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 struct scsi_device *sdev = cmd->device;
146 if (cmd->request->rq_flags & RQF_DONTPREP) {
147 cmd->request->rq_flags &= ~RQF_DONTPREP;
148 scsi_mq_uninit_cmd(cmd);
152 blk_mq_requeue_request(cmd->request, true);
153 put_device(&sdev->sdev_gendev);
157 * __scsi_queue_insert - private queue insertion
158 * @cmd: The SCSI command being requeued
159 * @reason: The reason for the requeue
160 * @unbusy: Whether the queue should be unbusied
162 * This is a private queue insertion. The public interface
163 * scsi_queue_insert() always assumes the queue should be unbusied
164 * because it's always called before the completion. This function is
165 * for a requeue after completion, which should only occur in this
168 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
170 struct scsi_device *device = cmd->device;
172 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
173 "Inserting command %p into mlqueue\n", cmd));
175 scsi_set_blocked(cmd, reason);
178 * Decrement the counters, since these commands are no longer
179 * active on the host/device.
182 scsi_device_unbusy(device);
185 * Requeue this command. It will go before all other commands
186 * that are already in the queue. Schedule requeue work under
187 * lock such that the kblockd_schedule_work() call happens
188 * before blk_cleanup_queue() finishes.
193 * Before a SCSI command is dispatched,
194 * get_device(&sdev->sdev_gendev) is called and the host,
195 * target and device busy counters are increased. Since
196 * requeuing a request causes these actions to be repeated and
197 * since scsi_device_unbusy() has already been called,
198 * put_device(&device->sdev_gendev) must still be called. Call
199 * put_device() after blk_mq_requeue_request() to avoid that
200 * removal of the SCSI device can start before requeueing has
203 blk_mq_requeue_request(cmd->request, true);
204 put_device(&device->sdev_gendev);
208 * Function: scsi_queue_insert()
210 * Purpose: Insert a command in the midlevel queue.
212 * Arguments: cmd - command that we are adding to queue.
213 * reason - why we are inserting command to queue.
215 * Lock status: Assumed that lock is not held upon entry.
219 * Notes: We do this for one of two cases. Either the host is busy
220 * and it cannot accept any more commands for the time being,
221 * or the device returned QUEUE_FULL and can accept no more
223 * Notes: This could be called either from an interrupt context or a
224 * normal process context.
226 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
228 __scsi_queue_insert(cmd, reason, true);
233 * __scsi_execute - insert request and wait for the result
236 * @data_direction: data direction
237 * @buffer: data buffer
238 * @bufflen: len of buffer
239 * @sense: optional sense buffer
240 * @sshdr: optional decoded sense header
241 * @timeout: request timeout in seconds
242 * @retries: number of times to retry request
243 * @flags: flags for ->cmd_flags
244 * @rq_flags: flags for ->rq_flags
245 * @resid: optional residual length
247 * Returns the scsi_cmnd result field if a command was executed, or a negative
248 * Linux error code if we didn't get that far.
250 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
251 int data_direction, void *buffer, unsigned bufflen,
252 unsigned char *sense, struct scsi_sense_hdr *sshdr,
253 int timeout, int retries, u64 flags, req_flags_t rq_flags,
257 struct scsi_request *rq;
258 int ret = DRIVER_ERROR << 24;
260 req = blk_get_request(sdev->request_queue,
261 data_direction == DMA_TO_DEVICE ?
262 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
267 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
268 buffer, bufflen, GFP_NOIO))
271 rq->cmd_len = COMMAND_SIZE(cmd[0]);
272 memcpy(rq->cmd, cmd, rq->cmd_len);
273 rq->retries = retries;
274 req->timeout = timeout;
275 req->cmd_flags |= flags;
276 req->rq_flags |= rq_flags | RQF_QUIET;
279 * head injection *required* here otherwise quiesce won't work
281 blk_execute_rq(req->q, NULL, req, 1);
284 * Some devices (USB mass-storage in particular) may transfer
285 * garbage data together with a residue indicating that the data
286 * is invalid. Prevent the garbage from being misinterpreted
287 * and prevent security leaks by zeroing out the excess data.
289 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
290 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
293 *resid = rq->resid_len;
294 if (sense && rq->sense_len)
295 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
297 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
300 blk_put_request(req);
304 EXPORT_SYMBOL(__scsi_execute);
307 * Function: scsi_init_cmd_errh()
309 * Purpose: Initialize cmd fields related to error handling.
311 * Arguments: cmd - command that is ready to be queued.
313 * Notes: This function has the job of initializing a number of
314 * fields related to error handling. Typically this will
315 * be called once for each command, as required.
317 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
319 cmd->serial_number = 0;
320 scsi_set_resid(cmd, 0);
321 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
322 if (cmd->cmd_len == 0)
323 cmd->cmd_len = scsi_command_size(cmd->cmnd);
327 * Decrement the host_busy counter and wake up the error handler if necessary.
328 * Avoid as follows that the error handler is not woken up if shost->host_busy
329 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
330 * with an RCU read lock in this function to ensure that this function in its
331 * entirety either finishes before scsi_eh_scmd_add() increases the
332 * host_failed counter or that it notices the shost state change made by
333 * scsi_eh_scmd_add().
335 static void scsi_dec_host_busy(struct Scsi_Host *shost)
340 atomic_dec(&shost->host_busy);
341 if (unlikely(scsi_host_in_recovery(shost))) {
342 spin_lock_irqsave(shost->host_lock, flags);
343 if (shost->host_failed || shost->host_eh_scheduled)
344 scsi_eh_wakeup(shost);
345 spin_unlock_irqrestore(shost->host_lock, flags);
350 void scsi_device_unbusy(struct scsi_device *sdev)
352 struct Scsi_Host *shost = sdev->host;
353 struct scsi_target *starget = scsi_target(sdev);
355 scsi_dec_host_busy(shost);
357 if (starget->can_queue > 0)
358 atomic_dec(&starget->target_busy);
360 atomic_dec(&sdev->device_busy);
363 static void scsi_kick_queue(struct request_queue *q)
365 blk_mq_run_hw_queues(q, false);
369 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
370 * and call blk_run_queue for all the scsi_devices on the target -
371 * including current_sdev first.
373 * Called with *no* scsi locks held.
375 static void scsi_single_lun_run(struct scsi_device *current_sdev)
377 struct Scsi_Host *shost = current_sdev->host;
378 struct scsi_device *sdev, *tmp;
379 struct scsi_target *starget = scsi_target(current_sdev);
382 spin_lock_irqsave(shost->host_lock, flags);
383 starget->starget_sdev_user = NULL;
384 spin_unlock_irqrestore(shost->host_lock, flags);
387 * Call blk_run_queue for all LUNs on the target, starting with
388 * current_sdev. We race with others (to set starget_sdev_user),
389 * but in most cases, we will be first. Ideally, each LU on the
390 * target would get some limited time or requests on the target.
392 scsi_kick_queue(current_sdev->request_queue);
394 spin_lock_irqsave(shost->host_lock, flags);
395 if (starget->starget_sdev_user)
397 list_for_each_entry_safe(sdev, tmp, &starget->devices,
398 same_target_siblings) {
399 if (sdev == current_sdev)
401 if (scsi_device_get(sdev))
404 spin_unlock_irqrestore(shost->host_lock, flags);
405 scsi_kick_queue(sdev->request_queue);
406 spin_lock_irqsave(shost->host_lock, flags);
408 scsi_device_put(sdev);
411 spin_unlock_irqrestore(shost->host_lock, flags);
414 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
416 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
418 if (atomic_read(&sdev->device_blocked) > 0)
423 static inline bool scsi_target_is_busy(struct scsi_target *starget)
425 if (starget->can_queue > 0) {
426 if (atomic_read(&starget->target_busy) >= starget->can_queue)
428 if (atomic_read(&starget->target_blocked) > 0)
434 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
436 if (shost->can_queue > 0 &&
437 atomic_read(&shost->host_busy) >= shost->can_queue)
439 if (atomic_read(&shost->host_blocked) > 0)
441 if (shost->host_self_blocked)
446 static void scsi_starved_list_run(struct Scsi_Host *shost)
448 LIST_HEAD(starved_list);
449 struct scsi_device *sdev;
452 spin_lock_irqsave(shost->host_lock, flags);
453 list_splice_init(&shost->starved_list, &starved_list);
455 while (!list_empty(&starved_list)) {
456 struct request_queue *slq;
459 * As long as shost is accepting commands and we have
460 * starved queues, call blk_run_queue. scsi_request_fn
461 * drops the queue_lock and can add us back to the
464 * host_lock protects the starved_list and starved_entry.
465 * scsi_request_fn must get the host_lock before checking
466 * or modifying starved_list or starved_entry.
468 if (scsi_host_is_busy(shost))
471 sdev = list_entry(starved_list.next,
472 struct scsi_device, starved_entry);
473 list_del_init(&sdev->starved_entry);
474 if (scsi_target_is_busy(scsi_target(sdev))) {
475 list_move_tail(&sdev->starved_entry,
476 &shost->starved_list);
481 * Once we drop the host lock, a racing scsi_remove_device()
482 * call may remove the sdev from the starved list and destroy
483 * it and the queue. Mitigate by taking a reference to the
484 * queue and never touching the sdev again after we drop the
485 * host lock. Note: if __scsi_remove_device() invokes
486 * blk_cleanup_queue() before the queue is run from this
487 * function then blk_run_queue() will return immediately since
488 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
490 slq = sdev->request_queue;
491 if (!blk_get_queue(slq))
493 spin_unlock_irqrestore(shost->host_lock, flags);
495 scsi_kick_queue(slq);
498 spin_lock_irqsave(shost->host_lock, flags);
500 /* put any unprocessed entries back */
501 list_splice(&starved_list, &shost->starved_list);
502 spin_unlock_irqrestore(shost->host_lock, flags);
506 * Function: scsi_run_queue()
508 * Purpose: Select a proper request queue to serve next
510 * Arguments: q - last request's queue
514 * Notes: The previous command was completely finished, start
515 * a new one if possible.
517 static void scsi_run_queue(struct request_queue *q)
519 struct scsi_device *sdev = q->queuedata;
521 if (scsi_target(sdev)->single_lun)
522 scsi_single_lun_run(sdev);
523 if (!list_empty(&sdev->host->starved_list))
524 scsi_starved_list_run(sdev->host);
526 blk_mq_run_hw_queues(q, false);
529 void scsi_requeue_run_queue(struct work_struct *work)
531 struct scsi_device *sdev;
532 struct request_queue *q;
534 sdev = container_of(work, struct scsi_device, requeue_work);
535 q = sdev->request_queue;
539 void scsi_run_host_queues(struct Scsi_Host *shost)
541 struct scsi_device *sdev;
543 shost_for_each_device(sdev, shost)
544 scsi_run_queue(sdev->request_queue);
547 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
549 if (!blk_rq_is_passthrough(cmd->request)) {
550 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
552 if (drv->uninit_command)
553 drv->uninit_command(cmd);
557 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
559 struct scsi_data_buffer *sdb;
561 if (cmd->sdb.table.nents)
562 sg_free_table_chained(&cmd->sdb.table, true);
563 if (cmd->request->next_rq) {
564 sdb = cmd->request->next_rq->special;
566 sg_free_table_chained(&sdb->table, true);
568 if (scsi_prot_sg_count(cmd))
569 sg_free_table_chained(&cmd->prot_sdb->table, true);
572 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
574 scsi_mq_free_sgtables(cmd);
575 scsi_uninit_cmd(cmd);
576 scsi_del_cmd_from_list(cmd);
579 /* Returns false when no more bytes to process, true if there are more */
580 static bool scsi_end_request(struct request *req, blk_status_t error,
581 unsigned int bytes, unsigned int bidi_bytes)
583 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
584 struct scsi_device *sdev = cmd->device;
585 struct request_queue *q = sdev->request_queue;
587 if (blk_update_request(req, error, bytes))
590 /* Bidi request must be completed as a whole */
591 if (unlikely(bidi_bytes) &&
592 blk_update_request(req->next_rq, error, bidi_bytes))
595 if (blk_queue_add_random(q))
596 add_disk_randomness(req->rq_disk);
598 if (!blk_rq_is_scsi(req)) {
599 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
600 cmd->flags &= ~SCMD_INITIALIZED;
601 destroy_rcu_head(&cmd->rcu);
605 * In the MQ case the command gets freed by __blk_mq_end_request,
606 * so we have to do all cleanup that depends on it earlier.
608 * We also can't kick the queues from irq context, so we
609 * will have to defer it to a workqueue.
611 scsi_mq_uninit_cmd(cmd);
613 __blk_mq_end_request(req, error);
615 if (scsi_target(sdev)->single_lun ||
616 !list_empty(&sdev->host->starved_list))
617 kblockd_schedule_work(&sdev->requeue_work);
619 blk_mq_run_hw_queues(q, true);
621 put_device(&sdev->sdev_gendev);
626 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
628 * @result: scsi error code
630 * Translate a SCSI result code into a blk_status_t value. May reset the host
631 * byte of @cmd->result.
633 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
635 switch (host_byte(result)) {
638 * Also check the other bytes than the status byte in result
639 * to handle the case when a SCSI LLD sets result to
640 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
642 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
644 return BLK_STS_IOERR;
645 case DID_TRANSPORT_FAILFAST:
646 return BLK_STS_TRANSPORT;
647 case DID_TARGET_FAILURE:
648 set_host_byte(cmd, DID_OK);
649 return BLK_STS_TARGET;
650 case DID_NEXUS_FAILURE:
651 return BLK_STS_NEXUS;
652 case DID_ALLOC_FAILURE:
653 set_host_byte(cmd, DID_OK);
654 return BLK_STS_NOSPC;
655 case DID_MEDIUM_ERROR:
656 set_host_byte(cmd, DID_OK);
657 return BLK_STS_MEDIUM;
659 return BLK_STS_IOERR;
663 /* Helper for scsi_io_completion() when "reprep" action required. */
664 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
665 struct request_queue *q)
667 /* A new command will be prepared and issued. */
668 scsi_mq_requeue_cmd(cmd);
671 /* Helper for scsi_io_completion() when special action required. */
672 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
674 struct request_queue *q = cmd->device->request_queue;
675 struct request *req = cmd->request;
677 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
678 ACTION_DELAYED_RETRY} action;
679 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
680 struct scsi_sense_hdr sshdr;
682 bool sense_current = true; /* false implies "deferred sense" */
683 blk_status_t blk_stat;
685 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
687 sense_current = !scsi_sense_is_deferred(&sshdr);
689 blk_stat = scsi_result_to_blk_status(cmd, result);
691 if (host_byte(result) == DID_RESET) {
692 /* Third party bus reset or reset for error recovery
693 * reasons. Just retry the command and see what
696 action = ACTION_RETRY;
697 } else if (sense_valid && sense_current) {
698 switch (sshdr.sense_key) {
700 if (cmd->device->removable) {
701 /* Detected disc change. Set a bit
702 * and quietly refuse further access.
704 cmd->device->changed = 1;
705 action = ACTION_FAIL;
707 /* Must have been a power glitch, or a
708 * bus reset. Could not have been a
709 * media change, so we just retry the
710 * command and see what happens.
712 action = ACTION_RETRY;
715 case ILLEGAL_REQUEST:
716 /* If we had an ILLEGAL REQUEST returned, then
717 * we may have performed an unsupported
718 * command. The only thing this should be
719 * would be a ten byte read where only a six
720 * byte read was supported. Also, on a system
721 * where READ CAPACITY failed, we may have
722 * read past the end of the disk.
724 if ((cmd->device->use_10_for_rw &&
725 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
726 (cmd->cmnd[0] == READ_10 ||
727 cmd->cmnd[0] == WRITE_10)) {
728 /* This will issue a new 6-byte command. */
729 cmd->device->use_10_for_rw = 0;
730 action = ACTION_REPREP;
731 } else if (sshdr.asc == 0x10) /* DIX */ {
732 action = ACTION_FAIL;
733 blk_stat = BLK_STS_PROTECTION;
734 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
735 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
736 action = ACTION_FAIL;
737 blk_stat = BLK_STS_TARGET;
739 action = ACTION_FAIL;
741 case ABORTED_COMMAND:
742 action = ACTION_FAIL;
743 if (sshdr.asc == 0x10) /* DIF */
744 blk_stat = BLK_STS_PROTECTION;
747 /* If the device is in the process of becoming
748 * ready, or has a temporary blockage, retry.
750 if (sshdr.asc == 0x04) {
751 switch (sshdr.ascq) {
752 case 0x01: /* becoming ready */
753 case 0x04: /* format in progress */
754 case 0x05: /* rebuild in progress */
755 case 0x06: /* recalculation in progress */
756 case 0x07: /* operation in progress */
757 case 0x08: /* Long write in progress */
758 case 0x09: /* self test in progress */
759 case 0x14: /* space allocation in progress */
760 case 0x1a: /* start stop unit in progress */
761 case 0x1b: /* sanitize in progress */
762 case 0x1d: /* configuration in progress */
763 case 0x24: /* depopulation in progress */
764 action = ACTION_DELAYED_RETRY;
767 action = ACTION_FAIL;
771 action = ACTION_FAIL;
773 case VOLUME_OVERFLOW:
774 /* See SSC3rXX or current. */
775 action = ACTION_FAIL;
778 action = ACTION_FAIL;
782 action = ACTION_FAIL;
784 if (action != ACTION_FAIL &&
785 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
786 action = ACTION_FAIL;
790 /* Give up and fail the remainder of the request */
791 if (!(req->rq_flags & RQF_QUIET)) {
792 static DEFINE_RATELIMIT_STATE(_rs,
793 DEFAULT_RATELIMIT_INTERVAL,
794 DEFAULT_RATELIMIT_BURST);
796 if (unlikely(scsi_logging_level))
798 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
799 SCSI_LOG_MLCOMPLETE_BITS);
802 * if logging is enabled the failure will be printed
803 * in scsi_log_completion(), so avoid duplicate messages
805 if (!level && __ratelimit(&_rs)) {
806 scsi_print_result(cmd, NULL, FAILED);
807 if (driver_byte(result) == DRIVER_SENSE)
808 scsi_print_sense(cmd);
809 scsi_print_command(cmd);
812 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req), 0))
816 scsi_io_completion_reprep(cmd, q);
819 /* Retry the same command immediately */
820 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
822 case ACTION_DELAYED_RETRY:
823 /* Retry the same command after a delay */
824 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
830 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
831 * new result that may suppress further error checking. Also modifies
832 * *blk_statp in some cases.
834 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
835 blk_status_t *blk_statp)
838 bool sense_current = true; /* false implies "deferred sense" */
839 struct request *req = cmd->request;
840 struct scsi_sense_hdr sshdr;
842 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
844 sense_current = !scsi_sense_is_deferred(&sshdr);
846 if (blk_rq_is_passthrough(req)) {
849 * SG_IO wants current and deferred errors
851 scsi_req(req)->sense_len =
852 min(8 + cmd->sense_buffer[7],
853 SCSI_SENSE_BUFFERSIZE);
856 *blk_statp = scsi_result_to_blk_status(cmd, result);
857 } else if (blk_rq_bytes(req) == 0 && sense_current) {
859 * Flush commands do not transfers any data, and thus cannot use
860 * good_bytes != blk_rq_bytes(req) as the signal for an error.
861 * This sets *blk_statp explicitly for the problem case.
863 *blk_statp = scsi_result_to_blk_status(cmd, result);
866 * Recovered errors need reporting, but they're always treated as
867 * success, so fiddle the result code here. For passthrough requests
868 * we already took a copy of the original into sreq->result which
869 * is what gets returned to the user
871 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
872 bool do_print = true;
874 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
875 * skip print since caller wants ATA registers. Only occurs
876 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
878 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
880 else if (req->rq_flags & RQF_QUIET)
883 scsi_print_sense(cmd);
885 /* for passthrough, *blk_statp may be set */
886 *blk_statp = BLK_STS_OK;
889 * Another corner case: the SCSI status byte is non-zero but 'good'.
890 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
891 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
892 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
893 * intermediate statuses (both obsolete in SAM-4) as good.
895 if (status_byte(result) && scsi_status_is_good(result)) {
897 *blk_statp = BLK_STS_OK;
903 * Function: scsi_io_completion()
905 * Purpose: Completion processing for block device I/O requests.
907 * Arguments: cmd - command that is finished.
909 * Lock status: Assumed that no lock is held upon entry.
913 * Notes: We will finish off the specified number of sectors. If we
914 * are done, the command block will be released and the queue
915 * function will be goosed. If we are not done then we have to
916 * figure out what to do next:
918 * a) We can call scsi_requeue_command(). The request
919 * will be unprepared and put back on the queue. Then
920 * a new command will be created for it. This should
921 * be used if we made forward progress, or if we want
922 * to switch from READ(10) to READ(6) for example.
924 * b) We can call __scsi_queue_insert(). The request will
925 * be put back on the queue and retried using the same
926 * command as before, possibly after a delay.
928 * c) We can call scsi_end_request() with blk_stat other than
929 * BLK_STS_OK, to fail the remainder of the request.
931 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
933 int result = cmd->result;
934 struct request_queue *q = cmd->device->request_queue;
935 struct request *req = cmd->request;
936 blk_status_t blk_stat = BLK_STS_OK;
938 if (unlikely(result)) /* a nz result may or may not be an error */
939 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
941 if (unlikely(blk_rq_is_passthrough(req))) {
943 * scsi_result_to_blk_status may have reset the host_byte
945 scsi_req(req)->result = cmd->result;
946 scsi_req(req)->resid_len = scsi_get_resid(cmd);
948 if (unlikely(scsi_bidi_cmnd(cmd))) {
950 * Bidi commands Must be complete as a whole,
951 * both sides at once.
953 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
954 if (scsi_end_request(req, BLK_STS_OK, blk_rq_bytes(req),
955 blk_rq_bytes(req->next_rq)))
957 "Bidi command with remaining bytes");
962 /* no bidi support yet, other than in pass-through */
963 if (unlikely(blk_bidi_rq(req))) {
964 WARN_ONCE(true, "Only support bidi command in passthrough");
965 scmd_printk(KERN_ERR, cmd, "Killing bidi command\n");
966 if (scsi_end_request(req, BLK_STS_IOERR, blk_rq_bytes(req),
967 blk_rq_bytes(req->next_rq)))
968 WARN_ONCE(true, "Bidi command with remaining bytes");
973 * Next deal with any sectors which we were able to correctly
976 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
977 "%u sectors total, %d bytes done.\n",
978 blk_rq_sectors(req), good_bytes));
981 * Next deal with any sectors which we were able to correctly
982 * handle. Failed, zero length commands always need to drop down
983 * to retry code. Fast path should return in this block.
985 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
986 if (likely(!scsi_end_request(req, blk_stat, good_bytes, 0)))
987 return; /* no bytes remaining */
990 /* Kill remainder if no retries. */
991 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
992 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req), 0))
994 "Bytes remaining after failed, no-retry command");
999 * If there had been no error, but we have leftover bytes in the
1000 * requeues just queue the command up again.
1002 if (likely(result == 0))
1003 scsi_io_completion_reprep(cmd, q);
1005 scsi_io_completion_action(cmd, result);
1008 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1013 * If sg table allocation fails, requeue request later.
1015 if (unlikely(sg_alloc_table_chained(&sdb->table,
1016 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1017 return BLKPREP_DEFER;
1020 * Next, walk the list, and fill in the addresses and sizes of
1023 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1024 BUG_ON(count > sdb->table.nents);
1025 sdb->table.nents = count;
1026 sdb->length = blk_rq_payload_bytes(req);
1031 * Function: scsi_init_io()
1033 * Purpose: SCSI I/O initialize function.
1035 * Arguments: cmd - Command descriptor we wish to initialize
1037 * Returns: 0 on success
1038 * BLKPREP_DEFER if the failure is retryable
1039 * BLKPREP_KILL if the failure is fatal
1041 int scsi_init_io(struct scsi_cmnd *cmd)
1043 struct request *rq = cmd->request;
1044 int error = BLKPREP_KILL;
1046 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1049 error = scsi_init_sgtable(rq, &cmd->sdb);
1053 if (blk_bidi_rq(rq)) {
1054 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1059 if (blk_integrity_rq(rq)) {
1060 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1063 if (prot_sdb == NULL) {
1065 * This can happen if someone (e.g. multipath)
1066 * queues a command to a device on an adapter
1067 * that does not support DIX.
1070 error = BLKPREP_KILL;
1074 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1076 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1077 prot_sdb->table.sgl)) {
1078 error = BLKPREP_DEFER;
1082 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1083 prot_sdb->table.sgl);
1084 BUG_ON(count > ivecs);
1085 BUG_ON(count > queue_max_integrity_segments(rq->q));
1087 cmd->prot_sdb = prot_sdb;
1088 cmd->prot_sdb->table.nents = count;
1093 scsi_mq_free_sgtables(cmd);
1096 EXPORT_SYMBOL(scsi_init_io);
1099 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1100 * @rq: Request associated with the SCSI command to be initialized.
1102 * This function initializes the members of struct scsi_cmnd that must be
1103 * initialized before request processing starts and that won't be
1104 * reinitialized if a SCSI command is requeued.
1106 * Called from inside blk_get_request() for pass-through requests and from
1107 * inside scsi_init_command() for filesystem requests.
1109 static void scsi_initialize_rq(struct request *rq)
1111 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1113 scsi_req_init(&cmd->req);
1114 init_rcu_head(&cmd->rcu);
1115 cmd->jiffies_at_alloc = jiffies;
1119 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1120 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1122 struct scsi_device *sdev = cmd->device;
1123 struct Scsi_Host *shost = sdev->host;
1124 unsigned long flags;
1126 if (shost->use_cmd_list) {
1127 spin_lock_irqsave(&sdev->list_lock, flags);
1128 list_add_tail(&cmd->list, &sdev->cmd_list);
1129 spin_unlock_irqrestore(&sdev->list_lock, flags);
1133 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1134 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1136 struct scsi_device *sdev = cmd->device;
1137 struct Scsi_Host *shost = sdev->host;
1138 unsigned long flags;
1140 if (shost->use_cmd_list) {
1141 spin_lock_irqsave(&sdev->list_lock, flags);
1142 BUG_ON(list_empty(&cmd->list));
1143 list_del_init(&cmd->list);
1144 spin_unlock_irqrestore(&sdev->list_lock, flags);
1148 /* Called after a request has been started. */
1149 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1151 void *buf = cmd->sense_buffer;
1152 void *prot = cmd->prot_sdb;
1153 struct request *rq = blk_mq_rq_from_pdu(cmd);
1154 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1155 unsigned long jiffies_at_alloc;
1158 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1159 flags |= SCMD_INITIALIZED;
1160 scsi_initialize_rq(rq);
1163 jiffies_at_alloc = cmd->jiffies_at_alloc;
1164 retries = cmd->retries;
1165 /* zero out the cmd, except for the embedded scsi_request */
1166 memset((char *)cmd + sizeof(cmd->req), 0,
1167 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1170 cmd->sense_buffer = buf;
1171 cmd->prot_sdb = prot;
1173 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1174 cmd->jiffies_at_alloc = jiffies_at_alloc;
1175 cmd->retries = retries;
1177 scsi_add_cmd_to_list(cmd);
1180 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1182 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1185 * Passthrough requests may transfer data, in which case they must
1186 * a bio attached to them. Or they might contain a SCSI command
1187 * that does not transfer data, in which case they may optionally
1188 * submit a request without an attached bio.
1191 int ret = scsi_init_io(cmd);
1195 BUG_ON(blk_rq_bytes(req));
1197 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1200 cmd->cmd_len = scsi_req(req)->cmd_len;
1201 cmd->cmnd = scsi_req(req)->cmd;
1202 cmd->transfersize = blk_rq_bytes(req);
1203 cmd->allowed = scsi_req(req)->retries;
1208 * Setup a normal block command. These are simple request from filesystems
1209 * that still need to be translated to SCSI CDBs from the ULD.
1211 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1213 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1215 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1216 int ret = sdev->handler->prep_fn(sdev, req);
1217 if (ret != BLKPREP_OK)
1221 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1222 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1223 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1226 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1228 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1230 if (!blk_rq_bytes(req))
1231 cmd->sc_data_direction = DMA_NONE;
1232 else if (rq_data_dir(req) == WRITE)
1233 cmd->sc_data_direction = DMA_TO_DEVICE;
1235 cmd->sc_data_direction = DMA_FROM_DEVICE;
1237 if (blk_rq_is_scsi(req))
1238 return scsi_setup_scsi_cmnd(sdev, req);
1240 return scsi_setup_fs_cmnd(sdev, req);
1244 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1246 switch (sdev->sdev_state) {
1248 case SDEV_TRANSPORT_OFFLINE:
1250 * If the device is offline we refuse to process any
1251 * commands. The device must be brought online
1252 * before trying any recovery commands.
1254 sdev_printk(KERN_ERR, sdev,
1255 "rejecting I/O to offline device\n");
1256 return BLK_STS_IOERR;
1259 * If the device is fully deleted, we refuse to
1260 * process any commands as well.
1262 sdev_printk(KERN_ERR, sdev,
1263 "rejecting I/O to dead device\n");
1264 return BLK_STS_IOERR;
1266 case SDEV_CREATED_BLOCK:
1267 return BLK_STS_RESOURCE;
1270 * If the devices is blocked we defer normal commands.
1272 if (req && !(req->rq_flags & RQF_PREEMPT))
1273 return BLK_STS_RESOURCE;
1277 * For any other not fully online state we only allow
1278 * special commands. In particular any user initiated
1279 * command is not allowed.
1281 if (req && !(req->rq_flags & RQF_PREEMPT))
1282 return BLK_STS_IOERR;
1288 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1291 * Called with the queue_lock held.
1293 static inline int scsi_dev_queue_ready(struct request_queue *q,
1294 struct scsi_device *sdev)
1298 busy = atomic_inc_return(&sdev->device_busy) - 1;
1299 if (atomic_read(&sdev->device_blocked)) {
1304 * unblock after device_blocked iterates to zero
1306 if (atomic_dec_return(&sdev->device_blocked) > 0)
1308 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1309 "unblocking device at zero depth\n"));
1312 if (busy >= sdev->queue_depth)
1317 atomic_dec(&sdev->device_busy);
1322 * scsi_target_queue_ready: checks if there we can send commands to target
1323 * @sdev: scsi device on starget to check.
1325 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1326 struct scsi_device *sdev)
1328 struct scsi_target *starget = scsi_target(sdev);
1331 if (starget->single_lun) {
1332 spin_lock_irq(shost->host_lock);
1333 if (starget->starget_sdev_user &&
1334 starget->starget_sdev_user != sdev) {
1335 spin_unlock_irq(shost->host_lock);
1338 starget->starget_sdev_user = sdev;
1339 spin_unlock_irq(shost->host_lock);
1342 if (starget->can_queue <= 0)
1345 busy = atomic_inc_return(&starget->target_busy) - 1;
1346 if (atomic_read(&starget->target_blocked) > 0) {
1351 * unblock after target_blocked iterates to zero
1353 if (atomic_dec_return(&starget->target_blocked) > 0)
1356 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1357 "unblocking target at zero depth\n"));
1360 if (busy >= starget->can_queue)
1366 spin_lock_irq(shost->host_lock);
1367 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1368 spin_unlock_irq(shost->host_lock);
1370 if (starget->can_queue > 0)
1371 atomic_dec(&starget->target_busy);
1376 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1377 * return 0. We must end up running the queue again whenever 0 is
1378 * returned, else IO can hang.
1380 static inline int scsi_host_queue_ready(struct request_queue *q,
1381 struct Scsi_Host *shost,
1382 struct scsi_device *sdev)
1386 if (scsi_host_in_recovery(shost))
1389 busy = atomic_inc_return(&shost->host_busy) - 1;
1390 if (atomic_read(&shost->host_blocked) > 0) {
1395 * unblock after host_blocked iterates to zero
1397 if (atomic_dec_return(&shost->host_blocked) > 0)
1401 shost_printk(KERN_INFO, shost,
1402 "unblocking host at zero depth\n"));
1405 if (shost->can_queue > 0 && busy >= shost->can_queue)
1407 if (shost->host_self_blocked)
1410 /* We're OK to process the command, so we can't be starved */
1411 if (!list_empty(&sdev->starved_entry)) {
1412 spin_lock_irq(shost->host_lock);
1413 if (!list_empty(&sdev->starved_entry))
1414 list_del_init(&sdev->starved_entry);
1415 spin_unlock_irq(shost->host_lock);
1421 spin_lock_irq(shost->host_lock);
1422 if (list_empty(&sdev->starved_entry))
1423 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1424 spin_unlock_irq(shost->host_lock);
1426 scsi_dec_host_busy(shost);
1431 * Busy state exporting function for request stacking drivers.
1433 * For efficiency, no lock is taken to check the busy state of
1434 * shost/starget/sdev, since the returned value is not guaranteed and
1435 * may be changed after request stacking drivers call the function,
1436 * regardless of taking lock or not.
1438 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1439 * needs to return 'not busy'. Otherwise, request stacking drivers
1440 * may hold requests forever.
1442 static bool scsi_mq_lld_busy(struct request_queue *q)
1444 struct scsi_device *sdev = q->queuedata;
1445 struct Scsi_Host *shost;
1447 if (blk_queue_dying(q))
1453 * Ignore host/starget busy state.
1454 * Since block layer does not have a concept of fairness across
1455 * multiple queues, congestion of host/starget needs to be handled
1458 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1464 static void scsi_softirq_done(struct request *rq)
1466 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1467 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1470 INIT_LIST_HEAD(&cmd->eh_entry);
1472 atomic_inc(&cmd->device->iodone_cnt);
1474 atomic_inc(&cmd->device->ioerr_cnt);
1476 disposition = scsi_decide_disposition(cmd);
1477 if (disposition != SUCCESS &&
1478 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1479 sdev_printk(KERN_ERR, cmd->device,
1480 "timing out command, waited %lus\n",
1482 disposition = SUCCESS;
1485 scsi_log_completion(cmd, disposition);
1487 switch (disposition) {
1489 scsi_finish_command(cmd);
1492 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1494 case ADD_TO_MLQUEUE:
1495 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1498 scsi_eh_scmd_add(cmd);
1504 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1505 * @cmd: command block we are dispatching.
1507 * Return: nonzero return request was rejected and device's queue needs to be
1510 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1512 struct Scsi_Host *host = cmd->device->host;
1515 atomic_inc(&cmd->device->iorequest_cnt);
1517 /* check if the device is still usable */
1518 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1519 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1520 * returns an immediate error upwards, and signals
1521 * that the device is no longer present */
1522 cmd->result = DID_NO_CONNECT << 16;
1526 /* Check to see if the scsi lld made this device blocked. */
1527 if (unlikely(scsi_device_blocked(cmd->device))) {
1529 * in blocked state, the command is just put back on
1530 * the device queue. The suspend state has already
1531 * blocked the queue so future requests should not
1532 * occur until the device transitions out of the
1535 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1536 "queuecommand : device blocked\n"));
1537 return SCSI_MLQUEUE_DEVICE_BUSY;
1540 /* Store the LUN value in cmnd, if needed. */
1541 if (cmd->device->lun_in_cdb)
1542 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1543 (cmd->device->lun << 5 & 0xe0);
1548 * Before we queue this command, check if the command
1549 * length exceeds what the host adapter can handle.
1551 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1552 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1553 "queuecommand : command too long. "
1554 "cdb_size=%d host->max_cmd_len=%d\n",
1555 cmd->cmd_len, cmd->device->host->max_cmd_len));
1556 cmd->result = (DID_ABORT << 16);
1560 if (unlikely(host->shost_state == SHOST_DEL)) {
1561 cmd->result = (DID_NO_CONNECT << 16);
1566 trace_scsi_dispatch_cmd_start(cmd);
1567 rtn = host->hostt->queuecommand(host, cmd);
1569 trace_scsi_dispatch_cmd_error(cmd, rtn);
1570 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1571 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1572 rtn = SCSI_MLQUEUE_HOST_BUSY;
1574 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1575 "queuecommand : request rejected\n"));
1580 cmd->scsi_done(cmd);
1584 static inline blk_status_t prep_to_mq(int ret)
1590 return BLK_STS_RESOURCE;
1592 return BLK_STS_IOERR;
1596 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1597 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1599 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1600 sizeof(struct scatterlist);
1603 static int scsi_mq_prep_fn(struct request *req)
1605 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1606 struct scsi_device *sdev = req->q->queuedata;
1607 struct Scsi_Host *shost = sdev->host;
1608 struct scatterlist *sg;
1610 scsi_init_command(sdev, cmd);
1616 cmd->tag = req->tag;
1617 cmd->prot_op = SCSI_PROT_NORMAL;
1619 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1620 cmd->sdb.table.sgl = sg;
1622 if (scsi_host_get_prot(shost)) {
1623 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1625 cmd->prot_sdb->table.sgl =
1626 (struct scatterlist *)(cmd->prot_sdb + 1);
1629 if (blk_bidi_rq(req)) {
1630 struct request *next_rq = req->next_rq;
1631 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1633 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1634 bidi_sdb->table.sgl =
1635 (struct scatterlist *)(bidi_sdb + 1);
1637 next_rq->special = bidi_sdb;
1640 blk_mq_start_request(req);
1642 return scsi_setup_cmnd(sdev, req);
1645 static void scsi_mq_done(struct scsi_cmnd *cmd)
1647 trace_scsi_dispatch_cmd_done(cmd);
1648 blk_mq_complete_request(cmd->request);
1651 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1653 struct request_queue *q = hctx->queue;
1654 struct scsi_device *sdev = q->queuedata;
1656 atomic_dec(&sdev->device_busy);
1657 put_device(&sdev->sdev_gendev);
1660 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1662 struct request_queue *q = hctx->queue;
1663 struct scsi_device *sdev = q->queuedata;
1665 if (!get_device(&sdev->sdev_gendev))
1667 if (!scsi_dev_queue_ready(q, sdev))
1668 goto out_put_device;
1673 put_device(&sdev->sdev_gendev);
1675 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1676 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1680 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1681 const struct blk_mq_queue_data *bd)
1683 struct request *req = bd->rq;
1684 struct request_queue *q = req->q;
1685 struct scsi_device *sdev = q->queuedata;
1686 struct Scsi_Host *shost = sdev->host;
1687 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1692 * If the device is not in running state we will reject some or all
1695 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1696 ret = scsi_prep_state_check(sdev, req);
1697 if (ret != BLK_STS_OK)
1698 goto out_put_budget;
1701 ret = BLK_STS_RESOURCE;
1702 if (!scsi_target_queue_ready(shost, sdev))
1703 goto out_put_budget;
1704 if (!scsi_host_queue_ready(q, shost, sdev))
1705 goto out_dec_target_busy;
1707 if (!(req->rq_flags & RQF_DONTPREP)) {
1708 ret = prep_to_mq(scsi_mq_prep_fn(req));
1709 if (ret != BLK_STS_OK)
1710 goto out_dec_host_busy;
1711 req->rq_flags |= RQF_DONTPREP;
1713 blk_mq_start_request(req);
1716 if (sdev->simple_tags)
1717 cmd->flags |= SCMD_TAGGED;
1719 cmd->flags &= ~SCMD_TAGGED;
1721 scsi_init_cmd_errh(cmd);
1722 cmd->scsi_done = scsi_mq_done;
1724 reason = scsi_dispatch_cmd(cmd);
1726 scsi_set_blocked(cmd, reason);
1727 ret = BLK_STS_RESOURCE;
1728 goto out_dec_host_busy;
1734 scsi_dec_host_busy(shost);
1735 out_dec_target_busy:
1736 if (scsi_target(sdev)->can_queue > 0)
1737 atomic_dec(&scsi_target(sdev)->target_busy);
1739 scsi_mq_put_budget(hctx);
1743 case BLK_STS_RESOURCE:
1744 if (atomic_read(&sdev->device_busy) ||
1745 scsi_device_blocked(sdev))
1746 ret = BLK_STS_DEV_RESOURCE;
1750 * Make sure to release all allocated ressources when
1751 * we hit an error, as we will never see this command
1754 if (req->rq_flags & RQF_DONTPREP)
1755 scsi_mq_uninit_cmd(cmd);
1761 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1765 return BLK_EH_RESET_TIMER;
1766 return scsi_times_out(req);
1769 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1770 unsigned int hctx_idx, unsigned int numa_node)
1772 struct Scsi_Host *shost = set->driver_data;
1773 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1774 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1775 struct scatterlist *sg;
1777 if (unchecked_isa_dma)
1778 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1779 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1780 GFP_KERNEL, numa_node);
1781 if (!cmd->sense_buffer)
1783 cmd->req.sense = cmd->sense_buffer;
1785 if (scsi_host_get_prot(shost)) {
1786 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1787 shost->hostt->cmd_size;
1788 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
1794 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1795 unsigned int hctx_idx)
1797 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1799 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1803 static int scsi_map_queues(struct blk_mq_tag_set *set)
1805 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1807 if (shost->hostt->map_queues)
1808 return shost->hostt->map_queues(shost);
1809 return blk_mq_map_queues(&set->map[0]);
1812 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1814 struct device *dev = shost->dma_dev;
1817 * this limit is imposed by hardware restrictions
1819 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1822 if (scsi_host_prot_dma(shost)) {
1823 shost->sg_prot_tablesize =
1824 min_not_zero(shost->sg_prot_tablesize,
1825 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1826 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1827 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1830 blk_queue_max_hw_sectors(q, shost->max_sectors);
1831 if (shost->unchecked_isa_dma)
1832 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1833 blk_queue_segment_boundary(q, shost->dma_boundary);
1834 dma_set_seg_boundary(dev, shost->dma_boundary);
1836 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1838 if (!shost->use_clustering)
1839 q->limits.cluster = 0;
1842 * Set a reasonable default alignment: The larger of 32-byte (dword),
1843 * which is a common minimum for HBAs, and the minimum DMA alignment,
1844 * which is set by the platform.
1846 * Devices that require a bigger alignment can increase it later.
1848 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1850 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1852 static const struct blk_mq_ops scsi_mq_ops = {
1853 .get_budget = scsi_mq_get_budget,
1854 .put_budget = scsi_mq_put_budget,
1855 .queue_rq = scsi_queue_rq,
1856 .complete = scsi_softirq_done,
1857 .timeout = scsi_timeout,
1858 #ifdef CONFIG_BLK_DEBUG_FS
1859 .show_rq = scsi_show_rq,
1861 .init_request = scsi_mq_init_request,
1862 .exit_request = scsi_mq_exit_request,
1863 .initialize_rq_fn = scsi_initialize_rq,
1864 .busy = scsi_mq_lld_busy,
1865 .map_queues = scsi_map_queues,
1868 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1870 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1871 if (IS_ERR(sdev->request_queue))
1874 sdev->request_queue->queuedata = sdev;
1875 __scsi_init_queue(sdev->host, sdev->request_queue);
1876 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1877 return sdev->request_queue;
1880 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1882 unsigned int cmd_size, sgl_size;
1884 sgl_size = scsi_mq_sgl_size(shost);
1885 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1886 if (scsi_host_get_prot(shost))
1887 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
1889 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1890 shost->tag_set.ops = &scsi_mq_ops;
1891 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1892 shost->tag_set.queue_depth = shost->can_queue;
1893 shost->tag_set.cmd_size = cmd_size;
1894 shost->tag_set.numa_node = NUMA_NO_NODE;
1895 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1896 shost->tag_set.flags |=
1897 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1898 shost->tag_set.driver_data = shost;
1900 return blk_mq_alloc_tag_set(&shost->tag_set);
1903 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1905 blk_mq_free_tag_set(&shost->tag_set);
1909 * scsi_device_from_queue - return sdev associated with a request_queue
1910 * @q: The request queue to return the sdev from
1912 * Return the sdev associated with a request queue or NULL if the
1913 * request_queue does not reference a SCSI device.
1915 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1917 struct scsi_device *sdev = NULL;
1919 if (q->mq_ops == &scsi_mq_ops)
1920 sdev = q->queuedata;
1921 if (!sdev || !get_device(&sdev->sdev_gendev))
1926 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1929 * Function: scsi_block_requests()
1931 * Purpose: Utility function used by low-level drivers to prevent further
1932 * commands from being queued to the device.
1934 * Arguments: shost - Host in question
1938 * Lock status: No locks are assumed held.
1940 * Notes: There is no timer nor any other means by which the requests
1941 * get unblocked other than the low-level driver calling
1942 * scsi_unblock_requests().
1944 void scsi_block_requests(struct Scsi_Host *shost)
1946 shost->host_self_blocked = 1;
1948 EXPORT_SYMBOL(scsi_block_requests);
1951 * Function: scsi_unblock_requests()
1953 * Purpose: Utility function used by low-level drivers to allow further
1954 * commands from being queued to the device.
1956 * Arguments: shost - Host in question
1960 * Lock status: No locks are assumed held.
1962 * Notes: There is no timer nor any other means by which the requests
1963 * get unblocked other than the low-level driver calling
1964 * scsi_unblock_requests().
1966 * This is done as an API function so that changes to the
1967 * internals of the scsi mid-layer won't require wholesale
1968 * changes to drivers that use this feature.
1970 void scsi_unblock_requests(struct Scsi_Host *shost)
1972 shost->host_self_blocked = 0;
1973 scsi_run_host_queues(shost);
1975 EXPORT_SYMBOL(scsi_unblock_requests);
1977 int __init scsi_init_queue(void)
1979 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1980 sizeof(struct scsi_data_buffer),
1982 if (!scsi_sdb_cache) {
1983 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1990 void scsi_exit_queue(void)
1992 kmem_cache_destroy(scsi_sense_cache);
1993 kmem_cache_destroy(scsi_sense_isadma_cache);
1994 kmem_cache_destroy(scsi_sdb_cache);
1998 * scsi_mode_select - issue a mode select
1999 * @sdev: SCSI device to be queried
2000 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2001 * @sp: Save page bit (0 == don't save, 1 == save)
2002 * @modepage: mode page being requested
2003 * @buffer: request buffer (may not be smaller than eight bytes)
2004 * @len: length of request buffer.
2005 * @timeout: command timeout
2006 * @retries: number of retries before failing
2007 * @data: returns a structure abstracting the mode header data
2008 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2009 * must be SCSI_SENSE_BUFFERSIZE big.
2011 * Returns zero if successful; negative error number or scsi
2016 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2017 unsigned char *buffer, int len, int timeout, int retries,
2018 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2020 unsigned char cmd[10];
2021 unsigned char *real_buffer;
2024 memset(cmd, 0, sizeof(cmd));
2025 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2027 if (sdev->use_10_for_ms) {
2030 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2033 memcpy(real_buffer + 8, buffer, len);
2037 real_buffer[2] = data->medium_type;
2038 real_buffer[3] = data->device_specific;
2039 real_buffer[4] = data->longlba ? 0x01 : 0;
2041 real_buffer[6] = data->block_descriptor_length >> 8;
2042 real_buffer[7] = data->block_descriptor_length;
2044 cmd[0] = MODE_SELECT_10;
2048 if (len > 255 || data->block_descriptor_length > 255 ||
2052 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2055 memcpy(real_buffer + 4, buffer, len);
2058 real_buffer[1] = data->medium_type;
2059 real_buffer[2] = data->device_specific;
2060 real_buffer[3] = data->block_descriptor_length;
2063 cmd[0] = MODE_SELECT;
2067 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2068 sshdr, timeout, retries, NULL);
2072 EXPORT_SYMBOL_GPL(scsi_mode_select);
2075 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2076 * @sdev: SCSI device to be queried
2077 * @dbd: set if mode sense will allow block descriptors to be returned
2078 * @modepage: mode page being requested
2079 * @buffer: request buffer (may not be smaller than eight bytes)
2080 * @len: length of request buffer.
2081 * @timeout: command timeout
2082 * @retries: number of retries before failing
2083 * @data: returns a structure abstracting the mode header data
2084 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2085 * must be SCSI_SENSE_BUFFERSIZE big.
2087 * Returns zero if unsuccessful, or the header offset (either 4
2088 * or 8 depending on whether a six or ten byte command was
2089 * issued) if successful.
2092 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2093 unsigned char *buffer, int len, int timeout, int retries,
2094 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2096 unsigned char cmd[12];
2099 int result, retry_count = retries;
2100 struct scsi_sense_hdr my_sshdr;
2102 memset(data, 0, sizeof(*data));
2103 memset(&cmd[0], 0, 12);
2104 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2107 /* caller might not be interested in sense, but we need it */
2112 use_10_for_ms = sdev->use_10_for_ms;
2114 if (use_10_for_ms) {
2118 cmd[0] = MODE_SENSE_10;
2125 cmd[0] = MODE_SENSE;
2130 memset(buffer, 0, len);
2132 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2133 sshdr, timeout, retries, NULL);
2135 /* This code looks awful: what it's doing is making sure an
2136 * ILLEGAL REQUEST sense return identifies the actual command
2137 * byte as the problem. MODE_SENSE commands can return
2138 * ILLEGAL REQUEST if the code page isn't supported */
2140 if (use_10_for_ms && !scsi_status_is_good(result) &&
2141 driver_byte(result) == DRIVER_SENSE) {
2142 if (scsi_sense_valid(sshdr)) {
2143 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2144 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2146 * Invalid command operation code
2148 sdev->use_10_for_ms = 0;
2154 if(scsi_status_is_good(result)) {
2155 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2156 (modepage == 6 || modepage == 8))) {
2157 /* Initio breakage? */
2160 data->medium_type = 0;
2161 data->device_specific = 0;
2163 data->block_descriptor_length = 0;
2164 } else if(use_10_for_ms) {
2165 data->length = buffer[0]*256 + buffer[1] + 2;
2166 data->medium_type = buffer[2];
2167 data->device_specific = buffer[3];
2168 data->longlba = buffer[4] & 0x01;
2169 data->block_descriptor_length = buffer[6]*256
2172 data->length = buffer[0] + 1;
2173 data->medium_type = buffer[1];
2174 data->device_specific = buffer[2];
2175 data->block_descriptor_length = buffer[3];
2177 data->header_length = header_length;
2178 } else if ((status_byte(result) == CHECK_CONDITION) &&
2179 scsi_sense_valid(sshdr) &&
2180 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2187 EXPORT_SYMBOL(scsi_mode_sense);
2190 * scsi_test_unit_ready - test if unit is ready
2191 * @sdev: scsi device to change the state of.
2192 * @timeout: command timeout
2193 * @retries: number of retries before failing
2194 * @sshdr: outpout pointer for decoded sense information.
2196 * Returns zero if unsuccessful or an error if TUR failed. For
2197 * removable media, UNIT_ATTENTION sets ->changed flag.
2200 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2201 struct scsi_sense_hdr *sshdr)
2204 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2208 /* try to eat the UNIT_ATTENTION if there are enough retries */
2210 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2212 if (sdev->removable && scsi_sense_valid(sshdr) &&
2213 sshdr->sense_key == UNIT_ATTENTION)
2215 } while (scsi_sense_valid(sshdr) &&
2216 sshdr->sense_key == UNIT_ATTENTION && --retries);
2220 EXPORT_SYMBOL(scsi_test_unit_ready);
2223 * scsi_device_set_state - Take the given device through the device state model.
2224 * @sdev: scsi device to change the state of.
2225 * @state: state to change to.
2227 * Returns zero if successful or an error if the requested
2228 * transition is illegal.
2231 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2233 enum scsi_device_state oldstate = sdev->sdev_state;
2235 if (state == oldstate)
2241 case SDEV_CREATED_BLOCK:
2252 case SDEV_TRANSPORT_OFFLINE:
2265 case SDEV_TRANSPORT_OFFLINE:
2273 case SDEV_TRANSPORT_OFFLINE:
2288 case SDEV_CREATED_BLOCK:
2296 case SDEV_CREATED_BLOCK:
2311 case SDEV_TRANSPORT_OFFLINE:
2323 case SDEV_TRANSPORT_OFFLINE:
2326 case SDEV_CREATED_BLOCK:
2334 sdev->sdev_state = state;
2338 SCSI_LOG_ERROR_RECOVERY(1,
2339 sdev_printk(KERN_ERR, sdev,
2340 "Illegal state transition %s->%s",
2341 scsi_device_state_name(oldstate),
2342 scsi_device_state_name(state))
2346 EXPORT_SYMBOL(scsi_device_set_state);
2349 * sdev_evt_emit - emit a single SCSI device uevent
2350 * @sdev: associated SCSI device
2351 * @evt: event to emit
2353 * Send a single uevent (scsi_event) to the associated scsi_device.
2355 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2360 switch (evt->evt_type) {
2361 case SDEV_EVT_MEDIA_CHANGE:
2362 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2364 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2365 scsi_rescan_device(&sdev->sdev_gendev);
2366 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2368 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2369 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2371 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2372 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2374 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2375 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2377 case SDEV_EVT_LUN_CHANGE_REPORTED:
2378 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2380 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2381 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2383 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2384 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2393 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2397 * sdev_evt_thread - send a uevent for each scsi event
2398 * @work: work struct for scsi_device
2400 * Dispatch queued events to their associated scsi_device kobjects
2403 void scsi_evt_thread(struct work_struct *work)
2405 struct scsi_device *sdev;
2406 enum scsi_device_event evt_type;
2407 LIST_HEAD(event_list);
2409 sdev = container_of(work, struct scsi_device, event_work);
2411 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2412 if (test_and_clear_bit(evt_type, sdev->pending_events))
2413 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2416 struct scsi_event *evt;
2417 struct list_head *this, *tmp;
2418 unsigned long flags;
2420 spin_lock_irqsave(&sdev->list_lock, flags);
2421 list_splice_init(&sdev->event_list, &event_list);
2422 spin_unlock_irqrestore(&sdev->list_lock, flags);
2424 if (list_empty(&event_list))
2427 list_for_each_safe(this, tmp, &event_list) {
2428 evt = list_entry(this, struct scsi_event, node);
2429 list_del(&evt->node);
2430 scsi_evt_emit(sdev, evt);
2437 * sdev_evt_send - send asserted event to uevent thread
2438 * @sdev: scsi_device event occurred on
2439 * @evt: event to send
2441 * Assert scsi device event asynchronously.
2443 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2445 unsigned long flags;
2448 /* FIXME: currently this check eliminates all media change events
2449 * for polled devices. Need to update to discriminate between AN
2450 * and polled events */
2451 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2457 spin_lock_irqsave(&sdev->list_lock, flags);
2458 list_add_tail(&evt->node, &sdev->event_list);
2459 schedule_work(&sdev->event_work);
2460 spin_unlock_irqrestore(&sdev->list_lock, flags);
2462 EXPORT_SYMBOL_GPL(sdev_evt_send);
2465 * sdev_evt_alloc - allocate a new scsi event
2466 * @evt_type: type of event to allocate
2467 * @gfpflags: GFP flags for allocation
2469 * Allocates and returns a new scsi_event.
2471 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2474 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2478 evt->evt_type = evt_type;
2479 INIT_LIST_HEAD(&evt->node);
2481 /* evt_type-specific initialization, if any */
2483 case SDEV_EVT_MEDIA_CHANGE:
2484 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2485 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2486 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2487 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2488 case SDEV_EVT_LUN_CHANGE_REPORTED:
2489 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2490 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2498 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2501 * sdev_evt_send_simple - send asserted event to uevent thread
2502 * @sdev: scsi_device event occurred on
2503 * @evt_type: type of event to send
2504 * @gfpflags: GFP flags for allocation
2506 * Assert scsi device event asynchronously, given an event type.
2508 void sdev_evt_send_simple(struct scsi_device *sdev,
2509 enum scsi_device_event evt_type, gfp_t gfpflags)
2511 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2513 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2518 sdev_evt_send(sdev, evt);
2520 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2523 * scsi_device_quiesce - Block user issued commands.
2524 * @sdev: scsi device to quiesce.
2526 * This works by trying to transition to the SDEV_QUIESCE state
2527 * (which must be a legal transition). When the device is in this
2528 * state, only special requests will be accepted, all others will
2529 * be deferred. Since special requests may also be requeued requests,
2530 * a successful return doesn't guarantee the device will be
2531 * totally quiescent.
2533 * Must be called with user context, may sleep.
2535 * Returns zero if unsuccessful or an error if not.
2538 scsi_device_quiesce(struct scsi_device *sdev)
2540 struct request_queue *q = sdev->request_queue;
2544 * It is allowed to call scsi_device_quiesce() multiple times from
2545 * the same context but concurrent scsi_device_quiesce() calls are
2548 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2550 if (sdev->quiesced_by == current)
2555 blk_mq_freeze_queue(q);
2557 * Ensure that the effect of blk_set_pm_only() will be visible
2558 * for percpu_ref_tryget() callers that occur after the queue
2559 * unfreeze even if the queue was already frozen before this function
2560 * was called. See also https://lwn.net/Articles/573497/.
2563 blk_mq_unfreeze_queue(q);
2565 mutex_lock(&sdev->state_mutex);
2566 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2568 sdev->quiesced_by = current;
2570 blk_clear_pm_only(q);
2571 mutex_unlock(&sdev->state_mutex);
2575 EXPORT_SYMBOL(scsi_device_quiesce);
2578 * scsi_device_resume - Restart user issued commands to a quiesced device.
2579 * @sdev: scsi device to resume.
2581 * Moves the device from quiesced back to running and restarts the
2584 * Must be called with user context, may sleep.
2586 void scsi_device_resume(struct scsi_device *sdev)
2588 /* check if the device state was mutated prior to resume, and if
2589 * so assume the state is being managed elsewhere (for example
2590 * device deleted during suspend)
2592 mutex_lock(&sdev->state_mutex);
2593 WARN_ON_ONCE(!sdev->quiesced_by);
2594 sdev->quiesced_by = NULL;
2595 blk_clear_pm_only(sdev->request_queue);
2596 if (sdev->sdev_state == SDEV_QUIESCE)
2597 scsi_device_set_state(sdev, SDEV_RUNNING);
2598 mutex_unlock(&sdev->state_mutex);
2600 EXPORT_SYMBOL(scsi_device_resume);
2603 device_quiesce_fn(struct scsi_device *sdev, void *data)
2605 scsi_device_quiesce(sdev);
2609 scsi_target_quiesce(struct scsi_target *starget)
2611 starget_for_each_device(starget, NULL, device_quiesce_fn);
2613 EXPORT_SYMBOL(scsi_target_quiesce);
2616 device_resume_fn(struct scsi_device *sdev, void *data)
2618 scsi_device_resume(sdev);
2622 scsi_target_resume(struct scsi_target *starget)
2624 starget_for_each_device(starget, NULL, device_resume_fn);
2626 EXPORT_SYMBOL(scsi_target_resume);
2629 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2630 * @sdev: device to block
2632 * Pause SCSI command processing on the specified device. Does not sleep.
2634 * Returns zero if successful or a negative error code upon failure.
2637 * This routine transitions the device to the SDEV_BLOCK state (which must be
2638 * a legal transition). When the device is in this state, command processing
2639 * is paused until the device leaves the SDEV_BLOCK state. See also
2640 * scsi_internal_device_unblock_nowait().
2642 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2644 struct request_queue *q = sdev->request_queue;
2647 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2649 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2656 * The device has transitioned to SDEV_BLOCK. Stop the
2657 * block layer from calling the midlayer with this device's
2660 blk_mq_quiesce_queue_nowait(q);
2663 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2666 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2667 * @sdev: device to block
2669 * Pause SCSI command processing on the specified device and wait until all
2670 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2672 * Returns zero if successful or a negative error code upon failure.
2675 * This routine transitions the device to the SDEV_BLOCK state (which must be
2676 * a legal transition). When the device is in this state, command processing
2677 * is paused until the device leaves the SDEV_BLOCK state. See also
2678 * scsi_internal_device_unblock().
2680 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2681 * scsi_internal_device_block() has blocked a SCSI device and also
2682 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2684 static int scsi_internal_device_block(struct scsi_device *sdev)
2686 struct request_queue *q = sdev->request_queue;
2689 mutex_lock(&sdev->state_mutex);
2690 err = scsi_internal_device_block_nowait(sdev);
2692 blk_mq_quiesce_queue(q);
2693 mutex_unlock(&sdev->state_mutex);
2698 void scsi_start_queue(struct scsi_device *sdev)
2700 struct request_queue *q = sdev->request_queue;
2702 blk_mq_unquiesce_queue(q);
2706 * scsi_internal_device_unblock_nowait - resume a device after a block request
2707 * @sdev: device to resume
2708 * @new_state: state to set the device to after unblocking
2710 * Restart the device queue for a previously suspended SCSI device. Does not
2713 * Returns zero if successful or a negative error code upon failure.
2716 * This routine transitions the device to the SDEV_RUNNING state or to one of
2717 * the offline states (which must be a legal transition) allowing the midlayer
2718 * to goose the queue for this device.
2720 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2721 enum scsi_device_state new_state)
2724 * Try to transition the scsi device to SDEV_RUNNING or one of the
2725 * offlined states and goose the device queue if successful.
2727 switch (sdev->sdev_state) {
2729 case SDEV_TRANSPORT_OFFLINE:
2730 sdev->sdev_state = new_state;
2732 case SDEV_CREATED_BLOCK:
2733 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2734 new_state == SDEV_OFFLINE)
2735 sdev->sdev_state = new_state;
2737 sdev->sdev_state = SDEV_CREATED;
2745 scsi_start_queue(sdev);
2749 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2752 * scsi_internal_device_unblock - resume a device after a block request
2753 * @sdev: device to resume
2754 * @new_state: state to set the device to after unblocking
2756 * Restart the device queue for a previously suspended SCSI device. May sleep.
2758 * Returns zero if successful or a negative error code upon failure.
2761 * This routine transitions the device to the SDEV_RUNNING state or to one of
2762 * the offline states (which must be a legal transition) allowing the midlayer
2763 * to goose the queue for this device.
2765 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2766 enum scsi_device_state new_state)
2770 mutex_lock(&sdev->state_mutex);
2771 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2772 mutex_unlock(&sdev->state_mutex);
2778 device_block(struct scsi_device *sdev, void *data)
2780 scsi_internal_device_block(sdev);
2784 target_block(struct device *dev, void *data)
2786 if (scsi_is_target_device(dev))
2787 starget_for_each_device(to_scsi_target(dev), NULL,
2793 scsi_target_block(struct device *dev)
2795 if (scsi_is_target_device(dev))
2796 starget_for_each_device(to_scsi_target(dev), NULL,
2799 device_for_each_child(dev, NULL, target_block);
2801 EXPORT_SYMBOL_GPL(scsi_target_block);
2804 device_unblock(struct scsi_device *sdev, void *data)
2806 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2810 target_unblock(struct device *dev, void *data)
2812 if (scsi_is_target_device(dev))
2813 starget_for_each_device(to_scsi_target(dev), data,
2819 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2821 if (scsi_is_target_device(dev))
2822 starget_for_each_device(to_scsi_target(dev), &new_state,
2825 device_for_each_child(dev, &new_state, target_unblock);
2827 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2830 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2831 * @sgl: scatter-gather list
2832 * @sg_count: number of segments in sg
2833 * @offset: offset in bytes into sg, on return offset into the mapped area
2834 * @len: bytes to map, on return number of bytes mapped
2836 * Returns virtual address of the start of the mapped page
2838 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2839 size_t *offset, size_t *len)
2842 size_t sg_len = 0, len_complete = 0;
2843 struct scatterlist *sg;
2846 WARN_ON(!irqs_disabled());
2848 for_each_sg(sgl, sg, sg_count, i) {
2849 len_complete = sg_len; /* Complete sg-entries */
2850 sg_len += sg->length;
2851 if (sg_len > *offset)
2855 if (unlikely(i == sg_count)) {
2856 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2858 __func__, sg_len, *offset, sg_count);
2863 /* Offset starting from the beginning of first page in this sg-entry */
2864 *offset = *offset - len_complete + sg->offset;
2866 /* Assumption: contiguous pages can be accessed as "page + i" */
2867 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2868 *offset &= ~PAGE_MASK;
2870 /* Bytes in this sg-entry from *offset to the end of the page */
2871 sg_len = PAGE_SIZE - *offset;
2875 return kmap_atomic(page);
2877 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2880 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2881 * @virt: virtual address to be unmapped
2883 void scsi_kunmap_atomic_sg(void *virt)
2885 kunmap_atomic(virt);
2887 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2889 void sdev_disable_disk_events(struct scsi_device *sdev)
2891 atomic_inc(&sdev->disk_events_disable_depth);
2893 EXPORT_SYMBOL(sdev_disable_disk_events);
2895 void sdev_enable_disk_events(struct scsi_device *sdev)
2897 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2899 atomic_dec(&sdev->disk_events_disable_depth);
2901 EXPORT_SYMBOL(sdev_enable_disk_events);
2904 * scsi_vpd_lun_id - return a unique device identification
2905 * @sdev: SCSI device
2906 * @id: buffer for the identification
2907 * @id_len: length of the buffer
2909 * Copies a unique device identification into @id based
2910 * on the information in the VPD page 0x83 of the device.
2911 * The string will be formatted as a SCSI name string.
2913 * Returns the length of the identification or error on failure.
2914 * If the identifier is longer than the supplied buffer the actual
2915 * identifier length is returned and the buffer is not zero-padded.
2917 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2919 u8 cur_id_type = 0xff;
2921 const unsigned char *d, *cur_id_str;
2922 const struct scsi_vpd *vpd_pg83;
2923 int id_size = -EINVAL;
2926 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2933 * Look for the correct descriptor.
2934 * Order of preference for lun descriptor:
2935 * - SCSI name string
2936 * - NAA IEEE Registered Extended
2937 * - EUI-64 based 16-byte
2938 * - EUI-64 based 12-byte
2939 * - NAA IEEE Registered
2940 * - NAA IEEE Extended
2942 * as longer descriptors reduce the likelyhood
2943 * of identification clashes.
2946 /* The id string must be at least 20 bytes + terminating NULL byte */
2952 memset(id, 0, id_len);
2953 d = vpd_pg83->data + 4;
2954 while (d < vpd_pg83->data + vpd_pg83->len) {
2955 /* Skip designators not referring to the LUN */
2956 if ((d[1] & 0x30) != 0x00)
2959 switch (d[1] & 0xf) {
2962 if (cur_id_size > d[3])
2964 /* Prefer anything */
2965 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2968 if (cur_id_size + 4 > id_len)
2969 cur_id_size = id_len - 4;
2971 cur_id_type = d[1] & 0xf;
2972 id_size = snprintf(id, id_len, "t10.%*pE",
2973 cur_id_size, cur_id_str);
2977 if (cur_id_size > d[3])
2979 /* Prefer NAA IEEE Registered Extended */
2980 if (cur_id_type == 0x3 &&
2981 cur_id_size == d[3])
2985 cur_id_type = d[1] & 0xf;
2986 switch (cur_id_size) {
2988 id_size = snprintf(id, id_len,
2993 id_size = snprintf(id, id_len,
2998 id_size = snprintf(id, id_len,
3009 if (cur_id_size > d[3])
3013 cur_id_type = d[1] & 0xf;
3014 switch (cur_id_size) {
3016 id_size = snprintf(id, id_len,
3021 id_size = snprintf(id, id_len,
3031 /* SCSI name string */
3032 if (cur_id_size + 4 > d[3])
3034 /* Prefer others for truncated descriptor */
3035 if (cur_id_size && d[3] > id_len)
3037 cur_id_size = id_size = d[3];
3039 cur_id_type = d[1] & 0xf;
3040 if (cur_id_size >= id_len)
3041 cur_id_size = id_len - 1;
3042 memcpy(id, cur_id_str, cur_id_size);
3043 /* Decrease priority for truncated descriptor */
3044 if (cur_id_size != id_size)
3057 EXPORT_SYMBOL(scsi_vpd_lun_id);
3060 * scsi_vpd_tpg_id - return a target port group identifier
3061 * @sdev: SCSI device
3063 * Returns the Target Port Group identifier from the information
3064 * froom VPD page 0x83 of the device.
3066 * Returns the identifier or error on failure.
3068 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3070 const unsigned char *d;
3071 const struct scsi_vpd *vpd_pg83;
3072 int group_id = -EAGAIN, rel_port = -1;
3075 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3081 d = vpd_pg83->data + 4;
3082 while (d < vpd_pg83->data + vpd_pg83->len) {
3083 switch (d[1] & 0xf) {
3085 /* Relative target port */
3086 rel_port = get_unaligned_be16(&d[6]);
3089 /* Target port group */
3090 group_id = get_unaligned_be16(&d[6]);
3099 if (group_id >= 0 && rel_id && rel_port != -1)
3104 EXPORT_SYMBOL(scsi_vpd_tpg_id);