1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
36 #include <scsi/scsi_dh.h>
38 #include <trace/events/scsi.h>
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
45 * Size of integrity metadata is usually small, 1 inline sg should
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define SCSI_INLINE_PROT_SG_CNT 0
50 #define SCSI_INLINE_SG_CNT 0
52 #define SCSI_INLINE_PROT_SG_CNT 1
53 #define SCSI_INLINE_SG_CNT 2
56 static struct kmem_cache *scsi_sense_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 int scsi_init_sense_cache(struct Scsi_Host *shost)
65 mutex_lock(&scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
74 mutex_unlock(&scsi_sense_cache_mutex);
79 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
81 struct Scsi_Host *host = cmd->device->host;
82 struct scsi_device *device = cmd->device;
83 struct scsi_target *starget = scsi_target(device);
86 * Set the appropriate busy bit for the device/host.
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
99 case SCSI_MLQUEUE_HOST_BUSY:
100 atomic_set(&host->host_blocked, host->max_host_blocked);
102 case SCSI_MLQUEUE_DEVICE_BUSY:
103 case SCSI_MLQUEUE_EH_RETRY:
104 atomic_set(&device->device_blocked,
105 device->max_device_blocked);
107 case SCSI_MLQUEUE_TARGET_BUSY:
108 atomic_set(&starget->target_blocked,
109 starget->max_target_blocked);
114 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
116 struct request *rq = scsi_cmd_to_rq(cmd);
118 if (rq->rq_flags & RQF_DONTPREP) {
119 rq->rq_flags &= ~RQF_DONTPREP;
120 scsi_mq_uninit_cmd(cmd);
124 blk_mq_requeue_request(rq, true);
128 * __scsi_queue_insert - private queue insertion
129 * @cmd: The SCSI command being requeued
130 * @reason: The reason for the requeue
131 * @unbusy: Whether the queue should be unbusied
133 * This is a private queue insertion. The public interface
134 * scsi_queue_insert() always assumes the queue should be unbusied
135 * because it's always called before the completion. This function is
136 * for a requeue after completion, which should only occur in this
139 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
141 struct scsi_device *device = cmd->device;
143 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
144 "Inserting command %p into mlqueue\n", cmd));
146 scsi_set_blocked(cmd, reason);
149 * Decrement the counters, since these commands are no longer
150 * active on the host/device.
153 scsi_device_unbusy(device, cmd);
156 * Requeue this command. It will go before all other commands
157 * that are already in the queue. Schedule requeue work under
158 * lock such that the kblockd_schedule_work() call happens
159 * before blk_mq_destroy_queue() finishes.
163 blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
167 * scsi_queue_insert - Reinsert a command in the queue.
168 * @cmd: command that we are adding to queue.
169 * @reason: why we are inserting command to queue.
171 * We do this for one of two cases. Either the host is busy and it cannot accept
172 * any more commands for the time being, or the device returned QUEUE_FULL and
173 * can accept no more commands.
175 * Context: This could be called either from an interrupt context or a normal
178 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
180 __scsi_queue_insert(cmd, reason, true);
185 * __scsi_execute - insert request and wait for the result
188 * @data_direction: data direction
189 * @buffer: data buffer
190 * @bufflen: len of buffer
191 * @sense: optional sense buffer
192 * @sshdr: optional decoded sense header
193 * @timeout: request timeout in HZ
194 * @retries: number of times to retry request
195 * @flags: flags for ->cmd_flags
196 * @rq_flags: flags for ->rq_flags
197 * @resid: optional residual length
199 * Returns the scsi_cmnd result field if a command was executed, or a negative
200 * Linux error code if we didn't get that far.
202 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
203 int data_direction, void *buffer, unsigned bufflen,
204 unsigned char *sense, struct scsi_sense_hdr *sshdr,
205 int timeout, int retries, blk_opf_t flags,
206 req_flags_t rq_flags, int *resid)
209 struct scsi_cmnd *scmd;
212 req = scsi_alloc_request(sdev->request_queue,
213 data_direction == DMA_TO_DEVICE ?
214 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
215 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
220 ret = blk_rq_map_kern(sdev->request_queue, req,
221 buffer, bufflen, GFP_NOIO);
225 scmd = blk_mq_rq_to_pdu(req);
226 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
227 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
228 scmd->allowed = retries;
229 req->timeout = timeout;
230 req->cmd_flags |= flags;
231 req->rq_flags |= rq_flags | RQF_QUIET;
234 * head injection *required* here otherwise quiesce won't work
236 blk_execute_rq(req, true);
239 * Some devices (USB mass-storage in particular) may transfer
240 * garbage data together with a residue indicating that the data
241 * is invalid. Prevent the garbage from being misinterpreted
242 * and prevent security leaks by zeroing out the excess data.
244 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
245 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
248 *resid = scmd->resid_len;
249 if (sense && scmd->sense_len)
250 memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
252 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
256 blk_mq_free_request(req);
260 EXPORT_SYMBOL(__scsi_execute);
263 * Wake up the error handler if necessary. Avoid as follows that the error
264 * handler is not woken up if host in-flight requests number ==
265 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
266 * with an RCU read lock in this function to ensure that this function in
267 * its entirety either finishes before scsi_eh_scmd_add() increases the
268 * host_failed counter or that it notices the shost state change made by
269 * scsi_eh_scmd_add().
271 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
276 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
277 if (unlikely(scsi_host_in_recovery(shost))) {
278 spin_lock_irqsave(shost->host_lock, flags);
279 if (shost->host_failed || shost->host_eh_scheduled)
280 scsi_eh_wakeup(shost);
281 spin_unlock_irqrestore(shost->host_lock, flags);
286 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
288 struct Scsi_Host *shost = sdev->host;
289 struct scsi_target *starget = scsi_target(sdev);
291 scsi_dec_host_busy(shost, cmd);
293 if (starget->can_queue > 0)
294 atomic_dec(&starget->target_busy);
296 sbitmap_put(&sdev->budget_map, cmd->budget_token);
297 cmd->budget_token = -1;
300 static void scsi_kick_queue(struct request_queue *q)
302 blk_mq_run_hw_queues(q, false);
306 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
307 * and call blk_run_queue for all the scsi_devices on the target -
308 * including current_sdev first.
310 * Called with *no* scsi locks held.
312 static void scsi_single_lun_run(struct scsi_device *current_sdev)
314 struct Scsi_Host *shost = current_sdev->host;
315 struct scsi_device *sdev, *tmp;
316 struct scsi_target *starget = scsi_target(current_sdev);
319 spin_lock_irqsave(shost->host_lock, flags);
320 starget->starget_sdev_user = NULL;
321 spin_unlock_irqrestore(shost->host_lock, flags);
324 * Call blk_run_queue for all LUNs on the target, starting with
325 * current_sdev. We race with others (to set starget_sdev_user),
326 * but in most cases, we will be first. Ideally, each LU on the
327 * target would get some limited time or requests on the target.
329 scsi_kick_queue(current_sdev->request_queue);
331 spin_lock_irqsave(shost->host_lock, flags);
332 if (starget->starget_sdev_user)
334 list_for_each_entry_safe(sdev, tmp, &starget->devices,
335 same_target_siblings) {
336 if (sdev == current_sdev)
338 if (scsi_device_get(sdev))
341 spin_unlock_irqrestore(shost->host_lock, flags);
342 scsi_kick_queue(sdev->request_queue);
343 spin_lock_irqsave(shost->host_lock, flags);
345 scsi_device_put(sdev);
348 spin_unlock_irqrestore(shost->host_lock, flags);
351 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
353 if (scsi_device_busy(sdev) >= sdev->queue_depth)
355 if (atomic_read(&sdev->device_blocked) > 0)
360 static inline bool scsi_target_is_busy(struct scsi_target *starget)
362 if (starget->can_queue > 0) {
363 if (atomic_read(&starget->target_busy) >= starget->can_queue)
365 if (atomic_read(&starget->target_blocked) > 0)
371 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
373 if (atomic_read(&shost->host_blocked) > 0)
375 if (shost->host_self_blocked)
380 static void scsi_starved_list_run(struct Scsi_Host *shost)
382 LIST_HEAD(starved_list);
383 struct scsi_device *sdev;
386 spin_lock_irqsave(shost->host_lock, flags);
387 list_splice_init(&shost->starved_list, &starved_list);
389 while (!list_empty(&starved_list)) {
390 struct request_queue *slq;
393 * As long as shost is accepting commands and we have
394 * starved queues, call blk_run_queue. scsi_request_fn
395 * drops the queue_lock and can add us back to the
398 * host_lock protects the starved_list and starved_entry.
399 * scsi_request_fn must get the host_lock before checking
400 * or modifying starved_list or starved_entry.
402 if (scsi_host_is_busy(shost))
405 sdev = list_entry(starved_list.next,
406 struct scsi_device, starved_entry);
407 list_del_init(&sdev->starved_entry);
408 if (scsi_target_is_busy(scsi_target(sdev))) {
409 list_move_tail(&sdev->starved_entry,
410 &shost->starved_list);
415 * Once we drop the host lock, a racing scsi_remove_device()
416 * call may remove the sdev from the starved list and destroy
417 * it and the queue. Mitigate by taking a reference to the
418 * queue and never touching the sdev again after we drop the
419 * host lock. Note: if __scsi_remove_device() invokes
420 * blk_mq_destroy_queue() before the queue is run from this
421 * function then blk_run_queue() will return immediately since
422 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
424 slq = sdev->request_queue;
425 if (!blk_get_queue(slq))
427 spin_unlock_irqrestore(shost->host_lock, flags);
429 scsi_kick_queue(slq);
432 spin_lock_irqsave(shost->host_lock, flags);
434 /* put any unprocessed entries back */
435 list_splice(&starved_list, &shost->starved_list);
436 spin_unlock_irqrestore(shost->host_lock, flags);
440 * scsi_run_queue - Select a proper request queue to serve next.
441 * @q: last request's queue
443 * The previous command was completely finished, start a new one if possible.
445 static void scsi_run_queue(struct request_queue *q)
447 struct scsi_device *sdev = q->queuedata;
449 if (scsi_target(sdev)->single_lun)
450 scsi_single_lun_run(sdev);
451 if (!list_empty(&sdev->host->starved_list))
452 scsi_starved_list_run(sdev->host);
454 blk_mq_run_hw_queues(q, false);
457 void scsi_requeue_run_queue(struct work_struct *work)
459 struct scsi_device *sdev;
460 struct request_queue *q;
462 sdev = container_of(work, struct scsi_device, requeue_work);
463 q = sdev->request_queue;
467 void scsi_run_host_queues(struct Scsi_Host *shost)
469 struct scsi_device *sdev;
471 shost_for_each_device(sdev, shost)
472 scsi_run_queue(sdev->request_queue);
475 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
477 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
478 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
480 if (drv->uninit_command)
481 drv->uninit_command(cmd);
485 void scsi_free_sgtables(struct scsi_cmnd *cmd)
487 if (cmd->sdb.table.nents)
488 sg_free_table_chained(&cmd->sdb.table,
490 if (scsi_prot_sg_count(cmd))
491 sg_free_table_chained(&cmd->prot_sdb->table,
492 SCSI_INLINE_PROT_SG_CNT);
494 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
496 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
498 scsi_free_sgtables(cmd);
499 scsi_uninit_cmd(cmd);
502 static void scsi_run_queue_async(struct scsi_device *sdev)
504 if (scsi_target(sdev)->single_lun ||
505 !list_empty(&sdev->host->starved_list)) {
506 kblockd_schedule_work(&sdev->requeue_work);
509 * smp_mb() present in sbitmap_queue_clear() or implied in
510 * .end_io is for ordering writing .device_busy in
511 * scsi_device_unbusy() and reading sdev->restarts.
513 int old = atomic_read(&sdev->restarts);
516 * ->restarts has to be kept as non-zero if new budget
519 * No need to run queue when either another re-run
520 * queue wins in updating ->restarts or a new budget
523 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
524 blk_mq_run_hw_queues(sdev->request_queue, true);
528 /* Returns false when no more bytes to process, true if there are more */
529 static bool scsi_end_request(struct request *req, blk_status_t error,
532 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
533 struct scsi_device *sdev = cmd->device;
534 struct request_queue *q = sdev->request_queue;
536 if (blk_update_request(req, error, bytes))
540 if (blk_queue_add_random(q))
541 add_disk_randomness(req->q->disk);
543 if (!blk_rq_is_passthrough(req)) {
544 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
545 cmd->flags &= ~SCMD_INITIALIZED;
549 * Calling rcu_barrier() is not necessary here because the
550 * SCSI error handler guarantees that the function called by
551 * call_rcu() has been called before scsi_end_request() is
554 destroy_rcu_head(&cmd->rcu);
557 * In the MQ case the command gets freed by __blk_mq_end_request,
558 * so we have to do all cleanup that depends on it earlier.
560 * We also can't kick the queues from irq context, so we
561 * will have to defer it to a workqueue.
563 scsi_mq_uninit_cmd(cmd);
566 * queue is still alive, so grab the ref for preventing it
567 * from being cleaned up during running queue.
569 percpu_ref_get(&q->q_usage_counter);
571 __blk_mq_end_request(req, error);
573 scsi_run_queue_async(sdev);
575 percpu_ref_put(&q->q_usage_counter);
580 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
582 * @result: scsi error code
584 * Translate a SCSI result code into a blk_status_t value. May reset the host
585 * byte of @cmd->result.
587 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
589 switch (host_byte(result)) {
591 if (scsi_status_is_good(result))
593 return BLK_STS_IOERR;
594 case DID_TRANSPORT_FAILFAST:
595 case DID_TRANSPORT_MARGINAL:
596 return BLK_STS_TRANSPORT;
597 case DID_TARGET_FAILURE:
598 set_host_byte(cmd, DID_OK);
599 return BLK_STS_TARGET;
600 case DID_NEXUS_FAILURE:
601 set_host_byte(cmd, DID_OK);
602 return BLK_STS_NEXUS;
603 case DID_ALLOC_FAILURE:
604 set_host_byte(cmd, DID_OK);
605 return BLK_STS_NOSPC;
606 case DID_MEDIUM_ERROR:
607 set_host_byte(cmd, DID_OK);
608 return BLK_STS_MEDIUM;
610 return BLK_STS_IOERR;
615 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
616 * @rq: request to examine
619 * A request could be merge of IOs which require different failure
620 * handling. This function determines the number of bytes which
621 * can be failed from the beginning of the request without
622 * crossing into area which need to be retried further.
625 * The number of bytes to fail.
627 static unsigned int scsi_rq_err_bytes(const struct request *rq)
629 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
630 unsigned int bytes = 0;
633 if (!(rq->rq_flags & RQF_MIXED_MERGE))
634 return blk_rq_bytes(rq);
637 * Currently the only 'mixing' which can happen is between
638 * different fastfail types. We can safely fail portions
639 * which have all the failfast bits that the first one has -
640 * the ones which are at least as eager to fail as the first
643 for (bio = rq->bio; bio; bio = bio->bi_next) {
644 if ((bio->bi_opf & ff) != ff)
646 bytes += bio->bi_iter.bi_size;
649 /* this could lead to infinite loop */
650 BUG_ON(blk_rq_bytes(rq) && !bytes);
654 /* Helper for scsi_io_completion() when "reprep" action required. */
655 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
656 struct request_queue *q)
658 /* A new command will be prepared and issued. */
659 scsi_mq_requeue_cmd(cmd);
662 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
664 struct request *req = scsi_cmd_to_rq(cmd);
665 unsigned long wait_for;
667 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
670 wait_for = (cmd->allowed + 1) * req->timeout;
671 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
672 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
679 /* Helper for scsi_io_completion() when special action required. */
680 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
682 struct request_queue *q = cmd->device->request_queue;
683 struct request *req = scsi_cmd_to_rq(cmd);
685 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
686 ACTION_DELAYED_RETRY} action;
687 struct scsi_sense_hdr sshdr;
689 bool sense_current = true; /* false implies "deferred sense" */
690 blk_status_t blk_stat;
692 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
694 sense_current = !scsi_sense_is_deferred(&sshdr);
696 blk_stat = scsi_result_to_blk_status(cmd, result);
698 if (host_byte(result) == DID_RESET) {
699 /* Third party bus reset or reset for error recovery
700 * reasons. Just retry the command and see what
703 action = ACTION_RETRY;
704 } else if (sense_valid && sense_current) {
705 switch (sshdr.sense_key) {
707 if (cmd->device->removable) {
708 /* Detected disc change. Set a bit
709 * and quietly refuse further access.
711 cmd->device->changed = 1;
712 action = ACTION_FAIL;
714 /* Must have been a power glitch, or a
715 * bus reset. Could not have been a
716 * media change, so we just retry the
717 * command and see what happens.
719 action = ACTION_RETRY;
722 case ILLEGAL_REQUEST:
723 /* If we had an ILLEGAL REQUEST returned, then
724 * we may have performed an unsupported
725 * command. The only thing this should be
726 * would be a ten byte read where only a six
727 * byte read was supported. Also, on a system
728 * where READ CAPACITY failed, we may have
729 * read past the end of the disk.
731 if ((cmd->device->use_10_for_rw &&
732 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
733 (cmd->cmnd[0] == READ_10 ||
734 cmd->cmnd[0] == WRITE_10)) {
735 /* This will issue a new 6-byte command. */
736 cmd->device->use_10_for_rw = 0;
737 action = ACTION_REPREP;
738 } else if (sshdr.asc == 0x10) /* DIX */ {
739 action = ACTION_FAIL;
740 blk_stat = BLK_STS_PROTECTION;
741 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
742 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
743 action = ACTION_FAIL;
744 blk_stat = BLK_STS_TARGET;
746 action = ACTION_FAIL;
748 case ABORTED_COMMAND:
749 action = ACTION_FAIL;
750 if (sshdr.asc == 0x10) /* DIF */
751 blk_stat = BLK_STS_PROTECTION;
754 /* If the device is in the process of becoming
755 * ready, or has a temporary blockage, retry.
757 if (sshdr.asc == 0x04) {
758 switch (sshdr.ascq) {
759 case 0x01: /* becoming ready */
760 case 0x04: /* format in progress */
761 case 0x05: /* rebuild in progress */
762 case 0x06: /* recalculation in progress */
763 case 0x07: /* operation in progress */
764 case 0x08: /* Long write in progress */
765 case 0x09: /* self test in progress */
766 case 0x11: /* notify (enable spinup) required */
767 case 0x14: /* space allocation in progress */
768 case 0x1a: /* start stop unit in progress */
769 case 0x1b: /* sanitize in progress */
770 case 0x1d: /* configuration in progress */
771 case 0x24: /* depopulation in progress */
772 action = ACTION_DELAYED_RETRY;
774 case 0x0a: /* ALUA state transition */
775 blk_stat = BLK_STS_TRANSPORT;
778 action = ACTION_FAIL;
782 action = ACTION_FAIL;
784 case VOLUME_OVERFLOW:
785 /* See SSC3rXX or current. */
786 action = ACTION_FAIL;
789 action = ACTION_FAIL;
790 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
791 (sshdr.asc == 0x55 &&
792 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
793 /* Insufficient zone resources */
794 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
798 action = ACTION_FAIL;
802 action = ACTION_FAIL;
804 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
805 action = ACTION_FAIL;
809 /* Give up and fail the remainder of the request */
810 if (!(req->rq_flags & RQF_QUIET)) {
811 static DEFINE_RATELIMIT_STATE(_rs,
812 DEFAULT_RATELIMIT_INTERVAL,
813 DEFAULT_RATELIMIT_BURST);
815 if (unlikely(scsi_logging_level))
817 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
818 SCSI_LOG_MLCOMPLETE_BITS);
821 * if logging is enabled the failure will be printed
822 * in scsi_log_completion(), so avoid duplicate messages
824 if (!level && __ratelimit(&_rs)) {
825 scsi_print_result(cmd, NULL, FAILED);
827 scsi_print_sense(cmd);
828 scsi_print_command(cmd);
831 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
835 scsi_io_completion_reprep(cmd, q);
838 /* Retry the same command immediately */
839 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
841 case ACTION_DELAYED_RETRY:
842 /* Retry the same command after a delay */
843 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
849 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
850 * new result that may suppress further error checking. Also modifies
851 * *blk_statp in some cases.
853 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
854 blk_status_t *blk_statp)
857 bool sense_current = true; /* false implies "deferred sense" */
858 struct request *req = scsi_cmd_to_rq(cmd);
859 struct scsi_sense_hdr sshdr;
861 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
863 sense_current = !scsi_sense_is_deferred(&sshdr);
865 if (blk_rq_is_passthrough(req)) {
868 * SG_IO wants current and deferred errors
870 cmd->sense_len = min(8 + cmd->sense_buffer[7],
871 SCSI_SENSE_BUFFERSIZE);
874 *blk_statp = scsi_result_to_blk_status(cmd, result);
875 } else if (blk_rq_bytes(req) == 0 && sense_current) {
877 * Flush commands do not transfers any data, and thus cannot use
878 * good_bytes != blk_rq_bytes(req) as the signal for an error.
879 * This sets *blk_statp explicitly for the problem case.
881 *blk_statp = scsi_result_to_blk_status(cmd, result);
884 * Recovered errors need reporting, but they're always treated as
885 * success, so fiddle the result code here. For passthrough requests
886 * we already took a copy of the original into sreq->result which
887 * is what gets returned to the user
889 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
890 bool do_print = true;
892 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
893 * skip print since caller wants ATA registers. Only occurs
894 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
896 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
898 else if (req->rq_flags & RQF_QUIET)
901 scsi_print_sense(cmd);
903 /* for passthrough, *blk_statp may be set */
904 *blk_statp = BLK_STS_OK;
907 * Another corner case: the SCSI status byte is non-zero but 'good'.
908 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
909 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
910 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
911 * intermediate statuses (both obsolete in SAM-4) as good.
913 if ((result & 0xff) && scsi_status_is_good(result)) {
915 *blk_statp = BLK_STS_OK;
921 * scsi_io_completion - Completion processing for SCSI commands.
922 * @cmd: command that is finished.
923 * @good_bytes: number of processed bytes.
925 * We will finish off the specified number of sectors. If we are done, the
926 * command block will be released and the queue function will be goosed. If we
927 * are not done then we have to figure out what to do next:
929 * a) We can call scsi_io_completion_reprep(). The request will be
930 * unprepared and put back on the queue. Then a new command will
931 * be created for it. This should be used if we made forward
932 * progress, or if we want to switch from READ(10) to READ(6) for
935 * b) We can call scsi_io_completion_action(). The request will be
936 * put back on the queue and retried using the same command as
937 * before, possibly after a delay.
939 * c) We can call scsi_end_request() with blk_stat other than
940 * BLK_STS_OK, to fail the remainder of the request.
942 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
944 int result = cmd->result;
945 struct request_queue *q = cmd->device->request_queue;
946 struct request *req = scsi_cmd_to_rq(cmd);
947 blk_status_t blk_stat = BLK_STS_OK;
949 if (unlikely(result)) /* a nz result may or may not be an error */
950 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
953 * Next deal with any sectors which we were able to correctly
956 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
957 "%u sectors total, %d bytes done.\n",
958 blk_rq_sectors(req), good_bytes));
961 * Failed, zero length commands always need to drop down
962 * to retry code. Fast path should return in this block.
964 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
965 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
966 return; /* no bytes remaining */
969 /* Kill remainder if no retries. */
970 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
971 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
973 "Bytes remaining after failed, no-retry command");
978 * If there had been no error, but we have leftover bytes in the
979 * request just queue the command up again.
981 if (likely(result == 0))
982 scsi_io_completion_reprep(cmd, q);
984 scsi_io_completion_action(cmd, result);
987 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
990 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
991 !op_is_write(req_op(rq)) &&
992 sdev->host->hostt->dma_need_drain(rq);
996 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
997 * @cmd: SCSI command data structure to initialize.
999 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1003 * * BLK_STS_OK - on success
1004 * * BLK_STS_RESOURCE - if the failure is retryable
1005 * * BLK_STS_IOERR - if the failure is fatal
1007 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1009 struct scsi_device *sdev = cmd->device;
1010 struct request *rq = scsi_cmd_to_rq(cmd);
1011 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1012 struct scatterlist *last_sg = NULL;
1014 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1017 if (WARN_ON_ONCE(!nr_segs))
1018 return BLK_STS_IOERR;
1021 * Make sure there is space for the drain. The driver must adjust
1022 * max_hw_segments to be prepared for this.
1028 * If sg table allocation fails, requeue request later.
1030 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1031 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1032 return BLK_STS_RESOURCE;
1035 * Next, walk the list, and fill in the addresses and sizes of
1038 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1040 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1041 unsigned int pad_len =
1042 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1044 last_sg->length += pad_len;
1045 cmd->extra_len += pad_len;
1049 sg_unmark_end(last_sg);
1050 last_sg = sg_next(last_sg);
1051 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1052 sg_mark_end(last_sg);
1054 cmd->extra_len += sdev->dma_drain_len;
1058 BUG_ON(count > cmd->sdb.table.nents);
1059 cmd->sdb.table.nents = count;
1060 cmd->sdb.length = blk_rq_payload_bytes(rq);
1062 if (blk_integrity_rq(rq)) {
1063 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1066 if (WARN_ON_ONCE(!prot_sdb)) {
1068 * This can happen if someone (e.g. multipath)
1069 * queues a command to a device on an adapter
1070 * that does not support DIX.
1072 ret = BLK_STS_IOERR;
1073 goto out_free_sgtables;
1076 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1078 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1079 prot_sdb->table.sgl,
1080 SCSI_INLINE_PROT_SG_CNT)) {
1081 ret = BLK_STS_RESOURCE;
1082 goto out_free_sgtables;
1085 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1086 prot_sdb->table.sgl);
1087 BUG_ON(count > ivecs);
1088 BUG_ON(count > queue_max_integrity_segments(rq->q));
1090 cmd->prot_sdb = prot_sdb;
1091 cmd->prot_sdb->table.nents = count;
1096 scsi_free_sgtables(cmd);
1099 EXPORT_SYMBOL(scsi_alloc_sgtables);
1102 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1103 * @rq: Request associated with the SCSI command to be initialized.
1105 * This function initializes the members of struct scsi_cmnd that must be
1106 * initialized before request processing starts and that won't be
1107 * reinitialized if a SCSI command is requeued.
1109 static void scsi_initialize_rq(struct request *rq)
1111 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1113 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1114 cmd->cmd_len = MAX_COMMAND_SIZE;
1116 init_rcu_head(&cmd->rcu);
1117 cmd->jiffies_at_alloc = jiffies;
1121 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1122 blk_mq_req_flags_t flags)
1126 rq = blk_mq_alloc_request(q, opf, flags);
1128 scsi_initialize_rq(rq);
1131 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1134 * Only called when the request isn't completed by SCSI, and not freed by
1137 static void scsi_cleanup_rq(struct request *rq)
1139 if (rq->rq_flags & RQF_DONTPREP) {
1140 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1141 rq->rq_flags &= ~RQF_DONTPREP;
1145 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1146 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1148 struct request *rq = scsi_cmd_to_rq(cmd);
1150 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1151 cmd->flags |= SCMD_INITIALIZED;
1152 scsi_initialize_rq(rq);
1156 INIT_LIST_HEAD(&cmd->eh_entry);
1157 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1160 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1161 struct request *req)
1163 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1166 * Passthrough requests may transfer data, in which case they must
1167 * a bio attached to them. Or they might contain a SCSI command
1168 * that does not transfer data, in which case they may optionally
1169 * submit a request without an attached bio.
1172 blk_status_t ret = scsi_alloc_sgtables(cmd);
1173 if (unlikely(ret != BLK_STS_OK))
1176 BUG_ON(blk_rq_bytes(req));
1178 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1181 cmd->transfersize = blk_rq_bytes(req);
1186 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1188 switch (sdev->sdev_state) {
1192 case SDEV_TRANSPORT_OFFLINE:
1194 * If the device is offline we refuse to process any
1195 * commands. The device must be brought online
1196 * before trying any recovery commands.
1198 if (!sdev->offline_already) {
1199 sdev->offline_already = true;
1200 sdev_printk(KERN_ERR, sdev,
1201 "rejecting I/O to offline device\n");
1203 return BLK_STS_IOERR;
1206 * If the device is fully deleted, we refuse to
1207 * process any commands as well.
1209 sdev_printk(KERN_ERR, sdev,
1210 "rejecting I/O to dead device\n");
1211 return BLK_STS_IOERR;
1213 case SDEV_CREATED_BLOCK:
1214 return BLK_STS_RESOURCE;
1217 * If the device is blocked we only accept power management
1220 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1221 return BLK_STS_RESOURCE;
1225 * For any other not fully online state we only allow
1226 * power management commands.
1228 if (req && !(req->rq_flags & RQF_PM))
1229 return BLK_STS_OFFLINE;
1235 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1236 * and return the token else return -1.
1238 static inline int scsi_dev_queue_ready(struct request_queue *q,
1239 struct scsi_device *sdev)
1243 token = sbitmap_get(&sdev->budget_map);
1244 if (atomic_read(&sdev->device_blocked)) {
1248 if (scsi_device_busy(sdev) > 1)
1252 * unblock after device_blocked iterates to zero
1254 if (atomic_dec_return(&sdev->device_blocked) > 0)
1256 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1257 "unblocking device at zero depth\n"));
1263 sbitmap_put(&sdev->budget_map, token);
1269 * scsi_target_queue_ready: checks if there we can send commands to target
1270 * @sdev: scsi device on starget to check.
1272 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1273 struct scsi_device *sdev)
1275 struct scsi_target *starget = scsi_target(sdev);
1278 if (starget->single_lun) {
1279 spin_lock_irq(shost->host_lock);
1280 if (starget->starget_sdev_user &&
1281 starget->starget_sdev_user != sdev) {
1282 spin_unlock_irq(shost->host_lock);
1285 starget->starget_sdev_user = sdev;
1286 spin_unlock_irq(shost->host_lock);
1289 if (starget->can_queue <= 0)
1292 busy = atomic_inc_return(&starget->target_busy) - 1;
1293 if (atomic_read(&starget->target_blocked) > 0) {
1298 * unblock after target_blocked iterates to zero
1300 if (atomic_dec_return(&starget->target_blocked) > 0)
1303 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1304 "unblocking target at zero depth\n"));
1307 if (busy >= starget->can_queue)
1313 spin_lock_irq(shost->host_lock);
1314 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1315 spin_unlock_irq(shost->host_lock);
1317 if (starget->can_queue > 0)
1318 atomic_dec(&starget->target_busy);
1323 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1324 * return 0. We must end up running the queue again whenever 0 is
1325 * returned, else IO can hang.
1327 static inline int scsi_host_queue_ready(struct request_queue *q,
1328 struct Scsi_Host *shost,
1329 struct scsi_device *sdev,
1330 struct scsi_cmnd *cmd)
1332 if (scsi_host_in_recovery(shost))
1335 if (atomic_read(&shost->host_blocked) > 0) {
1336 if (scsi_host_busy(shost) > 0)
1340 * unblock after host_blocked iterates to zero
1342 if (atomic_dec_return(&shost->host_blocked) > 0)
1346 shost_printk(KERN_INFO, shost,
1347 "unblocking host at zero depth\n"));
1350 if (shost->host_self_blocked)
1353 /* We're OK to process the command, so we can't be starved */
1354 if (!list_empty(&sdev->starved_entry)) {
1355 spin_lock_irq(shost->host_lock);
1356 if (!list_empty(&sdev->starved_entry))
1357 list_del_init(&sdev->starved_entry);
1358 spin_unlock_irq(shost->host_lock);
1361 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1366 spin_lock_irq(shost->host_lock);
1367 if (list_empty(&sdev->starved_entry))
1368 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1369 spin_unlock_irq(shost->host_lock);
1371 scsi_dec_host_busy(shost, cmd);
1376 * Busy state exporting function for request stacking drivers.
1378 * For efficiency, no lock is taken to check the busy state of
1379 * shost/starget/sdev, since the returned value is not guaranteed and
1380 * may be changed after request stacking drivers call the function,
1381 * regardless of taking lock or not.
1383 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1384 * needs to return 'not busy'. Otherwise, request stacking drivers
1385 * may hold requests forever.
1387 static bool scsi_mq_lld_busy(struct request_queue *q)
1389 struct scsi_device *sdev = q->queuedata;
1390 struct Scsi_Host *shost;
1392 if (blk_queue_dying(q))
1398 * Ignore host/starget busy state.
1399 * Since block layer does not have a concept of fairness across
1400 * multiple queues, congestion of host/starget needs to be handled
1403 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1410 * Block layer request completion callback. May be called from interrupt
1413 static void scsi_complete(struct request *rq)
1415 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1416 enum scsi_disposition disposition;
1418 INIT_LIST_HEAD(&cmd->eh_entry);
1420 atomic_inc(&cmd->device->iodone_cnt);
1422 atomic_inc(&cmd->device->ioerr_cnt);
1424 disposition = scsi_decide_disposition(cmd);
1425 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1426 disposition = SUCCESS;
1428 scsi_log_completion(cmd, disposition);
1430 switch (disposition) {
1432 scsi_finish_command(cmd);
1435 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1437 case ADD_TO_MLQUEUE:
1438 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1441 scsi_eh_scmd_add(cmd);
1447 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1448 * @cmd: command block we are dispatching.
1450 * Return: nonzero return request was rejected and device's queue needs to be
1453 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1455 struct Scsi_Host *host = cmd->device->host;
1458 atomic_inc(&cmd->device->iorequest_cnt);
1460 /* check if the device is still usable */
1461 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1462 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1463 * returns an immediate error upwards, and signals
1464 * that the device is no longer present */
1465 cmd->result = DID_NO_CONNECT << 16;
1469 /* Check to see if the scsi lld made this device blocked. */
1470 if (unlikely(scsi_device_blocked(cmd->device))) {
1472 * in blocked state, the command is just put back on
1473 * the device queue. The suspend state has already
1474 * blocked the queue so future requests should not
1475 * occur until the device transitions out of the
1478 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1479 "queuecommand : device blocked\n"));
1480 return SCSI_MLQUEUE_DEVICE_BUSY;
1483 /* Store the LUN value in cmnd, if needed. */
1484 if (cmd->device->lun_in_cdb)
1485 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1486 (cmd->device->lun << 5 & 0xe0);
1491 * Before we queue this command, check if the command
1492 * length exceeds what the host adapter can handle.
1494 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1495 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1496 "queuecommand : command too long. "
1497 "cdb_size=%d host->max_cmd_len=%d\n",
1498 cmd->cmd_len, cmd->device->host->max_cmd_len));
1499 cmd->result = (DID_ABORT << 16);
1503 if (unlikely(host->shost_state == SHOST_DEL)) {
1504 cmd->result = (DID_NO_CONNECT << 16);
1509 trace_scsi_dispatch_cmd_start(cmd);
1510 rtn = host->hostt->queuecommand(host, cmd);
1512 trace_scsi_dispatch_cmd_error(cmd, rtn);
1513 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1514 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1515 rtn = SCSI_MLQUEUE_HOST_BUSY;
1517 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1518 "queuecommand : request rejected\n"));
1527 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1528 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1530 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1531 sizeof(struct scatterlist);
1534 static blk_status_t scsi_prepare_cmd(struct request *req)
1536 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1537 struct scsi_device *sdev = req->q->queuedata;
1538 struct Scsi_Host *shost = sdev->host;
1539 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1540 struct scatterlist *sg;
1542 scsi_init_command(sdev, cmd);
1547 cmd->prot_flags = 0;
1549 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1551 cmd->transfersize = 0;
1552 cmd->host_scribble = NULL;
1557 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1560 * Only clear the driver-private command data if the LLD does not supply
1561 * a function to initialize that data.
1563 if (!shost->hostt->init_cmd_priv)
1564 memset(cmd + 1, 0, shost->hostt->cmd_size);
1566 cmd->prot_op = SCSI_PROT_NORMAL;
1567 if (blk_rq_bytes(req))
1568 cmd->sc_data_direction = rq_dma_dir(req);
1570 cmd->sc_data_direction = DMA_NONE;
1572 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1573 cmd->sdb.table.sgl = sg;
1575 if (scsi_host_get_prot(shost)) {
1576 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1578 cmd->prot_sdb->table.sgl =
1579 (struct scatterlist *)(cmd->prot_sdb + 1);
1583 * Special handling for passthrough commands, which don't go to the ULP
1586 if (blk_rq_is_passthrough(req))
1587 return scsi_setup_scsi_cmnd(sdev, req);
1589 if (sdev->handler && sdev->handler->prep_fn) {
1590 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1592 if (ret != BLK_STS_OK)
1596 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1597 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1600 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1602 struct request *req = scsi_cmd_to_rq(cmd);
1604 switch (cmd->submitter) {
1605 case SUBMITTED_BY_BLOCK_LAYER:
1607 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1608 return scsi_eh_done(cmd);
1609 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1613 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1615 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1617 trace_scsi_dispatch_cmd_done(cmd);
1619 if (complete_directly)
1620 blk_mq_complete_request_direct(req, scsi_complete);
1622 blk_mq_complete_request(req);
1625 void scsi_done(struct scsi_cmnd *cmd)
1627 scsi_done_internal(cmd, false);
1629 EXPORT_SYMBOL(scsi_done);
1631 void scsi_done_direct(struct scsi_cmnd *cmd)
1633 scsi_done_internal(cmd, true);
1635 EXPORT_SYMBOL(scsi_done_direct);
1637 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1639 struct scsi_device *sdev = q->queuedata;
1641 sbitmap_put(&sdev->budget_map, budget_token);
1645 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1646 * not change behaviour from the previous unplug mechanism, experimentation
1647 * may prove this needs changing.
1649 #define SCSI_QUEUE_DELAY 3
1651 static int scsi_mq_get_budget(struct request_queue *q)
1653 struct scsi_device *sdev = q->queuedata;
1654 int token = scsi_dev_queue_ready(q, sdev);
1659 atomic_inc(&sdev->restarts);
1662 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1663 * .restarts must be incremented before .device_busy is read because the
1664 * code in scsi_run_queue_async() depends on the order of these operations.
1666 smp_mb__after_atomic();
1669 * If all in-flight requests originated from this LUN are completed
1670 * before reading .device_busy, sdev->device_busy will be observed as
1671 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1672 * soon. Otherwise, completion of one of these requests will observe
1673 * the .restarts flag, and the request queue will be run for handling
1674 * this request, see scsi_end_request().
1676 if (unlikely(scsi_device_busy(sdev) == 0 &&
1677 !scsi_device_blocked(sdev)))
1678 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1682 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1684 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1686 cmd->budget_token = token;
1689 static int scsi_mq_get_rq_budget_token(struct request *req)
1691 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1693 return cmd->budget_token;
1696 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1697 const struct blk_mq_queue_data *bd)
1699 struct request *req = bd->rq;
1700 struct request_queue *q = req->q;
1701 struct scsi_device *sdev = q->queuedata;
1702 struct Scsi_Host *shost = sdev->host;
1703 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1707 WARN_ON_ONCE(cmd->budget_token < 0);
1710 * If the device is not in running state we will reject some or all
1713 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1714 ret = scsi_device_state_check(sdev, req);
1715 if (ret != BLK_STS_OK)
1716 goto out_put_budget;
1719 ret = BLK_STS_RESOURCE;
1720 if (!scsi_target_queue_ready(shost, sdev))
1721 goto out_put_budget;
1722 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1723 goto out_dec_target_busy;
1725 if (!(req->rq_flags & RQF_DONTPREP)) {
1726 ret = scsi_prepare_cmd(req);
1727 if (ret != BLK_STS_OK)
1728 goto out_dec_host_busy;
1729 req->rq_flags |= RQF_DONTPREP;
1731 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1734 cmd->flags &= SCMD_PRESERVED_FLAGS;
1735 if (sdev->simple_tags)
1736 cmd->flags |= SCMD_TAGGED;
1738 cmd->flags |= SCMD_LAST;
1740 scsi_set_resid(cmd, 0);
1741 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1742 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1744 blk_mq_start_request(req);
1745 reason = scsi_dispatch_cmd(cmd);
1747 scsi_set_blocked(cmd, reason);
1748 ret = BLK_STS_RESOURCE;
1749 goto out_dec_host_busy;
1755 scsi_dec_host_busy(shost, cmd);
1756 out_dec_target_busy:
1757 if (scsi_target(sdev)->can_queue > 0)
1758 atomic_dec(&scsi_target(sdev)->target_busy);
1760 scsi_mq_put_budget(q, cmd->budget_token);
1761 cmd->budget_token = -1;
1765 case BLK_STS_RESOURCE:
1766 case BLK_STS_ZONE_RESOURCE:
1767 if (scsi_device_blocked(sdev))
1768 ret = BLK_STS_DEV_RESOURCE;
1771 cmd->result = DID_BUS_BUSY << 16;
1772 if (req->rq_flags & RQF_DONTPREP)
1773 scsi_mq_uninit_cmd(cmd);
1776 if (unlikely(!scsi_device_online(sdev)))
1777 cmd->result = DID_NO_CONNECT << 16;
1779 cmd->result = DID_ERROR << 16;
1781 * Make sure to release all allocated resources when
1782 * we hit an error, as we will never see this command
1785 if (req->rq_flags & RQF_DONTPREP)
1786 scsi_mq_uninit_cmd(cmd);
1787 scsi_run_queue_async(sdev);
1793 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1794 unsigned int hctx_idx, unsigned int numa_node)
1796 struct Scsi_Host *shost = set->driver_data;
1797 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1798 struct scatterlist *sg;
1802 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1803 if (!cmd->sense_buffer)
1806 if (scsi_host_get_prot(shost)) {
1807 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1808 shost->hostt->cmd_size;
1809 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1812 if (shost->hostt->init_cmd_priv) {
1813 ret = shost->hostt->init_cmd_priv(shost, cmd);
1815 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1821 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1822 unsigned int hctx_idx)
1824 struct Scsi_Host *shost = set->driver_data;
1825 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1827 if (shost->hostt->exit_cmd_priv)
1828 shost->hostt->exit_cmd_priv(shost, cmd);
1829 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1833 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1835 struct Scsi_Host *shost = hctx->driver_data;
1837 if (shost->hostt->mq_poll)
1838 return shost->hostt->mq_poll(shost, hctx->queue_num);
1843 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1844 unsigned int hctx_idx)
1846 struct Scsi_Host *shost = data;
1848 hctx->driver_data = shost;
1852 static int scsi_map_queues(struct blk_mq_tag_set *set)
1854 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1856 if (shost->hostt->map_queues)
1857 return shost->hostt->map_queues(shost);
1858 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1861 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1863 struct device *dev = shost->dma_dev;
1866 * this limit is imposed by hardware restrictions
1868 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1871 if (scsi_host_prot_dma(shost)) {
1872 shost->sg_prot_tablesize =
1873 min_not_zero(shost->sg_prot_tablesize,
1874 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1875 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1876 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1879 blk_queue_max_hw_sectors(q, shost->max_sectors);
1880 blk_queue_segment_boundary(q, shost->dma_boundary);
1881 dma_set_seg_boundary(dev, shost->dma_boundary);
1883 blk_queue_max_segment_size(q, shost->max_segment_size);
1884 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1885 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1888 * Set a reasonable default alignment: The larger of 32-byte (dword),
1889 * which is a common minimum for HBAs, and the minimum DMA alignment,
1890 * which is set by the platform.
1892 * Devices that require a bigger alignment can increase it later.
1894 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1896 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1898 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1899 .get_budget = scsi_mq_get_budget,
1900 .put_budget = scsi_mq_put_budget,
1901 .queue_rq = scsi_queue_rq,
1902 .complete = scsi_complete,
1903 .timeout = scsi_timeout,
1904 #ifdef CONFIG_BLK_DEBUG_FS
1905 .show_rq = scsi_show_rq,
1907 .init_request = scsi_mq_init_request,
1908 .exit_request = scsi_mq_exit_request,
1909 .cleanup_rq = scsi_cleanup_rq,
1910 .busy = scsi_mq_lld_busy,
1911 .map_queues = scsi_map_queues,
1912 .init_hctx = scsi_init_hctx,
1913 .poll = scsi_mq_poll,
1914 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1915 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1919 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1921 struct Scsi_Host *shost = hctx->driver_data;
1923 shost->hostt->commit_rqs(shost, hctx->queue_num);
1926 static const struct blk_mq_ops scsi_mq_ops = {
1927 .get_budget = scsi_mq_get_budget,
1928 .put_budget = scsi_mq_put_budget,
1929 .queue_rq = scsi_queue_rq,
1930 .commit_rqs = scsi_commit_rqs,
1931 .complete = scsi_complete,
1932 .timeout = scsi_timeout,
1933 #ifdef CONFIG_BLK_DEBUG_FS
1934 .show_rq = scsi_show_rq,
1936 .init_request = scsi_mq_init_request,
1937 .exit_request = scsi_mq_exit_request,
1938 .cleanup_rq = scsi_cleanup_rq,
1939 .busy = scsi_mq_lld_busy,
1940 .map_queues = scsi_map_queues,
1941 .init_hctx = scsi_init_hctx,
1942 .poll = scsi_mq_poll,
1943 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1944 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1947 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1949 unsigned int cmd_size, sgl_size;
1950 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1952 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1953 scsi_mq_inline_sgl_size(shost));
1954 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1955 if (scsi_host_get_prot(shost))
1956 cmd_size += sizeof(struct scsi_data_buffer) +
1957 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1959 memset(tag_set, 0, sizeof(*tag_set));
1960 if (shost->hostt->commit_rqs)
1961 tag_set->ops = &scsi_mq_ops;
1963 tag_set->ops = &scsi_mq_ops_no_commit;
1964 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1965 tag_set->nr_maps = shost->nr_maps ? : 1;
1966 tag_set->queue_depth = shost->can_queue;
1967 tag_set->cmd_size = cmd_size;
1968 tag_set->numa_node = dev_to_node(shost->dma_dev);
1969 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1971 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1972 tag_set->driver_data = shost;
1973 if (shost->host_tagset)
1974 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1976 return blk_mq_alloc_tag_set(tag_set);
1979 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1981 blk_mq_free_tag_set(&shost->tag_set);
1985 * scsi_device_from_queue - return sdev associated with a request_queue
1986 * @q: The request queue to return the sdev from
1988 * Return the sdev associated with a request queue or NULL if the
1989 * request_queue does not reference a SCSI device.
1991 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1993 struct scsi_device *sdev = NULL;
1995 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1996 q->mq_ops == &scsi_mq_ops)
1997 sdev = q->queuedata;
1998 if (!sdev || !get_device(&sdev->sdev_gendev))
2004 * pktcdvd should have been integrated into the SCSI layers, but for historical
2005 * reasons like the old IDE driver it isn't. This export allows it to safely
2006 * probe if a given device is a SCSI one and only attach to that.
2008 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2009 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2013 * scsi_block_requests - Utility function used by low-level drivers to prevent
2014 * further commands from being queued to the device.
2015 * @shost: host in question
2017 * There is no timer nor any other means by which the requests get unblocked
2018 * other than the low-level driver calling scsi_unblock_requests().
2020 void scsi_block_requests(struct Scsi_Host *shost)
2022 shost->host_self_blocked = 1;
2024 EXPORT_SYMBOL(scsi_block_requests);
2027 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2028 * further commands to be queued to the device.
2029 * @shost: host in question
2031 * There is no timer nor any other means by which the requests get unblocked
2032 * other than the low-level driver calling scsi_unblock_requests(). This is done
2033 * as an API function so that changes to the internals of the scsi mid-layer
2034 * won't require wholesale changes to drivers that use this feature.
2036 void scsi_unblock_requests(struct Scsi_Host *shost)
2038 shost->host_self_blocked = 0;
2039 scsi_run_host_queues(shost);
2041 EXPORT_SYMBOL(scsi_unblock_requests);
2043 void scsi_exit_queue(void)
2045 kmem_cache_destroy(scsi_sense_cache);
2049 * scsi_mode_select - issue a mode select
2050 * @sdev: SCSI device to be queried
2051 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2052 * @sp: Save page bit (0 == don't save, 1 == save)
2053 * @buffer: request buffer (may not be smaller than eight bytes)
2054 * @len: length of request buffer.
2055 * @timeout: command timeout
2056 * @retries: number of retries before failing
2057 * @data: returns a structure abstracting the mode header data
2058 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2059 * must be SCSI_SENSE_BUFFERSIZE big.
2061 * Returns zero if successful; negative error number or scsi
2065 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2066 unsigned char *buffer, int len, int timeout, int retries,
2067 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2069 unsigned char cmd[10];
2070 unsigned char *real_buffer;
2073 memset(cmd, 0, sizeof(cmd));
2074 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2077 * Use MODE SELECT(10) if the device asked for it or if the mode page
2078 * and the mode select header cannot fit within the maximumm 255 bytes
2079 * of the MODE SELECT(6) command.
2081 if (sdev->use_10_for_ms ||
2083 data->block_descriptor_length > 255) {
2084 if (len > 65535 - 8)
2086 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2089 memcpy(real_buffer + 8, buffer, len);
2093 real_buffer[2] = data->medium_type;
2094 real_buffer[3] = data->device_specific;
2095 real_buffer[4] = data->longlba ? 0x01 : 0;
2097 put_unaligned_be16(data->block_descriptor_length,
2100 cmd[0] = MODE_SELECT_10;
2101 put_unaligned_be16(len, &cmd[7]);
2106 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2109 memcpy(real_buffer + 4, buffer, len);
2112 real_buffer[1] = data->medium_type;
2113 real_buffer[2] = data->device_specific;
2114 real_buffer[3] = data->block_descriptor_length;
2116 cmd[0] = MODE_SELECT;
2120 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2121 sshdr, timeout, retries, NULL);
2125 EXPORT_SYMBOL_GPL(scsi_mode_select);
2128 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2129 * @sdev: SCSI device to be queried
2130 * @dbd: set to prevent mode sense from returning block descriptors
2131 * @modepage: mode page being requested
2132 * @buffer: request buffer (may not be smaller than eight bytes)
2133 * @len: length of request buffer.
2134 * @timeout: command timeout
2135 * @retries: number of retries before failing
2136 * @data: returns a structure abstracting the mode header data
2137 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2138 * must be SCSI_SENSE_BUFFERSIZE big.
2140 * Returns zero if successful, or a negative error number on failure
2143 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2144 unsigned char *buffer, int len, int timeout, int retries,
2145 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2147 unsigned char cmd[12];
2150 int result, retry_count = retries;
2151 struct scsi_sense_hdr my_sshdr;
2153 memset(data, 0, sizeof(*data));
2154 memset(&cmd[0], 0, 12);
2156 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2157 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2160 /* caller might not be interested in sense, but we need it */
2165 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2167 if (use_10_for_ms) {
2168 if (len < 8 || len > 65535)
2171 cmd[0] = MODE_SENSE_10;
2172 put_unaligned_be16(len, &cmd[7]);
2178 cmd[0] = MODE_SENSE;
2183 memset(buffer, 0, len);
2185 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2186 sshdr, timeout, retries, NULL);
2190 /* This code looks awful: what it's doing is making sure an
2191 * ILLEGAL REQUEST sense return identifies the actual command
2192 * byte as the problem. MODE_SENSE commands can return
2193 * ILLEGAL REQUEST if the code page isn't supported */
2195 if (!scsi_status_is_good(result)) {
2196 if (scsi_sense_valid(sshdr)) {
2197 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2198 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2200 * Invalid command operation code: retry using
2201 * MODE SENSE(6) if this was a MODE SENSE(10)
2202 * request, except if the request mode page is
2203 * too large for MODE SENSE single byte
2204 * allocation length field.
2206 if (use_10_for_ms) {
2209 sdev->use_10_for_ms = 0;
2213 if (scsi_status_is_check_condition(result) &&
2214 sshdr->sense_key == UNIT_ATTENTION &&
2222 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2223 (modepage == 6 || modepage == 8))) {
2224 /* Initio breakage? */
2227 data->medium_type = 0;
2228 data->device_specific = 0;
2230 data->block_descriptor_length = 0;
2231 } else if (use_10_for_ms) {
2232 data->length = get_unaligned_be16(&buffer[0]) + 2;
2233 data->medium_type = buffer[2];
2234 data->device_specific = buffer[3];
2235 data->longlba = buffer[4] & 0x01;
2236 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2238 data->length = buffer[0] + 1;
2239 data->medium_type = buffer[1];
2240 data->device_specific = buffer[2];
2241 data->block_descriptor_length = buffer[3];
2243 data->header_length = header_length;
2247 EXPORT_SYMBOL(scsi_mode_sense);
2250 * scsi_test_unit_ready - test if unit is ready
2251 * @sdev: scsi device to change the state of.
2252 * @timeout: command timeout
2253 * @retries: number of retries before failing
2254 * @sshdr: outpout pointer for decoded sense information.
2256 * Returns zero if unsuccessful or an error if TUR failed. For
2257 * removable media, UNIT_ATTENTION sets ->changed flag.
2260 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2261 struct scsi_sense_hdr *sshdr)
2264 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2268 /* try to eat the UNIT_ATTENTION if there are enough retries */
2270 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2272 if (sdev->removable && scsi_sense_valid(sshdr) &&
2273 sshdr->sense_key == UNIT_ATTENTION)
2275 } while (scsi_sense_valid(sshdr) &&
2276 sshdr->sense_key == UNIT_ATTENTION && --retries);
2280 EXPORT_SYMBOL(scsi_test_unit_ready);
2283 * scsi_device_set_state - Take the given device through the device state model.
2284 * @sdev: scsi device to change the state of.
2285 * @state: state to change to.
2287 * Returns zero if successful or an error if the requested
2288 * transition is illegal.
2291 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2293 enum scsi_device_state oldstate = sdev->sdev_state;
2295 if (state == oldstate)
2301 case SDEV_CREATED_BLOCK:
2312 case SDEV_TRANSPORT_OFFLINE:
2325 case SDEV_TRANSPORT_OFFLINE:
2333 case SDEV_TRANSPORT_OFFLINE:
2348 case SDEV_CREATED_BLOCK:
2357 case SDEV_CREATED_BLOCK:
2372 case SDEV_TRANSPORT_OFFLINE:
2384 case SDEV_TRANSPORT_OFFLINE:
2387 case SDEV_CREATED_BLOCK:
2395 sdev->offline_already = false;
2396 sdev->sdev_state = state;
2400 SCSI_LOG_ERROR_RECOVERY(1,
2401 sdev_printk(KERN_ERR, sdev,
2402 "Illegal state transition %s->%s",
2403 scsi_device_state_name(oldstate),
2404 scsi_device_state_name(state))
2408 EXPORT_SYMBOL(scsi_device_set_state);
2411 * scsi_evt_emit - emit a single SCSI device uevent
2412 * @sdev: associated SCSI device
2413 * @evt: event to emit
2415 * Send a single uevent (scsi_event) to the associated scsi_device.
2417 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2422 switch (evt->evt_type) {
2423 case SDEV_EVT_MEDIA_CHANGE:
2424 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2426 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2427 scsi_rescan_device(&sdev->sdev_gendev);
2428 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2430 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2431 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2433 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2434 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2436 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2437 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2439 case SDEV_EVT_LUN_CHANGE_REPORTED:
2440 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2442 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2443 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2445 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2446 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2455 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2459 * scsi_evt_thread - send a uevent for each scsi event
2460 * @work: work struct for scsi_device
2462 * Dispatch queued events to their associated scsi_device kobjects
2465 void scsi_evt_thread(struct work_struct *work)
2467 struct scsi_device *sdev;
2468 enum scsi_device_event evt_type;
2469 LIST_HEAD(event_list);
2471 sdev = container_of(work, struct scsi_device, event_work);
2473 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2474 if (test_and_clear_bit(evt_type, sdev->pending_events))
2475 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2478 struct scsi_event *evt;
2479 struct list_head *this, *tmp;
2480 unsigned long flags;
2482 spin_lock_irqsave(&sdev->list_lock, flags);
2483 list_splice_init(&sdev->event_list, &event_list);
2484 spin_unlock_irqrestore(&sdev->list_lock, flags);
2486 if (list_empty(&event_list))
2489 list_for_each_safe(this, tmp, &event_list) {
2490 evt = list_entry(this, struct scsi_event, node);
2491 list_del(&evt->node);
2492 scsi_evt_emit(sdev, evt);
2499 * sdev_evt_send - send asserted event to uevent thread
2500 * @sdev: scsi_device event occurred on
2501 * @evt: event to send
2503 * Assert scsi device event asynchronously.
2505 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2507 unsigned long flags;
2510 /* FIXME: currently this check eliminates all media change events
2511 * for polled devices. Need to update to discriminate between AN
2512 * and polled events */
2513 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2519 spin_lock_irqsave(&sdev->list_lock, flags);
2520 list_add_tail(&evt->node, &sdev->event_list);
2521 schedule_work(&sdev->event_work);
2522 spin_unlock_irqrestore(&sdev->list_lock, flags);
2524 EXPORT_SYMBOL_GPL(sdev_evt_send);
2527 * sdev_evt_alloc - allocate a new scsi event
2528 * @evt_type: type of event to allocate
2529 * @gfpflags: GFP flags for allocation
2531 * Allocates and returns a new scsi_event.
2533 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2536 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2540 evt->evt_type = evt_type;
2541 INIT_LIST_HEAD(&evt->node);
2543 /* evt_type-specific initialization, if any */
2545 case SDEV_EVT_MEDIA_CHANGE:
2546 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2547 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2548 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2549 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2550 case SDEV_EVT_LUN_CHANGE_REPORTED:
2551 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2552 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2560 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2563 * sdev_evt_send_simple - send asserted event to uevent thread
2564 * @sdev: scsi_device event occurred on
2565 * @evt_type: type of event to send
2566 * @gfpflags: GFP flags for allocation
2568 * Assert scsi device event asynchronously, given an event type.
2570 void sdev_evt_send_simple(struct scsi_device *sdev,
2571 enum scsi_device_event evt_type, gfp_t gfpflags)
2573 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2575 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2580 sdev_evt_send(sdev, evt);
2582 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2585 * scsi_device_quiesce - Block all commands except power management.
2586 * @sdev: scsi device to quiesce.
2588 * This works by trying to transition to the SDEV_QUIESCE state
2589 * (which must be a legal transition). When the device is in this
2590 * state, only power management requests will be accepted, all others will
2593 * Must be called with user context, may sleep.
2595 * Returns zero if unsuccessful or an error if not.
2598 scsi_device_quiesce(struct scsi_device *sdev)
2600 struct request_queue *q = sdev->request_queue;
2604 * It is allowed to call scsi_device_quiesce() multiple times from
2605 * the same context but concurrent scsi_device_quiesce() calls are
2608 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2610 if (sdev->quiesced_by == current)
2615 blk_mq_freeze_queue(q);
2617 * Ensure that the effect of blk_set_pm_only() will be visible
2618 * for percpu_ref_tryget() callers that occur after the queue
2619 * unfreeze even if the queue was already frozen before this function
2620 * was called. See also https://lwn.net/Articles/573497/.
2623 blk_mq_unfreeze_queue(q);
2625 mutex_lock(&sdev->state_mutex);
2626 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2628 sdev->quiesced_by = current;
2630 blk_clear_pm_only(q);
2631 mutex_unlock(&sdev->state_mutex);
2635 EXPORT_SYMBOL(scsi_device_quiesce);
2638 * scsi_device_resume - Restart user issued commands to a quiesced device.
2639 * @sdev: scsi device to resume.
2641 * Moves the device from quiesced back to running and restarts the
2644 * Must be called with user context, may sleep.
2646 void scsi_device_resume(struct scsi_device *sdev)
2648 /* check if the device state was mutated prior to resume, and if
2649 * so assume the state is being managed elsewhere (for example
2650 * device deleted during suspend)
2652 mutex_lock(&sdev->state_mutex);
2653 if (sdev->sdev_state == SDEV_QUIESCE)
2654 scsi_device_set_state(sdev, SDEV_RUNNING);
2655 if (sdev->quiesced_by) {
2656 sdev->quiesced_by = NULL;
2657 blk_clear_pm_only(sdev->request_queue);
2659 mutex_unlock(&sdev->state_mutex);
2661 EXPORT_SYMBOL(scsi_device_resume);
2664 device_quiesce_fn(struct scsi_device *sdev, void *data)
2666 scsi_device_quiesce(sdev);
2670 scsi_target_quiesce(struct scsi_target *starget)
2672 starget_for_each_device(starget, NULL, device_quiesce_fn);
2674 EXPORT_SYMBOL(scsi_target_quiesce);
2677 device_resume_fn(struct scsi_device *sdev, void *data)
2679 scsi_device_resume(sdev);
2683 scsi_target_resume(struct scsi_target *starget)
2685 starget_for_each_device(starget, NULL, device_resume_fn);
2687 EXPORT_SYMBOL(scsi_target_resume);
2689 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2691 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2692 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2697 void scsi_start_queue(struct scsi_device *sdev)
2699 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2700 blk_mq_unquiesce_queue(sdev->request_queue);
2703 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2706 * The atomic variable of ->queue_stopped covers that
2707 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2709 * However, we still need to wait until quiesce is done
2710 * in case that queue has been stopped.
2712 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2714 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2716 blk_mq_quiesce_queue(sdev->request_queue);
2719 blk_mq_wait_quiesce_done(sdev->request_queue);
2724 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2725 * @sdev: device to block
2727 * Pause SCSI command processing on the specified device. Does not sleep.
2729 * Returns zero if successful or a negative error code upon failure.
2732 * This routine transitions the device to the SDEV_BLOCK state (which must be
2733 * a legal transition). When the device is in this state, command processing
2734 * is paused until the device leaves the SDEV_BLOCK state. See also
2735 * scsi_internal_device_unblock_nowait().
2737 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2739 int ret = __scsi_internal_device_block_nowait(sdev);
2742 * The device has transitioned to SDEV_BLOCK. Stop the
2743 * block layer from calling the midlayer with this device's
2747 scsi_stop_queue(sdev, true);
2750 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2753 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2754 * @sdev: device to block
2756 * Pause SCSI command processing on the specified device and wait until all
2757 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2759 * Returns zero if successful or a negative error code upon failure.
2762 * This routine transitions the device to the SDEV_BLOCK state (which must be
2763 * a legal transition). When the device is in this state, command processing
2764 * is paused until the device leaves the SDEV_BLOCK state. See also
2765 * scsi_internal_device_unblock().
2767 static int scsi_internal_device_block(struct scsi_device *sdev)
2771 mutex_lock(&sdev->state_mutex);
2772 err = __scsi_internal_device_block_nowait(sdev);
2774 scsi_stop_queue(sdev, false);
2775 mutex_unlock(&sdev->state_mutex);
2781 * scsi_internal_device_unblock_nowait - resume a device after a block request
2782 * @sdev: device to resume
2783 * @new_state: state to set the device to after unblocking
2785 * Restart the device queue for a previously suspended SCSI device. Does not
2788 * Returns zero if successful or a negative error code upon failure.
2791 * This routine transitions the device to the SDEV_RUNNING state or to one of
2792 * the offline states (which must be a legal transition) allowing the midlayer
2793 * to goose the queue for this device.
2795 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2796 enum scsi_device_state new_state)
2798 switch (new_state) {
2800 case SDEV_TRANSPORT_OFFLINE:
2807 * Try to transition the scsi device to SDEV_RUNNING or one of the
2808 * offlined states and goose the device queue if successful.
2810 switch (sdev->sdev_state) {
2812 case SDEV_TRANSPORT_OFFLINE:
2813 sdev->sdev_state = new_state;
2815 case SDEV_CREATED_BLOCK:
2816 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2817 new_state == SDEV_OFFLINE)
2818 sdev->sdev_state = new_state;
2820 sdev->sdev_state = SDEV_CREATED;
2828 scsi_start_queue(sdev);
2832 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2835 * scsi_internal_device_unblock - resume a device after a block request
2836 * @sdev: device to resume
2837 * @new_state: state to set the device to after unblocking
2839 * Restart the device queue for a previously suspended SCSI device. May sleep.
2841 * Returns zero if successful or a negative error code upon failure.
2844 * This routine transitions the device to the SDEV_RUNNING state or to one of
2845 * the offline states (which must be a legal transition) allowing the midlayer
2846 * to goose the queue for this device.
2848 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2849 enum scsi_device_state new_state)
2853 mutex_lock(&sdev->state_mutex);
2854 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2855 mutex_unlock(&sdev->state_mutex);
2861 device_block(struct scsi_device *sdev, void *data)
2865 ret = scsi_internal_device_block(sdev);
2867 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2868 dev_name(&sdev->sdev_gendev), ret);
2872 target_block(struct device *dev, void *data)
2874 if (scsi_is_target_device(dev))
2875 starget_for_each_device(to_scsi_target(dev), NULL,
2881 scsi_target_block(struct device *dev)
2883 if (scsi_is_target_device(dev))
2884 starget_for_each_device(to_scsi_target(dev), NULL,
2887 device_for_each_child(dev, NULL, target_block);
2889 EXPORT_SYMBOL_GPL(scsi_target_block);
2892 device_unblock(struct scsi_device *sdev, void *data)
2894 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2898 target_unblock(struct device *dev, void *data)
2900 if (scsi_is_target_device(dev))
2901 starget_for_each_device(to_scsi_target(dev), data,
2907 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2909 if (scsi_is_target_device(dev))
2910 starget_for_each_device(to_scsi_target(dev), &new_state,
2913 device_for_each_child(dev, &new_state, target_unblock);
2915 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2918 scsi_host_block(struct Scsi_Host *shost)
2920 struct scsi_device *sdev;
2924 * Call scsi_internal_device_block_nowait so we can avoid
2925 * calling synchronize_rcu() for each LUN.
2927 shost_for_each_device(sdev, shost) {
2928 mutex_lock(&sdev->state_mutex);
2929 ret = scsi_internal_device_block_nowait(sdev);
2930 mutex_unlock(&sdev->state_mutex);
2932 scsi_device_put(sdev);
2938 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2939 * calling synchronize_rcu() once is enough.
2941 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2948 EXPORT_SYMBOL_GPL(scsi_host_block);
2951 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2953 struct scsi_device *sdev;
2956 shost_for_each_device(sdev, shost) {
2957 ret = scsi_internal_device_unblock(sdev, new_state);
2959 scsi_device_put(sdev);
2965 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2968 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2969 * @sgl: scatter-gather list
2970 * @sg_count: number of segments in sg
2971 * @offset: offset in bytes into sg, on return offset into the mapped area
2972 * @len: bytes to map, on return number of bytes mapped
2974 * Returns virtual address of the start of the mapped page
2976 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2977 size_t *offset, size_t *len)
2980 size_t sg_len = 0, len_complete = 0;
2981 struct scatterlist *sg;
2984 WARN_ON(!irqs_disabled());
2986 for_each_sg(sgl, sg, sg_count, i) {
2987 len_complete = sg_len; /* Complete sg-entries */
2988 sg_len += sg->length;
2989 if (sg_len > *offset)
2993 if (unlikely(i == sg_count)) {
2994 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2996 __func__, sg_len, *offset, sg_count);
3001 /* Offset starting from the beginning of first page in this sg-entry */
3002 *offset = *offset - len_complete + sg->offset;
3004 /* Assumption: contiguous pages can be accessed as "page + i" */
3005 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3006 *offset &= ~PAGE_MASK;
3008 /* Bytes in this sg-entry from *offset to the end of the page */
3009 sg_len = PAGE_SIZE - *offset;
3013 return kmap_atomic(page);
3015 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3018 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3019 * @virt: virtual address to be unmapped
3021 void scsi_kunmap_atomic_sg(void *virt)
3023 kunmap_atomic(virt);
3025 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3027 void sdev_disable_disk_events(struct scsi_device *sdev)
3029 atomic_inc(&sdev->disk_events_disable_depth);
3031 EXPORT_SYMBOL(sdev_disable_disk_events);
3033 void sdev_enable_disk_events(struct scsi_device *sdev)
3035 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3037 atomic_dec(&sdev->disk_events_disable_depth);
3039 EXPORT_SYMBOL(sdev_enable_disk_events);
3041 static unsigned char designator_prio(const unsigned char *d)
3044 /* not associated with LUN */
3048 /* invalid length */
3052 * Order of preference for lun descriptor:
3053 * - SCSI name string
3054 * - NAA IEEE Registered Extended
3055 * - EUI-64 based 16-byte
3056 * - EUI-64 based 12-byte
3057 * - NAA IEEE Registered
3058 * - NAA IEEE Extended
3059 * - EUI-64 based 8-byte
3060 * - SCSI name string (truncated)
3062 * as longer descriptors reduce the likelyhood
3063 * of identification clashes.
3066 switch (d[1] & 0xf) {
3068 /* SCSI name string, variable-length UTF-8 */
3071 switch (d[4] >> 4) {
3073 /* NAA registered extended */
3076 /* NAA registered */
3082 /* NAA locally assigned */
3091 /* EUI64-based, 16 byte */
3094 /* EUI64-based, 12 byte */
3097 /* EUI64-based, 8 byte */
3114 * scsi_vpd_lun_id - return a unique device identification
3115 * @sdev: SCSI device
3116 * @id: buffer for the identification
3117 * @id_len: length of the buffer
3119 * Copies a unique device identification into @id based
3120 * on the information in the VPD page 0x83 of the device.
3121 * The string will be formatted as a SCSI name string.
3123 * Returns the length of the identification or error on failure.
3124 * If the identifier is longer than the supplied buffer the actual
3125 * identifier length is returned and the buffer is not zero-padded.
3127 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3131 const unsigned char *d, *cur_id_str;
3132 const struct scsi_vpd *vpd_pg83;
3133 int id_size = -EINVAL;
3136 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3142 /* The id string must be at least 20 bytes + terminating NULL byte */
3148 memset(id, 0, id_len);
3149 for (d = vpd_pg83->data + 4;
3150 d < vpd_pg83->data + vpd_pg83->len;
3152 u8 prio = designator_prio(d);
3154 if (prio == 0 || cur_id_prio > prio)
3157 switch (d[1] & 0xf) {
3160 if (cur_id_size > d[3])
3164 if (cur_id_size + 4 > id_len)
3165 cur_id_size = id_len - 4;
3167 id_size = snprintf(id, id_len, "t10.%*pE",
3168 cur_id_size, cur_id_str);
3175 switch (cur_id_size) {
3177 id_size = snprintf(id, id_len,
3182 id_size = snprintf(id, id_len,
3187 id_size = snprintf(id, id_len,
3200 switch (cur_id_size) {
3202 id_size = snprintf(id, id_len,
3207 id_size = snprintf(id, id_len,
3216 /* SCSI name string */
3217 if (cur_id_size > d[3])
3219 /* Prefer others for truncated descriptor */
3220 if (d[3] > id_len) {
3222 if (cur_id_prio > prio)
3226 cur_id_size = id_size = d[3];
3228 if (cur_id_size >= id_len)
3229 cur_id_size = id_len - 1;
3230 memcpy(id, cur_id_str, cur_id_size);
3240 EXPORT_SYMBOL(scsi_vpd_lun_id);
3243 * scsi_vpd_tpg_id - return a target port group identifier
3244 * @sdev: SCSI device
3246 * Returns the Target Port Group identifier from the information
3247 * froom VPD page 0x83 of the device.
3249 * Returns the identifier or error on failure.
3251 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3253 const unsigned char *d;
3254 const struct scsi_vpd *vpd_pg83;
3255 int group_id = -EAGAIN, rel_port = -1;
3258 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3264 d = vpd_pg83->data + 4;
3265 while (d < vpd_pg83->data + vpd_pg83->len) {
3266 switch (d[1] & 0xf) {
3268 /* Relative target port */
3269 rel_port = get_unaligned_be16(&d[6]);
3272 /* Target port group */
3273 group_id = get_unaligned_be16(&d[6]);
3282 if (group_id >= 0 && rel_id && rel_port != -1)
3287 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3290 * scsi_build_sense - build sense data for a command
3291 * @scmd: scsi command for which the sense should be formatted
3292 * @desc: Sense format (non-zero == descriptor format,
3293 * 0 == fixed format)
3295 * @asc: Additional sense code
3296 * @ascq: Additional sense code qualifier
3299 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3301 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3302 scmd->result = SAM_STAT_CHECK_CONDITION;
3304 EXPORT_SYMBOL_GPL(scsi_build_sense);