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, unsigned long msecs)
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);
126 blk_mq_requeue_request(rq, false);
127 blk_mq_delay_kick_requeue_list(rq->q, msecs);
129 blk_mq_requeue_request(rq, true);
133 * __scsi_queue_insert - private queue insertion
134 * @cmd: The SCSI command being requeued
135 * @reason: The reason for the requeue
136 * @unbusy: Whether the queue should be unbusied
138 * This is a private queue insertion. The public interface
139 * scsi_queue_insert() always assumes the queue should be unbusied
140 * because it's always called before the completion. This function is
141 * for a requeue after completion, which should only occur in this
144 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
146 struct scsi_device *device = cmd->device;
148 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
149 "Inserting command %p into mlqueue\n", cmd));
151 scsi_set_blocked(cmd, reason);
154 * Decrement the counters, since these commands are no longer
155 * active on the host/device.
158 scsi_device_unbusy(device, cmd);
161 * Requeue this command. It will go before all other commands
162 * that are already in the queue. Schedule requeue work under
163 * lock such that the kblockd_schedule_work() call happens
164 * before blk_mq_destroy_queue() finishes.
168 blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
172 * scsi_queue_insert - Reinsert a command in the queue.
173 * @cmd: command that we are adding to queue.
174 * @reason: why we are inserting command to queue.
176 * We do this for one of two cases. Either the host is busy and it cannot accept
177 * any more commands for the time being, or the device returned QUEUE_FULL and
178 * can accept no more commands.
180 * Context: This could be called either from an interrupt context or a normal
183 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
185 __scsi_queue_insert(cmd, reason, true);
190 * __scsi_execute - insert request and wait for the result
193 * @data_direction: data direction
194 * @buffer: data buffer
195 * @bufflen: len of buffer
196 * @sense: optional sense buffer
197 * @sshdr: optional decoded sense header
198 * @timeout: request timeout in HZ
199 * @retries: number of times to retry request
200 * @flags: flags for ->cmd_flags
201 * @rq_flags: flags for ->rq_flags
202 * @resid: optional residual length
204 * Returns the scsi_cmnd result field if a command was executed, or a negative
205 * Linux error code if we didn't get that far.
207 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
208 int data_direction, void *buffer, unsigned bufflen,
209 unsigned char *sense, struct scsi_sense_hdr *sshdr,
210 int timeout, int retries, blk_opf_t flags,
211 req_flags_t rq_flags, int *resid)
214 struct scsi_cmnd *scmd;
217 req = scsi_alloc_request(sdev->request_queue,
218 data_direction == DMA_TO_DEVICE ?
219 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
220 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
225 ret = blk_rq_map_kern(sdev->request_queue, req,
226 buffer, bufflen, GFP_NOIO);
230 scmd = blk_mq_rq_to_pdu(req);
231 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
232 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
233 scmd->allowed = retries;
234 req->timeout = timeout;
235 req->cmd_flags |= flags;
236 req->rq_flags |= rq_flags | RQF_QUIET;
239 * head injection *required* here otherwise quiesce won't work
241 blk_execute_rq(req, true);
244 * Some devices (USB mass-storage in particular) may transfer
245 * garbage data together with a residue indicating that the data
246 * is invalid. Prevent the garbage from being misinterpreted
247 * and prevent security leaks by zeroing out the excess data.
249 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
250 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
253 *resid = scmd->resid_len;
254 if (sense && scmd->sense_len)
255 memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
257 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
261 blk_mq_free_request(req);
265 EXPORT_SYMBOL(__scsi_execute);
268 * Wake up the error handler if necessary. Avoid as follows that the error
269 * handler is not woken up if host in-flight requests number ==
270 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
271 * with an RCU read lock in this function to ensure that this function in
272 * its entirety either finishes before scsi_eh_scmd_add() increases the
273 * host_failed counter or that it notices the shost state change made by
274 * scsi_eh_scmd_add().
276 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
281 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
282 if (unlikely(scsi_host_in_recovery(shost))) {
283 spin_lock_irqsave(shost->host_lock, flags);
284 if (shost->host_failed || shost->host_eh_scheduled)
285 scsi_eh_wakeup(shost);
286 spin_unlock_irqrestore(shost->host_lock, flags);
291 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
293 struct Scsi_Host *shost = sdev->host;
294 struct scsi_target *starget = scsi_target(sdev);
296 scsi_dec_host_busy(shost, cmd);
298 if (starget->can_queue > 0)
299 atomic_dec(&starget->target_busy);
301 sbitmap_put(&sdev->budget_map, cmd->budget_token);
302 cmd->budget_token = -1;
305 static void scsi_kick_queue(struct request_queue *q)
307 blk_mq_run_hw_queues(q, false);
311 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
312 * and call blk_run_queue for all the scsi_devices on the target -
313 * including current_sdev first.
315 * Called with *no* scsi locks held.
317 static void scsi_single_lun_run(struct scsi_device *current_sdev)
319 struct Scsi_Host *shost = current_sdev->host;
320 struct scsi_device *sdev, *tmp;
321 struct scsi_target *starget = scsi_target(current_sdev);
324 spin_lock_irqsave(shost->host_lock, flags);
325 starget->starget_sdev_user = NULL;
326 spin_unlock_irqrestore(shost->host_lock, flags);
329 * Call blk_run_queue for all LUNs on the target, starting with
330 * current_sdev. We race with others (to set starget_sdev_user),
331 * but in most cases, we will be first. Ideally, each LU on the
332 * target would get some limited time or requests on the target.
334 scsi_kick_queue(current_sdev->request_queue);
336 spin_lock_irqsave(shost->host_lock, flags);
337 if (starget->starget_sdev_user)
339 list_for_each_entry_safe(sdev, tmp, &starget->devices,
340 same_target_siblings) {
341 if (sdev == current_sdev)
343 if (scsi_device_get(sdev))
346 spin_unlock_irqrestore(shost->host_lock, flags);
347 scsi_kick_queue(sdev->request_queue);
348 spin_lock_irqsave(shost->host_lock, flags);
350 scsi_device_put(sdev);
353 spin_unlock_irqrestore(shost->host_lock, flags);
356 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
358 if (scsi_device_busy(sdev) >= sdev->queue_depth)
360 if (atomic_read(&sdev->device_blocked) > 0)
365 static inline bool scsi_target_is_busy(struct scsi_target *starget)
367 if (starget->can_queue > 0) {
368 if (atomic_read(&starget->target_busy) >= starget->can_queue)
370 if (atomic_read(&starget->target_blocked) > 0)
376 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
378 if (atomic_read(&shost->host_blocked) > 0)
380 if (shost->host_self_blocked)
385 static void scsi_starved_list_run(struct Scsi_Host *shost)
387 LIST_HEAD(starved_list);
388 struct scsi_device *sdev;
391 spin_lock_irqsave(shost->host_lock, flags);
392 list_splice_init(&shost->starved_list, &starved_list);
394 while (!list_empty(&starved_list)) {
395 struct request_queue *slq;
398 * As long as shost is accepting commands and we have
399 * starved queues, call blk_run_queue. scsi_request_fn
400 * drops the queue_lock and can add us back to the
403 * host_lock protects the starved_list and starved_entry.
404 * scsi_request_fn must get the host_lock before checking
405 * or modifying starved_list or starved_entry.
407 if (scsi_host_is_busy(shost))
410 sdev = list_entry(starved_list.next,
411 struct scsi_device, starved_entry);
412 list_del_init(&sdev->starved_entry);
413 if (scsi_target_is_busy(scsi_target(sdev))) {
414 list_move_tail(&sdev->starved_entry,
415 &shost->starved_list);
420 * Once we drop the host lock, a racing scsi_remove_device()
421 * call may remove the sdev from the starved list and destroy
422 * it and the queue. Mitigate by taking a reference to the
423 * queue and never touching the sdev again after we drop the
424 * host lock. Note: if __scsi_remove_device() invokes
425 * blk_mq_destroy_queue() before the queue is run from this
426 * function then blk_run_queue() will return immediately since
427 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
429 slq = sdev->request_queue;
430 if (!blk_get_queue(slq))
432 spin_unlock_irqrestore(shost->host_lock, flags);
434 scsi_kick_queue(slq);
437 spin_lock_irqsave(shost->host_lock, flags);
439 /* put any unprocessed entries back */
440 list_splice(&starved_list, &shost->starved_list);
441 spin_unlock_irqrestore(shost->host_lock, flags);
445 * scsi_run_queue - Select a proper request queue to serve next.
446 * @q: last request's queue
448 * The previous command was completely finished, start a new one if possible.
450 static void scsi_run_queue(struct request_queue *q)
452 struct scsi_device *sdev = q->queuedata;
454 if (scsi_target(sdev)->single_lun)
455 scsi_single_lun_run(sdev);
456 if (!list_empty(&sdev->host->starved_list))
457 scsi_starved_list_run(sdev->host);
459 blk_mq_run_hw_queues(q, false);
462 void scsi_requeue_run_queue(struct work_struct *work)
464 struct scsi_device *sdev;
465 struct request_queue *q;
467 sdev = container_of(work, struct scsi_device, requeue_work);
468 q = sdev->request_queue;
472 void scsi_run_host_queues(struct Scsi_Host *shost)
474 struct scsi_device *sdev;
476 shost_for_each_device(sdev, shost)
477 scsi_run_queue(sdev->request_queue);
480 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
482 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
483 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
485 if (drv->uninit_command)
486 drv->uninit_command(cmd);
490 void scsi_free_sgtables(struct scsi_cmnd *cmd)
492 if (cmd->sdb.table.nents)
493 sg_free_table_chained(&cmd->sdb.table,
495 if (scsi_prot_sg_count(cmd))
496 sg_free_table_chained(&cmd->prot_sdb->table,
497 SCSI_INLINE_PROT_SG_CNT);
499 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
501 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
503 scsi_free_sgtables(cmd);
504 scsi_uninit_cmd(cmd);
507 static void scsi_run_queue_async(struct scsi_device *sdev)
509 if (scsi_target(sdev)->single_lun ||
510 !list_empty(&sdev->host->starved_list)) {
511 kblockd_schedule_work(&sdev->requeue_work);
514 * smp_mb() present in sbitmap_queue_clear() or implied in
515 * .end_io is for ordering writing .device_busy in
516 * scsi_device_unbusy() and reading sdev->restarts.
518 int old = atomic_read(&sdev->restarts);
521 * ->restarts has to be kept as non-zero if new budget
524 * No need to run queue when either another re-run
525 * queue wins in updating ->restarts or a new budget
528 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
529 blk_mq_run_hw_queues(sdev->request_queue, true);
533 /* Returns false when no more bytes to process, true if there are more */
534 static bool scsi_end_request(struct request *req, blk_status_t error,
537 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
538 struct scsi_device *sdev = cmd->device;
539 struct request_queue *q = sdev->request_queue;
541 if (blk_update_request(req, error, bytes))
545 if (blk_queue_add_random(q))
546 add_disk_randomness(req->q->disk);
548 if (!blk_rq_is_passthrough(req)) {
549 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
550 cmd->flags &= ~SCMD_INITIALIZED;
554 * Calling rcu_barrier() is not necessary here because the
555 * SCSI error handler guarantees that the function called by
556 * call_rcu() has been called before scsi_end_request() is
559 destroy_rcu_head(&cmd->rcu);
562 * In the MQ case the command gets freed by __blk_mq_end_request,
563 * so we have to do all cleanup that depends on it earlier.
565 * We also can't kick the queues from irq context, so we
566 * will have to defer it to a workqueue.
568 scsi_mq_uninit_cmd(cmd);
571 * queue is still alive, so grab the ref for preventing it
572 * from being cleaned up during running queue.
574 percpu_ref_get(&q->q_usage_counter);
576 __blk_mq_end_request(req, error);
578 scsi_run_queue_async(sdev);
580 percpu_ref_put(&q->q_usage_counter);
584 static inline u8 get_scsi_ml_byte(int result)
586 return (result >> 8) & 0xff;
590 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
591 * @result: scsi error code
593 * Translate a SCSI result code into a blk_status_t value.
595 static blk_status_t scsi_result_to_blk_status(int result)
598 * Check the scsi-ml byte first in case we converted a host or status
601 switch (get_scsi_ml_byte(result)) {
604 case SCSIML_STAT_RESV_CONFLICT:
605 return BLK_STS_NEXUS;
606 case SCSIML_STAT_NOSPC:
607 return BLK_STS_NOSPC;
608 case SCSIML_STAT_MED_ERROR:
609 return BLK_STS_MEDIUM;
610 case SCSIML_STAT_TGT_FAILURE:
611 return BLK_STS_TARGET;
614 switch (host_byte(result)) {
616 if (scsi_status_is_good(result))
618 return BLK_STS_IOERR;
619 case DID_TRANSPORT_FAILFAST:
620 case DID_TRANSPORT_MARGINAL:
621 return BLK_STS_TRANSPORT;
623 return BLK_STS_IOERR;
628 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
629 * @rq: request to examine
632 * A request could be merge of IOs which require different failure
633 * handling. This function determines the number of bytes which
634 * can be failed from the beginning of the request without
635 * crossing into area which need to be retried further.
638 * The number of bytes to fail.
640 static unsigned int scsi_rq_err_bytes(const struct request *rq)
642 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
643 unsigned int bytes = 0;
646 if (!(rq->rq_flags & RQF_MIXED_MERGE))
647 return blk_rq_bytes(rq);
650 * Currently the only 'mixing' which can happen is between
651 * different fastfail types. We can safely fail portions
652 * which have all the failfast bits that the first one has -
653 * the ones which are at least as eager to fail as the first
656 for (bio = rq->bio; bio; bio = bio->bi_next) {
657 if ((bio->bi_opf & ff) != ff)
659 bytes += bio->bi_iter.bi_size;
662 /* this could lead to infinite loop */
663 BUG_ON(blk_rq_bytes(rq) && !bytes);
667 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
669 struct request *req = scsi_cmd_to_rq(cmd);
670 unsigned long wait_for;
672 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
675 wait_for = (cmd->allowed + 1) * req->timeout;
676 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
677 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
685 * When ALUA transition state is returned, reprep the cmd to
686 * use the ALUA handler's transition timeout. Delay the reprep
687 * 1 sec to avoid aggressive retries of the target in that
690 #define ALUA_TRANSITION_REPREP_DELAY 1000
692 /* Helper for scsi_io_completion() when special action required. */
693 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
695 struct request *req = scsi_cmd_to_rq(cmd);
697 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
698 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
699 struct scsi_sense_hdr sshdr;
701 bool sense_current = true; /* false implies "deferred sense" */
702 blk_status_t blk_stat;
704 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
706 sense_current = !scsi_sense_is_deferred(&sshdr);
708 blk_stat = scsi_result_to_blk_status(result);
710 if (host_byte(result) == DID_RESET) {
711 /* Third party bus reset or reset for error recovery
712 * reasons. Just retry the command and see what
715 action = ACTION_RETRY;
716 } else if (sense_valid && sense_current) {
717 switch (sshdr.sense_key) {
719 if (cmd->device->removable) {
720 /* Detected disc change. Set a bit
721 * and quietly refuse further access.
723 cmd->device->changed = 1;
724 action = ACTION_FAIL;
726 /* Must have been a power glitch, or a
727 * bus reset. Could not have been a
728 * media change, so we just retry the
729 * command and see what happens.
731 action = ACTION_RETRY;
734 case ILLEGAL_REQUEST:
735 /* If we had an ILLEGAL REQUEST returned, then
736 * we may have performed an unsupported
737 * command. The only thing this should be
738 * would be a ten byte read where only a six
739 * byte read was supported. Also, on a system
740 * where READ CAPACITY failed, we may have
741 * read past the end of the disk.
743 if ((cmd->device->use_10_for_rw &&
744 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
745 (cmd->cmnd[0] == READ_10 ||
746 cmd->cmnd[0] == WRITE_10)) {
747 /* This will issue a new 6-byte command. */
748 cmd->device->use_10_for_rw = 0;
749 action = ACTION_REPREP;
750 } else if (sshdr.asc == 0x10) /* DIX */ {
751 action = ACTION_FAIL;
752 blk_stat = BLK_STS_PROTECTION;
753 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
754 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
755 action = ACTION_FAIL;
756 blk_stat = BLK_STS_TARGET;
758 action = ACTION_FAIL;
760 case ABORTED_COMMAND:
761 action = ACTION_FAIL;
762 if (sshdr.asc == 0x10) /* DIF */
763 blk_stat = BLK_STS_PROTECTION;
766 /* If the device is in the process of becoming
767 * ready, or has a temporary blockage, retry.
769 if (sshdr.asc == 0x04) {
770 switch (sshdr.ascq) {
771 case 0x01: /* becoming ready */
772 case 0x04: /* format in progress */
773 case 0x05: /* rebuild in progress */
774 case 0x06: /* recalculation in progress */
775 case 0x07: /* operation in progress */
776 case 0x08: /* Long write in progress */
777 case 0x09: /* self test in progress */
778 case 0x11: /* notify (enable spinup) required */
779 case 0x14: /* space allocation in progress */
780 case 0x1a: /* start stop unit in progress */
781 case 0x1b: /* sanitize in progress */
782 case 0x1d: /* configuration in progress */
783 case 0x24: /* depopulation in progress */
784 action = ACTION_DELAYED_RETRY;
786 case 0x0a: /* ALUA state transition */
787 action = ACTION_DELAYED_REPREP;
790 action = ACTION_FAIL;
794 action = ACTION_FAIL;
796 case VOLUME_OVERFLOW:
797 /* See SSC3rXX or current. */
798 action = ACTION_FAIL;
801 action = ACTION_FAIL;
802 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
803 (sshdr.asc == 0x55 &&
804 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
805 /* Insufficient zone resources */
806 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
810 action = ACTION_FAIL;
814 action = ACTION_FAIL;
816 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
817 action = ACTION_FAIL;
821 /* Give up and fail the remainder of the request */
822 if (!(req->rq_flags & RQF_QUIET)) {
823 static DEFINE_RATELIMIT_STATE(_rs,
824 DEFAULT_RATELIMIT_INTERVAL,
825 DEFAULT_RATELIMIT_BURST);
827 if (unlikely(scsi_logging_level))
829 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
830 SCSI_LOG_MLCOMPLETE_BITS);
833 * if logging is enabled the failure will be printed
834 * in scsi_log_completion(), so avoid duplicate messages
836 if (!level && __ratelimit(&_rs)) {
837 scsi_print_result(cmd, NULL, FAILED);
839 scsi_print_sense(cmd);
840 scsi_print_command(cmd);
843 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
847 scsi_mq_requeue_cmd(cmd, 0);
849 case ACTION_DELAYED_REPREP:
850 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
853 /* Retry the same command immediately */
854 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
856 case ACTION_DELAYED_RETRY:
857 /* Retry the same command after a delay */
858 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
864 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
865 * new result that may suppress further error checking. Also modifies
866 * *blk_statp in some cases.
868 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
869 blk_status_t *blk_statp)
872 bool sense_current = true; /* false implies "deferred sense" */
873 struct request *req = scsi_cmd_to_rq(cmd);
874 struct scsi_sense_hdr sshdr;
876 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
878 sense_current = !scsi_sense_is_deferred(&sshdr);
880 if (blk_rq_is_passthrough(req)) {
883 * SG_IO wants current and deferred errors
885 cmd->sense_len = min(8 + cmd->sense_buffer[7],
886 SCSI_SENSE_BUFFERSIZE);
889 *blk_statp = scsi_result_to_blk_status(result);
890 } else if (blk_rq_bytes(req) == 0 && sense_current) {
892 * Flush commands do not transfers any data, and thus cannot use
893 * good_bytes != blk_rq_bytes(req) as the signal for an error.
894 * This sets *blk_statp explicitly for the problem case.
896 *blk_statp = scsi_result_to_blk_status(result);
899 * Recovered errors need reporting, but they're always treated as
900 * success, so fiddle the result code here. For passthrough requests
901 * we already took a copy of the original into sreq->result which
902 * is what gets returned to the user
904 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
905 bool do_print = true;
907 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
908 * skip print since caller wants ATA registers. Only occurs
909 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
911 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
913 else if (req->rq_flags & RQF_QUIET)
916 scsi_print_sense(cmd);
918 /* for passthrough, *blk_statp may be set */
919 *blk_statp = BLK_STS_OK;
922 * Another corner case: the SCSI status byte is non-zero but 'good'.
923 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
924 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
925 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
926 * intermediate statuses (both obsolete in SAM-4) as good.
928 if ((result & 0xff) && scsi_status_is_good(result)) {
930 *blk_statp = BLK_STS_OK;
936 * scsi_io_completion - Completion processing for SCSI commands.
937 * @cmd: command that is finished.
938 * @good_bytes: number of processed bytes.
940 * We will finish off the specified number of sectors. If we are done, the
941 * command block will be released and the queue function will be goosed. If we
942 * are not done then we have to figure out what to do next:
944 * a) We can call scsi_mq_requeue_cmd(). The request will be
945 * unprepared and put back on the queue. Then a new command will
946 * be created for it. This should be used if we made forward
947 * progress, or if we want to switch from READ(10) to READ(6) for
950 * b) We can call scsi_io_completion_action(). The request will be
951 * put back on the queue and retried using the same command as
952 * before, possibly after a delay.
954 * c) We can call scsi_end_request() with blk_stat other than
955 * BLK_STS_OK, to fail the remainder of the request.
957 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
959 int result = cmd->result;
960 struct request *req = scsi_cmd_to_rq(cmd);
961 blk_status_t blk_stat = BLK_STS_OK;
963 if (unlikely(result)) /* a nz result may or may not be an error */
964 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
967 * Next deal with any sectors which we were able to correctly
970 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
971 "%u sectors total, %d bytes done.\n",
972 blk_rq_sectors(req), good_bytes));
975 * Failed, zero length commands always need to drop down
976 * to retry code. Fast path should return in this block.
978 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
979 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
980 return; /* no bytes remaining */
983 /* Kill remainder if no retries. */
984 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
985 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
987 "Bytes remaining after failed, no-retry command");
992 * If there had been no error, but we have leftover bytes in the
993 * request just queue the command up again.
995 if (likely(result == 0))
996 scsi_mq_requeue_cmd(cmd, 0);
998 scsi_io_completion_action(cmd, result);
1001 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1004 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1005 !op_is_write(req_op(rq)) &&
1006 sdev->host->hostt->dma_need_drain(rq);
1010 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1011 * @cmd: SCSI command data structure to initialize.
1013 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1017 * * BLK_STS_OK - on success
1018 * * BLK_STS_RESOURCE - if the failure is retryable
1019 * * BLK_STS_IOERR - if the failure is fatal
1021 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1023 struct scsi_device *sdev = cmd->device;
1024 struct request *rq = scsi_cmd_to_rq(cmd);
1025 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1026 struct scatterlist *last_sg = NULL;
1028 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1031 if (WARN_ON_ONCE(!nr_segs))
1032 return BLK_STS_IOERR;
1035 * Make sure there is space for the drain. The driver must adjust
1036 * max_hw_segments to be prepared for this.
1042 * If sg table allocation fails, requeue request later.
1044 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1045 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1046 return BLK_STS_RESOURCE;
1049 * Next, walk the list, and fill in the addresses and sizes of
1052 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1054 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1055 unsigned int pad_len =
1056 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1058 last_sg->length += pad_len;
1059 cmd->extra_len += pad_len;
1063 sg_unmark_end(last_sg);
1064 last_sg = sg_next(last_sg);
1065 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1066 sg_mark_end(last_sg);
1068 cmd->extra_len += sdev->dma_drain_len;
1072 BUG_ON(count > cmd->sdb.table.nents);
1073 cmd->sdb.table.nents = count;
1074 cmd->sdb.length = blk_rq_payload_bytes(rq);
1076 if (blk_integrity_rq(rq)) {
1077 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1080 if (WARN_ON_ONCE(!prot_sdb)) {
1082 * This can happen if someone (e.g. multipath)
1083 * queues a command to a device on an adapter
1084 * that does not support DIX.
1086 ret = BLK_STS_IOERR;
1087 goto out_free_sgtables;
1090 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1092 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1093 prot_sdb->table.sgl,
1094 SCSI_INLINE_PROT_SG_CNT)) {
1095 ret = BLK_STS_RESOURCE;
1096 goto out_free_sgtables;
1099 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1100 prot_sdb->table.sgl);
1101 BUG_ON(count > ivecs);
1102 BUG_ON(count > queue_max_integrity_segments(rq->q));
1104 cmd->prot_sdb = prot_sdb;
1105 cmd->prot_sdb->table.nents = count;
1110 scsi_free_sgtables(cmd);
1113 EXPORT_SYMBOL(scsi_alloc_sgtables);
1116 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1117 * @rq: Request associated with the SCSI command to be initialized.
1119 * This function initializes the members of struct scsi_cmnd that must be
1120 * initialized before request processing starts and that won't be
1121 * reinitialized if a SCSI command is requeued.
1123 static void scsi_initialize_rq(struct request *rq)
1125 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1127 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1128 cmd->cmd_len = MAX_COMMAND_SIZE;
1130 init_rcu_head(&cmd->rcu);
1131 cmd->jiffies_at_alloc = jiffies;
1135 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1136 blk_mq_req_flags_t flags)
1140 rq = blk_mq_alloc_request(q, opf, flags);
1142 scsi_initialize_rq(rq);
1145 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1148 * Only called when the request isn't completed by SCSI, and not freed by
1151 static void scsi_cleanup_rq(struct request *rq)
1153 if (rq->rq_flags & RQF_DONTPREP) {
1154 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1155 rq->rq_flags &= ~RQF_DONTPREP;
1159 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1160 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1162 struct request *rq = scsi_cmd_to_rq(cmd);
1164 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1165 cmd->flags |= SCMD_INITIALIZED;
1166 scsi_initialize_rq(rq);
1170 INIT_LIST_HEAD(&cmd->eh_entry);
1171 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1174 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1175 struct request *req)
1177 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1180 * Passthrough requests may transfer data, in which case they must
1181 * a bio attached to them. Or they might contain a SCSI command
1182 * that does not transfer data, in which case they may optionally
1183 * submit a request without an attached bio.
1186 blk_status_t ret = scsi_alloc_sgtables(cmd);
1187 if (unlikely(ret != BLK_STS_OK))
1190 BUG_ON(blk_rq_bytes(req));
1192 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1195 cmd->transfersize = blk_rq_bytes(req);
1200 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1202 switch (sdev->sdev_state) {
1206 case SDEV_TRANSPORT_OFFLINE:
1208 * If the device is offline we refuse to process any
1209 * commands. The device must be brought online
1210 * before trying any recovery commands.
1212 if (!sdev->offline_already) {
1213 sdev->offline_already = true;
1214 sdev_printk(KERN_ERR, sdev,
1215 "rejecting I/O to offline device\n");
1217 return BLK_STS_IOERR;
1220 * If the device is fully deleted, we refuse to
1221 * process any commands as well.
1223 sdev_printk(KERN_ERR, sdev,
1224 "rejecting I/O to dead device\n");
1225 return BLK_STS_IOERR;
1227 case SDEV_CREATED_BLOCK:
1228 return BLK_STS_RESOURCE;
1231 * If the device is blocked we only accept power management
1234 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1235 return BLK_STS_RESOURCE;
1239 * For any other not fully online state we only allow
1240 * power management commands.
1242 if (req && !(req->rq_flags & RQF_PM))
1243 return BLK_STS_OFFLINE;
1249 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1250 * and return the token else return -1.
1252 static inline int scsi_dev_queue_ready(struct request_queue *q,
1253 struct scsi_device *sdev)
1257 token = sbitmap_get(&sdev->budget_map);
1258 if (atomic_read(&sdev->device_blocked)) {
1262 if (scsi_device_busy(sdev) > 1)
1266 * unblock after device_blocked iterates to zero
1268 if (atomic_dec_return(&sdev->device_blocked) > 0)
1270 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1271 "unblocking device at zero depth\n"));
1277 sbitmap_put(&sdev->budget_map, token);
1283 * scsi_target_queue_ready: checks if there we can send commands to target
1284 * @sdev: scsi device on starget to check.
1286 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1287 struct scsi_device *sdev)
1289 struct scsi_target *starget = scsi_target(sdev);
1292 if (starget->single_lun) {
1293 spin_lock_irq(shost->host_lock);
1294 if (starget->starget_sdev_user &&
1295 starget->starget_sdev_user != sdev) {
1296 spin_unlock_irq(shost->host_lock);
1299 starget->starget_sdev_user = sdev;
1300 spin_unlock_irq(shost->host_lock);
1303 if (starget->can_queue <= 0)
1306 busy = atomic_inc_return(&starget->target_busy) - 1;
1307 if (atomic_read(&starget->target_blocked) > 0) {
1312 * unblock after target_blocked iterates to zero
1314 if (atomic_dec_return(&starget->target_blocked) > 0)
1317 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1318 "unblocking target at zero depth\n"));
1321 if (busy >= starget->can_queue)
1327 spin_lock_irq(shost->host_lock);
1328 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1329 spin_unlock_irq(shost->host_lock);
1331 if (starget->can_queue > 0)
1332 atomic_dec(&starget->target_busy);
1337 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1338 * return 0. We must end up running the queue again whenever 0 is
1339 * returned, else IO can hang.
1341 static inline int scsi_host_queue_ready(struct request_queue *q,
1342 struct Scsi_Host *shost,
1343 struct scsi_device *sdev,
1344 struct scsi_cmnd *cmd)
1346 if (scsi_host_in_recovery(shost))
1349 if (atomic_read(&shost->host_blocked) > 0) {
1350 if (scsi_host_busy(shost) > 0)
1354 * unblock after host_blocked iterates to zero
1356 if (atomic_dec_return(&shost->host_blocked) > 0)
1360 shost_printk(KERN_INFO, shost,
1361 "unblocking host at zero depth\n"));
1364 if (shost->host_self_blocked)
1367 /* We're OK to process the command, so we can't be starved */
1368 if (!list_empty(&sdev->starved_entry)) {
1369 spin_lock_irq(shost->host_lock);
1370 if (!list_empty(&sdev->starved_entry))
1371 list_del_init(&sdev->starved_entry);
1372 spin_unlock_irq(shost->host_lock);
1375 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1380 spin_lock_irq(shost->host_lock);
1381 if (list_empty(&sdev->starved_entry))
1382 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1383 spin_unlock_irq(shost->host_lock);
1385 scsi_dec_host_busy(shost, cmd);
1390 * Busy state exporting function for request stacking drivers.
1392 * For efficiency, no lock is taken to check the busy state of
1393 * shost/starget/sdev, since the returned value is not guaranteed and
1394 * may be changed after request stacking drivers call the function,
1395 * regardless of taking lock or not.
1397 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1398 * needs to return 'not busy'. Otherwise, request stacking drivers
1399 * may hold requests forever.
1401 static bool scsi_mq_lld_busy(struct request_queue *q)
1403 struct scsi_device *sdev = q->queuedata;
1404 struct Scsi_Host *shost;
1406 if (blk_queue_dying(q))
1412 * Ignore host/starget busy state.
1413 * Since block layer does not have a concept of fairness across
1414 * multiple queues, congestion of host/starget needs to be handled
1417 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1424 * Block layer request completion callback. May be called from interrupt
1427 static void scsi_complete(struct request *rq)
1429 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1430 enum scsi_disposition disposition;
1432 INIT_LIST_HEAD(&cmd->eh_entry);
1434 atomic_inc(&cmd->device->iodone_cnt);
1436 atomic_inc(&cmd->device->ioerr_cnt);
1438 disposition = scsi_decide_disposition(cmd);
1439 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1440 disposition = SUCCESS;
1442 scsi_log_completion(cmd, disposition);
1444 switch (disposition) {
1446 scsi_finish_command(cmd);
1449 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1451 case ADD_TO_MLQUEUE:
1452 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1455 scsi_eh_scmd_add(cmd);
1461 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1462 * @cmd: command block we are dispatching.
1464 * Return: nonzero return request was rejected and device's queue needs to be
1467 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1469 struct Scsi_Host *host = cmd->device->host;
1472 atomic_inc(&cmd->device->iorequest_cnt);
1474 /* check if the device is still usable */
1475 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1476 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1477 * returns an immediate error upwards, and signals
1478 * that the device is no longer present */
1479 cmd->result = DID_NO_CONNECT << 16;
1483 /* Check to see if the scsi lld made this device blocked. */
1484 if (unlikely(scsi_device_blocked(cmd->device))) {
1486 * in blocked state, the command is just put back on
1487 * the device queue. The suspend state has already
1488 * blocked the queue so future requests should not
1489 * occur until the device transitions out of the
1492 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1493 "queuecommand : device blocked\n"));
1494 return SCSI_MLQUEUE_DEVICE_BUSY;
1497 /* Store the LUN value in cmnd, if needed. */
1498 if (cmd->device->lun_in_cdb)
1499 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1500 (cmd->device->lun << 5 & 0xe0);
1505 * Before we queue this command, check if the command
1506 * length exceeds what the host adapter can handle.
1508 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1509 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1510 "queuecommand : command too long. "
1511 "cdb_size=%d host->max_cmd_len=%d\n",
1512 cmd->cmd_len, cmd->device->host->max_cmd_len));
1513 cmd->result = (DID_ABORT << 16);
1517 if (unlikely(host->shost_state == SHOST_DEL)) {
1518 cmd->result = (DID_NO_CONNECT << 16);
1523 trace_scsi_dispatch_cmd_start(cmd);
1524 rtn = host->hostt->queuecommand(host, cmd);
1526 trace_scsi_dispatch_cmd_error(cmd, rtn);
1527 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1528 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1529 rtn = SCSI_MLQUEUE_HOST_BUSY;
1531 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1532 "queuecommand : request rejected\n"));
1541 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1542 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1544 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1545 sizeof(struct scatterlist);
1548 static blk_status_t scsi_prepare_cmd(struct request *req)
1550 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1551 struct scsi_device *sdev = req->q->queuedata;
1552 struct Scsi_Host *shost = sdev->host;
1553 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1554 struct scatterlist *sg;
1556 scsi_init_command(sdev, cmd);
1560 cmd->prot_flags = 0;
1562 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1564 cmd->transfersize = 0;
1565 cmd->host_scribble = NULL;
1570 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1573 * Only clear the driver-private command data if the LLD does not supply
1574 * a function to initialize that data.
1576 if (!shost->hostt->init_cmd_priv)
1577 memset(cmd + 1, 0, shost->hostt->cmd_size);
1579 cmd->prot_op = SCSI_PROT_NORMAL;
1580 if (blk_rq_bytes(req))
1581 cmd->sc_data_direction = rq_dma_dir(req);
1583 cmd->sc_data_direction = DMA_NONE;
1585 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1586 cmd->sdb.table.sgl = sg;
1588 if (scsi_host_get_prot(shost)) {
1589 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1591 cmd->prot_sdb->table.sgl =
1592 (struct scatterlist *)(cmd->prot_sdb + 1);
1596 * Special handling for passthrough commands, which don't go to the ULP
1599 if (blk_rq_is_passthrough(req))
1600 return scsi_setup_scsi_cmnd(sdev, req);
1602 if (sdev->handler && sdev->handler->prep_fn) {
1603 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1605 if (ret != BLK_STS_OK)
1609 /* Usually overridden by the ULP */
1611 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1612 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1615 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1617 struct request *req = scsi_cmd_to_rq(cmd);
1619 switch (cmd->submitter) {
1620 case SUBMITTED_BY_BLOCK_LAYER:
1622 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1623 return scsi_eh_done(cmd);
1624 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1628 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1630 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1632 trace_scsi_dispatch_cmd_done(cmd);
1634 if (complete_directly)
1635 blk_mq_complete_request_direct(req, scsi_complete);
1637 blk_mq_complete_request(req);
1640 void scsi_done(struct scsi_cmnd *cmd)
1642 scsi_done_internal(cmd, false);
1644 EXPORT_SYMBOL(scsi_done);
1646 void scsi_done_direct(struct scsi_cmnd *cmd)
1648 scsi_done_internal(cmd, true);
1650 EXPORT_SYMBOL(scsi_done_direct);
1652 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1654 struct scsi_device *sdev = q->queuedata;
1656 sbitmap_put(&sdev->budget_map, budget_token);
1660 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1661 * not change behaviour from the previous unplug mechanism, experimentation
1662 * may prove this needs changing.
1664 #define SCSI_QUEUE_DELAY 3
1666 static int scsi_mq_get_budget(struct request_queue *q)
1668 struct scsi_device *sdev = q->queuedata;
1669 int token = scsi_dev_queue_ready(q, sdev);
1674 atomic_inc(&sdev->restarts);
1677 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1678 * .restarts must be incremented before .device_busy is read because the
1679 * code in scsi_run_queue_async() depends on the order of these operations.
1681 smp_mb__after_atomic();
1684 * If all in-flight requests originated from this LUN are completed
1685 * before reading .device_busy, sdev->device_busy will be observed as
1686 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1687 * soon. Otherwise, completion of one of these requests will observe
1688 * the .restarts flag, and the request queue will be run for handling
1689 * this request, see scsi_end_request().
1691 if (unlikely(scsi_device_busy(sdev) == 0 &&
1692 !scsi_device_blocked(sdev)))
1693 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1697 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1699 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1701 cmd->budget_token = token;
1704 static int scsi_mq_get_rq_budget_token(struct request *req)
1706 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1708 return cmd->budget_token;
1711 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1712 const struct blk_mq_queue_data *bd)
1714 struct request *req = bd->rq;
1715 struct request_queue *q = req->q;
1716 struct scsi_device *sdev = q->queuedata;
1717 struct Scsi_Host *shost = sdev->host;
1718 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1722 WARN_ON_ONCE(cmd->budget_token < 0);
1725 * If the device is not in running state we will reject some or all
1728 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1729 ret = scsi_device_state_check(sdev, req);
1730 if (ret != BLK_STS_OK)
1731 goto out_put_budget;
1734 ret = BLK_STS_RESOURCE;
1735 if (!scsi_target_queue_ready(shost, sdev))
1736 goto out_put_budget;
1737 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1738 goto out_dec_target_busy;
1740 if (!(req->rq_flags & RQF_DONTPREP)) {
1741 ret = scsi_prepare_cmd(req);
1742 if (ret != BLK_STS_OK)
1743 goto out_dec_host_busy;
1744 req->rq_flags |= RQF_DONTPREP;
1746 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1749 cmd->flags &= SCMD_PRESERVED_FLAGS;
1750 if (sdev->simple_tags)
1751 cmd->flags |= SCMD_TAGGED;
1753 cmd->flags |= SCMD_LAST;
1755 scsi_set_resid(cmd, 0);
1756 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1757 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1759 blk_mq_start_request(req);
1760 reason = scsi_dispatch_cmd(cmd);
1762 scsi_set_blocked(cmd, reason);
1763 ret = BLK_STS_RESOURCE;
1764 goto out_dec_host_busy;
1770 scsi_dec_host_busy(shost, cmd);
1771 out_dec_target_busy:
1772 if (scsi_target(sdev)->can_queue > 0)
1773 atomic_dec(&scsi_target(sdev)->target_busy);
1775 scsi_mq_put_budget(q, cmd->budget_token);
1776 cmd->budget_token = -1;
1780 case BLK_STS_RESOURCE:
1781 case BLK_STS_ZONE_RESOURCE:
1782 if (scsi_device_blocked(sdev))
1783 ret = BLK_STS_DEV_RESOURCE;
1786 cmd->result = DID_BUS_BUSY << 16;
1787 if (req->rq_flags & RQF_DONTPREP)
1788 scsi_mq_uninit_cmd(cmd);
1791 if (unlikely(!scsi_device_online(sdev)))
1792 cmd->result = DID_NO_CONNECT << 16;
1794 cmd->result = DID_ERROR << 16;
1796 * Make sure to release all allocated resources when
1797 * we hit an error, as we will never see this command
1800 if (req->rq_flags & RQF_DONTPREP)
1801 scsi_mq_uninit_cmd(cmd);
1802 scsi_run_queue_async(sdev);
1808 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1809 unsigned int hctx_idx, unsigned int numa_node)
1811 struct Scsi_Host *shost = set->driver_data;
1812 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1813 struct scatterlist *sg;
1817 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1818 if (!cmd->sense_buffer)
1821 if (scsi_host_get_prot(shost)) {
1822 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1823 shost->hostt->cmd_size;
1824 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1827 if (shost->hostt->init_cmd_priv) {
1828 ret = shost->hostt->init_cmd_priv(shost, cmd);
1830 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1836 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1837 unsigned int hctx_idx)
1839 struct Scsi_Host *shost = set->driver_data;
1840 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1842 if (shost->hostt->exit_cmd_priv)
1843 shost->hostt->exit_cmd_priv(shost, cmd);
1844 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1848 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1850 struct Scsi_Host *shost = hctx->driver_data;
1852 if (shost->hostt->mq_poll)
1853 return shost->hostt->mq_poll(shost, hctx->queue_num);
1858 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1859 unsigned int hctx_idx)
1861 struct Scsi_Host *shost = data;
1863 hctx->driver_data = shost;
1867 static void scsi_map_queues(struct blk_mq_tag_set *set)
1869 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1871 if (shost->hostt->map_queues)
1872 return shost->hostt->map_queues(shost);
1873 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1876 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1878 struct device *dev = shost->dma_dev;
1881 * this limit is imposed by hardware restrictions
1883 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1886 if (scsi_host_prot_dma(shost)) {
1887 shost->sg_prot_tablesize =
1888 min_not_zero(shost->sg_prot_tablesize,
1889 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1890 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1891 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1894 blk_queue_max_hw_sectors(q, shost->max_sectors);
1895 blk_queue_segment_boundary(q, shost->dma_boundary);
1896 dma_set_seg_boundary(dev, shost->dma_boundary);
1898 blk_queue_max_segment_size(q, shost->max_segment_size);
1899 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1900 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1903 * Set a reasonable default alignment: The larger of 32-byte (dword),
1904 * which is a common minimum for HBAs, and the minimum DMA alignment,
1905 * which is set by the platform.
1907 * Devices that require a bigger alignment can increase it later.
1909 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1911 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1913 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1914 .get_budget = scsi_mq_get_budget,
1915 .put_budget = scsi_mq_put_budget,
1916 .queue_rq = scsi_queue_rq,
1917 .complete = scsi_complete,
1918 .timeout = scsi_timeout,
1919 #ifdef CONFIG_BLK_DEBUG_FS
1920 .show_rq = scsi_show_rq,
1922 .init_request = scsi_mq_init_request,
1923 .exit_request = scsi_mq_exit_request,
1924 .cleanup_rq = scsi_cleanup_rq,
1925 .busy = scsi_mq_lld_busy,
1926 .map_queues = scsi_map_queues,
1927 .init_hctx = scsi_init_hctx,
1928 .poll = scsi_mq_poll,
1929 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1930 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1934 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1936 struct Scsi_Host *shost = hctx->driver_data;
1938 shost->hostt->commit_rqs(shost, hctx->queue_num);
1941 static const struct blk_mq_ops scsi_mq_ops = {
1942 .get_budget = scsi_mq_get_budget,
1943 .put_budget = scsi_mq_put_budget,
1944 .queue_rq = scsi_queue_rq,
1945 .commit_rqs = scsi_commit_rqs,
1946 .complete = scsi_complete,
1947 .timeout = scsi_timeout,
1948 #ifdef CONFIG_BLK_DEBUG_FS
1949 .show_rq = scsi_show_rq,
1951 .init_request = scsi_mq_init_request,
1952 .exit_request = scsi_mq_exit_request,
1953 .cleanup_rq = scsi_cleanup_rq,
1954 .busy = scsi_mq_lld_busy,
1955 .map_queues = scsi_map_queues,
1956 .init_hctx = scsi_init_hctx,
1957 .poll = scsi_mq_poll,
1958 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1959 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1962 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1964 unsigned int cmd_size, sgl_size;
1965 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1967 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1968 scsi_mq_inline_sgl_size(shost));
1969 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1970 if (scsi_host_get_prot(shost))
1971 cmd_size += sizeof(struct scsi_data_buffer) +
1972 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1974 memset(tag_set, 0, sizeof(*tag_set));
1975 if (shost->hostt->commit_rqs)
1976 tag_set->ops = &scsi_mq_ops;
1978 tag_set->ops = &scsi_mq_ops_no_commit;
1979 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1980 tag_set->nr_maps = shost->nr_maps ? : 1;
1981 tag_set->queue_depth = shost->can_queue;
1982 tag_set->cmd_size = cmd_size;
1983 tag_set->numa_node = dev_to_node(shost->dma_dev);
1984 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1986 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1987 tag_set->driver_data = shost;
1988 if (shost->host_tagset)
1989 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1991 return blk_mq_alloc_tag_set(tag_set);
1994 void scsi_mq_free_tags(struct kref *kref)
1996 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
1999 blk_mq_free_tag_set(&shost->tag_set);
2000 complete(&shost->tagset_freed);
2004 * scsi_device_from_queue - return sdev associated with a request_queue
2005 * @q: The request queue to return the sdev from
2007 * Return the sdev associated with a request queue or NULL if the
2008 * request_queue does not reference a SCSI device.
2010 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2012 struct scsi_device *sdev = NULL;
2014 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2015 q->mq_ops == &scsi_mq_ops)
2016 sdev = q->queuedata;
2017 if (!sdev || !get_device(&sdev->sdev_gendev))
2023 * pktcdvd should have been integrated into the SCSI layers, but for historical
2024 * reasons like the old IDE driver it isn't. This export allows it to safely
2025 * probe if a given device is a SCSI one and only attach to that.
2027 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2028 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2032 * scsi_block_requests - Utility function used by low-level drivers to prevent
2033 * further commands from being queued to the device.
2034 * @shost: host in question
2036 * There is no timer nor any other means by which the requests get unblocked
2037 * other than the low-level driver calling scsi_unblock_requests().
2039 void scsi_block_requests(struct Scsi_Host *shost)
2041 shost->host_self_blocked = 1;
2043 EXPORT_SYMBOL(scsi_block_requests);
2046 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2047 * further commands to be queued to the device.
2048 * @shost: host in question
2050 * There is no timer nor any other means by which the requests get unblocked
2051 * other than the low-level driver calling scsi_unblock_requests(). This is done
2052 * as an API function so that changes to the internals of the scsi mid-layer
2053 * won't require wholesale changes to drivers that use this feature.
2055 void scsi_unblock_requests(struct Scsi_Host *shost)
2057 shost->host_self_blocked = 0;
2058 scsi_run_host_queues(shost);
2060 EXPORT_SYMBOL(scsi_unblock_requests);
2062 void scsi_exit_queue(void)
2064 kmem_cache_destroy(scsi_sense_cache);
2068 * scsi_mode_select - issue a mode select
2069 * @sdev: SCSI device to be queried
2070 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2071 * @sp: Save page bit (0 == don't save, 1 == save)
2072 * @buffer: request buffer (may not be smaller than eight bytes)
2073 * @len: length of request buffer.
2074 * @timeout: command timeout
2075 * @retries: number of retries before failing
2076 * @data: returns a structure abstracting the mode header data
2077 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2078 * must be SCSI_SENSE_BUFFERSIZE big.
2080 * Returns zero if successful; negative error number or scsi
2084 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2085 unsigned char *buffer, int len, int timeout, int retries,
2086 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2088 unsigned char cmd[10];
2089 unsigned char *real_buffer;
2092 memset(cmd, 0, sizeof(cmd));
2093 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2096 * Use MODE SELECT(10) if the device asked for it or if the mode page
2097 * and the mode select header cannot fit within the maximumm 255 bytes
2098 * of the MODE SELECT(6) command.
2100 if (sdev->use_10_for_ms ||
2102 data->block_descriptor_length > 255) {
2103 if (len > 65535 - 8)
2105 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2108 memcpy(real_buffer + 8, buffer, len);
2112 real_buffer[2] = data->medium_type;
2113 real_buffer[3] = data->device_specific;
2114 real_buffer[4] = data->longlba ? 0x01 : 0;
2116 put_unaligned_be16(data->block_descriptor_length,
2119 cmd[0] = MODE_SELECT_10;
2120 put_unaligned_be16(len, &cmd[7]);
2125 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2128 memcpy(real_buffer + 4, buffer, len);
2131 real_buffer[1] = data->medium_type;
2132 real_buffer[2] = data->device_specific;
2133 real_buffer[3] = data->block_descriptor_length;
2135 cmd[0] = MODE_SELECT;
2139 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2140 sshdr, timeout, retries, NULL);
2144 EXPORT_SYMBOL_GPL(scsi_mode_select);
2147 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2148 * @sdev: SCSI device to be queried
2149 * @dbd: set to prevent mode sense from returning block descriptors
2150 * @modepage: mode page being requested
2151 * @buffer: request buffer (may not be smaller than eight bytes)
2152 * @len: length of request buffer.
2153 * @timeout: command timeout
2154 * @retries: number of retries before failing
2155 * @data: returns a structure abstracting the mode header data
2156 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2157 * must be SCSI_SENSE_BUFFERSIZE big.
2159 * Returns zero if successful, or a negative error number on failure
2162 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2163 unsigned char *buffer, int len, int timeout, int retries,
2164 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2166 unsigned char cmd[12];
2169 int result, retry_count = retries;
2170 struct scsi_sense_hdr my_sshdr;
2172 memset(data, 0, sizeof(*data));
2173 memset(&cmd[0], 0, 12);
2175 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2176 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2179 /* caller might not be interested in sense, but we need it */
2184 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2186 if (use_10_for_ms) {
2187 if (len < 8 || len > 65535)
2190 cmd[0] = MODE_SENSE_10;
2191 put_unaligned_be16(len, &cmd[7]);
2197 cmd[0] = MODE_SENSE;
2202 memset(buffer, 0, len);
2204 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2205 sshdr, timeout, retries, NULL);
2209 /* This code looks awful: what it's doing is making sure an
2210 * ILLEGAL REQUEST sense return identifies the actual command
2211 * byte as the problem. MODE_SENSE commands can return
2212 * ILLEGAL REQUEST if the code page isn't supported */
2214 if (!scsi_status_is_good(result)) {
2215 if (scsi_sense_valid(sshdr)) {
2216 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2217 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2219 * Invalid command operation code: retry using
2220 * MODE SENSE(6) if this was a MODE SENSE(10)
2221 * request, except if the request mode page is
2222 * too large for MODE SENSE single byte
2223 * allocation length field.
2225 if (use_10_for_ms) {
2228 sdev->use_10_for_ms = 0;
2232 if (scsi_status_is_check_condition(result) &&
2233 sshdr->sense_key == UNIT_ATTENTION &&
2241 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2242 (modepage == 6 || modepage == 8))) {
2243 /* Initio breakage? */
2246 data->medium_type = 0;
2247 data->device_specific = 0;
2249 data->block_descriptor_length = 0;
2250 } else if (use_10_for_ms) {
2251 data->length = get_unaligned_be16(&buffer[0]) + 2;
2252 data->medium_type = buffer[2];
2253 data->device_specific = buffer[3];
2254 data->longlba = buffer[4] & 0x01;
2255 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2257 data->length = buffer[0] + 1;
2258 data->medium_type = buffer[1];
2259 data->device_specific = buffer[2];
2260 data->block_descriptor_length = buffer[3];
2262 data->header_length = header_length;
2266 EXPORT_SYMBOL(scsi_mode_sense);
2269 * scsi_test_unit_ready - test if unit is ready
2270 * @sdev: scsi device to change the state of.
2271 * @timeout: command timeout
2272 * @retries: number of retries before failing
2273 * @sshdr: outpout pointer for decoded sense information.
2275 * Returns zero if unsuccessful or an error if TUR failed. For
2276 * removable media, UNIT_ATTENTION sets ->changed flag.
2279 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2280 struct scsi_sense_hdr *sshdr)
2283 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2287 /* try to eat the UNIT_ATTENTION if there are enough retries */
2289 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2291 if (sdev->removable && scsi_sense_valid(sshdr) &&
2292 sshdr->sense_key == UNIT_ATTENTION)
2294 } while (scsi_sense_valid(sshdr) &&
2295 sshdr->sense_key == UNIT_ATTENTION && --retries);
2299 EXPORT_SYMBOL(scsi_test_unit_ready);
2302 * scsi_device_set_state - Take the given device through the device state model.
2303 * @sdev: scsi device to change the state of.
2304 * @state: state to change to.
2306 * Returns zero if successful or an error if the requested
2307 * transition is illegal.
2310 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2312 enum scsi_device_state oldstate = sdev->sdev_state;
2314 if (state == oldstate)
2320 case SDEV_CREATED_BLOCK:
2331 case SDEV_TRANSPORT_OFFLINE:
2344 case SDEV_TRANSPORT_OFFLINE:
2352 case SDEV_TRANSPORT_OFFLINE:
2367 case SDEV_CREATED_BLOCK:
2376 case SDEV_CREATED_BLOCK:
2391 case SDEV_TRANSPORT_OFFLINE:
2403 case SDEV_TRANSPORT_OFFLINE:
2406 case SDEV_CREATED_BLOCK:
2414 sdev->offline_already = false;
2415 sdev->sdev_state = state;
2419 SCSI_LOG_ERROR_RECOVERY(1,
2420 sdev_printk(KERN_ERR, sdev,
2421 "Illegal state transition %s->%s",
2422 scsi_device_state_name(oldstate),
2423 scsi_device_state_name(state))
2427 EXPORT_SYMBOL(scsi_device_set_state);
2430 * scsi_evt_emit - emit a single SCSI device uevent
2431 * @sdev: associated SCSI device
2432 * @evt: event to emit
2434 * Send a single uevent (scsi_event) to the associated scsi_device.
2436 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2441 switch (evt->evt_type) {
2442 case SDEV_EVT_MEDIA_CHANGE:
2443 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2445 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2446 scsi_rescan_device(&sdev->sdev_gendev);
2447 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2449 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2450 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2452 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2453 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2455 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2456 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2458 case SDEV_EVT_LUN_CHANGE_REPORTED:
2459 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2461 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2462 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2464 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2465 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2474 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2478 * scsi_evt_thread - send a uevent for each scsi event
2479 * @work: work struct for scsi_device
2481 * Dispatch queued events to their associated scsi_device kobjects
2484 void scsi_evt_thread(struct work_struct *work)
2486 struct scsi_device *sdev;
2487 enum scsi_device_event evt_type;
2488 LIST_HEAD(event_list);
2490 sdev = container_of(work, struct scsi_device, event_work);
2492 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2493 if (test_and_clear_bit(evt_type, sdev->pending_events))
2494 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2497 struct scsi_event *evt;
2498 struct list_head *this, *tmp;
2499 unsigned long flags;
2501 spin_lock_irqsave(&sdev->list_lock, flags);
2502 list_splice_init(&sdev->event_list, &event_list);
2503 spin_unlock_irqrestore(&sdev->list_lock, flags);
2505 if (list_empty(&event_list))
2508 list_for_each_safe(this, tmp, &event_list) {
2509 evt = list_entry(this, struct scsi_event, node);
2510 list_del(&evt->node);
2511 scsi_evt_emit(sdev, evt);
2518 * sdev_evt_send - send asserted event to uevent thread
2519 * @sdev: scsi_device event occurred on
2520 * @evt: event to send
2522 * Assert scsi device event asynchronously.
2524 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2526 unsigned long flags;
2529 /* FIXME: currently this check eliminates all media change events
2530 * for polled devices. Need to update to discriminate between AN
2531 * and polled events */
2532 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2538 spin_lock_irqsave(&sdev->list_lock, flags);
2539 list_add_tail(&evt->node, &sdev->event_list);
2540 schedule_work(&sdev->event_work);
2541 spin_unlock_irqrestore(&sdev->list_lock, flags);
2543 EXPORT_SYMBOL_GPL(sdev_evt_send);
2546 * sdev_evt_alloc - allocate a new scsi event
2547 * @evt_type: type of event to allocate
2548 * @gfpflags: GFP flags for allocation
2550 * Allocates and returns a new scsi_event.
2552 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2555 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2559 evt->evt_type = evt_type;
2560 INIT_LIST_HEAD(&evt->node);
2562 /* evt_type-specific initialization, if any */
2564 case SDEV_EVT_MEDIA_CHANGE:
2565 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2566 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2567 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2568 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2569 case SDEV_EVT_LUN_CHANGE_REPORTED:
2570 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2571 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2579 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2582 * sdev_evt_send_simple - send asserted event to uevent thread
2583 * @sdev: scsi_device event occurred on
2584 * @evt_type: type of event to send
2585 * @gfpflags: GFP flags for allocation
2587 * Assert scsi device event asynchronously, given an event type.
2589 void sdev_evt_send_simple(struct scsi_device *sdev,
2590 enum scsi_device_event evt_type, gfp_t gfpflags)
2592 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2594 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2599 sdev_evt_send(sdev, evt);
2601 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2604 * scsi_device_quiesce - Block all commands except power management.
2605 * @sdev: scsi device to quiesce.
2607 * This works by trying to transition to the SDEV_QUIESCE state
2608 * (which must be a legal transition). When the device is in this
2609 * state, only power management requests will be accepted, all others will
2612 * Must be called with user context, may sleep.
2614 * Returns zero if unsuccessful or an error if not.
2617 scsi_device_quiesce(struct scsi_device *sdev)
2619 struct request_queue *q = sdev->request_queue;
2623 * It is allowed to call scsi_device_quiesce() multiple times from
2624 * the same context but concurrent scsi_device_quiesce() calls are
2627 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2629 if (sdev->quiesced_by == current)
2634 blk_mq_freeze_queue(q);
2636 * Ensure that the effect of blk_set_pm_only() will be visible
2637 * for percpu_ref_tryget() callers that occur after the queue
2638 * unfreeze even if the queue was already frozen before this function
2639 * was called. See also https://lwn.net/Articles/573497/.
2642 blk_mq_unfreeze_queue(q);
2644 mutex_lock(&sdev->state_mutex);
2645 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2647 sdev->quiesced_by = current;
2649 blk_clear_pm_only(q);
2650 mutex_unlock(&sdev->state_mutex);
2654 EXPORT_SYMBOL(scsi_device_quiesce);
2657 * scsi_device_resume - Restart user issued commands to a quiesced device.
2658 * @sdev: scsi device to resume.
2660 * Moves the device from quiesced back to running and restarts the
2663 * Must be called with user context, may sleep.
2665 void scsi_device_resume(struct scsi_device *sdev)
2667 /* check if the device state was mutated prior to resume, and if
2668 * so assume the state is being managed elsewhere (for example
2669 * device deleted during suspend)
2671 mutex_lock(&sdev->state_mutex);
2672 if (sdev->sdev_state == SDEV_QUIESCE)
2673 scsi_device_set_state(sdev, SDEV_RUNNING);
2674 if (sdev->quiesced_by) {
2675 sdev->quiesced_by = NULL;
2676 blk_clear_pm_only(sdev->request_queue);
2678 mutex_unlock(&sdev->state_mutex);
2680 EXPORT_SYMBOL(scsi_device_resume);
2683 device_quiesce_fn(struct scsi_device *sdev, void *data)
2685 scsi_device_quiesce(sdev);
2689 scsi_target_quiesce(struct scsi_target *starget)
2691 starget_for_each_device(starget, NULL, device_quiesce_fn);
2693 EXPORT_SYMBOL(scsi_target_quiesce);
2696 device_resume_fn(struct scsi_device *sdev, void *data)
2698 scsi_device_resume(sdev);
2702 scsi_target_resume(struct scsi_target *starget)
2704 starget_for_each_device(starget, NULL, device_resume_fn);
2706 EXPORT_SYMBOL(scsi_target_resume);
2708 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2710 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2711 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2716 void scsi_start_queue(struct scsi_device *sdev)
2718 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2719 blk_mq_unquiesce_queue(sdev->request_queue);
2722 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2725 * The atomic variable of ->queue_stopped covers that
2726 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2728 * However, we still need to wait until quiesce is done
2729 * in case that queue has been stopped.
2731 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2733 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2735 blk_mq_quiesce_queue(sdev->request_queue);
2738 blk_mq_wait_quiesce_done(sdev->request_queue);
2743 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2744 * @sdev: device to block
2746 * Pause SCSI command processing on the specified device. Does not sleep.
2748 * Returns zero if successful or a negative error code upon failure.
2751 * This routine transitions the device to the SDEV_BLOCK state (which must be
2752 * a legal transition). When the device is in this state, command processing
2753 * is paused until the device leaves the SDEV_BLOCK state. See also
2754 * scsi_internal_device_unblock_nowait().
2756 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2758 int ret = __scsi_internal_device_block_nowait(sdev);
2761 * The device has transitioned to SDEV_BLOCK. Stop the
2762 * block layer from calling the midlayer with this device's
2766 scsi_stop_queue(sdev, true);
2769 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2772 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2773 * @sdev: device to block
2775 * Pause SCSI command processing on the specified device and wait until all
2776 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2778 * Returns zero if successful or a negative error code upon failure.
2781 * This routine transitions the device to the SDEV_BLOCK state (which must be
2782 * a legal transition). When the device is in this state, command processing
2783 * is paused until the device leaves the SDEV_BLOCK state. See also
2784 * scsi_internal_device_unblock().
2786 static int scsi_internal_device_block(struct scsi_device *sdev)
2790 mutex_lock(&sdev->state_mutex);
2791 err = __scsi_internal_device_block_nowait(sdev);
2793 scsi_stop_queue(sdev, false);
2794 mutex_unlock(&sdev->state_mutex);
2800 * scsi_internal_device_unblock_nowait - resume a device after a block request
2801 * @sdev: device to resume
2802 * @new_state: state to set the device to after unblocking
2804 * Restart the device queue for a previously suspended SCSI device. Does not
2807 * Returns zero if successful or a negative error code upon failure.
2810 * This routine transitions the device to the SDEV_RUNNING state or to one of
2811 * the offline states (which must be a legal transition) allowing the midlayer
2812 * to goose the queue for this device.
2814 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2815 enum scsi_device_state new_state)
2817 switch (new_state) {
2819 case SDEV_TRANSPORT_OFFLINE:
2826 * Try to transition the scsi device to SDEV_RUNNING or one of the
2827 * offlined states and goose the device queue if successful.
2829 switch (sdev->sdev_state) {
2831 case SDEV_TRANSPORT_OFFLINE:
2832 sdev->sdev_state = new_state;
2834 case SDEV_CREATED_BLOCK:
2835 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2836 new_state == SDEV_OFFLINE)
2837 sdev->sdev_state = new_state;
2839 sdev->sdev_state = SDEV_CREATED;
2847 scsi_start_queue(sdev);
2851 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2854 * scsi_internal_device_unblock - resume a device after a block request
2855 * @sdev: device to resume
2856 * @new_state: state to set the device to after unblocking
2858 * Restart the device queue for a previously suspended SCSI device. May sleep.
2860 * Returns zero if successful or a negative error code upon failure.
2863 * This routine transitions the device to the SDEV_RUNNING state or to one of
2864 * the offline states (which must be a legal transition) allowing the midlayer
2865 * to goose the queue for this device.
2867 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2868 enum scsi_device_state new_state)
2872 mutex_lock(&sdev->state_mutex);
2873 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2874 mutex_unlock(&sdev->state_mutex);
2880 device_block(struct scsi_device *sdev, void *data)
2884 ret = scsi_internal_device_block(sdev);
2886 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2887 dev_name(&sdev->sdev_gendev), ret);
2891 target_block(struct device *dev, void *data)
2893 if (scsi_is_target_device(dev))
2894 starget_for_each_device(to_scsi_target(dev), NULL,
2900 scsi_target_block(struct device *dev)
2902 if (scsi_is_target_device(dev))
2903 starget_for_each_device(to_scsi_target(dev), NULL,
2906 device_for_each_child(dev, NULL, target_block);
2908 EXPORT_SYMBOL_GPL(scsi_target_block);
2911 device_unblock(struct scsi_device *sdev, void *data)
2913 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2917 target_unblock(struct device *dev, void *data)
2919 if (scsi_is_target_device(dev))
2920 starget_for_each_device(to_scsi_target(dev), data,
2926 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2928 if (scsi_is_target_device(dev))
2929 starget_for_each_device(to_scsi_target(dev), &new_state,
2932 device_for_each_child(dev, &new_state, target_unblock);
2934 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2937 scsi_host_block(struct Scsi_Host *shost)
2939 struct scsi_device *sdev;
2943 * Call scsi_internal_device_block_nowait so we can avoid
2944 * calling synchronize_rcu() for each LUN.
2946 shost_for_each_device(sdev, shost) {
2947 mutex_lock(&sdev->state_mutex);
2948 ret = scsi_internal_device_block_nowait(sdev);
2949 mutex_unlock(&sdev->state_mutex);
2951 scsi_device_put(sdev);
2957 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2958 * calling synchronize_rcu() once is enough.
2960 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2967 EXPORT_SYMBOL_GPL(scsi_host_block);
2970 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2972 struct scsi_device *sdev;
2975 shost_for_each_device(sdev, shost) {
2976 ret = scsi_internal_device_unblock(sdev, new_state);
2978 scsi_device_put(sdev);
2984 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2987 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2988 * @sgl: scatter-gather list
2989 * @sg_count: number of segments in sg
2990 * @offset: offset in bytes into sg, on return offset into the mapped area
2991 * @len: bytes to map, on return number of bytes mapped
2993 * Returns virtual address of the start of the mapped page
2995 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2996 size_t *offset, size_t *len)
2999 size_t sg_len = 0, len_complete = 0;
3000 struct scatterlist *sg;
3003 WARN_ON(!irqs_disabled());
3005 for_each_sg(sgl, sg, sg_count, i) {
3006 len_complete = sg_len; /* Complete sg-entries */
3007 sg_len += sg->length;
3008 if (sg_len > *offset)
3012 if (unlikely(i == sg_count)) {
3013 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3015 __func__, sg_len, *offset, sg_count);
3020 /* Offset starting from the beginning of first page in this sg-entry */
3021 *offset = *offset - len_complete + sg->offset;
3023 /* Assumption: contiguous pages can be accessed as "page + i" */
3024 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3025 *offset &= ~PAGE_MASK;
3027 /* Bytes in this sg-entry from *offset to the end of the page */
3028 sg_len = PAGE_SIZE - *offset;
3032 return kmap_atomic(page);
3034 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3037 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3038 * @virt: virtual address to be unmapped
3040 void scsi_kunmap_atomic_sg(void *virt)
3042 kunmap_atomic(virt);
3044 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3046 void sdev_disable_disk_events(struct scsi_device *sdev)
3048 atomic_inc(&sdev->disk_events_disable_depth);
3050 EXPORT_SYMBOL(sdev_disable_disk_events);
3052 void sdev_enable_disk_events(struct scsi_device *sdev)
3054 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3056 atomic_dec(&sdev->disk_events_disable_depth);
3058 EXPORT_SYMBOL(sdev_enable_disk_events);
3060 static unsigned char designator_prio(const unsigned char *d)
3063 /* not associated with LUN */
3067 /* invalid length */
3071 * Order of preference for lun descriptor:
3072 * - SCSI name string
3073 * - NAA IEEE Registered Extended
3074 * - EUI-64 based 16-byte
3075 * - EUI-64 based 12-byte
3076 * - NAA IEEE Registered
3077 * - NAA IEEE Extended
3078 * - EUI-64 based 8-byte
3079 * - SCSI name string (truncated)
3081 * as longer descriptors reduce the likelyhood
3082 * of identification clashes.
3085 switch (d[1] & 0xf) {
3087 /* SCSI name string, variable-length UTF-8 */
3090 switch (d[4] >> 4) {
3092 /* NAA registered extended */
3095 /* NAA registered */
3101 /* NAA locally assigned */
3110 /* EUI64-based, 16 byte */
3113 /* EUI64-based, 12 byte */
3116 /* EUI64-based, 8 byte */
3133 * scsi_vpd_lun_id - return a unique device identification
3134 * @sdev: SCSI device
3135 * @id: buffer for the identification
3136 * @id_len: length of the buffer
3138 * Copies a unique device identification into @id based
3139 * on the information in the VPD page 0x83 of the device.
3140 * The string will be formatted as a SCSI name string.
3142 * Returns the length of the identification or error on failure.
3143 * If the identifier is longer than the supplied buffer the actual
3144 * identifier length is returned and the buffer is not zero-padded.
3146 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3150 const unsigned char *d, *cur_id_str;
3151 const struct scsi_vpd *vpd_pg83;
3152 int id_size = -EINVAL;
3155 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3161 /* The id string must be at least 20 bytes + terminating NULL byte */
3167 memset(id, 0, id_len);
3168 for (d = vpd_pg83->data + 4;
3169 d < vpd_pg83->data + vpd_pg83->len;
3171 u8 prio = designator_prio(d);
3173 if (prio == 0 || cur_id_prio > prio)
3176 switch (d[1] & 0xf) {
3179 if (cur_id_size > d[3])
3183 if (cur_id_size + 4 > id_len)
3184 cur_id_size = id_len - 4;
3186 id_size = snprintf(id, id_len, "t10.%*pE",
3187 cur_id_size, cur_id_str);
3194 switch (cur_id_size) {
3196 id_size = snprintf(id, id_len,
3201 id_size = snprintf(id, id_len,
3206 id_size = snprintf(id, id_len,
3219 switch (cur_id_size) {
3221 id_size = snprintf(id, id_len,
3226 id_size = snprintf(id, id_len,
3235 /* SCSI name string */
3236 if (cur_id_size > d[3])
3238 /* Prefer others for truncated descriptor */
3239 if (d[3] > id_len) {
3241 if (cur_id_prio > prio)
3245 cur_id_size = id_size = d[3];
3247 if (cur_id_size >= id_len)
3248 cur_id_size = id_len - 1;
3249 memcpy(id, cur_id_str, cur_id_size);
3259 EXPORT_SYMBOL(scsi_vpd_lun_id);
3262 * scsi_vpd_tpg_id - return a target port group identifier
3263 * @sdev: SCSI device
3265 * Returns the Target Port Group identifier from the information
3266 * froom VPD page 0x83 of the device.
3268 * Returns the identifier or error on failure.
3270 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3272 const unsigned char *d;
3273 const struct scsi_vpd *vpd_pg83;
3274 int group_id = -EAGAIN, rel_port = -1;
3277 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3283 d = vpd_pg83->data + 4;
3284 while (d < vpd_pg83->data + vpd_pg83->len) {
3285 switch (d[1] & 0xf) {
3287 /* Relative target port */
3288 rel_port = get_unaligned_be16(&d[6]);
3291 /* Target port group */
3292 group_id = get_unaligned_be16(&d[6]);
3301 if (group_id >= 0 && rel_id && rel_port != -1)
3306 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3309 * scsi_build_sense - build sense data for a command
3310 * @scmd: scsi command for which the sense should be formatted
3311 * @desc: Sense format (non-zero == descriptor format,
3312 * 0 == fixed format)
3314 * @asc: Additional sense code
3315 * @ascq: Additional sense code qualifier
3318 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3320 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3321 scmd->result = SAM_STAT_CHECK_CONDITION;
3323 EXPORT_SYMBOL_GPL(scsi_build_sense);