2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
47 static inline struct kmem_cache *
48 scsi_select_sense_cache(struct Scsi_Host *shost)
50 return shost->unchecked_isa_dma ?
51 scsi_sense_isadma_cache : scsi_sense_cache;
54 static void scsi_free_sense_buffer(struct Scsi_Host *shost,
55 unsigned char *sense_buffer)
57 kmem_cache_free(scsi_select_sense_cache(shost), sense_buffer);
60 static unsigned char *scsi_alloc_sense_buffer(struct Scsi_Host *shost,
61 gfp_t gfp_mask, int numa_node)
63 return kmem_cache_alloc_node(scsi_select_sense_cache(shost), gfp_mask,
67 int scsi_init_sense_cache(struct Scsi_Host *shost)
69 struct kmem_cache *cache;
72 cache = scsi_select_sense_cache(shost);
76 mutex_lock(&scsi_sense_cache_mutex);
77 if (shost->unchecked_isa_dma) {
78 scsi_sense_isadma_cache =
79 kmem_cache_create("scsi_sense_cache(DMA)",
80 SCSI_SENSE_BUFFERSIZE, 0,
81 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
82 if (!scsi_sense_isadma_cache)
86 kmem_cache_create("scsi_sense_cache",
87 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN, NULL);
88 if (!scsi_sense_cache)
92 mutex_unlock(&scsi_sense_cache_mutex);
97 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
98 * not change behaviour from the previous unplug mechanism, experimentation
99 * may prove this needs changing.
101 #define SCSI_QUEUE_DELAY 3
104 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
106 struct Scsi_Host *host = cmd->device->host;
107 struct scsi_device *device = cmd->device;
108 struct scsi_target *starget = scsi_target(device);
111 * Set the appropriate busy bit for the device/host.
113 * If the host/device isn't busy, assume that something actually
114 * completed, and that we should be able to queue a command now.
116 * Note that the prior mid-layer assumption that any host could
117 * always queue at least one command is now broken. The mid-layer
118 * will implement a user specifiable stall (see
119 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
120 * if a command is requeued with no other commands outstanding
121 * either for the device or for the host.
124 case SCSI_MLQUEUE_HOST_BUSY:
125 atomic_set(&host->host_blocked, host->max_host_blocked);
127 case SCSI_MLQUEUE_DEVICE_BUSY:
128 case SCSI_MLQUEUE_EH_RETRY:
129 atomic_set(&device->device_blocked,
130 device->max_device_blocked);
132 case SCSI_MLQUEUE_TARGET_BUSY:
133 atomic_set(&starget->target_blocked,
134 starget->max_target_blocked);
139 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
141 struct scsi_device *sdev = cmd->device;
143 blk_mq_requeue_request(cmd->request, true);
144 put_device(&sdev->sdev_gendev);
148 * __scsi_queue_insert - private queue insertion
149 * @cmd: The SCSI command being requeued
150 * @reason: The reason for the requeue
151 * @unbusy: Whether the queue should be unbusied
153 * This is a private queue insertion. The public interface
154 * scsi_queue_insert() always assumes the queue should be unbusied
155 * because it's always called before the completion. This function is
156 * for a requeue after completion, which should only occur in this
159 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
161 struct scsi_device *device = cmd->device;
162 struct request_queue *q = device->request_queue;
165 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
166 "Inserting command %p into mlqueue\n", cmd));
168 scsi_set_blocked(cmd, reason);
171 * Decrement the counters, since these commands are no longer
172 * active on the host/device.
175 scsi_device_unbusy(device);
178 * Requeue this command. It will go before all other commands
179 * that are already in the queue. Schedule requeue work under
180 * lock such that the kblockd_schedule_work() call happens
181 * before blk_cleanup_queue() finishes.
185 scsi_mq_requeue_cmd(cmd);
188 spin_lock_irqsave(q->queue_lock, flags);
189 blk_requeue_request(q, cmd->request);
190 kblockd_schedule_work(&device->requeue_work);
191 spin_unlock_irqrestore(q->queue_lock, flags);
195 * Function: scsi_queue_insert()
197 * Purpose: Insert a command in the midlevel queue.
199 * Arguments: cmd - command that we are adding to queue.
200 * reason - why we are inserting command to queue.
202 * Lock status: Assumed that lock is not held upon entry.
206 * Notes: We do this for one of two cases. Either the host is busy
207 * and it cannot accept any more commands for the time being,
208 * or the device returned QUEUE_FULL and can accept no more
210 * Notes: This could be called either from an interrupt context or a
211 * normal process context.
213 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
215 __scsi_queue_insert(cmd, reason, 1);
220 * scsi_execute - insert request and wait for the result
223 * @data_direction: data direction
224 * @buffer: data buffer
225 * @bufflen: len of buffer
226 * @sense: optional sense buffer
227 * @sshdr: optional decoded sense header
228 * @timeout: request timeout in seconds
229 * @retries: number of times to retry request
230 * @flags: flags for ->cmd_flags
231 * @rq_flags: flags for ->rq_flags
232 * @resid: optional residual length
234 * Returns the scsi_cmnd result field if a command was executed, or a negative
235 * Linux error code if we didn't get that far.
237 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
238 int data_direction, void *buffer, unsigned bufflen,
239 unsigned char *sense, struct scsi_sense_hdr *sshdr,
240 int timeout, int retries, u64 flags, req_flags_t rq_flags,
244 struct scsi_request *rq;
245 int ret = DRIVER_ERROR << 24;
247 req = blk_get_request(sdev->request_queue,
248 data_direction == DMA_TO_DEVICE ?
249 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, __GFP_RECLAIM);
254 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
255 buffer, bufflen, __GFP_RECLAIM))
258 rq->cmd_len = COMMAND_SIZE(cmd[0]);
259 memcpy(rq->cmd, cmd, rq->cmd_len);
260 rq->retries = retries;
261 req->timeout = timeout;
262 req->cmd_flags |= flags;
263 req->rq_flags |= rq_flags | RQF_QUIET | RQF_PREEMPT;
266 * head injection *required* here otherwise quiesce won't work
268 blk_execute_rq(req->q, NULL, req, 1);
271 * Some devices (USB mass-storage in particular) may transfer
272 * garbage data together with a residue indicating that the data
273 * is invalid. Prevent the garbage from being misinterpreted
274 * and prevent security leaks by zeroing out the excess data.
276 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
277 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
280 *resid = rq->resid_len;
281 if (sense && rq->sense_len)
282 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
284 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
287 blk_put_request(req);
291 EXPORT_SYMBOL(scsi_execute);
294 * Function: scsi_init_cmd_errh()
296 * Purpose: Initialize cmd fields related to error handling.
298 * Arguments: cmd - command that is ready to be queued.
300 * Notes: This function has the job of initializing a number of
301 * fields related to error handling. Typically this will
302 * be called once for each command, as required.
304 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
306 cmd->serial_number = 0;
307 scsi_set_resid(cmd, 0);
308 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
309 if (cmd->cmd_len == 0)
310 cmd->cmd_len = scsi_command_size(cmd->cmnd);
313 void scsi_device_unbusy(struct scsi_device *sdev)
315 struct Scsi_Host *shost = sdev->host;
316 struct scsi_target *starget = scsi_target(sdev);
319 atomic_dec(&shost->host_busy);
320 if (starget->can_queue > 0)
321 atomic_dec(&starget->target_busy);
323 if (unlikely(scsi_host_in_recovery(shost) &&
324 (shost->host_failed || shost->host_eh_scheduled))) {
325 spin_lock_irqsave(shost->host_lock, flags);
326 scsi_eh_wakeup(shost);
327 spin_unlock_irqrestore(shost->host_lock, flags);
330 atomic_dec(&sdev->device_busy);
333 static void scsi_kick_queue(struct request_queue *q)
336 blk_mq_start_hw_queues(q);
342 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
343 * and call blk_run_queue for all the scsi_devices on the target -
344 * including current_sdev first.
346 * Called with *no* scsi locks held.
348 static void scsi_single_lun_run(struct scsi_device *current_sdev)
350 struct Scsi_Host *shost = current_sdev->host;
351 struct scsi_device *sdev, *tmp;
352 struct scsi_target *starget = scsi_target(current_sdev);
355 spin_lock_irqsave(shost->host_lock, flags);
356 starget->starget_sdev_user = NULL;
357 spin_unlock_irqrestore(shost->host_lock, flags);
360 * Call blk_run_queue for all LUNs on the target, starting with
361 * current_sdev. We race with others (to set starget_sdev_user),
362 * but in most cases, we will be first. Ideally, each LU on the
363 * target would get some limited time or requests on the target.
365 scsi_kick_queue(current_sdev->request_queue);
367 spin_lock_irqsave(shost->host_lock, flags);
368 if (starget->starget_sdev_user)
370 list_for_each_entry_safe(sdev, tmp, &starget->devices,
371 same_target_siblings) {
372 if (sdev == current_sdev)
374 if (scsi_device_get(sdev))
377 spin_unlock_irqrestore(shost->host_lock, flags);
378 scsi_kick_queue(sdev->request_queue);
379 spin_lock_irqsave(shost->host_lock, flags);
381 scsi_device_put(sdev);
384 spin_unlock_irqrestore(shost->host_lock, flags);
387 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
389 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
391 if (atomic_read(&sdev->device_blocked) > 0)
396 static inline bool scsi_target_is_busy(struct scsi_target *starget)
398 if (starget->can_queue > 0) {
399 if (atomic_read(&starget->target_busy) >= starget->can_queue)
401 if (atomic_read(&starget->target_blocked) > 0)
407 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
409 if (shost->can_queue > 0 &&
410 atomic_read(&shost->host_busy) >= shost->can_queue)
412 if (atomic_read(&shost->host_blocked) > 0)
414 if (shost->host_self_blocked)
419 static void scsi_starved_list_run(struct Scsi_Host *shost)
421 LIST_HEAD(starved_list);
422 struct scsi_device *sdev;
425 spin_lock_irqsave(shost->host_lock, flags);
426 list_splice_init(&shost->starved_list, &starved_list);
428 while (!list_empty(&starved_list)) {
429 struct request_queue *slq;
432 * As long as shost is accepting commands and we have
433 * starved queues, call blk_run_queue. scsi_request_fn
434 * drops the queue_lock and can add us back to the
437 * host_lock protects the starved_list and starved_entry.
438 * scsi_request_fn must get the host_lock before checking
439 * or modifying starved_list or starved_entry.
441 if (scsi_host_is_busy(shost))
444 sdev = list_entry(starved_list.next,
445 struct scsi_device, starved_entry);
446 list_del_init(&sdev->starved_entry);
447 if (scsi_target_is_busy(scsi_target(sdev))) {
448 list_move_tail(&sdev->starved_entry,
449 &shost->starved_list);
454 * Once we drop the host lock, a racing scsi_remove_device()
455 * call may remove the sdev from the starved list and destroy
456 * it and the queue. Mitigate by taking a reference to the
457 * queue and never touching the sdev again after we drop the
458 * host lock. Note: if __scsi_remove_device() invokes
459 * blk_cleanup_queue() before the queue is run from this
460 * function then blk_run_queue() will return immediately since
461 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
463 slq = sdev->request_queue;
464 if (!blk_get_queue(slq))
466 spin_unlock_irqrestore(shost->host_lock, flags);
468 scsi_kick_queue(slq);
471 spin_lock_irqsave(shost->host_lock, flags);
473 /* put any unprocessed entries back */
474 list_splice(&starved_list, &shost->starved_list);
475 spin_unlock_irqrestore(shost->host_lock, flags);
479 * Function: scsi_run_queue()
481 * Purpose: Select a proper request queue to serve next
483 * Arguments: q - last request's queue
487 * Notes: The previous command was completely finished, start
488 * a new one if possible.
490 static void scsi_run_queue(struct request_queue *q)
492 struct scsi_device *sdev = q->queuedata;
494 if (scsi_target(sdev)->single_lun)
495 scsi_single_lun_run(sdev);
496 if (!list_empty(&sdev->host->starved_list))
497 scsi_starved_list_run(sdev->host);
500 blk_mq_run_hw_queues(q, false);
505 void scsi_requeue_run_queue(struct work_struct *work)
507 struct scsi_device *sdev;
508 struct request_queue *q;
510 sdev = container_of(work, struct scsi_device, requeue_work);
511 q = sdev->request_queue;
516 * Function: scsi_requeue_command()
518 * Purpose: Handle post-processing of completed commands.
520 * Arguments: q - queue to operate on
521 * cmd - command that may need to be requeued.
525 * Notes: After command completion, there may be blocks left
526 * over which weren't finished by the previous command
527 * this can be for a number of reasons - the main one is
528 * I/O errors in the middle of the request, in which case
529 * we need to request the blocks that come after the bad
531 * Notes: Upon return, cmd is a stale pointer.
533 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
535 struct scsi_device *sdev = cmd->device;
536 struct request *req = cmd->request;
539 spin_lock_irqsave(q->queue_lock, flags);
540 blk_unprep_request(req);
542 scsi_put_command(cmd);
543 blk_requeue_request(q, req);
544 spin_unlock_irqrestore(q->queue_lock, flags);
548 put_device(&sdev->sdev_gendev);
551 void scsi_run_host_queues(struct Scsi_Host *shost)
553 struct scsi_device *sdev;
555 shost_for_each_device(sdev, shost)
556 scsi_run_queue(sdev->request_queue);
559 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
561 if (!blk_rq_is_passthrough(cmd->request)) {
562 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
564 if (drv->uninit_command)
565 drv->uninit_command(cmd);
569 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
571 struct scsi_data_buffer *sdb;
573 if (cmd->sdb.table.nents)
574 sg_free_table_chained(&cmd->sdb.table, true);
575 if (cmd->request->next_rq) {
576 sdb = cmd->request->next_rq->special;
578 sg_free_table_chained(&sdb->table, true);
580 if (scsi_prot_sg_count(cmd))
581 sg_free_table_chained(&cmd->prot_sdb->table, true);
584 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
586 struct scsi_device *sdev = cmd->device;
587 struct Scsi_Host *shost = sdev->host;
590 scsi_mq_free_sgtables(cmd);
591 scsi_uninit_cmd(cmd);
593 if (shost->use_cmd_list) {
594 BUG_ON(list_empty(&cmd->list));
595 spin_lock_irqsave(&sdev->list_lock, flags);
596 list_del_init(&cmd->list);
597 spin_unlock_irqrestore(&sdev->list_lock, flags);
602 * Function: scsi_release_buffers()
604 * Purpose: Free resources allocate for a scsi_command.
606 * Arguments: cmd - command that we are bailing.
608 * Lock status: Assumed that no lock is held upon entry.
612 * Notes: In the event that an upper level driver rejects a
613 * command, we must release resources allocated during
614 * the __init_io() function. Primarily this would involve
615 * the scatter-gather table.
617 static void scsi_release_buffers(struct scsi_cmnd *cmd)
619 if (cmd->sdb.table.nents)
620 sg_free_table_chained(&cmd->sdb.table, false);
622 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
624 if (scsi_prot_sg_count(cmd))
625 sg_free_table_chained(&cmd->prot_sdb->table, false);
628 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
630 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
632 sg_free_table_chained(&bidi_sdb->table, false);
633 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
634 cmd->request->next_rq->special = NULL;
637 static bool scsi_end_request(struct request *req, blk_status_t error,
638 unsigned int bytes, unsigned int bidi_bytes)
640 struct scsi_cmnd *cmd = req->special;
641 struct scsi_device *sdev = cmd->device;
642 struct request_queue *q = sdev->request_queue;
644 if (blk_update_request(req, error, bytes))
647 /* Bidi request must be completed as a whole */
648 if (unlikely(bidi_bytes) &&
649 blk_update_request(req->next_rq, error, bidi_bytes))
652 if (blk_queue_add_random(q))
653 add_disk_randomness(req->rq_disk);
657 * In the MQ case the command gets freed by __blk_mq_end_request,
658 * so we have to do all cleanup that depends on it earlier.
660 * We also can't kick the queues from irq context, so we
661 * will have to defer it to a workqueue.
663 scsi_mq_uninit_cmd(cmd);
665 __blk_mq_end_request(req, error);
667 if (scsi_target(sdev)->single_lun ||
668 !list_empty(&sdev->host->starved_list))
669 kblockd_schedule_work(&sdev->requeue_work);
671 blk_mq_run_hw_queues(q, true);
676 scsi_release_bidi_buffers(cmd);
677 scsi_release_buffers(cmd);
678 scsi_put_command(cmd);
680 spin_lock_irqsave(q->queue_lock, flags);
681 blk_finish_request(req, error);
682 spin_unlock_irqrestore(q->queue_lock, flags);
687 put_device(&sdev->sdev_gendev);
692 * __scsi_error_from_host_byte - translate SCSI error code into errno
693 * @cmd: SCSI command (unused)
694 * @result: scsi error code
696 * Translate SCSI error code into block errors.
698 static blk_status_t __scsi_error_from_host_byte(struct scsi_cmnd *cmd,
701 switch (host_byte(result)) {
702 case DID_TRANSPORT_FAILFAST:
703 return BLK_STS_TRANSPORT;
704 case DID_TARGET_FAILURE:
705 set_host_byte(cmd, DID_OK);
706 return BLK_STS_TARGET;
707 case DID_NEXUS_FAILURE:
708 return BLK_STS_NEXUS;
709 case DID_ALLOC_FAILURE:
710 set_host_byte(cmd, DID_OK);
711 return BLK_STS_NOSPC;
712 case DID_MEDIUM_ERROR:
713 set_host_byte(cmd, DID_OK);
714 return BLK_STS_MEDIUM;
716 return BLK_STS_IOERR;
721 * Function: scsi_io_completion()
723 * Purpose: Completion processing for block device I/O requests.
725 * Arguments: cmd - command that is finished.
727 * Lock status: Assumed that no lock is held upon entry.
731 * Notes: We will finish off the specified number of sectors. If we
732 * are done, the command block will be released and the queue
733 * function will be goosed. If we are not done then we have to
734 * figure out what to do next:
736 * a) We can call scsi_requeue_command(). The request
737 * will be unprepared and put back on the queue. Then
738 * a new command will be created for it. This should
739 * be used if we made forward progress, or if we want
740 * to switch from READ(10) to READ(6) for example.
742 * b) We can call __scsi_queue_insert(). The request will
743 * be put back on the queue and retried using the same
744 * command as before, possibly after a delay.
746 * c) We can call scsi_end_request() with -EIO to fail
747 * the remainder of the request.
749 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
751 int result = cmd->result;
752 struct request_queue *q = cmd->device->request_queue;
753 struct request *req = cmd->request;
754 blk_status_t error = BLK_STS_OK;
755 struct scsi_sense_hdr sshdr;
756 bool sense_valid = false;
757 int sense_deferred = 0, level = 0;
758 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
759 ACTION_DELAYED_RETRY} action;
760 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
763 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
765 sense_deferred = scsi_sense_is_deferred(&sshdr);
768 if (blk_rq_is_passthrough(req)) {
772 * SG_IO wants current and deferred errors
774 scsi_req(req)->sense_len =
775 min(8 + cmd->sense_buffer[7],
776 SCSI_SENSE_BUFFERSIZE);
779 error = __scsi_error_from_host_byte(cmd, result);
782 * __scsi_error_from_host_byte may have reset the host_byte
784 scsi_req(req)->result = cmd->result;
785 scsi_req(req)->resid_len = scsi_get_resid(cmd);
787 if (scsi_bidi_cmnd(cmd)) {
789 * Bidi commands Must be complete as a whole,
790 * both sides at once.
792 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
793 if (scsi_end_request(req, BLK_STS_OK, blk_rq_bytes(req),
794 blk_rq_bytes(req->next_rq)))
798 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
800 * Flush commands do not transfers any data, and thus cannot use
801 * good_bytes != blk_rq_bytes(req) as the signal for an error.
802 * This sets the error explicitly for the problem case.
804 error = __scsi_error_from_host_byte(cmd, result);
807 /* no bidi support for !blk_rq_is_passthrough yet */
808 BUG_ON(blk_bidi_rq(req));
811 * Next deal with any sectors which we were able to correctly
814 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
815 "%u sectors total, %d bytes done.\n",
816 blk_rq_sectors(req), good_bytes));
819 * Recovered errors need reporting, but they're always treated as
820 * success, so fiddle the result code here. For passthrough requests
821 * we already took a copy of the original into sreq->result which
822 * is what gets returned to the user
824 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
825 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
826 * print since caller wants ATA registers. Only occurs on
827 * SCSI ATA PASS_THROUGH commands when CK_COND=1
829 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
831 else if (!(req->rq_flags & RQF_QUIET))
832 scsi_print_sense(cmd);
834 /* for passthrough error may be set */
839 * special case: failed zero length commands always need to
840 * drop down into the retry code. Otherwise, if we finished
841 * all bytes in the request we are done now.
843 if (!(blk_rq_bytes(req) == 0 && error) &&
844 !scsi_end_request(req, error, good_bytes, 0))
848 * Kill remainder if no retrys.
850 if (error && scsi_noretry_cmd(cmd)) {
851 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
857 * If there had been no error, but we have leftover bytes in the
858 * requeues just queue the command up again.
863 error = __scsi_error_from_host_byte(cmd, result);
865 if (host_byte(result) == DID_RESET) {
866 /* Third party bus reset or reset for error recovery
867 * reasons. Just retry the command and see what
870 action = ACTION_RETRY;
871 } else if (sense_valid && !sense_deferred) {
872 switch (sshdr.sense_key) {
874 if (cmd->device->removable) {
875 /* Detected disc change. Set a bit
876 * and quietly refuse further access.
878 cmd->device->changed = 1;
879 action = ACTION_FAIL;
881 /* Must have been a power glitch, or a
882 * bus reset. Could not have been a
883 * media change, so we just retry the
884 * command and see what happens.
886 action = ACTION_RETRY;
889 case ILLEGAL_REQUEST:
890 /* If we had an ILLEGAL REQUEST returned, then
891 * we may have performed an unsupported
892 * command. The only thing this should be
893 * would be a ten byte read where only a six
894 * byte read was supported. Also, on a system
895 * where READ CAPACITY failed, we may have
896 * read past the end of the disk.
898 if ((cmd->device->use_10_for_rw &&
899 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
900 (cmd->cmnd[0] == READ_10 ||
901 cmd->cmnd[0] == WRITE_10)) {
902 /* This will issue a new 6-byte command. */
903 cmd->device->use_10_for_rw = 0;
904 action = ACTION_REPREP;
905 } else if (sshdr.asc == 0x10) /* DIX */ {
906 action = ACTION_FAIL;
907 error = BLK_STS_PROTECTION;
908 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
909 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
910 action = ACTION_FAIL;
911 error = BLK_STS_TARGET;
913 action = ACTION_FAIL;
915 case ABORTED_COMMAND:
916 action = ACTION_FAIL;
917 if (sshdr.asc == 0x10) /* DIF */
918 error = BLK_STS_PROTECTION;
921 /* If the device is in the process of becoming
922 * ready, or has a temporary blockage, retry.
924 if (sshdr.asc == 0x04) {
925 switch (sshdr.ascq) {
926 case 0x01: /* becoming ready */
927 case 0x04: /* format in progress */
928 case 0x05: /* rebuild in progress */
929 case 0x06: /* recalculation in progress */
930 case 0x07: /* operation in progress */
931 case 0x08: /* Long write in progress */
932 case 0x09: /* self test in progress */
933 case 0x14: /* space allocation in progress */
934 action = ACTION_DELAYED_RETRY;
937 action = ACTION_FAIL;
941 action = ACTION_FAIL;
943 case VOLUME_OVERFLOW:
944 /* See SSC3rXX or current. */
945 action = ACTION_FAIL;
948 action = ACTION_FAIL;
952 action = ACTION_FAIL;
954 if (action != ACTION_FAIL &&
955 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
956 action = ACTION_FAIL;
960 /* Give up and fail the remainder of the request */
961 if (!(req->rq_flags & RQF_QUIET)) {
962 static DEFINE_RATELIMIT_STATE(_rs,
963 DEFAULT_RATELIMIT_INTERVAL,
964 DEFAULT_RATELIMIT_BURST);
966 if (unlikely(scsi_logging_level))
967 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
968 SCSI_LOG_MLCOMPLETE_BITS);
971 * if logging is enabled the failure will be printed
972 * in scsi_log_completion(), so avoid duplicate messages
974 if (!level && __ratelimit(&_rs)) {
975 scsi_print_result(cmd, NULL, FAILED);
976 if (driver_byte(result) & DRIVER_SENSE)
977 scsi_print_sense(cmd);
978 scsi_print_command(cmd);
981 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
986 /* Unprep the request and put it back at the head of the queue.
987 * A new command will be prepared and issued.
990 cmd->request->rq_flags &= ~RQF_DONTPREP;
991 scsi_mq_uninit_cmd(cmd);
992 scsi_mq_requeue_cmd(cmd);
994 scsi_release_buffers(cmd);
995 scsi_requeue_command(q, cmd);
999 /* Retry the same command immediately */
1000 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1002 case ACTION_DELAYED_RETRY:
1003 /* Retry the same command after a delay */
1004 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1009 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1014 * If sg table allocation fails, requeue request later.
1016 if (unlikely(sg_alloc_table_chained(&sdb->table,
1017 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1018 return BLKPREP_DEFER;
1021 * Next, walk the list, and fill in the addresses and sizes of
1024 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1025 BUG_ON(count > sdb->table.nents);
1026 sdb->table.nents = count;
1027 sdb->length = blk_rq_payload_bytes(req);
1032 * Function: scsi_init_io()
1034 * Purpose: SCSI I/O initialize function.
1036 * Arguments: cmd - Command descriptor we wish to initialize
1038 * Returns: 0 on success
1039 * BLKPREP_DEFER if the failure is retryable
1040 * BLKPREP_KILL if the failure is fatal
1042 int scsi_init_io(struct scsi_cmnd *cmd)
1044 struct scsi_device *sdev = cmd->device;
1045 struct request *rq = cmd->request;
1046 bool is_mq = (rq->mq_ctx != NULL);
1047 int error = BLKPREP_KILL;
1049 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1052 error = scsi_init_sgtable(rq, &cmd->sdb);
1056 if (blk_bidi_rq(rq)) {
1057 if (!rq->q->mq_ops) {
1058 struct scsi_data_buffer *bidi_sdb =
1059 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1061 error = BLKPREP_DEFER;
1065 rq->next_rq->special = bidi_sdb;
1068 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1073 if (blk_integrity_rq(rq)) {
1074 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1077 if (prot_sdb == NULL) {
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.
1084 error = BLKPREP_KILL;
1088 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1090 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1091 prot_sdb->table.sgl)) {
1092 error = BLKPREP_DEFER;
1096 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1097 prot_sdb->table.sgl);
1098 BUG_ON(unlikely(count > ivecs));
1099 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1101 cmd->prot_sdb = prot_sdb;
1102 cmd->prot_sdb->table.nents = count;
1108 scsi_mq_free_sgtables(cmd);
1110 scsi_release_buffers(cmd);
1111 cmd->request->special = NULL;
1112 scsi_put_command(cmd);
1113 put_device(&sdev->sdev_gendev);
1117 EXPORT_SYMBOL(scsi_init_io);
1120 * scsi_initialize_rq - initialize struct scsi_cmnd.req
1122 * Called from inside blk_get_request().
1124 void scsi_initialize_rq(struct request *rq)
1126 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1128 scsi_req_init(&cmd->req);
1130 EXPORT_SYMBOL(scsi_initialize_rq);
1132 /* Called after a request has been started. */
1133 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1135 void *buf = cmd->sense_buffer;
1136 void *prot = cmd->prot_sdb;
1137 unsigned long flags;
1139 /* zero out the cmd, except for the embedded scsi_request */
1140 memset((char *)cmd + sizeof(cmd->req), 0,
1141 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1144 cmd->sense_buffer = buf;
1145 cmd->prot_sdb = prot;
1146 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1147 cmd->jiffies_at_alloc = jiffies;
1149 spin_lock_irqsave(&dev->list_lock, flags);
1150 list_add_tail(&cmd->list, &dev->cmd_list);
1151 spin_unlock_irqrestore(&dev->list_lock, flags);
1154 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1156 struct scsi_cmnd *cmd = req->special;
1159 * Passthrough requests may transfer data, in which case they must
1160 * a bio attached to them. Or they might contain a SCSI command
1161 * that does not transfer data, in which case they may optionally
1162 * submit a request without an attached bio.
1165 int ret = scsi_init_io(cmd);
1169 BUG_ON(blk_rq_bytes(req));
1171 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1174 cmd->cmd_len = scsi_req(req)->cmd_len;
1175 cmd->cmnd = scsi_req(req)->cmd;
1176 cmd->transfersize = blk_rq_bytes(req);
1177 cmd->allowed = scsi_req(req)->retries;
1182 * Setup a normal block command. These are simple request from filesystems
1183 * that still need to be translated to SCSI CDBs from the ULD.
1185 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1187 struct scsi_cmnd *cmd = req->special;
1189 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1190 int ret = sdev->handler->prep_fn(sdev, req);
1191 if (ret != BLKPREP_OK)
1195 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1196 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1197 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1200 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1202 struct scsi_cmnd *cmd = req->special;
1204 if (!blk_rq_bytes(req))
1205 cmd->sc_data_direction = DMA_NONE;
1206 else if (rq_data_dir(req) == WRITE)
1207 cmd->sc_data_direction = DMA_TO_DEVICE;
1209 cmd->sc_data_direction = DMA_FROM_DEVICE;
1211 if (blk_rq_is_scsi(req))
1212 return scsi_setup_scsi_cmnd(sdev, req);
1214 return scsi_setup_fs_cmnd(sdev, req);
1218 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1220 int ret = BLKPREP_OK;
1223 * If the device is not in running state we will reject some
1226 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1227 switch (sdev->sdev_state) {
1229 case SDEV_TRANSPORT_OFFLINE:
1231 * If the device is offline we refuse to process any
1232 * commands. The device must be brought online
1233 * before trying any recovery commands.
1235 sdev_printk(KERN_ERR, sdev,
1236 "rejecting I/O to offline device\n");
1241 * If the device is fully deleted, we refuse to
1242 * process any commands as well.
1244 sdev_printk(KERN_ERR, sdev,
1245 "rejecting I/O to dead device\n");
1249 case SDEV_CREATED_BLOCK:
1250 ret = BLKPREP_DEFER;
1254 * If the devices is blocked we defer normal commands.
1256 if (!(req->rq_flags & RQF_PREEMPT))
1257 ret = BLKPREP_DEFER;
1261 * For any other not fully online state we only allow
1262 * special commands. In particular any user initiated
1263 * command is not allowed.
1265 if (!(req->rq_flags & RQF_PREEMPT))
1274 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1276 struct scsi_device *sdev = q->queuedata;
1280 case BLKPREP_INVALID:
1281 scsi_req(req)->result = DID_NO_CONNECT << 16;
1282 /* release the command and kill it */
1284 struct scsi_cmnd *cmd = req->special;
1285 scsi_release_buffers(cmd);
1286 scsi_put_command(cmd);
1287 put_device(&sdev->sdev_gendev);
1288 req->special = NULL;
1293 * If we defer, the blk_peek_request() returns NULL, but the
1294 * queue must be restarted, so we schedule a callback to happen
1297 if (atomic_read(&sdev->device_busy) == 0)
1298 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1301 req->rq_flags |= RQF_DONTPREP;
1307 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1309 struct scsi_device *sdev = q->queuedata;
1310 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1313 ret = scsi_prep_state_check(sdev, req);
1314 if (ret != BLKPREP_OK)
1317 if (!req->special) {
1318 /* Bail if we can't get a reference to the device */
1319 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1320 ret = BLKPREP_DEFER;
1324 scsi_init_command(sdev, cmd);
1328 cmd->tag = req->tag;
1330 cmd->prot_op = SCSI_PROT_NORMAL;
1332 ret = scsi_setup_cmnd(sdev, req);
1334 return scsi_prep_return(q, req, ret);
1337 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1339 scsi_uninit_cmd(req->special);
1343 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1346 * Called with the queue_lock held.
1348 static inline int scsi_dev_queue_ready(struct request_queue *q,
1349 struct scsi_device *sdev)
1353 busy = atomic_inc_return(&sdev->device_busy) - 1;
1354 if (atomic_read(&sdev->device_blocked)) {
1359 * unblock after device_blocked iterates to zero
1361 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1363 * For the MQ case we take care of this in the caller.
1366 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1369 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1370 "unblocking device at zero depth\n"));
1373 if (busy >= sdev->queue_depth)
1378 atomic_dec(&sdev->device_busy);
1383 * scsi_target_queue_ready: checks if there we can send commands to target
1384 * @sdev: scsi device on starget to check.
1386 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1387 struct scsi_device *sdev)
1389 struct scsi_target *starget = scsi_target(sdev);
1392 if (starget->single_lun) {
1393 spin_lock_irq(shost->host_lock);
1394 if (starget->starget_sdev_user &&
1395 starget->starget_sdev_user != sdev) {
1396 spin_unlock_irq(shost->host_lock);
1399 starget->starget_sdev_user = sdev;
1400 spin_unlock_irq(shost->host_lock);
1403 if (starget->can_queue <= 0)
1406 busy = atomic_inc_return(&starget->target_busy) - 1;
1407 if (atomic_read(&starget->target_blocked) > 0) {
1412 * unblock after target_blocked iterates to zero
1414 if (atomic_dec_return(&starget->target_blocked) > 0)
1417 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1418 "unblocking target at zero depth\n"));
1421 if (busy >= starget->can_queue)
1427 spin_lock_irq(shost->host_lock);
1428 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1429 spin_unlock_irq(shost->host_lock);
1431 if (starget->can_queue > 0)
1432 atomic_dec(&starget->target_busy);
1437 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1438 * return 0. We must end up running the queue again whenever 0 is
1439 * returned, else IO can hang.
1441 static inline int scsi_host_queue_ready(struct request_queue *q,
1442 struct Scsi_Host *shost,
1443 struct scsi_device *sdev)
1447 if (scsi_host_in_recovery(shost))
1450 busy = atomic_inc_return(&shost->host_busy) - 1;
1451 if (atomic_read(&shost->host_blocked) > 0) {
1456 * unblock after host_blocked iterates to zero
1458 if (atomic_dec_return(&shost->host_blocked) > 0)
1462 shost_printk(KERN_INFO, shost,
1463 "unblocking host at zero depth\n"));
1466 if (shost->can_queue > 0 && busy >= shost->can_queue)
1468 if (shost->host_self_blocked)
1471 /* We're OK to process the command, so we can't be starved */
1472 if (!list_empty(&sdev->starved_entry)) {
1473 spin_lock_irq(shost->host_lock);
1474 if (!list_empty(&sdev->starved_entry))
1475 list_del_init(&sdev->starved_entry);
1476 spin_unlock_irq(shost->host_lock);
1482 spin_lock_irq(shost->host_lock);
1483 if (list_empty(&sdev->starved_entry))
1484 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1485 spin_unlock_irq(shost->host_lock);
1487 atomic_dec(&shost->host_busy);
1492 * Busy state exporting function for request stacking drivers.
1494 * For efficiency, no lock is taken to check the busy state of
1495 * shost/starget/sdev, since the returned value is not guaranteed and
1496 * may be changed after request stacking drivers call the function,
1497 * regardless of taking lock or not.
1499 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1500 * needs to return 'not busy'. Otherwise, request stacking drivers
1501 * may hold requests forever.
1503 static int scsi_lld_busy(struct request_queue *q)
1505 struct scsi_device *sdev = q->queuedata;
1506 struct Scsi_Host *shost;
1508 if (blk_queue_dying(q))
1514 * Ignore host/starget busy state.
1515 * Since block layer does not have a concept of fairness across
1516 * multiple queues, congestion of host/starget needs to be handled
1519 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1526 * Kill a request for a dead device
1528 static void scsi_kill_request(struct request *req, struct request_queue *q)
1530 struct scsi_cmnd *cmd = req->special;
1531 struct scsi_device *sdev;
1532 struct scsi_target *starget;
1533 struct Scsi_Host *shost;
1535 blk_start_request(req);
1537 scmd_printk(KERN_INFO, cmd, "killing request\n");
1540 starget = scsi_target(sdev);
1542 scsi_init_cmd_errh(cmd);
1543 cmd->result = DID_NO_CONNECT << 16;
1544 atomic_inc(&cmd->device->iorequest_cnt);
1547 * SCSI request completion path will do scsi_device_unbusy(),
1548 * bump busy counts. To bump the counters, we need to dance
1549 * with the locks as normal issue path does.
1551 atomic_inc(&sdev->device_busy);
1552 atomic_inc(&shost->host_busy);
1553 if (starget->can_queue > 0)
1554 atomic_inc(&starget->target_busy);
1556 blk_complete_request(req);
1559 static void scsi_softirq_done(struct request *rq)
1561 struct scsi_cmnd *cmd = rq->special;
1562 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1565 INIT_LIST_HEAD(&cmd->eh_entry);
1567 atomic_inc(&cmd->device->iodone_cnt);
1569 atomic_inc(&cmd->device->ioerr_cnt);
1571 disposition = scsi_decide_disposition(cmd);
1572 if (disposition != SUCCESS &&
1573 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1574 sdev_printk(KERN_ERR, cmd->device,
1575 "timing out command, waited %lus\n",
1577 disposition = SUCCESS;
1580 scsi_log_completion(cmd, disposition);
1582 switch (disposition) {
1584 scsi_finish_command(cmd);
1587 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1589 case ADD_TO_MLQUEUE:
1590 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1593 scsi_eh_scmd_add(cmd);
1599 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1600 * @cmd: command block we are dispatching.
1602 * Return: nonzero return request was rejected and device's queue needs to be
1605 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1607 struct Scsi_Host *host = cmd->device->host;
1610 atomic_inc(&cmd->device->iorequest_cnt);
1612 /* check if the device is still usable */
1613 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1614 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1615 * returns an immediate error upwards, and signals
1616 * that the device is no longer present */
1617 cmd->result = DID_NO_CONNECT << 16;
1621 /* Check to see if the scsi lld made this device blocked. */
1622 if (unlikely(scsi_device_blocked(cmd->device))) {
1624 * in blocked state, the command is just put back on
1625 * the device queue. The suspend state has already
1626 * blocked the queue so future requests should not
1627 * occur until the device transitions out of the
1630 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1631 "queuecommand : device blocked\n"));
1632 return SCSI_MLQUEUE_DEVICE_BUSY;
1635 /* Store the LUN value in cmnd, if needed. */
1636 if (cmd->device->lun_in_cdb)
1637 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1638 (cmd->device->lun << 5 & 0xe0);
1643 * Before we queue this command, check if the command
1644 * length exceeds what the host adapter can handle.
1646 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1647 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1648 "queuecommand : command too long. "
1649 "cdb_size=%d host->max_cmd_len=%d\n",
1650 cmd->cmd_len, cmd->device->host->max_cmd_len));
1651 cmd->result = (DID_ABORT << 16);
1655 if (unlikely(host->shost_state == SHOST_DEL)) {
1656 cmd->result = (DID_NO_CONNECT << 16);
1661 trace_scsi_dispatch_cmd_start(cmd);
1662 rtn = host->hostt->queuecommand(host, cmd);
1664 trace_scsi_dispatch_cmd_error(cmd, rtn);
1665 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1666 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1667 rtn = SCSI_MLQUEUE_HOST_BUSY;
1669 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1670 "queuecommand : request rejected\n"));
1675 cmd->scsi_done(cmd);
1680 * scsi_done - Invoke completion on finished SCSI command.
1681 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1682 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1684 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1685 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1686 * calls blk_complete_request() for further processing.
1688 * This function is interrupt context safe.
1690 static void scsi_done(struct scsi_cmnd *cmd)
1692 trace_scsi_dispatch_cmd_done(cmd);
1693 blk_complete_request(cmd->request);
1697 * Function: scsi_request_fn()
1699 * Purpose: Main strategy routine for SCSI.
1701 * Arguments: q - Pointer to actual queue.
1705 * Lock status: IO request lock assumed to be held when called.
1707 static void scsi_request_fn(struct request_queue *q)
1708 __releases(q->queue_lock)
1709 __acquires(q->queue_lock)
1711 struct scsi_device *sdev = q->queuedata;
1712 struct Scsi_Host *shost;
1713 struct scsi_cmnd *cmd;
1714 struct request *req;
1717 * To start with, we keep looping until the queue is empty, or until
1718 * the host is no longer able to accept any more requests.
1724 * get next queueable request. We do this early to make sure
1725 * that the request is fully prepared even if we cannot
1728 req = blk_peek_request(q);
1732 if (unlikely(!scsi_device_online(sdev))) {
1733 sdev_printk(KERN_ERR, sdev,
1734 "rejecting I/O to offline device\n");
1735 scsi_kill_request(req, q);
1739 if (!scsi_dev_queue_ready(q, sdev))
1743 * Remove the request from the request list.
1745 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1746 blk_start_request(req);
1748 spin_unlock_irq(q->queue_lock);
1750 if (unlikely(cmd == NULL)) {
1751 printk(KERN_CRIT "impossible request in %s.\n"
1752 "please mail a stack trace to "
1753 "linux-scsi@vger.kernel.org\n",
1755 blk_dump_rq_flags(req, "foo");
1760 * We hit this when the driver is using a host wide
1761 * tag map. For device level tag maps the queue_depth check
1762 * in the device ready fn would prevent us from trying
1763 * to allocate a tag. Since the map is a shared host resource
1764 * we add the dev to the starved list so it eventually gets
1765 * a run when a tag is freed.
1767 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1768 spin_lock_irq(shost->host_lock);
1769 if (list_empty(&sdev->starved_entry))
1770 list_add_tail(&sdev->starved_entry,
1771 &shost->starved_list);
1772 spin_unlock_irq(shost->host_lock);
1776 if (!scsi_target_queue_ready(shost, sdev))
1779 if (!scsi_host_queue_ready(q, shost, sdev))
1780 goto host_not_ready;
1782 if (sdev->simple_tags)
1783 cmd->flags |= SCMD_TAGGED;
1785 cmd->flags &= ~SCMD_TAGGED;
1788 * Finally, initialize any error handling parameters, and set up
1789 * the timers for timeouts.
1791 scsi_init_cmd_errh(cmd);
1794 * Dispatch the command to the low-level driver.
1796 cmd->scsi_done = scsi_done;
1797 rtn = scsi_dispatch_cmd(cmd);
1799 scsi_queue_insert(cmd, rtn);
1800 spin_lock_irq(q->queue_lock);
1803 spin_lock_irq(q->queue_lock);
1809 if (scsi_target(sdev)->can_queue > 0)
1810 atomic_dec(&scsi_target(sdev)->target_busy);
1813 * lock q, handle tag, requeue req, and decrement device_busy. We
1814 * must return with queue_lock held.
1816 * Decrementing device_busy without checking it is OK, as all such
1817 * cases (host limits or settings) should run the queue at some
1820 spin_lock_irq(q->queue_lock);
1821 blk_requeue_request(q, req);
1822 atomic_dec(&sdev->device_busy);
1824 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1825 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1828 static inline blk_status_t prep_to_mq(int ret)
1834 return BLK_STS_RESOURCE;
1836 return BLK_STS_IOERR;
1840 static int scsi_mq_prep_fn(struct request *req)
1842 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1843 struct scsi_device *sdev = req->q->queuedata;
1844 struct Scsi_Host *shost = sdev->host;
1845 unsigned char *sense_buf = cmd->sense_buffer;
1846 struct scatterlist *sg;
1848 /* zero out the cmd, except for the embedded scsi_request */
1849 memset((char *)cmd + sizeof(cmd->req), 0,
1850 sizeof(*cmd) - sizeof(cmd->req) + shost->hostt->cmd_size);
1856 cmd->sense_buffer = sense_buf;
1858 cmd->tag = req->tag;
1860 cmd->prot_op = SCSI_PROT_NORMAL;
1862 INIT_LIST_HEAD(&cmd->list);
1863 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1864 cmd->jiffies_at_alloc = jiffies;
1866 if (shost->use_cmd_list) {
1867 spin_lock_irq(&sdev->list_lock);
1868 list_add_tail(&cmd->list, &sdev->cmd_list);
1869 spin_unlock_irq(&sdev->list_lock);
1872 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1873 cmd->sdb.table.sgl = sg;
1875 if (scsi_host_get_prot(shost)) {
1876 cmd->prot_sdb = (void *)sg +
1878 shost->sg_tablesize, SG_CHUNK_SIZE) *
1879 sizeof(struct scatterlist);
1880 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1882 cmd->prot_sdb->table.sgl =
1883 (struct scatterlist *)(cmd->prot_sdb + 1);
1886 if (blk_bidi_rq(req)) {
1887 struct request *next_rq = req->next_rq;
1888 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1890 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1891 bidi_sdb->table.sgl =
1892 (struct scatterlist *)(bidi_sdb + 1);
1894 next_rq->special = bidi_sdb;
1897 blk_mq_start_request(req);
1899 return scsi_setup_cmnd(sdev, req);
1902 static void scsi_mq_done(struct scsi_cmnd *cmd)
1904 trace_scsi_dispatch_cmd_done(cmd);
1905 blk_mq_complete_request(cmd->request);
1908 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1909 const struct blk_mq_queue_data *bd)
1911 struct request *req = bd->rq;
1912 struct request_queue *q = req->q;
1913 struct scsi_device *sdev = q->queuedata;
1914 struct Scsi_Host *shost = sdev->host;
1915 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1919 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1920 if (ret != BLK_STS_OK)
1923 ret = BLK_STS_RESOURCE;
1924 if (!get_device(&sdev->sdev_gendev))
1927 if (!scsi_dev_queue_ready(q, sdev))
1928 goto out_put_device;
1929 if (!scsi_target_queue_ready(shost, sdev))
1930 goto out_dec_device_busy;
1931 if (!scsi_host_queue_ready(q, shost, sdev))
1932 goto out_dec_target_busy;
1934 if (!(req->rq_flags & RQF_DONTPREP)) {
1935 ret = prep_to_mq(scsi_mq_prep_fn(req));
1936 if (ret != BLK_STS_OK)
1937 goto out_dec_host_busy;
1938 req->rq_flags |= RQF_DONTPREP;
1940 blk_mq_start_request(req);
1943 if (sdev->simple_tags)
1944 cmd->flags |= SCMD_TAGGED;
1946 cmd->flags &= ~SCMD_TAGGED;
1948 scsi_init_cmd_errh(cmd);
1949 cmd->scsi_done = scsi_mq_done;
1951 reason = scsi_dispatch_cmd(cmd);
1953 scsi_set_blocked(cmd, reason);
1954 ret = BLK_STS_RESOURCE;
1955 goto out_dec_host_busy;
1961 atomic_dec(&shost->host_busy);
1962 out_dec_target_busy:
1963 if (scsi_target(sdev)->can_queue > 0)
1964 atomic_dec(&scsi_target(sdev)->target_busy);
1965 out_dec_device_busy:
1966 atomic_dec(&sdev->device_busy);
1968 put_device(&sdev->sdev_gendev);
1973 case BLK_STS_RESOURCE:
1974 if (atomic_read(&sdev->device_busy) == 0 &&
1975 !scsi_device_blocked(sdev))
1976 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1980 * Make sure to release all allocated ressources when
1981 * we hit an error, as we will never see this command
1984 if (req->rq_flags & RQF_DONTPREP)
1985 scsi_mq_uninit_cmd(cmd);
1991 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1995 return BLK_EH_RESET_TIMER;
1996 return scsi_times_out(req);
1999 static int scsi_init_request(struct blk_mq_tag_set *set, struct request *rq,
2000 unsigned int hctx_idx, unsigned int numa_node)
2002 struct Scsi_Host *shost = set->driver_data;
2003 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2006 scsi_alloc_sense_buffer(shost, GFP_KERNEL, numa_node);
2007 if (!cmd->sense_buffer)
2009 cmd->req.sense = cmd->sense_buffer;
2013 static void scsi_exit_request(struct blk_mq_tag_set *set, struct request *rq,
2014 unsigned int hctx_idx)
2016 struct Scsi_Host *shost = set->driver_data;
2017 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2019 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2022 static int scsi_map_queues(struct blk_mq_tag_set *set)
2024 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2026 if (shost->hostt->map_queues)
2027 return shost->hostt->map_queues(shost);
2028 return blk_mq_map_queues(set);
2031 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2033 struct device *host_dev;
2034 u64 bounce_limit = 0xffffffff;
2036 if (shost->unchecked_isa_dma)
2037 return BLK_BOUNCE_ISA;
2039 * Platforms with virtual-DMA translation
2040 * hardware have no practical limit.
2042 if (!PCI_DMA_BUS_IS_PHYS)
2043 return BLK_BOUNCE_ANY;
2045 host_dev = scsi_get_device(shost);
2046 if (host_dev && host_dev->dma_mask)
2047 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2049 return bounce_limit;
2052 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2054 struct device *dev = shost->dma_dev;
2056 queue_flag_set_unlocked(QUEUE_FLAG_SCSI_PASSTHROUGH, q);
2059 * this limit is imposed by hardware restrictions
2061 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2064 if (scsi_host_prot_dma(shost)) {
2065 shost->sg_prot_tablesize =
2066 min_not_zero(shost->sg_prot_tablesize,
2067 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2068 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2069 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2072 blk_queue_max_hw_sectors(q, shost->max_sectors);
2073 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2074 blk_queue_segment_boundary(q, shost->dma_boundary);
2075 dma_set_seg_boundary(dev, shost->dma_boundary);
2077 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2079 if (!shost->use_clustering)
2080 q->limits.cluster = 0;
2083 * set a reasonable default alignment on word boundaries: the
2084 * host and device may alter it using
2085 * blk_queue_update_dma_alignment() later.
2087 blk_queue_dma_alignment(q, 0x03);
2089 EXPORT_SYMBOL_GPL(__scsi_init_queue);
2091 static int scsi_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp)
2093 struct Scsi_Host *shost = q->rq_alloc_data;
2094 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2096 memset(cmd, 0, sizeof(*cmd));
2098 cmd->sense_buffer = scsi_alloc_sense_buffer(shost, gfp, NUMA_NO_NODE);
2099 if (!cmd->sense_buffer)
2101 cmd->req.sense = cmd->sense_buffer;
2103 if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2104 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2106 goto fail_free_sense;
2112 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2117 static void scsi_exit_rq(struct request_queue *q, struct request *rq)
2119 struct Scsi_Host *shost = q->rq_alloc_data;
2120 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2123 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2124 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2127 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2129 struct Scsi_Host *shost = sdev->host;
2130 struct request_queue *q;
2132 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE);
2135 q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2136 q->rq_alloc_data = shost;
2137 q->request_fn = scsi_request_fn;
2138 q->init_rq_fn = scsi_init_rq;
2139 q->exit_rq_fn = scsi_exit_rq;
2140 q->initialize_rq_fn = scsi_initialize_rq;
2142 if (blk_init_allocated_queue(q) < 0) {
2143 blk_cleanup_queue(q);
2147 __scsi_init_queue(shost, q);
2148 blk_queue_prep_rq(q, scsi_prep_fn);
2149 blk_queue_unprep_rq(q, scsi_unprep_fn);
2150 blk_queue_softirq_done(q, scsi_softirq_done);
2151 blk_queue_rq_timed_out(q, scsi_times_out);
2152 blk_queue_lld_busy(q, scsi_lld_busy);
2156 static const struct blk_mq_ops scsi_mq_ops = {
2157 .queue_rq = scsi_queue_rq,
2158 .complete = scsi_softirq_done,
2159 .timeout = scsi_timeout,
2160 #ifdef CONFIG_BLK_DEBUG_FS
2161 .show_rq = scsi_show_rq,
2163 .init_request = scsi_init_request,
2164 .exit_request = scsi_exit_request,
2165 .initialize_rq_fn = scsi_initialize_rq,
2166 .map_queues = scsi_map_queues,
2169 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2171 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2172 if (IS_ERR(sdev->request_queue))
2175 sdev->request_queue->queuedata = sdev;
2176 __scsi_init_queue(sdev->host, sdev->request_queue);
2177 return sdev->request_queue;
2180 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2182 unsigned int cmd_size, sgl_size, tbl_size;
2184 tbl_size = shost->sg_tablesize;
2185 if (tbl_size > SG_CHUNK_SIZE)
2186 tbl_size = SG_CHUNK_SIZE;
2187 sgl_size = tbl_size * sizeof(struct scatterlist);
2188 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2189 if (scsi_host_get_prot(shost))
2190 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2192 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2193 shost->tag_set.ops = &scsi_mq_ops;
2194 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2195 shost->tag_set.queue_depth = shost->can_queue;
2196 shost->tag_set.cmd_size = cmd_size;
2197 shost->tag_set.numa_node = NUMA_NO_NODE;
2198 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2199 shost->tag_set.flags |=
2200 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2201 shost->tag_set.driver_data = shost;
2203 return blk_mq_alloc_tag_set(&shost->tag_set);
2206 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2208 blk_mq_free_tag_set(&shost->tag_set);
2212 * scsi_device_from_queue - return sdev associated with a request_queue
2213 * @q: The request queue to return the sdev from
2215 * Return the sdev associated with a request queue or NULL if the
2216 * request_queue does not reference a SCSI device.
2218 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2220 struct scsi_device *sdev = NULL;
2223 if (q->mq_ops == &scsi_mq_ops)
2224 sdev = q->queuedata;
2225 } else if (q->request_fn == scsi_request_fn)
2226 sdev = q->queuedata;
2227 if (!sdev || !get_device(&sdev->sdev_gendev))
2232 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2235 * Function: scsi_block_requests()
2237 * Purpose: Utility function used by low-level drivers to prevent further
2238 * commands from being queued to the device.
2240 * Arguments: shost - Host in question
2244 * Lock status: No locks are assumed held.
2246 * Notes: There is no timer nor any other means by which the requests
2247 * get unblocked other than the low-level driver calling
2248 * scsi_unblock_requests().
2250 void scsi_block_requests(struct Scsi_Host *shost)
2252 shost->host_self_blocked = 1;
2254 EXPORT_SYMBOL(scsi_block_requests);
2257 * Function: scsi_unblock_requests()
2259 * Purpose: Utility function used by low-level drivers to allow further
2260 * commands from being queued to the device.
2262 * Arguments: shost - Host in question
2266 * Lock status: No locks are assumed held.
2268 * Notes: There is no timer nor any other means by which the requests
2269 * get unblocked other than the low-level driver calling
2270 * scsi_unblock_requests().
2272 * This is done as an API function so that changes to the
2273 * internals of the scsi mid-layer won't require wholesale
2274 * changes to drivers that use this feature.
2276 void scsi_unblock_requests(struct Scsi_Host *shost)
2278 shost->host_self_blocked = 0;
2279 scsi_run_host_queues(shost);
2281 EXPORT_SYMBOL(scsi_unblock_requests);
2283 int __init scsi_init_queue(void)
2285 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2286 sizeof(struct scsi_data_buffer),
2288 if (!scsi_sdb_cache) {
2289 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2296 void scsi_exit_queue(void)
2298 kmem_cache_destroy(scsi_sense_cache);
2299 kmem_cache_destroy(scsi_sense_isadma_cache);
2300 kmem_cache_destroy(scsi_sdb_cache);
2304 * scsi_mode_select - issue a mode select
2305 * @sdev: SCSI device to be queried
2306 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2307 * @sp: Save page bit (0 == don't save, 1 == save)
2308 * @modepage: mode page being requested
2309 * @buffer: request buffer (may not be smaller than eight bytes)
2310 * @len: length of request buffer.
2311 * @timeout: command timeout
2312 * @retries: number of retries before failing
2313 * @data: returns a structure abstracting the mode header data
2314 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2315 * must be SCSI_SENSE_BUFFERSIZE big.
2317 * Returns zero if successful; negative error number or scsi
2322 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2323 unsigned char *buffer, int len, int timeout, int retries,
2324 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2326 unsigned char cmd[10];
2327 unsigned char *real_buffer;
2330 memset(cmd, 0, sizeof(cmd));
2331 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2333 if (sdev->use_10_for_ms) {
2336 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2339 memcpy(real_buffer + 8, buffer, len);
2343 real_buffer[2] = data->medium_type;
2344 real_buffer[3] = data->device_specific;
2345 real_buffer[4] = data->longlba ? 0x01 : 0;
2347 real_buffer[6] = data->block_descriptor_length >> 8;
2348 real_buffer[7] = data->block_descriptor_length;
2350 cmd[0] = MODE_SELECT_10;
2354 if (len > 255 || data->block_descriptor_length > 255 ||
2358 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2361 memcpy(real_buffer + 4, buffer, len);
2364 real_buffer[1] = data->medium_type;
2365 real_buffer[2] = data->device_specific;
2366 real_buffer[3] = data->block_descriptor_length;
2369 cmd[0] = MODE_SELECT;
2373 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2374 sshdr, timeout, retries, NULL);
2378 EXPORT_SYMBOL_GPL(scsi_mode_select);
2381 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2382 * @sdev: SCSI device to be queried
2383 * @dbd: set if mode sense will allow block descriptors to be returned
2384 * @modepage: mode page being requested
2385 * @buffer: request buffer (may not be smaller than eight bytes)
2386 * @len: length of request buffer.
2387 * @timeout: command timeout
2388 * @retries: number of retries before failing
2389 * @data: returns a structure abstracting the mode header data
2390 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2391 * must be SCSI_SENSE_BUFFERSIZE big.
2393 * Returns zero if unsuccessful, or the header offset (either 4
2394 * or 8 depending on whether a six or ten byte command was
2395 * issued) if successful.
2398 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2399 unsigned char *buffer, int len, int timeout, int retries,
2400 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2402 unsigned char cmd[12];
2405 int result, retry_count = retries;
2406 struct scsi_sense_hdr my_sshdr;
2408 memset(data, 0, sizeof(*data));
2409 memset(&cmd[0], 0, 12);
2410 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2413 /* caller might not be interested in sense, but we need it */
2418 use_10_for_ms = sdev->use_10_for_ms;
2420 if (use_10_for_ms) {
2424 cmd[0] = MODE_SENSE_10;
2431 cmd[0] = MODE_SENSE;
2436 memset(buffer, 0, len);
2438 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2439 sshdr, timeout, retries, NULL);
2441 /* This code looks awful: what it's doing is making sure an
2442 * ILLEGAL REQUEST sense return identifies the actual command
2443 * byte as the problem. MODE_SENSE commands can return
2444 * ILLEGAL REQUEST if the code page isn't supported */
2446 if (use_10_for_ms && !scsi_status_is_good(result) &&
2447 (driver_byte(result) & DRIVER_SENSE)) {
2448 if (scsi_sense_valid(sshdr)) {
2449 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2450 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2452 * Invalid command operation code
2454 sdev->use_10_for_ms = 0;
2460 if(scsi_status_is_good(result)) {
2461 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2462 (modepage == 6 || modepage == 8))) {
2463 /* Initio breakage? */
2466 data->medium_type = 0;
2467 data->device_specific = 0;
2469 data->block_descriptor_length = 0;
2470 } else if(use_10_for_ms) {
2471 data->length = buffer[0]*256 + buffer[1] + 2;
2472 data->medium_type = buffer[2];
2473 data->device_specific = buffer[3];
2474 data->longlba = buffer[4] & 0x01;
2475 data->block_descriptor_length = buffer[6]*256
2478 data->length = buffer[0] + 1;
2479 data->medium_type = buffer[1];
2480 data->device_specific = buffer[2];
2481 data->block_descriptor_length = buffer[3];
2483 data->header_length = header_length;
2484 } else if ((status_byte(result) == CHECK_CONDITION) &&
2485 scsi_sense_valid(sshdr) &&
2486 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2493 EXPORT_SYMBOL(scsi_mode_sense);
2496 * scsi_test_unit_ready - test if unit is ready
2497 * @sdev: scsi device to change the state of.
2498 * @timeout: command timeout
2499 * @retries: number of retries before failing
2500 * @sshdr: outpout pointer for decoded sense information.
2502 * Returns zero if unsuccessful or an error if TUR failed. For
2503 * removable media, UNIT_ATTENTION sets ->changed flag.
2506 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2507 struct scsi_sense_hdr *sshdr)
2510 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2514 /* try to eat the UNIT_ATTENTION if there are enough retries */
2516 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2517 timeout, retries, NULL);
2518 if (sdev->removable && scsi_sense_valid(sshdr) &&
2519 sshdr->sense_key == UNIT_ATTENTION)
2521 } while (scsi_sense_valid(sshdr) &&
2522 sshdr->sense_key == UNIT_ATTENTION && --retries);
2526 EXPORT_SYMBOL(scsi_test_unit_ready);
2529 * scsi_device_set_state - Take the given device through the device state model.
2530 * @sdev: scsi device to change the state of.
2531 * @state: state to change to.
2533 * Returns zero if unsuccessful or an error if the requested
2534 * transition is illegal.
2537 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2539 enum scsi_device_state oldstate = sdev->sdev_state;
2541 if (state == oldstate)
2547 case SDEV_CREATED_BLOCK:
2558 case SDEV_TRANSPORT_OFFLINE:
2571 case SDEV_TRANSPORT_OFFLINE:
2579 case SDEV_TRANSPORT_OFFLINE:
2594 case SDEV_CREATED_BLOCK:
2601 case SDEV_CREATED_BLOCK:
2616 case SDEV_TRANSPORT_OFFLINE:
2629 case SDEV_TRANSPORT_OFFLINE:
2631 case SDEV_CREATED_BLOCK:
2639 sdev->sdev_state = state;
2643 SCSI_LOG_ERROR_RECOVERY(1,
2644 sdev_printk(KERN_ERR, sdev,
2645 "Illegal state transition %s->%s",
2646 scsi_device_state_name(oldstate),
2647 scsi_device_state_name(state))
2651 EXPORT_SYMBOL(scsi_device_set_state);
2654 * sdev_evt_emit - emit a single SCSI device uevent
2655 * @sdev: associated SCSI device
2656 * @evt: event to emit
2658 * Send a single uevent (scsi_event) to the associated scsi_device.
2660 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2665 switch (evt->evt_type) {
2666 case SDEV_EVT_MEDIA_CHANGE:
2667 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2669 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2670 scsi_rescan_device(&sdev->sdev_gendev);
2671 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2673 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2674 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2676 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2677 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2679 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2680 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2682 case SDEV_EVT_LUN_CHANGE_REPORTED:
2683 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2685 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2686 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2695 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2699 * sdev_evt_thread - send a uevent for each scsi event
2700 * @work: work struct for scsi_device
2702 * Dispatch queued events to their associated scsi_device kobjects
2705 void scsi_evt_thread(struct work_struct *work)
2707 struct scsi_device *sdev;
2708 enum scsi_device_event evt_type;
2709 LIST_HEAD(event_list);
2711 sdev = container_of(work, struct scsi_device, event_work);
2713 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2714 if (test_and_clear_bit(evt_type, sdev->pending_events))
2715 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2718 struct scsi_event *evt;
2719 struct list_head *this, *tmp;
2720 unsigned long flags;
2722 spin_lock_irqsave(&sdev->list_lock, flags);
2723 list_splice_init(&sdev->event_list, &event_list);
2724 spin_unlock_irqrestore(&sdev->list_lock, flags);
2726 if (list_empty(&event_list))
2729 list_for_each_safe(this, tmp, &event_list) {
2730 evt = list_entry(this, struct scsi_event, node);
2731 list_del(&evt->node);
2732 scsi_evt_emit(sdev, evt);
2739 * sdev_evt_send - send asserted event to uevent thread
2740 * @sdev: scsi_device event occurred on
2741 * @evt: event to send
2743 * Assert scsi device event asynchronously.
2745 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2747 unsigned long flags;
2750 /* FIXME: currently this check eliminates all media change events
2751 * for polled devices. Need to update to discriminate between AN
2752 * and polled events */
2753 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2759 spin_lock_irqsave(&sdev->list_lock, flags);
2760 list_add_tail(&evt->node, &sdev->event_list);
2761 schedule_work(&sdev->event_work);
2762 spin_unlock_irqrestore(&sdev->list_lock, flags);
2764 EXPORT_SYMBOL_GPL(sdev_evt_send);
2767 * sdev_evt_alloc - allocate a new scsi event
2768 * @evt_type: type of event to allocate
2769 * @gfpflags: GFP flags for allocation
2771 * Allocates and returns a new scsi_event.
2773 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2776 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2780 evt->evt_type = evt_type;
2781 INIT_LIST_HEAD(&evt->node);
2783 /* evt_type-specific initialization, if any */
2785 case SDEV_EVT_MEDIA_CHANGE:
2786 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2787 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2788 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2789 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2790 case SDEV_EVT_LUN_CHANGE_REPORTED:
2791 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2799 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2802 * sdev_evt_send_simple - send asserted event to uevent thread
2803 * @sdev: scsi_device event occurred on
2804 * @evt_type: type of event to send
2805 * @gfpflags: GFP flags for allocation
2807 * Assert scsi device event asynchronously, given an event type.
2809 void sdev_evt_send_simple(struct scsi_device *sdev,
2810 enum scsi_device_event evt_type, gfp_t gfpflags)
2812 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2814 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2819 sdev_evt_send(sdev, evt);
2821 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2824 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2825 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2827 static int scsi_request_fn_active(struct scsi_device *sdev)
2829 struct request_queue *q = sdev->request_queue;
2830 int request_fn_active;
2832 WARN_ON_ONCE(sdev->host->use_blk_mq);
2834 spin_lock_irq(q->queue_lock);
2835 request_fn_active = q->request_fn_active;
2836 spin_unlock_irq(q->queue_lock);
2838 return request_fn_active;
2842 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2843 * @sdev: SCSI device pointer.
2845 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2846 * invoked from scsi_request_fn() have finished.
2848 static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2850 WARN_ON_ONCE(sdev->host->use_blk_mq);
2852 while (scsi_request_fn_active(sdev))
2857 * scsi_device_quiesce - Block user issued commands.
2858 * @sdev: scsi device to quiesce.
2860 * This works by trying to transition to the SDEV_QUIESCE state
2861 * (which must be a legal transition). When the device is in this
2862 * state, only special requests will be accepted, all others will
2863 * be deferred. Since special requests may also be requeued requests,
2864 * a successful return doesn't guarantee the device will be
2865 * totally quiescent.
2867 * Must be called with user context, may sleep.
2869 * Returns zero if unsuccessful or an error if not.
2872 scsi_device_quiesce(struct scsi_device *sdev)
2874 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2878 scsi_run_queue(sdev->request_queue);
2879 while (atomic_read(&sdev->device_busy)) {
2880 msleep_interruptible(200);
2881 scsi_run_queue(sdev->request_queue);
2885 EXPORT_SYMBOL(scsi_device_quiesce);
2888 * scsi_device_resume - Restart user issued commands to a quiesced device.
2889 * @sdev: scsi device to resume.
2891 * Moves the device from quiesced back to running and restarts the
2894 * Must be called with user context, may sleep.
2896 void scsi_device_resume(struct scsi_device *sdev)
2898 /* check if the device state was mutated prior to resume, and if
2899 * so assume the state is being managed elsewhere (for example
2900 * device deleted during suspend)
2902 if (sdev->sdev_state != SDEV_QUIESCE ||
2903 scsi_device_set_state(sdev, SDEV_RUNNING))
2905 scsi_run_queue(sdev->request_queue);
2907 EXPORT_SYMBOL(scsi_device_resume);
2910 device_quiesce_fn(struct scsi_device *sdev, void *data)
2912 scsi_device_quiesce(sdev);
2916 scsi_target_quiesce(struct scsi_target *starget)
2918 starget_for_each_device(starget, NULL, device_quiesce_fn);
2920 EXPORT_SYMBOL(scsi_target_quiesce);
2923 device_resume_fn(struct scsi_device *sdev, void *data)
2925 scsi_device_resume(sdev);
2929 scsi_target_resume(struct scsi_target *starget)
2931 starget_for_each_device(starget, NULL, device_resume_fn);
2933 EXPORT_SYMBOL(scsi_target_resume);
2936 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2937 * @sdev: device to block
2938 * @wait: Whether or not to wait until ongoing .queuecommand() /
2939 * .queue_rq() calls have finished.
2941 * Block request made by scsi lld's to temporarily stop all
2942 * scsi commands on the specified device. May sleep.
2944 * Returns zero if successful or error if not
2947 * This routine transitions the device to the SDEV_BLOCK state
2948 * (which must be a legal transition). When the device is in this
2949 * state, all commands are deferred until the scsi lld reenables
2950 * the device with scsi_device_unblock or device_block_tmo fires.
2952 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2953 * scsi_internal_device_block() has blocked a SCSI device and also
2954 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2957 scsi_internal_device_block(struct scsi_device *sdev, bool wait)
2959 struct request_queue *q = sdev->request_queue;
2960 unsigned long flags;
2963 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2965 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2972 * The device has transitioned to SDEV_BLOCK. Stop the
2973 * block layer from calling the midlayer with this device's
2978 blk_mq_quiesce_queue(q);
2980 blk_mq_quiesce_queue_nowait(q);
2982 spin_lock_irqsave(q->queue_lock, flags);
2984 spin_unlock_irqrestore(q->queue_lock, flags);
2986 scsi_wait_for_queuecommand(sdev);
2991 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2994 * scsi_internal_device_unblock - resume a device after a block request
2995 * @sdev: device to resume
2996 * @new_state: state to set devices to after unblocking
2998 * Called by scsi lld's or the midlayer to restart the device queue
2999 * for the previously suspended scsi device. Called from interrupt or
3000 * normal process context.
3002 * Returns zero if successful or error if not.
3005 * This routine transitions the device to the SDEV_RUNNING state
3006 * or to one of the offline states (which must be a legal transition)
3007 * allowing the midlayer to goose the queue for this device.
3010 scsi_internal_device_unblock(struct scsi_device *sdev,
3011 enum scsi_device_state new_state)
3013 struct request_queue *q = sdev->request_queue;
3014 unsigned long flags;
3017 * Try to transition the scsi device to SDEV_RUNNING or one of the
3018 * offlined states and goose the device queue if successful.
3020 if ((sdev->sdev_state == SDEV_BLOCK) ||
3021 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3022 sdev->sdev_state = new_state;
3023 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3024 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3025 new_state == SDEV_OFFLINE)
3026 sdev->sdev_state = new_state;
3028 sdev->sdev_state = SDEV_CREATED;
3029 } else if (sdev->sdev_state != SDEV_CANCEL &&
3030 sdev->sdev_state != SDEV_OFFLINE)
3034 blk_mq_unquiesce_queue(q);
3036 spin_lock_irqsave(q->queue_lock, flags);
3038 spin_unlock_irqrestore(q->queue_lock, flags);
3043 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3046 device_block(struct scsi_device *sdev, void *data)
3048 scsi_internal_device_block(sdev, true);
3052 target_block(struct device *dev, void *data)
3054 if (scsi_is_target_device(dev))
3055 starget_for_each_device(to_scsi_target(dev), NULL,
3061 scsi_target_block(struct device *dev)
3063 if (scsi_is_target_device(dev))
3064 starget_for_each_device(to_scsi_target(dev), NULL,
3067 device_for_each_child(dev, NULL, target_block);
3069 EXPORT_SYMBOL_GPL(scsi_target_block);
3072 device_unblock(struct scsi_device *sdev, void *data)
3074 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3078 target_unblock(struct device *dev, void *data)
3080 if (scsi_is_target_device(dev))
3081 starget_for_each_device(to_scsi_target(dev), data,
3087 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3089 if (scsi_is_target_device(dev))
3090 starget_for_each_device(to_scsi_target(dev), &new_state,
3093 device_for_each_child(dev, &new_state, target_unblock);
3095 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3098 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3099 * @sgl: scatter-gather list
3100 * @sg_count: number of segments in sg
3101 * @offset: offset in bytes into sg, on return offset into the mapped area
3102 * @len: bytes to map, on return number of bytes mapped
3104 * Returns virtual address of the start of the mapped page
3106 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3107 size_t *offset, size_t *len)
3110 size_t sg_len = 0, len_complete = 0;
3111 struct scatterlist *sg;
3114 WARN_ON(!irqs_disabled());
3116 for_each_sg(sgl, sg, sg_count, i) {
3117 len_complete = sg_len; /* Complete sg-entries */
3118 sg_len += sg->length;
3119 if (sg_len > *offset)
3123 if (unlikely(i == sg_count)) {
3124 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3126 __func__, sg_len, *offset, sg_count);
3131 /* Offset starting from the beginning of first page in this sg-entry */
3132 *offset = *offset - len_complete + sg->offset;
3134 /* Assumption: contiguous pages can be accessed as "page + i" */
3135 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3136 *offset &= ~PAGE_MASK;
3138 /* Bytes in this sg-entry from *offset to the end of the page */
3139 sg_len = PAGE_SIZE - *offset;
3143 return kmap_atomic(page);
3145 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3148 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3149 * @virt: virtual address to be unmapped
3151 void scsi_kunmap_atomic_sg(void *virt)
3153 kunmap_atomic(virt);
3155 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3157 void sdev_disable_disk_events(struct scsi_device *sdev)
3159 atomic_inc(&sdev->disk_events_disable_depth);
3161 EXPORT_SYMBOL(sdev_disable_disk_events);
3163 void sdev_enable_disk_events(struct scsi_device *sdev)
3165 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3167 atomic_dec(&sdev->disk_events_disable_depth);
3169 EXPORT_SYMBOL(sdev_enable_disk_events);
3172 * scsi_vpd_lun_id - return a unique device identification
3173 * @sdev: SCSI device
3174 * @id: buffer for the identification
3175 * @id_len: length of the buffer
3177 * Copies a unique device identification into @id based
3178 * on the information in the VPD page 0x83 of the device.
3179 * The string will be formatted as a SCSI name string.
3181 * Returns the length of the identification or error on failure.
3182 * If the identifier is longer than the supplied buffer the actual
3183 * identifier length is returned and the buffer is not zero-padded.
3185 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3187 u8 cur_id_type = 0xff;
3189 unsigned char *d, *cur_id_str;
3190 unsigned char __rcu *vpd_pg83;
3191 int id_size = -EINVAL;
3194 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3201 * Look for the correct descriptor.
3202 * Order of preference for lun descriptor:
3203 * - SCSI name string
3204 * - NAA IEEE Registered Extended
3205 * - EUI-64 based 16-byte
3206 * - EUI-64 based 12-byte
3207 * - NAA IEEE Registered
3208 * - NAA IEEE Extended
3210 * as longer descriptors reduce the likelyhood
3211 * of identification clashes.
3214 /* The id string must be at least 20 bytes + terminating NULL byte */
3220 memset(id, 0, id_len);
3222 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3223 /* Skip designators not referring to the LUN */
3224 if ((d[1] & 0x30) != 0x00)
3227 switch (d[1] & 0xf) {
3230 if (cur_id_size > d[3])
3232 /* Prefer anything */
3233 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3236 if (cur_id_size + 4 > id_len)
3237 cur_id_size = id_len - 4;
3239 cur_id_type = d[1] & 0xf;
3240 id_size = snprintf(id, id_len, "t10.%*pE",
3241 cur_id_size, cur_id_str);
3245 if (cur_id_size > d[3])
3247 /* Prefer NAA IEEE Registered Extended */
3248 if (cur_id_type == 0x3 &&
3249 cur_id_size == d[3])
3253 cur_id_type = d[1] & 0xf;
3254 switch (cur_id_size) {
3256 id_size = snprintf(id, id_len,
3261 id_size = snprintf(id, id_len,
3266 id_size = snprintf(id, id_len,
3277 if (cur_id_size > d[3])
3281 cur_id_type = d[1] & 0xf;
3282 switch (cur_id_size) {
3284 id_size = snprintf(id, id_len,
3289 id_size = snprintf(id, id_len,
3299 /* SCSI name string */
3300 if (cur_id_size + 4 > d[3])
3302 /* Prefer others for truncated descriptor */
3303 if (cur_id_size && d[3] > id_len)
3305 cur_id_size = id_size = d[3];
3307 cur_id_type = d[1] & 0xf;
3308 if (cur_id_size >= id_len)
3309 cur_id_size = id_len - 1;
3310 memcpy(id, cur_id_str, cur_id_size);
3311 /* Decrease priority for truncated descriptor */
3312 if (cur_id_size != id_size)
3325 EXPORT_SYMBOL(scsi_vpd_lun_id);
3328 * scsi_vpd_tpg_id - return a target port group identifier
3329 * @sdev: SCSI device
3331 * Returns the Target Port Group identifier from the information
3332 * froom VPD page 0x83 of the device.
3334 * Returns the identifier or error on failure.
3336 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3339 unsigned char __rcu *vpd_pg83;
3340 int group_id = -EAGAIN, rel_port = -1;
3343 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3349 d = sdev->vpd_pg83 + 4;
3350 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3351 switch (d[1] & 0xf) {
3353 /* Relative target port */
3354 rel_port = get_unaligned_be16(&d[6]);
3357 /* Target port group */
3358 group_id = get_unaligned_be16(&d[6]);
3367 if (group_id >= 0 && rel_id && rel_port != -1)
3372 EXPORT_SYMBOL(scsi_vpd_tpg_id);