 if ((d[1] & 0x30) != 0x00)
2911 switch (d[1] & 0xf) {
2914 if (cur_id_size > d[3])
2916 /* Prefer anything */
2917 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2920 if (cur_id_size + 4 > id_len)
2921 cur_id_size = id_len - 4;
2923 cur_id_type = d[1] & 0xf;
2924 id_size = snprintf(id, id_len, "t10.%*pE",
2925 cur_id_size, cur_id_str);
2929 if (cur_id_size > d[3])
2931 /* Prefer NAA IEEE Registered Extended */
2932 if (cur_id_type == 0x3 &&
2933 cur_id_size == d[3])
2937 cur_id_type = d[1] & 0xf;
2938 switch (cur_id_size) {
2940 id_size = snprintf(id, id_len,
2945 id_size = snprintf(id, id_len,
2950 id_size = snprintf(id, id_len,
2961 if (cur_id_size > d[3])
2965 cur_id_type = d[1] & 0xf;
2966 switch (cur_id_size) {
2968 id_size = snprintf(id, id_len,
2973 id_size = snprintf(id, id_len,
2983 /* SCSI name string */
2984 if (cur_id_size + 4 > d[3])
2986 /* Prefer others for truncated descriptor */
2987 if (cur_id_size && d[3] > id_len)
2989 cur_id_size = id_size = d[3];
2991 cur_id_type = d[1] & 0xf;
2992 if (cur_id_size >= id_len)
2993 cur_id_size = id_len - 1;
2994 memcpy(id, cur_id_str, cur_id_size);
2995 /* Decrease priority for truncated descriptor */
2996 if (cur_id_size != id_size)
3009 EXPORT_SYMBOL(scsi_vpd_lun_id);
3012 * scsi_vpd_tpg_id - return a target port group identifier
3013 * @sdev: SCSI device
3015 * Returns the Target Port Group identifier from the information
3016 * froom VPD page 0x83 of the device.
3018 * Returns the identifier or error on failure.
3020 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3022 const unsigned char *d;
3023 const struct scsi_vpd *vpd_pg83;
3024 int group_id = -EAGAIN, rel_port = -1;
3027 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3033 d = vpd_pg83->data + 4;
3034 while (d < vpd_pg83->data + vpd_pg83->len) {
3035 switch (d[1] & 0xf) {
3037 /* Relative target port */
3038 rel_port = get_unaligned_be16(&d[6]);
3041 /* Target port group */
3042 group_id = get_unaligned_be16(&d[6]);
3051 if (group_id >= 0 && rel_id && rel_port != -1)
3056 EXPORT_SYMBOL(scsi_vpd_tpg_id);