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);
189 * scsi_execute_cmd - insert request and wait for the result
192 * @opf: block layer request cmd_flags
193 * @buffer: data buffer
194 * @bufflen: len of buffer
195 * @timeout: request timeout in HZ
196 * @retries: number of times to retry request
197 * @args: Optional args. See struct definition for field descriptions
199 * Returns the scsi_cmnd result field if a command was executed, or a negative
200 * Linux error code if we didn't get that far.
202 int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
203 blk_opf_t opf, void *buffer, unsigned int bufflen,
204 int timeout, int retries,
205 const struct scsi_exec_args *args)
207 static const struct scsi_exec_args default_args;
209 struct scsi_cmnd *scmd;
213 args = &default_args;
214 else if (WARN_ON_ONCE(args->sense &&
215 args->sense_len != SCSI_SENSE_BUFFERSIZE))
218 req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
223 ret = blk_rq_map_kern(sdev->request_queue, req,
224 buffer, bufflen, GFP_NOIO);
228 scmd = blk_mq_rq_to_pdu(req);
229 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
230 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
231 scmd->allowed = retries;
232 scmd->flags |= args->scmd_flags;
233 req->timeout = timeout;
234 req->rq_flags |= RQF_QUIET;
237 * head injection *required* here otherwise quiesce won't work
239 blk_execute_rq(req, true);
242 * Some devices (USB mass-storage in particular) may transfer
243 * garbage data together with a residue indicating that the data
244 * is invalid. Prevent the garbage from being misinterpreted
245 * and prevent security leaks by zeroing out the excess data.
247 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
248 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
251 *args->resid = scmd->resid_len;
253 memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
255 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
260 blk_mq_free_request(req);
264 EXPORT_SYMBOL(scsi_execute_cmd);
267 * Wake up the error handler if necessary. Avoid as follows that the error
268 * handler is not woken up if host in-flight requests number ==
269 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
270 * with an RCU read lock in this function to ensure that this function in
271 * its entirety either finishes before scsi_eh_scmd_add() increases the
272 * host_failed counter or that it notices the shost state change made by
273 * scsi_eh_scmd_add().
275 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
280 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
281 if (unlikely(scsi_host_in_recovery(shost))) {
282 spin_lock_irqsave(shost->host_lock, flags);
283 if (shost->host_failed || shost->host_eh_scheduled)
284 scsi_eh_wakeup(shost);
285 spin_unlock_irqrestore(shost->host_lock, flags);
290 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
292 struct Scsi_Host *shost = sdev->host;
293 struct scsi_target *starget = scsi_target(sdev);
295 scsi_dec_host_busy(shost, cmd);
297 if (starget->can_queue > 0)
298 atomic_dec(&starget->target_busy);
300 sbitmap_put(&sdev->budget_map, cmd->budget_token);
301 cmd->budget_token = -1;
304 static void scsi_kick_queue(struct request_queue *q)
306 blk_mq_run_hw_queues(q, false);
310 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
311 * interrupts disabled.
313 static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
315 struct scsi_device *current_sdev = data;
317 if (sdev != current_sdev)
318 blk_mq_run_hw_queues(sdev->request_queue, true);
322 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
323 * and call blk_run_queue for all the scsi_devices on the target -
324 * including current_sdev first.
326 * Called with *no* scsi locks held.
328 static void scsi_single_lun_run(struct scsi_device *current_sdev)
330 struct Scsi_Host *shost = current_sdev->host;
331 struct scsi_target *starget = scsi_target(current_sdev);
334 spin_lock_irqsave(shost->host_lock, flags);
335 starget->starget_sdev_user = NULL;
336 spin_unlock_irqrestore(shost->host_lock, flags);
339 * Call blk_run_queue for all LUNs on the target, starting with
340 * current_sdev. We race with others (to set starget_sdev_user),
341 * but in most cases, we will be first. Ideally, each LU on the
342 * target would get some limited time or requests on the target.
344 scsi_kick_queue(current_sdev->request_queue);
346 spin_lock_irqsave(shost->host_lock, flags);
347 if (!starget->starget_sdev_user)
348 __starget_for_each_device(starget, current_sdev,
349 scsi_kick_sdev_queue);
350 spin_unlock_irqrestore(shost->host_lock, flags);
353 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
355 if (scsi_device_busy(sdev) >= sdev->queue_depth)
357 if (atomic_read(&sdev->device_blocked) > 0)
362 static inline bool scsi_target_is_busy(struct scsi_target *starget)
364 if (starget->can_queue > 0) {
365 if (atomic_read(&starget->target_busy) >= starget->can_queue)
367 if (atomic_read(&starget->target_blocked) > 0)
373 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
375 if (atomic_read(&shost->host_blocked) > 0)
377 if (shost->host_self_blocked)
382 static void scsi_starved_list_run(struct Scsi_Host *shost)
384 LIST_HEAD(starved_list);
385 struct scsi_device *sdev;
388 spin_lock_irqsave(shost->host_lock, flags);
389 list_splice_init(&shost->starved_list, &starved_list);
391 while (!list_empty(&starved_list)) {
392 struct request_queue *slq;
395 * As long as shost is accepting commands and we have
396 * starved queues, call blk_run_queue. scsi_request_fn
397 * drops the queue_lock and can add us back to the
400 * host_lock protects the starved_list and starved_entry.
401 * scsi_request_fn must get the host_lock before checking
402 * or modifying starved_list or starved_entry.
404 if (scsi_host_is_busy(shost))
407 sdev = list_entry(starved_list.next,
408 struct scsi_device, starved_entry);
409 list_del_init(&sdev->starved_entry);
410 if (scsi_target_is_busy(scsi_target(sdev))) {
411 list_move_tail(&sdev->starved_entry,
412 &shost->starved_list);
417 * Once we drop the host lock, a racing scsi_remove_device()
418 * call may remove the sdev from the starved list and destroy
419 * it and the queue. Mitigate by taking a reference to the
420 * queue and never touching the sdev again after we drop the
421 * host lock. Note: if __scsi_remove_device() invokes
422 * blk_mq_destroy_queue() before the queue is run from this
423 * function then blk_run_queue() will return immediately since
424 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
426 slq = sdev->request_queue;
427 if (!blk_get_queue(slq))
429 spin_unlock_irqrestore(shost->host_lock, flags);
431 scsi_kick_queue(slq);
434 spin_lock_irqsave(shost->host_lock, flags);
436 /* put any unprocessed entries back */
437 list_splice(&starved_list, &shost->starved_list);
438 spin_unlock_irqrestore(shost->host_lock, flags);
442 * scsi_run_queue - Select a proper request queue to serve next.
443 * @q: last request's queue
445 * The previous command was completely finished, start a new one if possible.
447 static void scsi_run_queue(struct request_queue *q)
449 struct scsi_device *sdev = q->queuedata;
451 if (scsi_target(sdev)->single_lun)
452 scsi_single_lun_run(sdev);
453 if (!list_empty(&sdev->host->starved_list))
454 scsi_starved_list_run(sdev->host);
456 blk_mq_run_hw_queues(q, false);
459 void scsi_requeue_run_queue(struct work_struct *work)
461 struct scsi_device *sdev;
462 struct request_queue *q;
464 sdev = container_of(work, struct scsi_device, requeue_work);
465 q = sdev->request_queue;
469 void scsi_run_host_queues(struct Scsi_Host *shost)
471 struct scsi_device *sdev;
473 shost_for_each_device(sdev, shost)
474 scsi_run_queue(sdev->request_queue);
477 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
479 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
480 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
482 if (drv->uninit_command)
483 drv->uninit_command(cmd);
487 void scsi_free_sgtables(struct scsi_cmnd *cmd)
489 if (cmd->sdb.table.nents)
490 sg_free_table_chained(&cmd->sdb.table,
492 if (scsi_prot_sg_count(cmd))
493 sg_free_table_chained(&cmd->prot_sdb->table,
494 SCSI_INLINE_PROT_SG_CNT);
496 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
498 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
500 scsi_free_sgtables(cmd);
501 scsi_uninit_cmd(cmd);
504 static void scsi_run_queue_async(struct scsi_device *sdev)
506 if (scsi_target(sdev)->single_lun ||
507 !list_empty(&sdev->host->starved_list)) {
508 kblockd_schedule_work(&sdev->requeue_work);
511 * smp_mb() present in sbitmap_queue_clear() or implied in
512 * .end_io is for ordering writing .device_busy in
513 * scsi_device_unbusy() and reading sdev->restarts.
515 int old = atomic_read(&sdev->restarts);
518 * ->restarts has to be kept as non-zero if new budget
521 * No need to run queue when either another re-run
522 * queue wins in updating ->restarts or a new budget
525 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
526 blk_mq_run_hw_queues(sdev->request_queue, true);
530 /* Returns false when no more bytes to process, true if there are more */
531 static bool scsi_end_request(struct request *req, blk_status_t error,
534 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
535 struct scsi_device *sdev = cmd->device;
536 struct request_queue *q = sdev->request_queue;
538 if (blk_update_request(req, error, bytes))
542 if (blk_queue_add_random(q))
543 add_disk_randomness(req->q->disk);
545 if (!blk_rq_is_passthrough(req)) {
546 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
547 cmd->flags &= ~SCMD_INITIALIZED;
551 * Calling rcu_barrier() is not necessary here because the
552 * SCSI error handler guarantees that the function called by
553 * call_rcu() has been called before scsi_end_request() is
556 destroy_rcu_head(&cmd->rcu);
559 * In the MQ case the command gets freed by __blk_mq_end_request,
560 * so we have to do all cleanup that depends on it earlier.
562 * We also can't kick the queues from irq context, so we
563 * will have to defer it to a workqueue.
565 scsi_mq_uninit_cmd(cmd);
568 * queue is still alive, so grab the ref for preventing it
569 * from being cleaned up during running queue.
571 percpu_ref_get(&q->q_usage_counter);
573 __blk_mq_end_request(req, error);
575 scsi_run_queue_async(sdev);
577 percpu_ref_put(&q->q_usage_counter);
581 static inline u8 get_scsi_ml_byte(int result)
583 return (result >> 8) & 0xff;
587 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
588 * @result: scsi error code
590 * Translate a SCSI result code into a blk_status_t value.
592 static blk_status_t scsi_result_to_blk_status(int result)
595 * Check the scsi-ml byte first in case we converted a host or status
598 switch (get_scsi_ml_byte(result)) {
601 case SCSIML_STAT_RESV_CONFLICT:
602 return BLK_STS_NEXUS;
603 case SCSIML_STAT_NOSPC:
604 return BLK_STS_NOSPC;
605 case SCSIML_STAT_MED_ERROR:
606 return BLK_STS_MEDIUM;
607 case SCSIML_STAT_TGT_FAILURE:
608 return BLK_STS_TARGET;
611 switch (host_byte(result)) {
613 if (scsi_status_is_good(result))
615 return BLK_STS_IOERR;
616 case DID_TRANSPORT_FAILFAST:
617 case DID_TRANSPORT_MARGINAL:
618 return BLK_STS_TRANSPORT;
620 return BLK_STS_IOERR;
625 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
626 * @rq: request to examine
629 * A request could be merge of IOs which require different failure
630 * handling. This function determines the number of bytes which
631 * can be failed from the beginning of the request without
632 * crossing into area which need to be retried further.
635 * The number of bytes to fail.
637 static unsigned int scsi_rq_err_bytes(const struct request *rq)
639 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
640 unsigned int bytes = 0;
643 if (!(rq->rq_flags & RQF_MIXED_MERGE))
644 return blk_rq_bytes(rq);
647 * Currently the only 'mixing' which can happen is between
648 * different fastfail types. We can safely fail portions
649 * which have all the failfast bits that the first one has -
650 * the ones which are at least as eager to fail as the first
653 for (bio = rq->bio; bio; bio = bio->bi_next) {
654 if ((bio->bi_opf & ff) != ff)
656 bytes += bio->bi_iter.bi_size;
659 /* this could lead to infinite loop */
660 BUG_ON(blk_rq_bytes(rq) && !bytes);
664 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
666 struct request *req = scsi_cmd_to_rq(cmd);
667 unsigned long wait_for;
669 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
672 wait_for = (cmd->allowed + 1) * req->timeout;
673 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
674 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
682 * When ALUA transition state is returned, reprep the cmd to
683 * use the ALUA handler's transition timeout. Delay the reprep
684 * 1 sec to avoid aggressive retries of the target in that
687 #define ALUA_TRANSITION_REPREP_DELAY 1000
689 /* Helper for scsi_io_completion() when special action required. */
690 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
692 struct request *req = scsi_cmd_to_rq(cmd);
694 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
695 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
696 struct scsi_sense_hdr sshdr;
698 bool sense_current = true; /* false implies "deferred sense" */
699 blk_status_t blk_stat;
701 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
703 sense_current = !scsi_sense_is_deferred(&sshdr);
705 blk_stat = scsi_result_to_blk_status(result);
707 if (host_byte(result) == DID_RESET) {
708 /* Third party bus reset or reset for error recovery
709 * reasons. Just retry the command and see what
712 action = ACTION_RETRY;
713 } else if (sense_valid && sense_current) {
714 switch (sshdr.sense_key) {
716 if (cmd->device->removable) {
717 /* Detected disc change. Set a bit
718 * and quietly refuse further access.
720 cmd->device->changed = 1;
721 action = ACTION_FAIL;
723 /* Must have been a power glitch, or a
724 * bus reset. Could not have been a
725 * media change, so we just retry the
726 * command and see what happens.
728 action = ACTION_RETRY;
731 case ILLEGAL_REQUEST:
732 /* If we had an ILLEGAL REQUEST returned, then
733 * we may have performed an unsupported
734 * command. The only thing this should be
735 * would be a ten byte read where only a six
736 * byte read was supported. Also, on a system
737 * where READ CAPACITY failed, we may have
738 * read past the end of the disk.
740 if ((cmd->device->use_10_for_rw &&
741 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
742 (cmd->cmnd[0] == READ_10 ||
743 cmd->cmnd[0] == WRITE_10)) {
744 /* This will issue a new 6-byte command. */
745 cmd->device->use_10_for_rw = 0;
746 action = ACTION_REPREP;
747 } else if (sshdr.asc == 0x10) /* DIX */ {
748 action = ACTION_FAIL;
749 blk_stat = BLK_STS_PROTECTION;
750 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
751 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
752 action = ACTION_FAIL;
753 blk_stat = BLK_STS_TARGET;
755 action = ACTION_FAIL;
757 case ABORTED_COMMAND:
758 action = ACTION_FAIL;
759 if (sshdr.asc == 0x10) /* DIF */
760 blk_stat = BLK_STS_PROTECTION;
763 /* If the device is in the process of becoming
764 * ready, or has a temporary blockage, retry.
766 if (sshdr.asc == 0x04) {
767 switch (sshdr.ascq) {
768 case 0x01: /* becoming ready */
769 case 0x04: /* format in progress */
770 case 0x05: /* rebuild in progress */
771 case 0x06: /* recalculation in progress */
772 case 0x07: /* operation in progress */
773 case 0x08: /* Long write in progress */
774 case 0x09: /* self test in progress */
775 case 0x11: /* notify (enable spinup) required */
776 case 0x14: /* space allocation in progress */
777 case 0x1a: /* start stop unit in progress */
778 case 0x1b: /* sanitize in progress */
779 case 0x1d: /* configuration in progress */
780 case 0x24: /* depopulation in progress */
781 action = ACTION_DELAYED_RETRY;
783 case 0x0a: /* ALUA state transition */
784 action = ACTION_DELAYED_REPREP;
787 action = ACTION_FAIL;
791 action = ACTION_FAIL;
793 case VOLUME_OVERFLOW:
794 /* See SSC3rXX or current. */
795 action = ACTION_FAIL;
798 action = ACTION_FAIL;
799 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
800 (sshdr.asc == 0x55 &&
801 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
802 /* Insufficient zone resources */
803 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
807 action = ACTION_FAIL;
811 action = ACTION_FAIL;
813 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
814 action = ACTION_FAIL;
818 /* Give up and fail the remainder of the request */
819 if (!(req->rq_flags & RQF_QUIET)) {
820 static DEFINE_RATELIMIT_STATE(_rs,
821 DEFAULT_RATELIMIT_INTERVAL,
822 DEFAULT_RATELIMIT_BURST);
824 if (unlikely(scsi_logging_level))
826 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
827 SCSI_LOG_MLCOMPLETE_BITS);
830 * if logging is enabled the failure will be printed
831 * in scsi_log_completion(), so avoid duplicate messages
833 if (!level && __ratelimit(&_rs)) {
834 scsi_print_result(cmd, NULL, FAILED);
836 scsi_print_sense(cmd);
837 scsi_print_command(cmd);
840 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
844 scsi_mq_requeue_cmd(cmd, 0);
846 case ACTION_DELAYED_REPREP:
847 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
850 /* Retry the same command immediately */
851 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
853 case ACTION_DELAYED_RETRY:
854 /* Retry the same command after a delay */
855 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
861 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
862 * new result that may suppress further error checking. Also modifies
863 * *blk_statp in some cases.
865 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
866 blk_status_t *blk_statp)
869 bool sense_current = true; /* false implies "deferred sense" */
870 struct request *req = scsi_cmd_to_rq(cmd);
871 struct scsi_sense_hdr sshdr;
873 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
875 sense_current = !scsi_sense_is_deferred(&sshdr);
877 if (blk_rq_is_passthrough(req)) {
880 * SG_IO wants current and deferred errors
882 cmd->sense_len = min(8 + cmd->sense_buffer[7],
883 SCSI_SENSE_BUFFERSIZE);
886 *blk_statp = scsi_result_to_blk_status(result);
887 } else if (blk_rq_bytes(req) == 0 && sense_current) {
889 * Flush commands do not transfers any data, and thus cannot use
890 * good_bytes != blk_rq_bytes(req) as the signal for an error.
891 * This sets *blk_statp explicitly for the problem case.
893 *blk_statp = scsi_result_to_blk_status(result);
896 * Recovered errors need reporting, but they're always treated as
897 * success, so fiddle the result code here. For passthrough requests
898 * we already took a copy of the original into sreq->result which
899 * is what gets returned to the user
901 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
902 bool do_print = true;
904 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
905 * skip print since caller wants ATA registers. Only occurs
906 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
908 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
910 else if (req->rq_flags & RQF_QUIET)
913 scsi_print_sense(cmd);
915 /* for passthrough, *blk_statp may be set */
916 *blk_statp = BLK_STS_OK;
919 * Another corner case: the SCSI status byte is non-zero but 'good'.
920 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
921 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
922 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
923 * intermediate statuses (both obsolete in SAM-4) as good.
925 if ((result & 0xff) && scsi_status_is_good(result)) {
927 *blk_statp = BLK_STS_OK;
933 * scsi_io_completion - Completion processing for SCSI commands.
934 * @cmd: command that is finished.
935 * @good_bytes: number of processed bytes.
937 * We will finish off the specified number of sectors. If we are done, the
938 * command block will be released and the queue function will be goosed. If we
939 * are not done then we have to figure out what to do next:
941 * a) We can call scsi_mq_requeue_cmd(). The request will be
942 * unprepared and put back on the queue. Then a new command will
943 * be created for it. This should be used if we made forward
944 * progress, or if we want to switch from READ(10) to READ(6) for
947 * b) We can call scsi_io_completion_action(). The request will be
948 * put back on the queue and retried using the same command as
949 * before, possibly after a delay.
951 * c) We can call scsi_end_request() with blk_stat other than
952 * BLK_STS_OK, to fail the remainder of the request.
954 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
956 int result = cmd->result;
957 struct request *req = scsi_cmd_to_rq(cmd);
958 blk_status_t blk_stat = BLK_STS_OK;
960 if (unlikely(result)) /* a nz result may or may not be an error */
961 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
964 * Next deal with any sectors which we were able to correctly
967 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
968 "%u sectors total, %d bytes done.\n",
969 blk_rq_sectors(req), good_bytes));
972 * Failed, zero length commands always need to drop down
973 * to retry code. Fast path should return in this block.
975 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
976 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
977 return; /* no bytes remaining */
980 /* Kill remainder if no retries. */
981 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
982 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
984 "Bytes remaining after failed, no-retry command");
989 * If there had been no error, but we have leftover bytes in the
990 * request just queue the command up again.
992 if (likely(result == 0))
993 scsi_mq_requeue_cmd(cmd, 0);
995 scsi_io_completion_action(cmd, result);
998 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1001 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1002 !op_is_write(req_op(rq)) &&
1003 sdev->host->hostt->dma_need_drain(rq);
1007 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1008 * @cmd: SCSI command data structure to initialize.
1010 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1014 * * BLK_STS_OK - on success
1015 * * BLK_STS_RESOURCE - if the failure is retryable
1016 * * BLK_STS_IOERR - if the failure is fatal
1018 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1020 struct scsi_device *sdev = cmd->device;
1021 struct request *rq = scsi_cmd_to_rq(cmd);
1022 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1023 struct scatterlist *last_sg = NULL;
1025 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1028 if (WARN_ON_ONCE(!nr_segs))
1029 return BLK_STS_IOERR;
1032 * Make sure there is space for the drain. The driver must adjust
1033 * max_hw_segments to be prepared for this.
1039 * If sg table allocation fails, requeue request later.
1041 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1042 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1043 return BLK_STS_RESOURCE;
1046 * Next, walk the list, and fill in the addresses and sizes of
1049 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1051 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1052 unsigned int pad_len =
1053 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1055 last_sg->length += pad_len;
1056 cmd->extra_len += pad_len;
1060 sg_unmark_end(last_sg);
1061 last_sg = sg_next(last_sg);
1062 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1063 sg_mark_end(last_sg);
1065 cmd->extra_len += sdev->dma_drain_len;
1069 BUG_ON(count > cmd->sdb.table.nents);
1070 cmd->sdb.table.nents = count;
1071 cmd->sdb.length = blk_rq_payload_bytes(rq);
1073 if (blk_integrity_rq(rq)) {
1074 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1077 if (WARN_ON_ONCE(!prot_sdb)) {
1079 * This can happen if someone (e.g. multipath)
1080 * queues a command to a device on an adapter
1081 * that does not support DIX.
1083 ret = BLK_STS_IOERR;
1084 goto out_free_sgtables;
1087 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1089 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1090 prot_sdb->table.sgl,
1091 SCSI_INLINE_PROT_SG_CNT)) {
1092 ret = BLK_STS_RESOURCE;
1093 goto out_free_sgtables;
1096 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1097 prot_sdb->table.sgl);
1098 BUG_ON(count > ivecs);
1099 BUG_ON(count > queue_max_integrity_segments(rq->q));
1101 cmd->prot_sdb = prot_sdb;
1102 cmd->prot_sdb->table.nents = count;
1107 scsi_free_sgtables(cmd);
1110 EXPORT_SYMBOL(scsi_alloc_sgtables);
1113 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1114 * @rq: Request associated with the SCSI command to be initialized.
1116 * This function initializes the members of struct scsi_cmnd that must be
1117 * initialized before request processing starts and that won't be
1118 * reinitialized if a SCSI command is requeued.
1120 static void scsi_initialize_rq(struct request *rq)
1122 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1124 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1125 cmd->cmd_len = MAX_COMMAND_SIZE;
1127 init_rcu_head(&cmd->rcu);
1128 cmd->jiffies_at_alloc = jiffies;
1132 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1133 blk_mq_req_flags_t flags)
1137 rq = blk_mq_alloc_request(q, opf, flags);
1139 scsi_initialize_rq(rq);
1142 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1145 * Only called when the request isn't completed by SCSI, and not freed by
1148 static void scsi_cleanup_rq(struct request *rq)
1150 if (rq->rq_flags & RQF_DONTPREP) {
1151 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1152 rq->rq_flags &= ~RQF_DONTPREP;
1156 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1157 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1159 struct request *rq = scsi_cmd_to_rq(cmd);
1161 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1162 cmd->flags |= SCMD_INITIALIZED;
1163 scsi_initialize_rq(rq);
1167 INIT_LIST_HEAD(&cmd->eh_entry);
1168 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1171 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1172 struct request *req)
1174 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1177 * Passthrough requests may transfer data, in which case they must
1178 * a bio attached to them. Or they might contain a SCSI command
1179 * that does not transfer data, in which case they may optionally
1180 * submit a request without an attached bio.
1183 blk_status_t ret = scsi_alloc_sgtables(cmd);
1184 if (unlikely(ret != BLK_STS_OK))
1187 BUG_ON(blk_rq_bytes(req));
1189 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1192 cmd->transfersize = blk_rq_bytes(req);
1197 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1199 switch (sdev->sdev_state) {
1203 case SDEV_TRANSPORT_OFFLINE:
1205 * If the device is offline we refuse to process any
1206 * commands. The device must be brought online
1207 * before trying any recovery commands.
1209 if (!sdev->offline_already) {
1210 sdev->offline_already = true;
1211 sdev_printk(KERN_ERR, sdev,
1212 "rejecting I/O to offline device\n");
1214 return BLK_STS_IOERR;
1217 * If the device is fully deleted, we refuse to
1218 * process any commands as well.
1220 sdev_printk(KERN_ERR, sdev,
1221 "rejecting I/O to dead device\n");
1222 return BLK_STS_IOERR;
1224 case SDEV_CREATED_BLOCK:
1225 return BLK_STS_RESOURCE;
1228 * If the device is blocked we only accept power management
1231 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1232 return BLK_STS_RESOURCE;
1236 * For any other not fully online state we only allow
1237 * power management commands.
1239 if (req && !(req->rq_flags & RQF_PM))
1240 return BLK_STS_OFFLINE;
1246 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1247 * and return the token else return -1.
1249 static inline int scsi_dev_queue_ready(struct request_queue *q,
1250 struct scsi_device *sdev)
1254 token = sbitmap_get(&sdev->budget_map);
1255 if (atomic_read(&sdev->device_blocked)) {
1259 if (scsi_device_busy(sdev) > 1)
1263 * unblock after device_blocked iterates to zero
1265 if (atomic_dec_return(&sdev->device_blocked) > 0)
1267 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1268 "unblocking device at zero depth\n"));
1274 sbitmap_put(&sdev->budget_map, token);
1280 * scsi_target_queue_ready: checks if there we can send commands to target
1281 * @sdev: scsi device on starget to check.
1283 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1284 struct scsi_device *sdev)
1286 struct scsi_target *starget = scsi_target(sdev);
1289 if (starget->single_lun) {
1290 spin_lock_irq(shost->host_lock);
1291 if (starget->starget_sdev_user &&
1292 starget->starget_sdev_user != sdev) {
1293 spin_unlock_irq(shost->host_lock);
1296 starget->starget_sdev_user = sdev;
1297 spin_unlock_irq(shost->host_lock);
1300 if (starget->can_queue <= 0)
1303 busy = atomic_inc_return(&starget->target_busy) - 1;
1304 if (atomic_read(&starget->target_blocked) > 0) {
1309 * unblock after target_blocked iterates to zero
1311 if (atomic_dec_return(&starget->target_blocked) > 0)
1314 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1315 "unblocking target at zero depth\n"));
1318 if (busy >= starget->can_queue)
1324 spin_lock_irq(shost->host_lock);
1325 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1326 spin_unlock_irq(shost->host_lock);
1328 if (starget->can_queue > 0)
1329 atomic_dec(&starget->target_busy);
1334 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1335 * return 0. We must end up running the queue again whenever 0 is
1336 * returned, else IO can hang.
1338 static inline int scsi_host_queue_ready(struct request_queue *q,
1339 struct Scsi_Host *shost,
1340 struct scsi_device *sdev,
1341 struct scsi_cmnd *cmd)
1343 if (atomic_read(&shost->host_blocked) > 0) {
1344 if (scsi_host_busy(shost) > 0)
1348 * unblock after host_blocked iterates to zero
1350 if (atomic_dec_return(&shost->host_blocked) > 0)
1354 shost_printk(KERN_INFO, shost,
1355 "unblocking host at zero depth\n"));
1358 if (shost->host_self_blocked)
1361 /* We're OK to process the command, so we can't be starved */
1362 if (!list_empty(&sdev->starved_entry)) {
1363 spin_lock_irq(shost->host_lock);
1364 if (!list_empty(&sdev->starved_entry))
1365 list_del_init(&sdev->starved_entry);
1366 spin_unlock_irq(shost->host_lock);
1369 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1374 spin_lock_irq(shost->host_lock);
1375 if (list_empty(&sdev->starved_entry))
1376 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1377 spin_unlock_irq(shost->host_lock);
1379 scsi_dec_host_busy(shost, cmd);
1384 * Busy state exporting function for request stacking drivers.
1386 * For efficiency, no lock is taken to check the busy state of
1387 * shost/starget/sdev, since the returned value is not guaranteed and
1388 * may be changed after request stacking drivers call the function,
1389 * regardless of taking lock or not.
1391 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1392 * needs to return 'not busy'. Otherwise, request stacking drivers
1393 * may hold requests forever.
1395 static bool scsi_mq_lld_busy(struct request_queue *q)
1397 struct scsi_device *sdev = q->queuedata;
1398 struct Scsi_Host *shost;
1400 if (blk_queue_dying(q))
1406 * Ignore host/starget busy state.
1407 * Since block layer does not have a concept of fairness across
1408 * multiple queues, congestion of host/starget needs to be handled
1411 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1418 * Block layer request completion callback. May be called from interrupt
1421 static void scsi_complete(struct request *rq)
1423 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1424 enum scsi_disposition disposition;
1426 INIT_LIST_HEAD(&cmd->eh_entry);
1428 atomic_inc(&cmd->device->iodone_cnt);
1430 atomic_inc(&cmd->device->ioerr_cnt);
1432 disposition = scsi_decide_disposition(cmd);
1433 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1434 disposition = SUCCESS;
1436 scsi_log_completion(cmd, disposition);
1438 switch (disposition) {
1440 scsi_finish_command(cmd);
1443 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1445 case ADD_TO_MLQUEUE:
1446 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1449 scsi_eh_scmd_add(cmd);
1455 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1456 * @cmd: command block we are dispatching.
1458 * Return: nonzero return request was rejected and device's queue needs to be
1461 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1463 struct Scsi_Host *host = cmd->device->host;
1466 /* check if the device is still usable */
1467 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1468 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1469 * returns an immediate error upwards, and signals
1470 * that the device is no longer present */
1471 cmd->result = DID_NO_CONNECT << 16;
1475 /* Check to see if the scsi lld made this device blocked. */
1476 if (unlikely(scsi_device_blocked(cmd->device))) {
1478 * in blocked state, the command is just put back on
1479 * the device queue. The suspend state has already
1480 * blocked the queue so future requests should not
1481 * occur until the device transitions out of the
1484 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1485 "queuecommand : device blocked\n"));
1486 return SCSI_MLQUEUE_DEVICE_BUSY;
1489 /* Store the LUN value in cmnd, if needed. */
1490 if (cmd->device->lun_in_cdb)
1491 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1492 (cmd->device->lun << 5 & 0xe0);
1497 * Before we queue this command, check if the command
1498 * length exceeds what the host adapter can handle.
1500 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1501 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1502 "queuecommand : command too long. "
1503 "cdb_size=%d host->max_cmd_len=%d\n",
1504 cmd->cmd_len, cmd->device->host->max_cmd_len));
1505 cmd->result = (DID_ABORT << 16);
1509 if (unlikely(host->shost_state == SHOST_DEL)) {
1510 cmd->result = (DID_NO_CONNECT << 16);
1515 trace_scsi_dispatch_cmd_start(cmd);
1516 rtn = host->hostt->queuecommand(host, cmd);
1518 trace_scsi_dispatch_cmd_error(cmd, rtn);
1519 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1520 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1521 rtn = SCSI_MLQUEUE_HOST_BUSY;
1523 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1524 "queuecommand : request rejected\n"));
1533 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1534 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1536 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1537 sizeof(struct scatterlist);
1540 static blk_status_t scsi_prepare_cmd(struct request *req)
1542 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1543 struct scsi_device *sdev = req->q->queuedata;
1544 struct Scsi_Host *shost = sdev->host;
1545 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1546 struct scatterlist *sg;
1548 scsi_init_command(sdev, cmd);
1552 cmd->prot_flags = 0;
1554 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1556 cmd->transfersize = 0;
1557 cmd->host_scribble = NULL;
1562 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1565 * Only clear the driver-private command data if the LLD does not supply
1566 * a function to initialize that data.
1568 if (!shost->hostt->init_cmd_priv)
1569 memset(cmd + 1, 0, shost->hostt->cmd_size);
1571 cmd->prot_op = SCSI_PROT_NORMAL;
1572 if (blk_rq_bytes(req))
1573 cmd->sc_data_direction = rq_dma_dir(req);
1575 cmd->sc_data_direction = DMA_NONE;
1577 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1578 cmd->sdb.table.sgl = sg;
1580 if (scsi_host_get_prot(shost)) {
1581 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1583 cmd->prot_sdb->table.sgl =
1584 (struct scatterlist *)(cmd->prot_sdb + 1);
1588 * Special handling for passthrough commands, which don't go to the ULP
1591 if (blk_rq_is_passthrough(req))
1592 return scsi_setup_scsi_cmnd(sdev, req);
1594 if (sdev->handler && sdev->handler->prep_fn) {
1595 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1597 if (ret != BLK_STS_OK)
1601 /* Usually overridden by the ULP */
1603 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1604 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1607 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1609 struct request *req = scsi_cmd_to_rq(cmd);
1611 switch (cmd->submitter) {
1612 case SUBMITTED_BY_BLOCK_LAYER:
1614 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1615 return scsi_eh_done(cmd);
1616 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1620 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1622 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1624 trace_scsi_dispatch_cmd_done(cmd);
1626 if (complete_directly)
1627 blk_mq_complete_request_direct(req, scsi_complete);
1629 blk_mq_complete_request(req);
1632 void scsi_done(struct scsi_cmnd *cmd)
1634 scsi_done_internal(cmd, false);
1636 EXPORT_SYMBOL(scsi_done);
1638 void scsi_done_direct(struct scsi_cmnd *cmd)
1640 scsi_done_internal(cmd, true);
1642 EXPORT_SYMBOL(scsi_done_direct);
1644 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1646 struct scsi_device *sdev = q->queuedata;
1648 sbitmap_put(&sdev->budget_map, budget_token);
1652 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1653 * not change behaviour from the previous unplug mechanism, experimentation
1654 * may prove this needs changing.
1656 #define SCSI_QUEUE_DELAY 3
1658 static int scsi_mq_get_budget(struct request_queue *q)
1660 struct scsi_device *sdev = q->queuedata;
1661 int token = scsi_dev_queue_ready(q, sdev);
1666 atomic_inc(&sdev->restarts);
1669 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1670 * .restarts must be incremented before .device_busy is read because the
1671 * code in scsi_run_queue_async() depends on the order of these operations.
1673 smp_mb__after_atomic();
1676 * If all in-flight requests originated from this LUN are completed
1677 * before reading .device_busy, sdev->device_busy will be observed as
1678 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1679 * soon. Otherwise, completion of one of these requests will observe
1680 * the .restarts flag, and the request queue will be run for handling
1681 * this request, see scsi_end_request().
1683 if (unlikely(scsi_device_busy(sdev) == 0 &&
1684 !scsi_device_blocked(sdev)))
1685 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1689 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1691 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1693 cmd->budget_token = token;
1696 static int scsi_mq_get_rq_budget_token(struct request *req)
1698 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1700 return cmd->budget_token;
1703 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1704 const struct blk_mq_queue_data *bd)
1706 struct request *req = bd->rq;
1707 struct request_queue *q = req->q;
1708 struct scsi_device *sdev = q->queuedata;
1709 struct Scsi_Host *shost = sdev->host;
1710 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1714 WARN_ON_ONCE(cmd->budget_token < 0);
1717 * If the device is not in running state we will reject some or all
1720 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1721 ret = scsi_device_state_check(sdev, req);
1722 if (ret != BLK_STS_OK)
1723 goto out_put_budget;
1726 ret = BLK_STS_RESOURCE;
1727 if (!scsi_target_queue_ready(shost, sdev))
1728 goto out_put_budget;
1729 if (unlikely(scsi_host_in_recovery(shost))) {
1730 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1731 ret = BLK_STS_OFFLINE;
1732 goto out_dec_target_busy;
1734 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1735 goto out_dec_target_busy;
1737 if (!(req->rq_flags & RQF_DONTPREP)) {
1738 ret = scsi_prepare_cmd(req);
1739 if (ret != BLK_STS_OK)
1740 goto out_dec_host_busy;
1741 req->rq_flags |= RQF_DONTPREP;
1743 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1746 cmd->flags &= SCMD_PRESERVED_FLAGS;
1747 if (sdev->simple_tags)
1748 cmd->flags |= SCMD_TAGGED;
1750 cmd->flags |= SCMD_LAST;
1752 scsi_set_resid(cmd, 0);
1753 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1754 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1756 blk_mq_start_request(req);
1757 reason = scsi_dispatch_cmd(cmd);
1759 scsi_set_blocked(cmd, reason);
1760 ret = BLK_STS_RESOURCE;
1761 goto out_dec_host_busy;
1764 atomic_inc(&cmd->device->iorequest_cnt);
1768 scsi_dec_host_busy(shost, cmd);
1769 out_dec_target_busy:
1770 if (scsi_target(sdev)->can_queue > 0)
1771 atomic_dec(&scsi_target(sdev)->target_busy);
1773 scsi_mq_put_budget(q, cmd->budget_token);
1774 cmd->budget_token = -1;
1778 case BLK_STS_RESOURCE:
1779 case BLK_STS_ZONE_RESOURCE:
1780 if (scsi_device_blocked(sdev))
1781 ret = BLK_STS_DEV_RESOURCE;
1784 cmd->result = DID_BUS_BUSY << 16;
1785 if (req->rq_flags & RQF_DONTPREP)
1786 scsi_mq_uninit_cmd(cmd);
1789 if (unlikely(!scsi_device_online(sdev)))
1790 cmd->result = DID_NO_CONNECT << 16;
1792 cmd->result = DID_ERROR << 16;
1794 * Make sure to release all allocated resources when
1795 * we hit an error, as we will never see this command
1798 if (req->rq_flags & RQF_DONTPREP)
1799 scsi_mq_uninit_cmd(cmd);
1800 scsi_run_queue_async(sdev);
1806 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1807 unsigned int hctx_idx, unsigned int numa_node)
1809 struct Scsi_Host *shost = set->driver_data;
1810 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1811 struct scatterlist *sg;
1815 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1816 if (!cmd->sense_buffer)
1819 if (scsi_host_get_prot(shost)) {
1820 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1821 shost->hostt->cmd_size;
1822 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1825 if (shost->hostt->init_cmd_priv) {
1826 ret = shost->hostt->init_cmd_priv(shost, cmd);
1828 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1834 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1835 unsigned int hctx_idx)
1837 struct Scsi_Host *shost = set->driver_data;
1838 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1840 if (shost->hostt->exit_cmd_priv)
1841 shost->hostt->exit_cmd_priv(shost, cmd);
1842 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1846 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1848 struct Scsi_Host *shost = hctx->driver_data;
1850 if (shost->hostt->mq_poll)
1851 return shost->hostt->mq_poll(shost, hctx->queue_num);
1856 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1857 unsigned int hctx_idx)
1859 struct Scsi_Host *shost = data;
1861 hctx->driver_data = shost;
1865 static void scsi_map_queues(struct blk_mq_tag_set *set)
1867 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1869 if (shost->hostt->map_queues)
1870 return shost->hostt->map_queues(shost);
1871 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1874 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1876 struct device *dev = shost->dma_dev;
1879 * this limit is imposed by hardware restrictions
1881 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1884 if (scsi_host_prot_dma(shost)) {
1885 shost->sg_prot_tablesize =
1886 min_not_zero(shost->sg_prot_tablesize,
1887 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1888 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1889 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1892 blk_queue_max_hw_sectors(q, shost->max_sectors);
1893 blk_queue_segment_boundary(q, shost->dma_boundary);
1894 dma_set_seg_boundary(dev, shost->dma_boundary);
1896 blk_queue_max_segment_size(q, shost->max_segment_size);
1897 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1898 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1901 * Set a reasonable default alignment: The larger of 32-byte (dword),
1902 * which is a common minimum for HBAs, and the minimum DMA alignment,
1903 * which is set by the platform.
1905 * Devices that require a bigger alignment can increase it later.
1907 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1909 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1911 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1912 .get_budget = scsi_mq_get_budget,
1913 .put_budget = scsi_mq_put_budget,
1914 .queue_rq = scsi_queue_rq,
1915 .complete = scsi_complete,
1916 .timeout = scsi_timeout,
1917 #ifdef CONFIG_BLK_DEBUG_FS
1918 .show_rq = scsi_show_rq,
1920 .init_request = scsi_mq_init_request,
1921 .exit_request = scsi_mq_exit_request,
1922 .cleanup_rq = scsi_cleanup_rq,
1923 .busy = scsi_mq_lld_busy,
1924 .map_queues = scsi_map_queues,
1925 .init_hctx = scsi_init_hctx,
1926 .poll = scsi_mq_poll,
1927 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1928 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1932 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1934 struct Scsi_Host *shost = hctx->driver_data;
1936 shost->hostt->commit_rqs(shost, hctx->queue_num);
1939 static const struct blk_mq_ops scsi_mq_ops = {
1940 .get_budget = scsi_mq_get_budget,
1941 .put_budget = scsi_mq_put_budget,
1942 .queue_rq = scsi_queue_rq,
1943 .commit_rqs = scsi_commit_rqs,
1944 .complete = scsi_complete,
1945 .timeout = scsi_timeout,
1946 #ifdef CONFIG_BLK_DEBUG_FS
1947 .show_rq = scsi_show_rq,
1949 .init_request = scsi_mq_init_request,
1950 .exit_request = scsi_mq_exit_request,
1951 .cleanup_rq = scsi_cleanup_rq,
1952 .busy = scsi_mq_lld_busy,
1953 .map_queues = scsi_map_queues,
1954 .init_hctx = scsi_init_hctx,
1955 .poll = scsi_mq_poll,
1956 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1957 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1960 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1962 unsigned int cmd_size, sgl_size;
1963 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1965 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1966 scsi_mq_inline_sgl_size(shost));
1967 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1968 if (scsi_host_get_prot(shost))
1969 cmd_size += sizeof(struct scsi_data_buffer) +
1970 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1972 memset(tag_set, 0, sizeof(*tag_set));
1973 if (shost->hostt->commit_rqs)
1974 tag_set->ops = &scsi_mq_ops;
1976 tag_set->ops = &scsi_mq_ops_no_commit;
1977 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1978 tag_set->nr_maps = shost->nr_maps ? : 1;
1979 tag_set->queue_depth = shost->can_queue;
1980 tag_set->cmd_size = cmd_size;
1981 tag_set->numa_node = dev_to_node(shost->dma_dev);
1982 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1984 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1985 tag_set->driver_data = shost;
1986 if (shost->host_tagset)
1987 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1989 return blk_mq_alloc_tag_set(tag_set);
1992 void scsi_mq_free_tags(struct kref *kref)
1994 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
1997 blk_mq_free_tag_set(&shost->tag_set);
1998 complete(&shost->tagset_freed);
2002 * scsi_device_from_queue - return sdev associated with a request_queue
2003 * @q: The request queue to return the sdev from
2005 * Return the sdev associated with a request queue or NULL if the
2006 * request_queue does not reference a SCSI device.
2008 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2010 struct scsi_device *sdev = NULL;
2012 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2013 q->mq_ops == &scsi_mq_ops)
2014 sdev = q->queuedata;
2015 if (!sdev || !get_device(&sdev->sdev_gendev))
2021 * pktcdvd should have been integrated into the SCSI layers, but for historical
2022 * reasons like the old IDE driver it isn't. This export allows it to safely
2023 * probe if a given device is a SCSI one and only attach to that.
2025 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2026 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2030 * scsi_block_requests - Utility function used by low-level drivers to prevent
2031 * further commands from being queued to the device.
2032 * @shost: host in question
2034 * There is no timer nor any other means by which the requests get unblocked
2035 * other than the low-level driver calling scsi_unblock_requests().
2037 void scsi_block_requests(struct Scsi_Host *shost)
2039 shost->host_self_blocked = 1;
2041 EXPORT_SYMBOL(scsi_block_requests);
2044 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2045 * further commands to be queued to the device.
2046 * @shost: host in question
2048 * There is no timer nor any other means by which the requests get unblocked
2049 * other than the low-level driver calling scsi_unblock_requests(). This is done
2050 * as an API function so that changes to the internals of the scsi mid-layer
2051 * won't require wholesale changes to drivers that use this feature.
2053 void scsi_unblock_requests(struct Scsi_Host *shost)
2055 shost->host_self_blocked = 0;
2056 scsi_run_host_queues(shost);
2058 EXPORT_SYMBOL(scsi_unblock_requests);
2060 void scsi_exit_queue(void)
2062 kmem_cache_destroy(scsi_sense_cache);
2066 * scsi_mode_select - issue a mode select
2067 * @sdev: SCSI device to be queried
2068 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2069 * @sp: Save page bit (0 == don't save, 1 == save)
2070 * @buffer: request buffer (may not be smaller than eight bytes)
2071 * @len: length of request buffer.
2072 * @timeout: command timeout
2073 * @retries: number of retries before failing
2074 * @data: returns a structure abstracting the mode header data
2075 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2076 * must be SCSI_SENSE_BUFFERSIZE big.
2078 * Returns zero if successful; negative error number or scsi
2082 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2083 unsigned char *buffer, int len, int timeout, int retries,
2084 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2086 unsigned char cmd[10];
2087 unsigned char *real_buffer;
2088 const struct scsi_exec_args exec_args = {
2093 memset(cmd, 0, sizeof(cmd));
2094 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2097 * Use MODE SELECT(10) if the device asked for it or if the mode page
2098 * and the mode select header cannot fit within the maximumm 255 bytes
2099 * of the MODE SELECT(6) command.
2101 if (sdev->use_10_for_ms ||
2103 data->block_descriptor_length > 255) {
2104 if (len > 65535 - 8)
2106 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2109 memcpy(real_buffer + 8, buffer, len);
2113 real_buffer[2] = data->medium_type;
2114 real_buffer[3] = data->device_specific;
2115 real_buffer[4] = data->longlba ? 0x01 : 0;
2117 put_unaligned_be16(data->block_descriptor_length,
2120 cmd[0] = MODE_SELECT_10;
2121 put_unaligned_be16(len, &cmd[7]);
2126 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2129 memcpy(real_buffer + 4, buffer, len);
2132 real_buffer[1] = data->medium_type;
2133 real_buffer[2] = data->device_specific;
2134 real_buffer[3] = data->block_descriptor_length;
2136 cmd[0] = MODE_SELECT;
2140 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2141 timeout, retries, &exec_args);
2145 EXPORT_SYMBOL_GPL(scsi_mode_select);
2148 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2149 * @sdev: SCSI device to be queried
2150 * @dbd: set to prevent mode sense from returning block descriptors
2151 * @modepage: mode page being requested
2152 * @buffer: request buffer (may not be smaller than eight bytes)
2153 * @len: length of request buffer.
2154 * @timeout: command timeout
2155 * @retries: number of retries before failing
2156 * @data: returns a structure abstracting the mode header data
2157 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2158 * must be SCSI_SENSE_BUFFERSIZE big.
2160 * Returns zero if successful, or a negative error number on failure
2163 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2164 unsigned char *buffer, int len, int timeout, int retries,
2165 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2167 unsigned char cmd[12];
2170 int result, retry_count = retries;
2171 struct scsi_sense_hdr my_sshdr;
2172 const struct scsi_exec_args exec_args = {
2173 /* caller might not be interested in sense, but we need it */
2174 .sshdr = sshdr ? : &my_sshdr,
2177 memset(data, 0, sizeof(*data));
2178 memset(&cmd[0], 0, 12);
2180 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2181 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2184 sshdr = exec_args.sshdr;
2187 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2189 if (use_10_for_ms) {
2190 if (len < 8 || len > 65535)
2193 cmd[0] = MODE_SENSE_10;
2194 put_unaligned_be16(len, &cmd[7]);
2200 cmd[0] = MODE_SENSE;
2205 memset(buffer, 0, len);
2207 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2208 timeout, retries, &exec_args);
2212 /* This code looks awful: what it's doing is making sure an
2213 * ILLEGAL REQUEST sense return identifies the actual command
2214 * byte as the problem. MODE_SENSE commands can return
2215 * ILLEGAL REQUEST if the code page isn't supported */
2217 if (!scsi_status_is_good(result)) {
2218 if (scsi_sense_valid(sshdr)) {
2219 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2220 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2222 * Invalid command operation code: retry using
2223 * MODE SENSE(6) if this was a MODE SENSE(10)
2224 * request, except if the request mode page is
2225 * too large for MODE SENSE single byte
2226 * allocation length field.
2228 if (use_10_for_ms) {
2231 sdev->use_10_for_ms = 0;
2235 if (scsi_status_is_check_condition(result) &&
2236 sshdr->sense_key == UNIT_ATTENTION &&
2244 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2245 (modepage == 6 || modepage == 8))) {
2246 /* Initio breakage? */
2249 data->medium_type = 0;
2250 data->device_specific = 0;
2252 data->block_descriptor_length = 0;
2253 } else if (use_10_for_ms) {
2254 data->length = get_unaligned_be16(&buffer[0]) + 2;
2255 data->medium_type = buffer[2];
2256 data->device_specific = buffer[3];
2257 data->longlba = buffer[4] & 0x01;
2258 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2260 data->length = buffer[0] + 1;
2261 data->medium_type = buffer[1];
2262 data->device_specific = buffer[2];
2263 data->block_descriptor_length = buffer[3];
2265 data->header_length = header_length;
2269 EXPORT_SYMBOL(scsi_mode_sense);
2272 * scsi_test_unit_ready - test if unit is ready
2273 * @sdev: scsi device to change the state of.
2274 * @timeout: command timeout
2275 * @retries: number of retries before failing
2276 * @sshdr: outpout pointer for decoded sense information.
2278 * Returns zero if unsuccessful or an error if TUR failed. For
2279 * removable media, UNIT_ATTENTION sets ->changed flag.
2282 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2283 struct scsi_sense_hdr *sshdr)
2286 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2288 const struct scsi_exec_args exec_args = {
2293 /* try to eat the UNIT_ATTENTION if there are enough retries */
2295 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2296 timeout, 1, &exec_args);
2297 if (sdev->removable && scsi_sense_valid(sshdr) &&
2298 sshdr->sense_key == UNIT_ATTENTION)
2300 } while (scsi_sense_valid(sshdr) &&
2301 sshdr->sense_key == UNIT_ATTENTION && --retries);
2305 EXPORT_SYMBOL(scsi_test_unit_ready);
2308 * scsi_device_set_state - Take the given device through the device state model.
2309 * @sdev: scsi device to change the state of.
2310 * @state: state to change to.
2312 * Returns zero if successful or an error if the requested
2313 * transition is illegal.
2316 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2318 enum scsi_device_state oldstate = sdev->sdev_state;
2320 if (state == oldstate)
2326 case SDEV_CREATED_BLOCK:
2337 case SDEV_TRANSPORT_OFFLINE:
2350 case SDEV_TRANSPORT_OFFLINE:
2358 case SDEV_TRANSPORT_OFFLINE:
2373 case SDEV_CREATED_BLOCK:
2382 case SDEV_CREATED_BLOCK:
2397 case SDEV_TRANSPORT_OFFLINE:
2409 case SDEV_TRANSPORT_OFFLINE:
2412 case SDEV_CREATED_BLOCK:
2420 sdev->offline_already = false;
2421 sdev->sdev_state = state;
2425 SCSI_LOG_ERROR_RECOVERY(1,
2426 sdev_printk(KERN_ERR, sdev,
2427 "Illegal state transition %s->%s",
2428 scsi_device_state_name(oldstate),
2429 scsi_device_state_name(state))
2433 EXPORT_SYMBOL(scsi_device_set_state);
2436 * scsi_evt_emit - emit a single SCSI device uevent
2437 * @sdev: associated SCSI device
2438 * @evt: event to emit
2440 * Send a single uevent (scsi_event) to the associated scsi_device.
2442 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2447 switch (evt->evt_type) {
2448 case SDEV_EVT_MEDIA_CHANGE:
2449 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2451 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2452 scsi_rescan_device(&sdev->sdev_gendev);
2453 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2455 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2456 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2458 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2459 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2461 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2462 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2464 case SDEV_EVT_LUN_CHANGE_REPORTED:
2465 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2467 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2468 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2470 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2471 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2480 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2484 * scsi_evt_thread - send a uevent for each scsi event
2485 * @work: work struct for scsi_device
2487 * Dispatch queued events to their associated scsi_device kobjects
2490 void scsi_evt_thread(struct work_struct *work)
2492 struct scsi_device *sdev;
2493 enum scsi_device_event evt_type;
2494 LIST_HEAD(event_list);
2496 sdev = container_of(work, struct scsi_device, event_work);
2498 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2499 if (test_and_clear_bit(evt_type, sdev->pending_events))
2500 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2503 struct scsi_event *evt;
2504 struct list_head *this, *tmp;
2505 unsigned long flags;
2507 spin_lock_irqsave(&sdev->list_lock, flags);
2508 list_splice_init(&sdev->event_list, &event_list);
2509 spin_unlock_irqrestore(&sdev->list_lock, flags);
2511 if (list_empty(&event_list))
2514 list_for_each_safe(this, tmp, &event_list) {
2515 evt = list_entry(this, struct scsi_event, node);
2516 list_del(&evt->node);
2517 scsi_evt_emit(sdev, evt);
2524 * sdev_evt_send - send asserted event to uevent thread
2525 * @sdev: scsi_device event occurred on
2526 * @evt: event to send
2528 * Assert scsi device event asynchronously.
2530 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2532 unsigned long flags;
2535 /* FIXME: currently this check eliminates all media change events
2536 * for polled devices. Need to update to discriminate between AN
2537 * and polled events */
2538 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2544 spin_lock_irqsave(&sdev->list_lock, flags);
2545 list_add_tail(&evt->node, &sdev->event_list);
2546 schedule_work(&sdev->event_work);
2547 spin_unlock_irqrestore(&sdev->list_lock, flags);
2549 EXPORT_SYMBOL_GPL(sdev_evt_send);
2552 * sdev_evt_alloc - allocate a new scsi event
2553 * @evt_type: type of event to allocate
2554 * @gfpflags: GFP flags for allocation
2556 * Allocates and returns a new scsi_event.
2558 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2561 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2565 evt->evt_type = evt_type;
2566 INIT_LIST_HEAD(&evt->node);
2568 /* evt_type-specific initialization, if any */
2570 case SDEV_EVT_MEDIA_CHANGE:
2571 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2572 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2573 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2574 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2575 case SDEV_EVT_LUN_CHANGE_REPORTED:
2576 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2577 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2585 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2588 * sdev_evt_send_simple - send asserted event to uevent thread
2589 * @sdev: scsi_device event occurred on
2590 * @evt_type: type of event to send
2591 * @gfpflags: GFP flags for allocation
2593 * Assert scsi device event asynchronously, given an event type.
2595 void sdev_evt_send_simple(struct scsi_device *sdev,
2596 enum scsi_device_event evt_type, gfp_t gfpflags)
2598 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2600 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2605 sdev_evt_send(sdev, evt);
2607 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2610 * scsi_device_quiesce - Block all commands except power management.
2611 * @sdev: scsi device to quiesce.
2613 * This works by trying to transition to the SDEV_QUIESCE state
2614 * (which must be a legal transition). When the device is in this
2615 * state, only power management requests will be accepted, all others will
2618 * Must be called with user context, may sleep.
2620 * Returns zero if unsuccessful or an error if not.
2623 scsi_device_quiesce(struct scsi_device *sdev)
2625 struct request_queue *q = sdev->request_queue;
2629 * It is allowed to call scsi_device_quiesce() multiple times from
2630 * the same context but concurrent scsi_device_quiesce() calls are
2633 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2635 if (sdev->quiesced_by == current)
2640 blk_mq_freeze_queue(q);
2642 * Ensure that the effect of blk_set_pm_only() will be visible
2643 * for percpu_ref_tryget() callers that occur after the queue
2644 * unfreeze even if the queue was already frozen before this function
2645 * was called. See also https://lwn.net/Articles/573497/.
2648 blk_mq_unfreeze_queue(q);
2650 mutex_lock(&sdev->state_mutex);
2651 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2653 sdev->quiesced_by = current;
2655 blk_clear_pm_only(q);
2656 mutex_unlock(&sdev->state_mutex);
2660 EXPORT_SYMBOL(scsi_device_quiesce);
2663 * scsi_device_resume - Restart user issued commands to a quiesced device.
2664 * @sdev: scsi device to resume.
2666 * Moves the device from quiesced back to running and restarts the
2669 * Must be called with user context, may sleep.
2671 void scsi_device_resume(struct scsi_device *sdev)
2673 /* check if the device state was mutated prior to resume, and if
2674 * so assume the state is being managed elsewhere (for example
2675 * device deleted during suspend)
2677 mutex_lock(&sdev->state_mutex);
2678 if (sdev->sdev_state == SDEV_QUIESCE)
2679 scsi_device_set_state(sdev, SDEV_RUNNING);
2680 if (sdev->quiesced_by) {
2681 sdev->quiesced_by = NULL;
2682 blk_clear_pm_only(sdev->request_queue);
2684 mutex_unlock(&sdev->state_mutex);
2686 EXPORT_SYMBOL(scsi_device_resume);
2689 device_quiesce_fn(struct scsi_device *sdev, void *data)
2691 scsi_device_quiesce(sdev);
2695 scsi_target_quiesce(struct scsi_target *starget)
2697 starget_for_each_device(starget, NULL, device_quiesce_fn);
2699 EXPORT_SYMBOL(scsi_target_quiesce);
2702 device_resume_fn(struct scsi_device *sdev, void *data)
2704 scsi_device_resume(sdev);
2708 scsi_target_resume(struct scsi_target *starget)
2710 starget_for_each_device(starget, NULL, device_resume_fn);
2712 EXPORT_SYMBOL(scsi_target_resume);
2714 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2716 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2717 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2722 void scsi_start_queue(struct scsi_device *sdev)
2724 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2725 blk_mq_unquiesce_queue(sdev->request_queue);
2728 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2731 * The atomic variable of ->queue_stopped covers that
2732 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2734 * However, we still need to wait until quiesce is done
2735 * in case that queue has been stopped.
2737 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2739 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2741 blk_mq_quiesce_queue(sdev->request_queue);
2744 blk_mq_wait_quiesce_done(sdev->request_queue->tag_set);
2749 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2750 * @sdev: device to block
2752 * Pause SCSI command processing on the specified device. Does not sleep.
2754 * Returns zero if successful or a negative error code upon failure.
2757 * This routine transitions the device to the SDEV_BLOCK state (which must be
2758 * a legal transition). When the device is in this state, command processing
2759 * is paused until the device leaves the SDEV_BLOCK state. See also
2760 * scsi_internal_device_unblock_nowait().
2762 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2764 int ret = __scsi_internal_device_block_nowait(sdev);
2767 * The device has transitioned to SDEV_BLOCK. Stop the
2768 * block layer from calling the midlayer with this device's
2772 scsi_stop_queue(sdev, true);
2775 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2778 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2779 * @sdev: device to block
2781 * Pause SCSI command processing on the specified device and wait until all
2782 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2784 * Returns zero if successful or a negative error code upon failure.
2787 * This routine transitions the device to the SDEV_BLOCK state (which must be
2788 * a legal transition). When the device is in this state, command processing
2789 * is paused until the device leaves the SDEV_BLOCK state. See also
2790 * scsi_internal_device_unblock().
2792 static int scsi_internal_device_block(struct scsi_device *sdev)
2796 mutex_lock(&sdev->state_mutex);
2797 err = __scsi_internal_device_block_nowait(sdev);
2799 scsi_stop_queue(sdev, false);
2800 mutex_unlock(&sdev->state_mutex);
2806 * scsi_internal_device_unblock_nowait - resume a device after a block request
2807 * @sdev: device to resume
2808 * @new_state: state to set the device to after unblocking
2810 * Restart the device queue for a previously suspended SCSI device. Does not
2813 * Returns zero if successful or a negative error code upon failure.
2816 * This routine transitions the device to the SDEV_RUNNING state or to one of
2817 * the offline states (which must be a legal transition) allowing the midlayer
2818 * to goose the queue for this device.
2820 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2821 enum scsi_device_state new_state)
2823 switch (new_state) {
2825 case SDEV_TRANSPORT_OFFLINE:
2832 * Try to transition the scsi device to SDEV_RUNNING or one of the
2833 * offlined states and goose the device queue if successful.
2835 switch (sdev->sdev_state) {
2837 case SDEV_TRANSPORT_OFFLINE:
2838 sdev->sdev_state = new_state;
2840 case SDEV_CREATED_BLOCK:
2841 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2842 new_state == SDEV_OFFLINE)
2843 sdev->sdev_state = new_state;
2845 sdev->sdev_state = SDEV_CREATED;
2853 scsi_start_queue(sdev);
2857 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2860 * scsi_internal_device_unblock - resume a device after a block request
2861 * @sdev: device to resume
2862 * @new_state: state to set the device to after unblocking
2864 * Restart the device queue for a previously suspended SCSI device. May sleep.
2866 * Returns zero if successful or a negative error code upon failure.
2869 * This routine transitions the device to the SDEV_RUNNING state or to one of
2870 * the offline states (which must be a legal transition) allowing the midlayer
2871 * to goose the queue for this device.
2873 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2874 enum scsi_device_state new_state)
2878 mutex_lock(&sdev->state_mutex);
2879 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2880 mutex_unlock(&sdev->state_mutex);
2886 device_block(struct scsi_device *sdev, void *data)
2890 ret = scsi_internal_device_block(sdev);
2892 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2893 dev_name(&sdev->sdev_gendev), ret);
2897 target_block(struct device *dev, void *data)
2899 if (scsi_is_target_device(dev))
2900 starget_for_each_device(to_scsi_target(dev), NULL,
2906 scsi_target_block(struct device *dev)
2908 if (scsi_is_target_device(dev))
2909 starget_for_each_device(to_scsi_target(dev), NULL,
2912 device_for_each_child(dev, NULL, target_block);
2914 EXPORT_SYMBOL_GPL(scsi_target_block);
2917 device_unblock(struct scsi_device *sdev, void *data)
2919 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2923 target_unblock(struct device *dev, void *data)
2925 if (scsi_is_target_device(dev))
2926 starget_for_each_device(to_scsi_target(dev), data,
2932 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2934 if (scsi_is_target_device(dev))
2935 starget_for_each_device(to_scsi_target(dev), &new_state,
2938 device_for_each_child(dev, &new_state, target_unblock);
2940 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2943 scsi_host_block(struct Scsi_Host *shost)
2945 struct scsi_device *sdev;
2949 * Call scsi_internal_device_block_nowait so we can avoid
2950 * calling synchronize_rcu() for each LUN.
2952 shost_for_each_device(sdev, shost) {
2953 mutex_lock(&sdev->state_mutex);
2954 ret = scsi_internal_device_block_nowait(sdev);
2955 mutex_unlock(&sdev->state_mutex);
2957 scsi_device_put(sdev);
2963 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2964 * calling synchronize_rcu() once is enough.
2966 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2973 EXPORT_SYMBOL_GPL(scsi_host_block);
2976 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2978 struct scsi_device *sdev;
2981 shost_for_each_device(sdev, shost) {
2982 ret = scsi_internal_device_unblock(sdev, new_state);
2984 scsi_device_put(sdev);
2990 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2993 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2994 * @sgl: scatter-gather list
2995 * @sg_count: number of segments in sg
2996 * @offset: offset in bytes into sg, on return offset into the mapped area
2997 * @len: bytes to map, on return number of bytes mapped
2999 * Returns virtual address of the start of the mapped page
3001 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3002 size_t *offset, size_t *len)
3005 size_t sg_len = 0, len_complete = 0;
3006 struct scatterlist *sg;
3009 WARN_ON(!irqs_disabled());
3011 for_each_sg(sgl, sg, sg_count, i) {
3012 len_complete = sg_len; /* Complete sg-entries */
3013 sg_len += sg->length;
3014 if (sg_len > *offset)
3018 if (unlikely(i == sg_count)) {
3019 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3021 __func__, sg_len, *offset, sg_count);
3026 /* Offset starting from the beginning of first page in this sg-entry */
3027 *offset = *offset - len_complete + sg->offset;
3029 /* Assumption: contiguous pages can be accessed as "page + i" */
3030 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3031 *offset &= ~PAGE_MASK;
3033 /* Bytes in this sg-entry from *offset to the end of the page */
3034 sg_len = PAGE_SIZE - *offset;
3038 return kmap_atomic(page);
3040 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3043 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3044 * @virt: virtual address to be unmapped
3046 void scsi_kunmap_atomic_sg(void *virt)
3048 kunmap_atomic(virt);
3050 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3052 void sdev_disable_disk_events(struct scsi_device *sdev)
3054 atomic_inc(&sdev->disk_events_disable_depth);
3056 EXPORT_SYMBOL(sdev_disable_disk_events);
3058 void sdev_enable_disk_events(struct scsi_device *sdev)
3060 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3062 atomic_dec(&sdev->disk_events_disable_depth);
3064 EXPORT_SYMBOL(sdev_enable_disk_events);
3066 static unsigned char designator_prio(const unsigned char *d)
3069 /* not associated with LUN */
3073 /* invalid length */
3077 * Order of preference for lun descriptor:
3078 * - SCSI name string
3079 * - NAA IEEE Registered Extended
3080 * - EUI-64 based 16-byte
3081 * - EUI-64 based 12-byte
3082 * - NAA IEEE Registered
3083 * - NAA IEEE Extended
3084 * - EUI-64 based 8-byte
3085 * - SCSI name string (truncated)
3087 * as longer descriptors reduce the likelyhood
3088 * of identification clashes.
3091 switch (d[1] & 0xf) {
3093 /* SCSI name string, variable-length UTF-8 */
3096 switch (d[4] >> 4) {
3098 /* NAA registered extended */
3101 /* NAA registered */
3107 /* NAA locally assigned */
3116 /* EUI64-based, 16 byte */
3119 /* EUI64-based, 12 byte */
3122 /* EUI64-based, 8 byte */
3139 * scsi_vpd_lun_id - return a unique device identification
3140 * @sdev: SCSI device
3141 * @id: buffer for the identification
3142 * @id_len: length of the buffer
3144 * Copies a unique device identification into @id based
3145 * on the information in the VPD page 0x83 of the device.
3146 * The string will be formatted as a SCSI name string.
3148 * Returns the length of the identification or error on failure.
3149 * If the identifier is longer than the supplied buffer the actual
3150 * identifier length is returned and the buffer is not zero-padded.
3152 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3156 const unsigned char *d, *cur_id_str;
3157 const struct scsi_vpd *vpd_pg83;
3158 int id_size = -EINVAL;
3161 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3167 /* The id string must be at least 20 bytes + terminating NULL byte */
3173 memset(id, 0, id_len);
3174 for (d = vpd_pg83->data + 4;
3175 d < vpd_pg83->data + vpd_pg83->len;
3177 u8 prio = designator_prio(d);
3179 if (prio == 0 || cur_id_prio > prio)
3182 switch (d[1] & 0xf) {
3185 if (cur_id_size > d[3])
3189 if (cur_id_size + 4 > id_len)
3190 cur_id_size = id_len - 4;
3192 id_size = snprintf(id, id_len, "t10.%*pE",
3193 cur_id_size, cur_id_str);
3200 switch (cur_id_size) {
3202 id_size = snprintf(id, id_len,
3207 id_size = snprintf(id, id_len,
3212 id_size = snprintf(id, id_len,
3225 switch (cur_id_size) {
3227 id_size = snprintf(id, id_len,
3232 id_size = snprintf(id, id_len,
3241 /* SCSI name string */
3242 if (cur_id_size > d[3])
3244 /* Prefer others for truncated descriptor */
3245 if (d[3] > id_len) {
3247 if (cur_id_prio > prio)
3251 cur_id_size = id_size = d[3];
3253 if (cur_id_size >= id_len)
3254 cur_id_size = id_len - 1;
3255 memcpy(id, cur_id_str, cur_id_size);
3265 EXPORT_SYMBOL(scsi_vpd_lun_id);
3268 * scsi_vpd_tpg_id - return a target port group identifier
3269 * @sdev: SCSI device
3271 * Returns the Target Port Group identifier from the information
3272 * froom VPD page 0x83 of the device.
3274 * Returns the identifier or error on failure.
3276 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3278 const unsigned char *d;
3279 const struct scsi_vpd *vpd_pg83;
3280 int group_id = -EAGAIN, rel_port = -1;
3283 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3289 d = vpd_pg83->data + 4;
3290 while (d < vpd_pg83->data + vpd_pg83->len) {
3291 switch (d[1] & 0xf) {
3293 /* Relative target port */
3294 rel_port = get_unaligned_be16(&d[6]);
3297 /* Target port group */
3298 group_id = get_unaligned_be16(&d[6]);
3307 if (group_id >= 0 && rel_id && rel_port != -1)
3312 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3315 * scsi_build_sense - build sense data for a command
3316 * @scmd: scsi command for which the sense should be formatted
3317 * @desc: Sense format (non-zero == descriptor format,
3318 * 0 == fixed format)
3320 * @asc: Additional sense code
3321 * @ascq: Additional sense code qualifier
3324 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3326 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3327 scmd->result = SAM_STAT_CHECK_CONDITION;
3329 EXPORT_SYMBOL_GPL(scsi_build_sense);