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/ratelimit.h>
25 #include <asm/unaligned.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_cmnd.h>
29 #include <scsi/scsi_dbg.h>
30 #include <scsi/scsi_device.h>
31 #include <scsi/scsi_driver.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_host.h>
34 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35 #include <scsi/scsi_dh.h>
37 #include <trace/events/scsi.h>
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
44 * Size of integrity metadata is usually small, 1 inline sg should
47 #ifdef CONFIG_ARCH_NO_SG_CHAIN
48 #define SCSI_INLINE_PROT_SG_CNT 0
49 #define SCSI_INLINE_SG_CNT 0
51 #define SCSI_INLINE_PROT_SG_CNT 1
52 #define SCSI_INLINE_SG_CNT 2
55 static struct kmem_cache *scsi_sense_cache;
56 static struct kmem_cache *scsi_sense_isadma_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61 static inline struct kmem_cache *
62 scsi_select_sense_cache(bool unchecked_isa_dma)
64 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
67 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
68 unsigned char *sense_buffer)
70 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
74 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
75 gfp_t gfp_mask, int numa_node)
77 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
81 int scsi_init_sense_cache(struct Scsi_Host *shost)
83 struct kmem_cache *cache;
86 mutex_lock(&scsi_sense_cache_mutex);
87 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
91 if (shost->unchecked_isa_dma) {
92 scsi_sense_isadma_cache =
93 kmem_cache_create("scsi_sense_cache(DMA)",
94 SCSI_SENSE_BUFFERSIZE, 0,
95 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
96 if (!scsi_sense_isadma_cache)
100 kmem_cache_create_usercopy("scsi_sense_cache",
101 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
102 0, SCSI_SENSE_BUFFERSIZE, NULL);
103 if (!scsi_sense_cache)
107 mutex_unlock(&scsi_sense_cache_mutex);
112 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
113 * not change behaviour from the previous unplug mechanism, experimentation
114 * may prove this needs changing.
116 #define SCSI_QUEUE_DELAY 3
119 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
121 struct Scsi_Host *host = cmd->device->host;
122 struct scsi_device *device = cmd->device;
123 struct scsi_target *starget = scsi_target(device);
126 * Set the appropriate busy bit for the device/host.
128 * If the host/device isn't busy, assume that something actually
129 * completed, and that we should be able to queue a command now.
131 * Note that the prior mid-layer assumption that any host could
132 * always queue at least one command is now broken. The mid-layer
133 * will implement a user specifiable stall (see
134 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
135 * if a command is requeued with no other commands outstanding
136 * either for the device or for the host.
139 case SCSI_MLQUEUE_HOST_BUSY:
140 atomic_set(&host->host_blocked, host->max_host_blocked);
142 case SCSI_MLQUEUE_DEVICE_BUSY:
143 case SCSI_MLQUEUE_EH_RETRY:
144 atomic_set(&device->device_blocked,
145 device->max_device_blocked);
147 case SCSI_MLQUEUE_TARGET_BUSY:
148 atomic_set(&starget->target_blocked,
149 starget->max_target_blocked);
154 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156 if (cmd->request->rq_flags & RQF_DONTPREP) {
157 cmd->request->rq_flags &= ~RQF_DONTPREP;
158 scsi_mq_uninit_cmd(cmd);
162 blk_mq_requeue_request(cmd->request, true);
166 * __scsi_queue_insert - private queue insertion
167 * @cmd: The SCSI command being requeued
168 * @reason: The reason for the requeue
169 * @unbusy: Whether the queue should be unbusied
171 * This is a private queue insertion. The public interface
172 * scsi_queue_insert() always assumes the queue should be unbusied
173 * because it's always called before the completion. This function is
174 * for a requeue after completion, which should only occur in this
177 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
179 struct scsi_device *device = cmd->device;
181 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
182 "Inserting command %p into mlqueue\n", cmd));
184 scsi_set_blocked(cmd, reason);
187 * Decrement the counters, since these commands are no longer
188 * active on the host/device.
191 scsi_device_unbusy(device, cmd);
194 * Requeue this command. It will go before all other commands
195 * that are already in the queue. Schedule requeue work under
196 * lock such that the kblockd_schedule_work() call happens
197 * before blk_cleanup_queue() finishes.
201 blk_mq_requeue_request(cmd->request, true);
205 * scsi_queue_insert - Reinsert a command in the queue.
206 * @cmd: command that we are adding to queue.
207 * @reason: why we are inserting command to queue.
209 * We do this for one of two cases. Either the host is busy and it cannot accept
210 * any more commands for the time being, or the device returned QUEUE_FULL and
211 * can accept no more commands.
213 * Context: This could be called either from an interrupt context or a normal
216 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
218 __scsi_queue_insert(cmd, reason, true);
223 * __scsi_execute - insert request and wait for the result
226 * @data_direction: data direction
227 * @buffer: data buffer
228 * @bufflen: len of buffer
229 * @sense: optional sense buffer
230 * @sshdr: optional decoded sense header
231 * @timeout: request timeout in seconds
232 * @retries: number of times to retry request
233 * @flags: flags for ->cmd_flags
234 * @rq_flags: flags for ->rq_flags
235 * @resid: optional residual length
237 * Returns the scsi_cmnd result field if a command was executed, or a negative
238 * Linux error code if we didn't get that far.
240 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
241 int data_direction, void *buffer, unsigned bufflen,
242 unsigned char *sense, struct scsi_sense_hdr *sshdr,
243 int timeout, int retries, u64 flags, req_flags_t rq_flags,
247 struct scsi_request *rq;
248 int ret = DRIVER_ERROR << 24;
250 req = blk_get_request(sdev->request_queue,
251 data_direction == DMA_TO_DEVICE ?
252 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN,
253 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
258 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
259 buffer, bufflen, GFP_NOIO))
262 rq->cmd_len = COMMAND_SIZE(cmd[0]);
263 memcpy(rq->cmd, cmd, rq->cmd_len);
264 rq->retries = retries;
265 req->timeout = timeout;
266 req->cmd_flags |= flags;
267 req->rq_flags |= rq_flags | RQF_QUIET;
270 * head injection *required* here otherwise quiesce won't work
272 blk_execute_rq(NULL, req, 1);
275 * Some devices (USB mass-storage in particular) may transfer
276 * garbage data together with a residue indicating that the data
277 * is invalid. Prevent the garbage from being misinterpreted
278 * and prevent security leaks by zeroing out the excess data.
280 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
281 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
284 *resid = rq->resid_len;
285 if (sense && rq->sense_len)
286 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
288 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
291 blk_put_request(req);
295 EXPORT_SYMBOL(__scsi_execute);
298 * Wake up the error handler if necessary. Avoid as follows that the error
299 * handler is not woken up if host in-flight requests number ==
300 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
301 * with an RCU read lock in this function to ensure that this function in
302 * its entirety either finishes before scsi_eh_scmd_add() increases the
303 * host_failed counter or that it notices the shost state change made by
304 * scsi_eh_scmd_add().
306 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
311 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
312 if (unlikely(scsi_host_in_recovery(shost))) {
313 spin_lock_irqsave(shost->host_lock, flags);
314 if (shost->host_failed || shost->host_eh_scheduled)
315 scsi_eh_wakeup(shost);
316 spin_unlock_irqrestore(shost->host_lock, flags);
321 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
323 struct Scsi_Host *shost = sdev->host;
324 struct scsi_target *starget = scsi_target(sdev);
326 scsi_dec_host_busy(shost, cmd);
328 if (starget->can_queue > 0)
329 atomic_dec(&starget->target_busy);
331 sbitmap_put(&sdev->budget_map, cmd->budget_token);
332 cmd->budget_token = -1;
335 static void scsi_kick_queue(struct request_queue *q)
337 blk_mq_run_hw_queues(q, false);
341 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
342 * and call blk_run_queue for all the scsi_devices on the target -
343 * including current_sdev first.
345 * Called with *no* scsi locks held.
347 static void scsi_single_lun_run(struct scsi_device *current_sdev)
349 struct Scsi_Host *shost = current_sdev->host;
350 struct scsi_device *sdev, *tmp;
351 struct scsi_target *starget = scsi_target(current_sdev);
354 spin_lock_irqsave(shost->host_lock, flags);
355 starget->starget_sdev_user = NULL;
356 spin_unlock_irqrestore(shost->host_lock, flags);
359 * Call blk_run_queue for all LUNs on the target, starting with
360 * current_sdev. We race with others (to set starget_sdev_user),
361 * but in most cases, we will be first. Ideally, each LU on the
362 * target would get some limited time or requests on the target.
364 scsi_kick_queue(current_sdev->request_queue);
366 spin_lock_irqsave(shost->host_lock, flags);
367 if (starget->starget_sdev_user)
369 list_for_each_entry_safe(sdev, tmp, &starget->devices,
370 same_target_siblings) {
371 if (sdev == current_sdev)
373 if (scsi_device_get(sdev))
376 spin_unlock_irqrestore(shost->host_lock, flags);
377 scsi_kick_queue(sdev->request_queue);
378 spin_lock_irqsave(shost->host_lock, flags);
380 scsi_device_put(sdev);
383 spin_unlock_irqrestore(shost->host_lock, flags);
386 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
388 if (scsi_device_busy(sdev) >= sdev->queue_depth)
390 if (atomic_read(&sdev->device_blocked) > 0)
395 static inline bool scsi_target_is_busy(struct scsi_target *starget)
397 if (starget->can_queue > 0) {
398 if (atomic_read(&starget->target_busy) >= starget->can_queue)
400 if (atomic_read(&starget->target_blocked) > 0)
406 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
408 if (atomic_read(&shost->host_blocked) > 0)
410 if (shost->host_self_blocked)
415 static void scsi_starved_list_run(struct Scsi_Host *shost)
417 LIST_HEAD(starved_list);
418 struct scsi_device *sdev;
421 spin_lock_irqsave(shost->host_lock, flags);
422 list_splice_init(&shost->starved_list, &starved_list);
424 while (!list_empty(&starved_list)) {
425 struct request_queue *slq;
428 * As long as shost is accepting commands and we have
429 * starved queues, call blk_run_queue. scsi_request_fn
430 * drops the queue_lock and can add us back to the
433 * host_lock protects the starved_list and starved_entry.
434 * scsi_request_fn must get the host_lock before checking
435 * or modifying starved_list or starved_entry.
437 if (scsi_host_is_busy(shost))
440 sdev = list_entry(starved_list.next,
441 struct scsi_device, starved_entry);
442 list_del_init(&sdev->starved_entry);
443 if (scsi_target_is_busy(scsi_target(sdev))) {
444 list_move_tail(&sdev->starved_entry,
445 &shost->starved_list);
450 * Once we drop the host lock, a racing scsi_remove_device()
451 * call may remove the sdev from the starved list and destroy
452 * it and the queue. Mitigate by taking a reference to the
453 * queue and never touching the sdev again after we drop the
454 * host lock. Note: if __scsi_remove_device() invokes
455 * blk_cleanup_queue() before the queue is run from this
456 * function then blk_run_queue() will return immediately since
457 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
459 slq = sdev->request_queue;
460 if (!blk_get_queue(slq))
462 spin_unlock_irqrestore(shost->host_lock, flags);
464 scsi_kick_queue(slq);
467 spin_lock_irqsave(shost->host_lock, flags);
469 /* put any unprocessed entries back */
470 list_splice(&starved_list, &shost->starved_list);
471 spin_unlock_irqrestore(shost->host_lock, flags);
475 * scsi_run_queue - Select a proper request queue to serve next.
476 * @q: last request's queue
478 * The previous command was completely finished, start a new one if possible.
480 static void scsi_run_queue(struct request_queue *q)
482 struct scsi_device *sdev = q->queuedata;
484 if (scsi_target(sdev)->single_lun)
485 scsi_single_lun_run(sdev);
486 if (!list_empty(&sdev->host->starved_list))
487 scsi_starved_list_run(sdev->host);
489 blk_mq_run_hw_queues(q, false);
492 void scsi_requeue_run_queue(struct work_struct *work)
494 struct scsi_device *sdev;
495 struct request_queue *q;
497 sdev = container_of(work, struct scsi_device, requeue_work);
498 q = sdev->request_queue;
502 void scsi_run_host_queues(struct Scsi_Host *shost)
504 struct scsi_device *sdev;
506 shost_for_each_device(sdev, shost)
507 scsi_run_queue(sdev->request_queue);
510 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
512 if (!blk_rq_is_passthrough(cmd->request)) {
513 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
515 if (drv->uninit_command)
516 drv->uninit_command(cmd);
520 void scsi_free_sgtables(struct scsi_cmnd *cmd)
522 if (cmd->sdb.table.nents)
523 sg_free_table_chained(&cmd->sdb.table,
525 if (scsi_prot_sg_count(cmd))
526 sg_free_table_chained(&cmd->prot_sdb->table,
527 SCSI_INLINE_PROT_SG_CNT);
529 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
531 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
533 scsi_free_sgtables(cmd);
534 scsi_uninit_cmd(cmd);
537 static void scsi_run_queue_async(struct scsi_device *sdev)
539 if (scsi_target(sdev)->single_lun ||
540 !list_empty(&sdev->host->starved_list)) {
541 kblockd_schedule_work(&sdev->requeue_work);
544 * smp_mb() present in sbitmap_queue_clear() or implied in
545 * .end_io is for ordering writing .device_busy in
546 * scsi_device_unbusy() and reading sdev->restarts.
548 int old = atomic_read(&sdev->restarts);
551 * ->restarts has to be kept as non-zero if new budget
554 * No need to run queue when either another re-run
555 * queue wins in updating ->restarts or a new budget
558 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
559 blk_mq_run_hw_queues(sdev->request_queue, true);
563 /* Returns false when no more bytes to process, true if there are more */
564 static bool scsi_end_request(struct request *req, blk_status_t error,
567 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
568 struct scsi_device *sdev = cmd->device;
569 struct request_queue *q = sdev->request_queue;
571 if (blk_update_request(req, error, bytes))
574 if (blk_queue_add_random(q))
575 add_disk_randomness(req->rq_disk);
577 if (!blk_rq_is_scsi(req)) {
578 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
579 cmd->flags &= ~SCMD_INITIALIZED;
583 * Calling rcu_barrier() is not necessary here because the
584 * SCSI error handler guarantees that the function called by
585 * call_rcu() has been called before scsi_end_request() is
588 destroy_rcu_head(&cmd->rcu);
591 * In the MQ case the command gets freed by __blk_mq_end_request,
592 * so we have to do all cleanup that depends on it earlier.
594 * We also can't kick the queues from irq context, so we
595 * will have to defer it to a workqueue.
597 scsi_mq_uninit_cmd(cmd);
600 * queue is still alive, so grab the ref for preventing it
601 * from being cleaned up during running queue.
603 percpu_ref_get(&q->q_usage_counter);
605 __blk_mq_end_request(req, error);
607 scsi_run_queue_async(sdev);
609 percpu_ref_put(&q->q_usage_counter);
614 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
616 * @result: scsi error code
618 * Translate a SCSI result code into a blk_status_t value. May reset the host
619 * byte of @cmd->result.
621 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
623 switch (host_byte(result)) {
626 * Also check the other bytes than the status byte in result
627 * to handle the case when a SCSI LLD sets result to
628 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
630 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
632 return BLK_STS_IOERR;
633 case DID_TRANSPORT_FAILFAST:
634 case DID_TRANSPORT_MARGINAL:
635 return BLK_STS_TRANSPORT;
636 case DID_TARGET_FAILURE:
637 set_host_byte(cmd, DID_OK);
638 return BLK_STS_TARGET;
639 case DID_NEXUS_FAILURE:
640 set_host_byte(cmd, DID_OK);
641 return BLK_STS_NEXUS;
642 case DID_ALLOC_FAILURE:
643 set_host_byte(cmd, DID_OK);
644 return BLK_STS_NOSPC;
645 case DID_MEDIUM_ERROR:
646 set_host_byte(cmd, DID_OK);
647 return BLK_STS_MEDIUM;
649 return BLK_STS_IOERR;
653 /* Helper for scsi_io_completion() when "reprep" action required. */
654 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
655 struct request_queue *q)
657 /* A new command will be prepared and issued. */
658 scsi_mq_requeue_cmd(cmd);
661 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
663 struct request *req = cmd->request;
664 unsigned long wait_for;
666 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
669 wait_for = (cmd->allowed + 1) * req->timeout;
670 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
671 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
678 /* Helper for scsi_io_completion() when special action required. */
679 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
681 struct request_queue *q = cmd->device->request_queue;
682 struct request *req = cmd->request;
684 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
685 ACTION_DELAYED_RETRY} action;
686 struct scsi_sense_hdr sshdr;
688 bool sense_current = true; /* false implies "deferred sense" */
689 blk_status_t blk_stat;
691 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
693 sense_current = !scsi_sense_is_deferred(&sshdr);
695 blk_stat = scsi_result_to_blk_status(cmd, result);
697 if (host_byte(result) == DID_RESET) {
698 /* Third party bus reset or reset for error recovery
699 * reasons. Just retry the command and see what
702 action = ACTION_RETRY;
703 } else if (sense_valid && sense_current) {
704 switch (sshdr.sense_key) {
706 if (cmd->device->removable) {
707 /* Detected disc change. Set a bit
708 * and quietly refuse further access.
710 cmd->device->changed = 1;
711 action = ACTION_FAIL;
713 /* Must have been a power glitch, or a
714 * bus reset. Could not have been a
715 * media change, so we just retry the
716 * command and see what happens.
718 action = ACTION_RETRY;
721 case ILLEGAL_REQUEST:
722 /* If we had an ILLEGAL REQUEST returned, then
723 * we may have performed an unsupported
724 * command. The only thing this should be
725 * would be a ten byte read where only a six
726 * byte read was supported. Also, on a system
727 * where READ CAPACITY failed, we may have
728 * read past the end of the disk.
730 if ((cmd->device->use_10_for_rw &&
731 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
732 (cmd->cmnd[0] == READ_10 ||
733 cmd->cmnd[0] == WRITE_10)) {
734 /* This will issue a new 6-byte command. */
735 cmd->device->use_10_for_rw = 0;
736 action = ACTION_REPREP;
737 } else if (sshdr.asc == 0x10) /* DIX */ {
738 action = ACTION_FAIL;
739 blk_stat = BLK_STS_PROTECTION;
740 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
741 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
742 action = ACTION_FAIL;
743 blk_stat = BLK_STS_TARGET;
745 action = ACTION_FAIL;
747 case ABORTED_COMMAND:
748 action = ACTION_FAIL;
749 if (sshdr.asc == 0x10) /* DIF */
750 blk_stat = BLK_STS_PROTECTION;
753 /* If the device is in the process of becoming
754 * ready, or has a temporary blockage, retry.
756 if (sshdr.asc == 0x04) {
757 switch (sshdr.ascq) {
758 case 0x01: /* becoming ready */
759 case 0x04: /* format in progress */
760 case 0x05: /* rebuild in progress */
761 case 0x06: /* recalculation in progress */
762 case 0x07: /* operation in progress */
763 case 0x08: /* Long write in progress */
764 case 0x09: /* self test in progress */
765 case 0x14: /* space allocation in progress */
766 case 0x1a: /* start stop unit in progress */
767 case 0x1b: /* sanitize in progress */
768 case 0x1d: /* configuration in progress */
769 case 0x24: /* depopulation in progress */
770 action = ACTION_DELAYED_RETRY;
772 case 0x0a: /* ALUA state transition */
773 blk_stat = BLK_STS_AGAIN;
776 action = ACTION_FAIL;
780 action = ACTION_FAIL;
782 case VOLUME_OVERFLOW:
783 /* See SSC3rXX or current. */
784 action = ACTION_FAIL;
787 action = ACTION_FAIL;
788 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
789 (sshdr.asc == 0x55 &&
790 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
791 /* Insufficient zone resources */
792 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
796 action = ACTION_FAIL;
800 action = ACTION_FAIL;
802 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
803 action = ACTION_FAIL;
807 /* Give up and fail the remainder of the request */
808 if (!(req->rq_flags & RQF_QUIET)) {
809 static DEFINE_RATELIMIT_STATE(_rs,
810 DEFAULT_RATELIMIT_INTERVAL,
811 DEFAULT_RATELIMIT_BURST);
813 if (unlikely(scsi_logging_level))
815 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
816 SCSI_LOG_MLCOMPLETE_BITS);
819 * if logging is enabled the failure will be printed
820 * in scsi_log_completion(), so avoid duplicate messages
822 if (!level && __ratelimit(&_rs)) {
823 scsi_print_result(cmd, NULL, FAILED);
824 if (driver_byte(result) == DRIVER_SENSE)
825 scsi_print_sense(cmd);
826 scsi_print_command(cmd);
829 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
833 scsi_io_completion_reprep(cmd, q);
836 /* Retry the same command immediately */
837 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
839 case ACTION_DELAYED_RETRY:
840 /* Retry the same command after a delay */
841 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
847 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
848 * new result that may suppress further error checking. Also modifies
849 * *blk_statp in some cases.
851 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
852 blk_status_t *blk_statp)
855 bool sense_current = true; /* false implies "deferred sense" */
856 struct request *req = cmd->request;
857 struct scsi_sense_hdr sshdr;
859 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
861 sense_current = !scsi_sense_is_deferred(&sshdr);
863 if (blk_rq_is_passthrough(req)) {
866 * SG_IO wants current and deferred errors
868 scsi_req(req)->sense_len =
869 min(8 + cmd->sense_buffer[7],
870 SCSI_SENSE_BUFFERSIZE);
873 *blk_statp = scsi_result_to_blk_status(cmd, result);
874 } else if (blk_rq_bytes(req) == 0 && sense_current) {
876 * Flush commands do not transfers any data, and thus cannot use
877 * good_bytes != blk_rq_bytes(req) as the signal for an error.
878 * This sets *blk_statp explicitly for the problem case.
880 *blk_statp = scsi_result_to_blk_status(cmd, result);
883 * Recovered errors need reporting, but they're always treated as
884 * success, so fiddle the result code here. For passthrough requests
885 * we already took a copy of the original into sreq->result which
886 * is what gets returned to the user
888 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
889 bool do_print = true;
891 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
892 * skip print since caller wants ATA registers. Only occurs
893 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
895 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
897 else if (req->rq_flags & RQF_QUIET)
900 scsi_print_sense(cmd);
902 /* for passthrough, *blk_statp may be set */
903 *blk_statp = BLK_STS_OK;
906 * Another corner case: the SCSI status byte is non-zero but 'good'.
907 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
908 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
909 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
910 * intermediate statuses (both obsolete in SAM-4) as good.
912 if (status_byte(result) && scsi_status_is_good(result)) {
914 *blk_statp = BLK_STS_OK;
920 * scsi_io_completion - Completion processing for SCSI commands.
921 * @cmd: command that is finished.
922 * @good_bytes: number of processed bytes.
924 * We will finish off the specified number of sectors. If we are done, the
925 * command block will be released and the queue function will be goosed. If we
926 * are not done then we have to figure out what to do next:
928 * a) We can call scsi_io_completion_reprep(). The request will be
929 * unprepared and put back on the queue. Then a new command will
930 * be created for it. This should be used if we made forward
931 * progress, or if we want to switch from READ(10) to READ(6) for
934 * b) We can call scsi_io_completion_action(). The request will be
935 * put back on the queue and retried using the same command as
936 * before, possibly after a delay.
938 * c) We can call scsi_end_request() with blk_stat other than
939 * BLK_STS_OK, to fail the remainder of the request.
941 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
943 int result = cmd->result;
944 struct request_queue *q = cmd->device->request_queue;
945 struct request *req = cmd->request;
946 blk_status_t blk_stat = BLK_STS_OK;
948 if (unlikely(result)) /* a nz result may or may not be an error */
949 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
951 if (unlikely(blk_rq_is_passthrough(req))) {
953 * scsi_result_to_blk_status may have reset the host_byte
955 scsi_req(req)->result = cmd->result;
959 * Next deal with any sectors which we were able to correctly
962 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
963 "%u sectors total, %d bytes done.\n",
964 blk_rq_sectors(req), good_bytes));
967 * Failed, zero length commands always need to drop down
968 * to retry code. Fast path should return in this block.
970 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
971 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
972 return; /* no bytes remaining */
975 /* Kill remainder if no retries. */
976 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
977 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
979 "Bytes remaining after failed, no-retry command");
984 * If there had been no error, but we have leftover bytes in the
985 * requeues just queue the command up again.
987 if (likely(result == 0))
988 scsi_io_completion_reprep(cmd, q);
990 scsi_io_completion_action(cmd, result);
993 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
996 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
997 !op_is_write(req_op(rq)) &&
998 sdev->host->hostt->dma_need_drain(rq);
1002 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1003 * @cmd: SCSI command data structure to initialize.
1005 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1009 * * BLK_STS_OK - on success
1010 * * BLK_STS_RESOURCE - if the failure is retryable
1011 * * BLK_STS_IOERR - if the failure is fatal
1013 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1015 struct scsi_device *sdev = cmd->device;
1016 struct request *rq = cmd->request;
1017 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1018 struct scatterlist *last_sg = NULL;
1020 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1023 if (WARN_ON_ONCE(!nr_segs))
1024 return BLK_STS_IOERR;
1027 * Make sure there is space for the drain. The driver must adjust
1028 * max_hw_segments to be prepared for this.
1034 * If sg table allocation fails, requeue request later.
1036 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1037 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1038 return BLK_STS_RESOURCE;
1041 * Next, walk the list, and fill in the addresses and sizes of
1044 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1046 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1047 unsigned int pad_len =
1048 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1050 last_sg->length += pad_len;
1051 cmd->extra_len += pad_len;
1055 sg_unmark_end(last_sg);
1056 last_sg = sg_next(last_sg);
1057 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1058 sg_mark_end(last_sg);
1060 cmd->extra_len += sdev->dma_drain_len;
1064 BUG_ON(count > cmd->sdb.table.nents);
1065 cmd->sdb.table.nents = count;
1066 cmd->sdb.length = blk_rq_payload_bytes(rq);
1068 if (blk_integrity_rq(rq)) {
1069 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1072 if (WARN_ON_ONCE(!prot_sdb)) {
1074 * This can happen if someone (e.g. multipath)
1075 * queues a command to a device on an adapter
1076 * that does not support DIX.
1078 ret = BLK_STS_IOERR;
1079 goto out_free_sgtables;
1082 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1084 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1085 prot_sdb->table.sgl,
1086 SCSI_INLINE_PROT_SG_CNT)) {
1087 ret = BLK_STS_RESOURCE;
1088 goto out_free_sgtables;
1091 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1092 prot_sdb->table.sgl);
1093 BUG_ON(count > ivecs);
1094 BUG_ON(count > queue_max_integrity_segments(rq->q));
1096 cmd->prot_sdb = prot_sdb;
1097 cmd->prot_sdb->table.nents = count;
1102 scsi_free_sgtables(cmd);
1105 EXPORT_SYMBOL(scsi_alloc_sgtables);
1108 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1109 * @rq: Request associated with the SCSI command to be initialized.
1111 * This function initializes the members of struct scsi_cmnd that must be
1112 * initialized before request processing starts and that won't be
1113 * reinitialized if a SCSI command is requeued.
1115 * Called from inside blk_get_request() for pass-through requests and from
1116 * inside scsi_init_command() for filesystem requests.
1118 static void scsi_initialize_rq(struct request *rq)
1120 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1122 scsi_req_init(&cmd->req);
1123 init_rcu_head(&cmd->rcu);
1124 cmd->jiffies_at_alloc = jiffies;
1129 * Only called when the request isn't completed by SCSI, and not freed by
1132 static void scsi_cleanup_rq(struct request *rq)
1134 if (rq->rq_flags & RQF_DONTPREP) {
1135 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1136 rq->rq_flags &= ~RQF_DONTPREP;
1140 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1141 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1143 void *buf = cmd->sense_buffer;
1144 void *prot = cmd->prot_sdb;
1145 struct request *rq = blk_mq_rq_from_pdu(cmd);
1146 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1147 unsigned long jiffies_at_alloc;
1148 int retries, to_clear;
1150 int budget_token = cmd->budget_token;
1152 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1153 flags |= SCMD_INITIALIZED;
1154 scsi_initialize_rq(rq);
1157 jiffies_at_alloc = cmd->jiffies_at_alloc;
1158 retries = cmd->retries;
1159 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1161 * Zero out the cmd, except for the embedded scsi_request. Only clear
1162 * the driver-private command data if the LLD does not supply a
1163 * function to initialize that data.
1165 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1166 if (!dev->host->hostt->init_cmd_priv)
1167 to_clear += dev->host->hostt->cmd_size;
1168 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1171 cmd->sense_buffer = buf;
1172 cmd->prot_sdb = prot;
1174 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1175 cmd->jiffies_at_alloc = jiffies_at_alloc;
1176 cmd->retries = retries;
1178 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1179 cmd->budget_token = budget_token;
1183 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1184 struct request *req)
1186 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1189 * Passthrough requests may transfer data, in which case they must
1190 * a bio attached to them. Or they might contain a SCSI command
1191 * that does not transfer data, in which case they may optionally
1192 * submit a request without an attached bio.
1195 blk_status_t ret = scsi_alloc_sgtables(cmd);
1196 if (unlikely(ret != BLK_STS_OK))
1199 BUG_ON(blk_rq_bytes(req));
1201 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1204 cmd->cmd_len = scsi_req(req)->cmd_len;
1205 if (cmd->cmd_len == 0)
1206 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1207 cmd->cmnd = scsi_req(req)->cmd;
1208 cmd->transfersize = blk_rq_bytes(req);
1209 cmd->allowed = scsi_req(req)->retries;
1214 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1216 switch (sdev->sdev_state) {
1220 case SDEV_TRANSPORT_OFFLINE:
1222 * If the device is offline we refuse to process any
1223 * commands. The device must be brought online
1224 * before trying any recovery commands.
1226 if (!sdev->offline_already) {
1227 sdev->offline_already = true;
1228 sdev_printk(KERN_ERR, sdev,
1229 "rejecting I/O to offline device\n");
1231 return BLK_STS_IOERR;
1234 * If the device is fully deleted, we refuse to
1235 * process any commands as well.
1237 sdev_printk(KERN_ERR, sdev,
1238 "rejecting I/O to dead device\n");
1239 return BLK_STS_IOERR;
1241 case SDEV_CREATED_BLOCK:
1242 return BLK_STS_RESOURCE;
1245 * If the device is blocked we only accept power management
1248 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1249 return BLK_STS_RESOURCE;
1253 * For any other not fully online state we only allow
1254 * power management commands.
1256 if (req && !(req->rq_flags & RQF_PM))
1257 return BLK_STS_IOERR;
1263 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1264 * and return the token else return -1.
1266 static inline int scsi_dev_queue_ready(struct request_queue *q,
1267 struct scsi_device *sdev)
1271 token = sbitmap_get(&sdev->budget_map);
1272 if (atomic_read(&sdev->device_blocked)) {
1276 if (scsi_device_busy(sdev) > 1)
1280 * unblock after device_blocked iterates to zero
1282 if (atomic_dec_return(&sdev->device_blocked) > 0)
1284 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1285 "unblocking device at zero depth\n"));
1291 sbitmap_put(&sdev->budget_map, token);
1297 * scsi_target_queue_ready: checks if there we can send commands to target
1298 * @sdev: scsi device on starget to check.
1300 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1301 struct scsi_device *sdev)
1303 struct scsi_target *starget = scsi_target(sdev);
1306 if (starget->single_lun) {
1307 spin_lock_irq(shost->host_lock);
1308 if (starget->starget_sdev_user &&
1309 starget->starget_sdev_user != sdev) {
1310 spin_unlock_irq(shost->host_lock);
1313 starget->starget_sdev_user = sdev;
1314 spin_unlock_irq(shost->host_lock);
1317 if (starget->can_queue <= 0)
1320 busy = atomic_inc_return(&starget->target_busy) - 1;
1321 if (atomic_read(&starget->target_blocked) > 0) {
1326 * unblock after target_blocked iterates to zero
1328 if (atomic_dec_return(&starget->target_blocked) > 0)
1331 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1332 "unblocking target at zero depth\n"));
1335 if (busy >= starget->can_queue)
1341 spin_lock_irq(shost->host_lock);
1342 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1343 spin_unlock_irq(shost->host_lock);
1345 if (starget->can_queue > 0)
1346 atomic_dec(&starget->target_busy);
1351 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1352 * return 0. We must end up running the queue again whenever 0 is
1353 * returned, else IO can hang.
1355 static inline int scsi_host_queue_ready(struct request_queue *q,
1356 struct Scsi_Host *shost,
1357 struct scsi_device *sdev,
1358 struct scsi_cmnd *cmd)
1360 if (scsi_host_in_recovery(shost))
1363 if (atomic_read(&shost->host_blocked) > 0) {
1364 if (scsi_host_busy(shost) > 0)
1368 * unblock after host_blocked iterates to zero
1370 if (atomic_dec_return(&shost->host_blocked) > 0)
1374 shost_printk(KERN_INFO, shost,
1375 "unblocking host at zero depth\n"));
1378 if (shost->host_self_blocked)
1381 /* We're OK to process the command, so we can't be starved */
1382 if (!list_empty(&sdev->starved_entry)) {
1383 spin_lock_irq(shost->host_lock);
1384 if (!list_empty(&sdev->starved_entry))
1385 list_del_init(&sdev->starved_entry);
1386 spin_unlock_irq(shost->host_lock);
1389 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1394 spin_lock_irq(shost->host_lock);
1395 if (list_empty(&sdev->starved_entry))
1396 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1397 spin_unlock_irq(shost->host_lock);
1399 scsi_dec_host_busy(shost, cmd);
1404 * Busy state exporting function for request stacking drivers.
1406 * For efficiency, no lock is taken to check the busy state of
1407 * shost/starget/sdev, since the returned value is not guaranteed and
1408 * may be changed after request stacking drivers call the function,
1409 * regardless of taking lock or not.
1411 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1412 * needs to return 'not busy'. Otherwise, request stacking drivers
1413 * may hold requests forever.
1415 static bool scsi_mq_lld_busy(struct request_queue *q)
1417 struct scsi_device *sdev = q->queuedata;
1418 struct Scsi_Host *shost;
1420 if (blk_queue_dying(q))
1426 * Ignore host/starget busy state.
1427 * Since block layer does not have a concept of fairness across
1428 * multiple queues, congestion of host/starget needs to be handled
1431 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1438 * Block layer request completion callback. May be called from interrupt
1441 static void scsi_complete(struct request *rq)
1443 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1444 enum scsi_disposition disposition;
1446 INIT_LIST_HEAD(&cmd->eh_entry);
1448 atomic_inc(&cmd->device->iodone_cnt);
1450 atomic_inc(&cmd->device->ioerr_cnt);
1452 disposition = scsi_decide_disposition(cmd);
1453 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1454 disposition = SUCCESS;
1456 scsi_log_completion(cmd, disposition);
1458 switch (disposition) {
1460 scsi_finish_command(cmd);
1463 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1465 case ADD_TO_MLQUEUE:
1466 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1469 scsi_eh_scmd_add(cmd);
1475 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1476 * @cmd: command block we are dispatching.
1478 * Return: nonzero return request was rejected and device's queue needs to be
1481 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1483 struct Scsi_Host *host = cmd->device->host;
1486 atomic_inc(&cmd->device->iorequest_cnt);
1488 /* check if the device is still usable */
1489 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1490 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1491 * returns an immediate error upwards, and signals
1492 * that the device is no longer present */
1493 cmd->result = DID_NO_CONNECT << 16;
1497 /* Check to see if the scsi lld made this device blocked. */
1498 if (unlikely(scsi_device_blocked(cmd->device))) {
1500 * in blocked state, the command is just put back on
1501 * the device queue. The suspend state has already
1502 * blocked the queue so future requests should not
1503 * occur until the device transitions out of the
1506 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1507 "queuecommand : device blocked\n"));
1508 return SCSI_MLQUEUE_DEVICE_BUSY;
1511 /* Store the LUN value in cmnd, if needed. */
1512 if (cmd->device->lun_in_cdb)
1513 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1514 (cmd->device->lun << 5 & 0xe0);
1519 * Before we queue this command, check if the command
1520 * length exceeds what the host adapter can handle.
1522 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1523 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1524 "queuecommand : command too long. "
1525 "cdb_size=%d host->max_cmd_len=%d\n",
1526 cmd->cmd_len, cmd->device->host->max_cmd_len));
1527 cmd->result = (DID_ABORT << 16);
1531 if (unlikely(host->shost_state == SHOST_DEL)) {
1532 cmd->result = (DID_NO_CONNECT << 16);
1537 trace_scsi_dispatch_cmd_start(cmd);
1538 rtn = host->hostt->queuecommand(host, cmd);
1540 trace_scsi_dispatch_cmd_error(cmd, rtn);
1541 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1542 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1543 rtn = SCSI_MLQUEUE_HOST_BUSY;
1545 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1546 "queuecommand : request rejected\n"));
1551 cmd->scsi_done(cmd);
1555 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1556 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1558 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1559 sizeof(struct scatterlist);
1562 static blk_status_t scsi_prepare_cmd(struct request *req)
1564 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1565 struct scsi_device *sdev = req->q->queuedata;
1566 struct Scsi_Host *shost = sdev->host;
1567 struct scatterlist *sg;
1569 scsi_init_command(sdev, cmd);
1572 cmd->tag = req->tag;
1573 cmd->prot_op = SCSI_PROT_NORMAL;
1574 if (blk_rq_bytes(req))
1575 cmd->sc_data_direction = rq_dma_dir(req);
1577 cmd->sc_data_direction = DMA_NONE;
1579 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 cmd->sdb.table.sgl = sg;
1582 if (scsi_host_get_prot(shost)) {
1583 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1585 cmd->prot_sdb->table.sgl =
1586 (struct scatterlist *)(cmd->prot_sdb + 1);
1590 * Special handling for passthrough commands, which don't go to the ULP
1593 if (blk_rq_is_scsi(req))
1594 return scsi_setup_scsi_cmnd(sdev, req);
1596 if (sdev->handler && sdev->handler->prep_fn) {
1597 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1599 if (ret != BLK_STS_OK)
1603 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1604 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1605 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1608 static void scsi_mq_done(struct scsi_cmnd *cmd)
1610 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1612 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1614 trace_scsi_dispatch_cmd_done(cmd);
1615 blk_mq_complete_request(cmd->request);
1618 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1620 struct scsi_device *sdev = q->queuedata;
1622 sbitmap_put(&sdev->budget_map, budget_token);
1625 static int scsi_mq_get_budget(struct request_queue *q)
1627 struct scsi_device *sdev = q->queuedata;
1628 int token = scsi_dev_queue_ready(q, sdev);
1633 atomic_inc(&sdev->restarts);
1636 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1637 * .restarts must be incremented before .device_busy is read because the
1638 * code in scsi_run_queue_async() depends on the order of these operations.
1640 smp_mb__after_atomic();
1643 * If all in-flight requests originated from this LUN are completed
1644 * before reading .device_busy, sdev->device_busy will be observed as
1645 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1646 * soon. Otherwise, completion of one of these requests will observe
1647 * the .restarts flag, and the request queue will be run for handling
1648 * this request, see scsi_end_request().
1650 if (unlikely(scsi_device_busy(sdev) == 0 &&
1651 !scsi_device_blocked(sdev)))
1652 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1656 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1658 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1660 cmd->budget_token = token;
1663 static int scsi_mq_get_rq_budget_token(struct request *req)
1665 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1667 return cmd->budget_token;
1670 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1671 const struct blk_mq_queue_data *bd)
1673 struct request *req = bd->rq;
1674 struct request_queue *q = req->q;
1675 struct scsi_device *sdev = q->queuedata;
1676 struct Scsi_Host *shost = sdev->host;
1677 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1681 WARN_ON_ONCE(cmd->budget_token < 0);
1684 * If the device is not in running state we will reject some or all
1687 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1688 ret = scsi_device_state_check(sdev, req);
1689 if (ret != BLK_STS_OK)
1690 goto out_put_budget;
1693 ret = BLK_STS_RESOURCE;
1694 if (!scsi_target_queue_ready(shost, sdev))
1695 goto out_put_budget;
1696 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1697 goto out_dec_target_busy;
1699 if (!(req->rq_flags & RQF_DONTPREP)) {
1700 ret = scsi_prepare_cmd(req);
1701 if (ret != BLK_STS_OK)
1702 goto out_dec_host_busy;
1703 req->rq_flags |= RQF_DONTPREP;
1705 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1708 cmd->flags &= SCMD_PRESERVED_FLAGS;
1709 if (sdev->simple_tags)
1710 cmd->flags |= SCMD_TAGGED;
1712 cmd->flags |= SCMD_LAST;
1714 scsi_set_resid(cmd, 0);
1715 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1716 cmd->scsi_done = scsi_mq_done;
1718 blk_mq_start_request(req);
1719 reason = scsi_dispatch_cmd(cmd);
1721 scsi_set_blocked(cmd, reason);
1722 ret = BLK_STS_RESOURCE;
1723 goto out_dec_host_busy;
1729 scsi_dec_host_busy(shost, cmd);
1730 out_dec_target_busy:
1731 if (scsi_target(sdev)->can_queue > 0)
1732 atomic_dec(&scsi_target(sdev)->target_busy);
1734 scsi_mq_put_budget(q, cmd->budget_token);
1735 cmd->budget_token = -1;
1739 case BLK_STS_RESOURCE:
1740 case BLK_STS_ZONE_RESOURCE:
1741 if (scsi_device_blocked(sdev))
1742 ret = BLK_STS_DEV_RESOURCE;
1745 scsi_req(req)->result = DID_BUS_BUSY << 16;
1746 if (req->rq_flags & RQF_DONTPREP)
1747 scsi_mq_uninit_cmd(cmd);
1750 if (unlikely(!scsi_device_online(sdev)))
1751 scsi_req(req)->result = DID_NO_CONNECT << 16;
1753 scsi_req(req)->result = DID_ERROR << 16;
1755 * Make sure to release all allocated resources when
1756 * we hit an error, as we will never see this command
1759 if (req->rq_flags & RQF_DONTPREP)
1760 scsi_mq_uninit_cmd(cmd);
1761 scsi_run_queue_async(sdev);
1767 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1771 return BLK_EH_RESET_TIMER;
1772 return scsi_times_out(req);
1775 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1776 unsigned int hctx_idx, unsigned int numa_node)
1778 struct Scsi_Host *shost = set->driver_data;
1779 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1780 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1781 struct scatterlist *sg;
1784 if (unchecked_isa_dma)
1785 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1786 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1787 GFP_KERNEL, numa_node);
1788 if (!cmd->sense_buffer)
1790 cmd->req.sense = cmd->sense_buffer;
1792 if (scsi_host_get_prot(shost)) {
1793 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1794 shost->hostt->cmd_size;
1795 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1798 if (shost->hostt->init_cmd_priv) {
1799 ret = shost->hostt->init_cmd_priv(shost, cmd);
1801 scsi_free_sense_buffer(unchecked_isa_dma,
1808 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1809 unsigned int hctx_idx)
1811 struct Scsi_Host *shost = set->driver_data;
1812 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1814 if (shost->hostt->exit_cmd_priv)
1815 shost->hostt->exit_cmd_priv(shost, cmd);
1816 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1821 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
1823 struct Scsi_Host *shost = hctx->driver_data;
1825 if (shost->hostt->mq_poll)
1826 return shost->hostt->mq_poll(shost, hctx->queue_num);
1831 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1832 unsigned int hctx_idx)
1834 struct Scsi_Host *shost = data;
1836 hctx->driver_data = shost;
1840 static int scsi_map_queues(struct blk_mq_tag_set *set)
1842 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1844 if (shost->hostt->map_queues)
1845 return shost->hostt->map_queues(shost);
1846 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1849 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1851 struct device *dev = shost->dma_dev;
1854 * this limit is imposed by hardware restrictions
1856 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1859 if (scsi_host_prot_dma(shost)) {
1860 shost->sg_prot_tablesize =
1861 min_not_zero(shost->sg_prot_tablesize,
1862 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1863 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1864 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1867 if (dev->dma_mask) {
1868 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1869 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1871 blk_queue_max_hw_sectors(q, shost->max_sectors);
1872 if (shost->unchecked_isa_dma)
1873 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1874 blk_queue_segment_boundary(q, shost->dma_boundary);
1875 dma_set_seg_boundary(dev, shost->dma_boundary);
1877 blk_queue_max_segment_size(q, shost->max_segment_size);
1878 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1879 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1882 * Set a reasonable default alignment: The larger of 32-byte (dword),
1883 * which is a common minimum for HBAs, and the minimum DMA alignment,
1884 * which is set by the platform.
1886 * Devices that require a bigger alignment can increase it later.
1888 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1890 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1892 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1893 .get_budget = scsi_mq_get_budget,
1894 .put_budget = scsi_mq_put_budget,
1895 .queue_rq = scsi_queue_rq,
1896 .complete = scsi_complete,
1897 .timeout = scsi_timeout,
1898 #ifdef CONFIG_BLK_DEBUG_FS
1899 .show_rq = scsi_show_rq,
1901 .init_request = scsi_mq_init_request,
1902 .exit_request = scsi_mq_exit_request,
1903 .initialize_rq_fn = scsi_initialize_rq,
1904 .cleanup_rq = scsi_cleanup_rq,
1905 .busy = scsi_mq_lld_busy,
1906 .map_queues = scsi_map_queues,
1907 .init_hctx = scsi_init_hctx,
1908 .poll = scsi_mq_poll,
1909 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1910 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1914 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1916 struct Scsi_Host *shost = hctx->driver_data;
1918 shost->hostt->commit_rqs(shost, hctx->queue_num);
1921 static const struct blk_mq_ops scsi_mq_ops = {
1922 .get_budget = scsi_mq_get_budget,
1923 .put_budget = scsi_mq_put_budget,
1924 .queue_rq = scsi_queue_rq,
1925 .commit_rqs = scsi_commit_rqs,
1926 .complete = scsi_complete,
1927 .timeout = scsi_timeout,
1928 #ifdef CONFIG_BLK_DEBUG_FS
1929 .show_rq = scsi_show_rq,
1931 .init_request = scsi_mq_init_request,
1932 .exit_request = scsi_mq_exit_request,
1933 .initialize_rq_fn = scsi_initialize_rq,
1934 .cleanup_rq = scsi_cleanup_rq,
1935 .busy = scsi_mq_lld_busy,
1936 .map_queues = scsi_map_queues,
1937 .init_hctx = scsi_init_hctx,
1938 .poll = scsi_mq_poll,
1939 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1940 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1943 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1945 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1946 if (IS_ERR(sdev->request_queue))
1949 sdev->request_queue->queuedata = sdev;
1950 __scsi_init_queue(sdev->host, sdev->request_queue);
1951 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1952 return sdev->request_queue;
1955 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1957 unsigned int cmd_size, sgl_size;
1958 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1960 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1961 scsi_mq_inline_sgl_size(shost));
1962 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1963 if (scsi_host_get_prot(shost))
1964 cmd_size += sizeof(struct scsi_data_buffer) +
1965 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1967 memset(tag_set, 0, sizeof(*tag_set));
1968 if (shost->hostt->commit_rqs)
1969 tag_set->ops = &scsi_mq_ops;
1971 tag_set->ops = &scsi_mq_ops_no_commit;
1972 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1973 tag_set->nr_maps = shost->nr_maps ? : 1;
1974 tag_set->queue_depth = shost->can_queue;
1975 tag_set->cmd_size = cmd_size;
1976 tag_set->numa_node = NUMA_NO_NODE;
1977 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1979 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1980 tag_set->driver_data = shost;
1981 if (shost->host_tagset)
1982 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1984 return blk_mq_alloc_tag_set(tag_set);
1987 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1989 blk_mq_free_tag_set(&shost->tag_set);
1993 * scsi_device_from_queue - return sdev associated with a request_queue
1994 * @q: The request queue to return the sdev from
1996 * Return the sdev associated with a request queue or NULL if the
1997 * request_queue does not reference a SCSI device.
1999 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2001 struct scsi_device *sdev = NULL;
2003 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2004 q->mq_ops == &scsi_mq_ops)
2005 sdev = q->queuedata;
2006 if (!sdev || !get_device(&sdev->sdev_gendev))
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);
2046 kmem_cache_destroy(scsi_sense_isadma_cache);
2050 * scsi_mode_select - issue a mode select
2051 * @sdev: SCSI device to be queried
2052 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2053 * @sp: Save page bit (0 == don't save, 1 == save)
2054 * @modepage: mode page being requested
2055 * @buffer: request buffer (may not be smaller than eight bytes)
2056 * @len: length of request buffer.
2057 * @timeout: command timeout
2058 * @retries: number of retries before failing
2059 * @data: returns a structure abstracting the mode header data
2060 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2061 * must be SCSI_SENSE_BUFFERSIZE big.
2063 * Returns zero if successful; negative error number or scsi
2068 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2069 unsigned char *buffer, int len, int timeout, int retries,
2070 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2072 unsigned char cmd[10];
2073 unsigned char *real_buffer;
2076 memset(cmd, 0, sizeof(cmd));
2077 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2079 if (sdev->use_10_for_ms) {
2082 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2085 memcpy(real_buffer + 8, buffer, len);
2089 real_buffer[2] = data->medium_type;
2090 real_buffer[3] = data->device_specific;
2091 real_buffer[4] = data->longlba ? 0x01 : 0;
2093 real_buffer[6] = data->block_descriptor_length >> 8;
2094 real_buffer[7] = data->block_descriptor_length;
2096 cmd[0] = MODE_SELECT_10;
2100 if (len > 255 || data->block_descriptor_length > 255 ||
2104 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2107 memcpy(real_buffer + 4, buffer, len);
2110 real_buffer[1] = data->medium_type;
2111 real_buffer[2] = data->device_specific;
2112 real_buffer[3] = data->block_descriptor_length;
2114 cmd[0] = MODE_SELECT;
2118 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2119 sshdr, timeout, retries, NULL);
2123 EXPORT_SYMBOL_GPL(scsi_mode_select);
2126 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2127 * @sdev: SCSI device to be queried
2128 * @dbd: set if mode sense will allow block descriptors to be returned
2129 * @modepage: mode page being requested
2130 * @buffer: request buffer (may not be smaller than eight bytes)
2131 * @len: length of request buffer.
2132 * @timeout: command timeout
2133 * @retries: number of retries before failing
2134 * @data: returns a structure abstracting the mode header data
2135 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2136 * must be SCSI_SENSE_BUFFERSIZE big.
2138 * Returns zero if unsuccessful, or the header offset (either 4
2139 * or 8 depending on whether a six or ten byte command was
2140 * issued) if successful.
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;
2167 if (use_10_for_ms) {
2171 cmd[0] = MODE_SENSE_10;
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);
2188 /* This code looks awful: what it's doing is making sure an
2189 * ILLEGAL REQUEST sense return identifies the actual command
2190 * byte as the problem. MODE_SENSE commands can return
2191 * ILLEGAL REQUEST if the code page isn't supported */
2193 if (use_10_for_ms && !scsi_status_is_good(result) &&
2194 driver_byte(result) == DRIVER_SENSE) {
2195 if (scsi_sense_valid(sshdr)) {
2196 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2197 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2199 * Invalid command operation code
2201 sdev->use_10_for_ms = 0;
2207 if (scsi_status_is_good(result)) {
2208 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2209 (modepage == 6 || modepage == 8))) {
2210 /* Initio breakage? */
2213 data->medium_type = 0;
2214 data->device_specific = 0;
2216 data->block_descriptor_length = 0;
2217 } else if (use_10_for_ms) {
2218 data->length = buffer[0]*256 + buffer[1] + 2;
2219 data->medium_type = buffer[2];
2220 data->device_specific = buffer[3];
2221 data->longlba = buffer[4] & 0x01;
2222 data->block_descriptor_length = buffer[6]*256
2225 data->length = buffer[0] + 1;
2226 data->medium_type = buffer[1];
2227 data->device_specific = buffer[2];
2228 data->block_descriptor_length = buffer[3];
2230 data->header_length = header_length;
2231 } else if ((status_byte(result) == CHECK_CONDITION) &&
2232 scsi_sense_valid(sshdr) &&
2233 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2240 EXPORT_SYMBOL(scsi_mode_sense);
2243 * scsi_test_unit_ready - test if unit is ready
2244 * @sdev: scsi device to change the state of.
2245 * @timeout: command timeout
2246 * @retries: number of retries before failing
2247 * @sshdr: outpout pointer for decoded sense information.
2249 * Returns zero if unsuccessful or an error if TUR failed. For
2250 * removable media, UNIT_ATTENTION sets ->changed flag.
2253 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2254 struct scsi_sense_hdr *sshdr)
2257 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2261 /* try to eat the UNIT_ATTENTION if there are enough retries */
2263 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2265 if (sdev->removable && scsi_sense_valid(sshdr) &&
2266 sshdr->sense_key == UNIT_ATTENTION)
2268 } while (scsi_sense_valid(sshdr) &&
2269 sshdr->sense_key == UNIT_ATTENTION && --retries);
2273 EXPORT_SYMBOL(scsi_test_unit_ready);
2276 * scsi_device_set_state - Take the given device through the device state model.
2277 * @sdev: scsi device to change the state of.
2278 * @state: state to change to.
2280 * Returns zero if successful or an error if the requested
2281 * transition is illegal.
2284 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2286 enum scsi_device_state oldstate = sdev->sdev_state;
2288 if (state == oldstate)
2294 case SDEV_CREATED_BLOCK:
2305 case SDEV_TRANSPORT_OFFLINE:
2318 case SDEV_TRANSPORT_OFFLINE:
2326 case SDEV_TRANSPORT_OFFLINE:
2341 case SDEV_CREATED_BLOCK:
2350 case SDEV_CREATED_BLOCK:
2365 case SDEV_TRANSPORT_OFFLINE:
2377 case SDEV_TRANSPORT_OFFLINE:
2380 case SDEV_CREATED_BLOCK:
2388 sdev->offline_already = false;
2389 sdev->sdev_state = state;
2393 SCSI_LOG_ERROR_RECOVERY(1,
2394 sdev_printk(KERN_ERR, sdev,
2395 "Illegal state transition %s->%s",
2396 scsi_device_state_name(oldstate),
2397 scsi_device_state_name(state))
2401 EXPORT_SYMBOL(scsi_device_set_state);
2404 * scsi_evt_emit - emit a single SCSI device uevent
2405 * @sdev: associated SCSI device
2406 * @evt: event to emit
2408 * Send a single uevent (scsi_event) to the associated scsi_device.
2410 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2415 switch (evt->evt_type) {
2416 case SDEV_EVT_MEDIA_CHANGE:
2417 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2419 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2420 scsi_rescan_device(&sdev->sdev_gendev);
2421 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2423 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2424 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2426 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2427 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2429 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2430 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2432 case SDEV_EVT_LUN_CHANGE_REPORTED:
2433 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2435 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2436 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2438 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2439 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2448 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2452 * scsi_evt_thread - send a uevent for each scsi event
2453 * @work: work struct for scsi_device
2455 * Dispatch queued events to their associated scsi_device kobjects
2458 void scsi_evt_thread(struct work_struct *work)
2460 struct scsi_device *sdev;
2461 enum scsi_device_event evt_type;
2462 LIST_HEAD(event_list);
2464 sdev = container_of(work, struct scsi_device, event_work);
2466 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2467 if (test_and_clear_bit(evt_type, sdev->pending_events))
2468 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2471 struct scsi_event *evt;
2472 struct list_head *this, *tmp;
2473 unsigned long flags;
2475 spin_lock_irqsave(&sdev->list_lock, flags);
2476 list_splice_init(&sdev->event_list, &event_list);
2477 spin_unlock_irqrestore(&sdev->list_lock, flags);
2479 if (list_empty(&event_list))
2482 list_for_each_safe(this, tmp, &event_list) {
2483 evt = list_entry(this, struct scsi_event, node);
2484 list_del(&evt->node);
2485 scsi_evt_emit(sdev, evt);
2492 * sdev_evt_send - send asserted event to uevent thread
2493 * @sdev: scsi_device event occurred on
2494 * @evt: event to send
2496 * Assert scsi device event asynchronously.
2498 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2500 unsigned long flags;
2503 /* FIXME: currently this check eliminates all media change events
2504 * for polled devices. Need to update to discriminate between AN
2505 * and polled events */
2506 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2512 spin_lock_irqsave(&sdev->list_lock, flags);
2513 list_add_tail(&evt->node, &sdev->event_list);
2514 schedule_work(&sdev->event_work);
2515 spin_unlock_irqrestore(&sdev->list_lock, flags);
2517 EXPORT_SYMBOL_GPL(sdev_evt_send);
2520 * sdev_evt_alloc - allocate a new scsi event
2521 * @evt_type: type of event to allocate
2522 * @gfpflags: GFP flags for allocation
2524 * Allocates and returns a new scsi_event.
2526 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2529 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2533 evt->evt_type = evt_type;
2534 INIT_LIST_HEAD(&evt->node);
2536 /* evt_type-specific initialization, if any */
2538 case SDEV_EVT_MEDIA_CHANGE:
2539 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2540 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2541 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2542 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2543 case SDEV_EVT_LUN_CHANGE_REPORTED:
2544 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2545 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2553 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2556 * sdev_evt_send_simple - send asserted event to uevent thread
2557 * @sdev: scsi_device event occurred on
2558 * @evt_type: type of event to send
2559 * @gfpflags: GFP flags for allocation
2561 * Assert scsi device event asynchronously, given an event type.
2563 void sdev_evt_send_simple(struct scsi_device *sdev,
2564 enum scsi_device_event evt_type, gfp_t gfpflags)
2566 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2568 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2573 sdev_evt_send(sdev, evt);
2575 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2578 * scsi_device_quiesce - Block all commands except power management.
2579 * @sdev: scsi device to quiesce.
2581 * This works by trying to transition to the SDEV_QUIESCE state
2582 * (which must be a legal transition). When the device is in this
2583 * state, only power management requests will be accepted, all others will
2586 * Must be called with user context, may sleep.
2588 * Returns zero if unsuccessful or an error if not.
2591 scsi_device_quiesce(struct scsi_device *sdev)
2593 struct request_queue *q = sdev->request_queue;
2597 * It is allowed to call scsi_device_quiesce() multiple times from
2598 * the same context but concurrent scsi_device_quiesce() calls are
2601 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2603 if (sdev->quiesced_by == current)
2608 blk_mq_freeze_queue(q);
2610 * Ensure that the effect of blk_set_pm_only() will be visible
2611 * for percpu_ref_tryget() callers that occur after the queue
2612 * unfreeze even if the queue was already frozen before this function
2613 * was called. See also https://lwn.net/Articles/573497/.
2616 blk_mq_unfreeze_queue(q);
2618 mutex_lock(&sdev->state_mutex);
2619 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2621 sdev->quiesced_by = current;
2623 blk_clear_pm_only(q);
2624 mutex_unlock(&sdev->state_mutex);
2628 EXPORT_SYMBOL(scsi_device_quiesce);
2631 * scsi_device_resume - Restart user issued commands to a quiesced device.
2632 * @sdev: scsi device to resume.
2634 * Moves the device from quiesced back to running and restarts the
2637 * Must be called with user context, may sleep.
2639 void scsi_device_resume(struct scsi_device *sdev)
2641 /* check if the device state was mutated prior to resume, and if
2642 * so assume the state is being managed elsewhere (for example
2643 * device deleted during suspend)
2645 mutex_lock(&sdev->state_mutex);
2646 if (sdev->sdev_state == SDEV_QUIESCE)
2647 scsi_device_set_state(sdev, SDEV_RUNNING);
2648 if (sdev->quiesced_by) {
2649 sdev->quiesced_by = NULL;
2650 blk_clear_pm_only(sdev->request_queue);
2652 mutex_unlock(&sdev->state_mutex);
2654 EXPORT_SYMBOL(scsi_device_resume);
2657 device_quiesce_fn(struct scsi_device *sdev, void *data)
2659 scsi_device_quiesce(sdev);
2663 scsi_target_quiesce(struct scsi_target *starget)
2665 starget_for_each_device(starget, NULL, device_quiesce_fn);
2667 EXPORT_SYMBOL(scsi_target_quiesce);
2670 device_resume_fn(struct scsi_device *sdev, void *data)
2672 scsi_device_resume(sdev);
2676 scsi_target_resume(struct scsi_target *starget)
2678 starget_for_each_device(starget, NULL, device_resume_fn);
2680 EXPORT_SYMBOL(scsi_target_resume);
2683 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2684 * @sdev: device to block
2686 * Pause SCSI command processing on the specified device. Does not sleep.
2688 * Returns zero if successful or a negative error code upon failure.
2691 * This routine transitions the device to the SDEV_BLOCK state (which must be
2692 * a legal transition). When the device is in this state, command processing
2693 * is paused until the device leaves the SDEV_BLOCK state. See also
2694 * scsi_internal_device_unblock_nowait().
2696 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2698 struct request_queue *q = sdev->request_queue;
2701 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2703 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2710 * The device has transitioned to SDEV_BLOCK. Stop the
2711 * block layer from calling the midlayer with this device's
2714 blk_mq_quiesce_queue_nowait(q);
2717 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2720 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2721 * @sdev: device to block
2723 * Pause SCSI command processing on the specified device and wait until all
2724 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2726 * Returns zero if successful or a negative error code upon failure.
2729 * This routine transitions the device to the SDEV_BLOCK state (which must be
2730 * a legal transition). When the device is in this state, command processing
2731 * is paused until the device leaves the SDEV_BLOCK state. See also
2732 * scsi_internal_device_unblock().
2734 static int scsi_internal_device_block(struct scsi_device *sdev)
2736 struct request_queue *q = sdev->request_queue;
2739 mutex_lock(&sdev->state_mutex);
2740 err = scsi_internal_device_block_nowait(sdev);
2742 blk_mq_quiesce_queue(q);
2743 mutex_unlock(&sdev->state_mutex);
2748 void scsi_start_queue(struct scsi_device *sdev)
2750 struct request_queue *q = sdev->request_queue;
2752 blk_mq_unquiesce_queue(q);
2756 * scsi_internal_device_unblock_nowait - resume a device after a block request
2757 * @sdev: device to resume
2758 * @new_state: state to set the device to after unblocking
2760 * Restart the device queue for a previously suspended SCSI device. Does not
2763 * Returns zero if successful or a negative error code upon failure.
2766 * This routine transitions the device to the SDEV_RUNNING state or to one of
2767 * the offline states (which must be a legal transition) allowing the midlayer
2768 * to goose the queue for this device.
2770 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2771 enum scsi_device_state new_state)
2773 switch (new_state) {
2775 case SDEV_TRANSPORT_OFFLINE:
2782 * Try to transition the scsi device to SDEV_RUNNING or one of the
2783 * offlined states and goose the device queue if successful.
2785 switch (sdev->sdev_state) {
2787 case SDEV_TRANSPORT_OFFLINE:
2788 sdev->sdev_state = new_state;
2790 case SDEV_CREATED_BLOCK:
2791 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2792 new_state == SDEV_OFFLINE)
2793 sdev->sdev_state = new_state;
2795 sdev->sdev_state = SDEV_CREATED;
2803 scsi_start_queue(sdev);
2807 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2810 * scsi_internal_device_unblock - resume a device after a block request
2811 * @sdev: device to resume
2812 * @new_state: state to set the device to after unblocking
2814 * Restart the device queue for a previously suspended SCSI device. May sleep.
2816 * Returns zero if successful or a negative error code upon failure.
2819 * This routine transitions the device to the SDEV_RUNNING state or to one of
2820 * the offline states (which must be a legal transition) allowing the midlayer
2821 * to goose the queue for this device.
2823 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2824 enum scsi_device_state new_state)
2828 mutex_lock(&sdev->state_mutex);
2829 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2830 mutex_unlock(&sdev->state_mutex);
2836 device_block(struct scsi_device *sdev, void *data)
2840 ret = scsi_internal_device_block(sdev);
2842 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2843 dev_name(&sdev->sdev_gendev), ret);
2847 target_block(struct device *dev, void *data)
2849 if (scsi_is_target_device(dev))
2850 starget_for_each_device(to_scsi_target(dev), NULL,
2856 scsi_target_block(struct device *dev)
2858 if (scsi_is_target_device(dev))
2859 starget_for_each_device(to_scsi_target(dev), NULL,
2862 device_for_each_child(dev, NULL, target_block);
2864 EXPORT_SYMBOL_GPL(scsi_target_block);
2867 device_unblock(struct scsi_device *sdev, void *data)
2869 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2873 target_unblock(struct device *dev, void *data)
2875 if (scsi_is_target_device(dev))
2876 starget_for_each_device(to_scsi_target(dev), data,
2882 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2884 if (scsi_is_target_device(dev))
2885 starget_for_each_device(to_scsi_target(dev), &new_state,
2888 device_for_each_child(dev, &new_state, target_unblock);
2890 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2893 scsi_host_block(struct Scsi_Host *shost)
2895 struct scsi_device *sdev;
2899 * Call scsi_internal_device_block_nowait so we can avoid
2900 * calling synchronize_rcu() for each LUN.
2902 shost_for_each_device(sdev, shost) {
2903 mutex_lock(&sdev->state_mutex);
2904 ret = scsi_internal_device_block_nowait(sdev);
2905 mutex_unlock(&sdev->state_mutex);
2907 scsi_device_put(sdev);
2913 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2914 * calling synchronize_rcu() once is enough.
2916 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2923 EXPORT_SYMBOL_GPL(scsi_host_block);
2926 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2928 struct scsi_device *sdev;
2931 shost_for_each_device(sdev, shost) {
2932 ret = scsi_internal_device_unblock(sdev, new_state);
2934 scsi_device_put(sdev);
2940 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2943 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2944 * @sgl: scatter-gather list
2945 * @sg_count: number of segments in sg
2946 * @offset: offset in bytes into sg, on return offset into the mapped area
2947 * @len: bytes to map, on return number of bytes mapped
2949 * Returns virtual address of the start of the mapped page
2951 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2952 size_t *offset, size_t *len)
2955 size_t sg_len = 0, len_complete = 0;
2956 struct scatterlist *sg;
2959 WARN_ON(!irqs_disabled());
2961 for_each_sg(sgl, sg, sg_count, i) {
2962 len_complete = sg_len; /* Complete sg-entries */
2963 sg_len += sg->length;
2964 if (sg_len > *offset)
2968 if (unlikely(i == sg_count)) {
2969 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2971 __func__, sg_len, *offset, sg_count);
2976 /* Offset starting from the beginning of first page in this sg-entry */
2977 *offset = *offset - len_complete + sg->offset;
2979 /* Assumption: contiguous pages can be accessed as "page + i" */
2980 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2981 *offset &= ~PAGE_MASK;
2983 /* Bytes in this sg-entry from *offset to the end of the page */
2984 sg_len = PAGE_SIZE - *offset;
2988 return kmap_atomic(page);
2990 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2993 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2994 * @virt: virtual address to be unmapped
2996 void scsi_kunmap_atomic_sg(void *virt)
2998 kunmap_atomic(virt);
3000 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3002 void sdev_disable_disk_events(struct scsi_device *sdev)
3004 atomic_inc(&sdev->disk_events_disable_depth);
3006 EXPORT_SYMBOL(sdev_disable_disk_events);
3008 void sdev_enable_disk_events(struct scsi_device *sdev)
3010 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3012 atomic_dec(&sdev->disk_events_disable_depth);
3014 EXPORT_SYMBOL(sdev_enable_disk_events);
3016 static unsigned char designator_prio(const unsigned char *d)
3019 /* not associated with LUN */
3023 /* invalid length */
3027 * Order of preference for lun descriptor:
3028 * - SCSI name string
3029 * - NAA IEEE Registered Extended
3030 * - EUI-64 based 16-byte
3031 * - EUI-64 based 12-byte
3032 * - NAA IEEE Registered
3033 * - NAA IEEE Extended
3034 * - EUI-64 based 8-byte
3035 * - SCSI name string (truncated)
3037 * as longer descriptors reduce the likelyhood
3038 * of identification clashes.
3041 switch (d[1] & 0xf) {
3043 /* SCSI name string, variable-length UTF-8 */
3046 switch (d[4] >> 4) {
3048 /* NAA registered extended */
3051 /* NAA registered */
3057 /* NAA locally assigned */
3066 /* EUI64-based, 16 byte */
3069 /* EUI64-based, 12 byte */
3072 /* EUI64-based, 8 byte */
3089 * scsi_vpd_lun_id - return a unique device identification
3090 * @sdev: SCSI device
3091 * @id: buffer for the identification
3092 * @id_len: length of the buffer
3094 * Copies a unique device identification into @id based
3095 * on the information in the VPD page 0x83 of the device.
3096 * The string will be formatted as a SCSI name string.
3098 * Returns the length of the identification or error on failure.
3099 * If the identifier is longer than the supplied buffer the actual
3100 * identifier length is returned and the buffer is not zero-padded.
3102 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3106 const unsigned char *d, *cur_id_str;
3107 const struct scsi_vpd *vpd_pg83;
3108 int id_size = -EINVAL;
3111 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3117 /* The id string must be at least 20 bytes + terminating NULL byte */
3123 memset(id, 0, id_len);
3124 for (d = vpd_pg83->data + 4;
3125 d < vpd_pg83->data + vpd_pg83->len;
3127 u8 prio = designator_prio(d);
3129 if (prio == 0 || cur_id_prio > prio)
3132 switch (d[1] & 0xf) {
3135 if (cur_id_size > d[3])
3139 if (cur_id_size + 4 > id_len)
3140 cur_id_size = id_len - 4;
3142 id_size = snprintf(id, id_len, "t10.%*pE",
3143 cur_id_size, cur_id_str);
3150 switch (cur_id_size) {
3152 id_size = snprintf(id, id_len,
3157 id_size = snprintf(id, id_len,
3162 id_size = snprintf(id, id_len,
3175 switch (cur_id_size) {
3177 id_size = snprintf(id, id_len,
3182 id_size = snprintf(id, id_len,
3191 /* SCSI name string */
3192 if (cur_id_size > d[3])
3194 /* Prefer others for truncated descriptor */
3195 if (d[3] > id_len) {
3197 if (cur_id_prio > prio)
3201 cur_id_size = id_size = d[3];
3203 if (cur_id_size >= id_len)
3204 cur_id_size = id_len - 1;
3205 memcpy(id, cur_id_str, cur_id_size);
3215 EXPORT_SYMBOL(scsi_vpd_lun_id);
3218 * scsi_vpd_tpg_id - return a target port group identifier
3219 * @sdev: SCSI device
3221 * Returns the Target Port Group identifier from the information
3222 * froom VPD page 0x83 of the device.
3224 * Returns the identifier or error on failure.
3226 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3228 const unsigned char *d;
3229 const struct scsi_vpd *vpd_pg83;
3230 int group_id = -EAGAIN, rel_port = -1;
3233 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3239 d = vpd_pg83->data + 4;
3240 while (d < vpd_pg83->data + vpd_pg83->len) {
3241 switch (d[1] & 0xf) {
3243 /* Relative target port */
3244 rel_port = get_unaligned_be16(&d[6]);
3247 /* Target port group */
3248 group_id = get_unaligned_be16(&d[6]);
3257 if (group_id >= 0 && rel_id && rel_port != -1)
3262 EXPORT_SYMBOL(scsi_vpd_tpg_id);