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_dh.h>
35 #include <trace/events/scsi.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
40 static struct kmem_cache *scsi_sdb_cache;
41 static struct kmem_cache *scsi_sense_cache;
42 static struct kmem_cache *scsi_sense_isadma_cache;
43 static DEFINE_MUTEX(scsi_sense_cache_mutex);
45 static inline struct kmem_cache *
46 scsi_select_sense_cache(struct Scsi_Host *shost)
48 return shost->unchecked_isa_dma ?
49 scsi_sense_isadma_cache : scsi_sense_cache;
52 static void scsi_free_sense_buffer(struct Scsi_Host *shost,
53 unsigned char *sense_buffer)
55 kmem_cache_free(scsi_select_sense_cache(shost), sense_buffer);
58 static unsigned char *scsi_alloc_sense_buffer(struct Scsi_Host *shost,
59 gfp_t gfp_mask, int numa_node)
61 return kmem_cache_alloc_node(scsi_select_sense_cache(shost), gfp_mask,
65 int scsi_init_sense_cache(struct Scsi_Host *shost)
67 struct kmem_cache *cache;
70 cache = scsi_select_sense_cache(shost);
74 mutex_lock(&scsi_sense_cache_mutex);
75 if (shost->unchecked_isa_dma) {
76 scsi_sense_isadma_cache =
77 kmem_cache_create("scsi_sense_cache(DMA)",
78 SCSI_SENSE_BUFFERSIZE, 0,
79 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
80 if (!scsi_sense_isadma_cache)
84 kmem_cache_create("scsi_sense_cache",
85 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN, NULL);
86 if (!scsi_sense_cache)
90 mutex_unlock(&scsi_sense_cache_mutex);
95 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
96 * not change behaviour from the previous unplug mechanism, experimentation
97 * may prove this needs changing.
99 #define SCSI_QUEUE_DELAY 3
102 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
104 struct Scsi_Host *host = cmd->device->host;
105 struct scsi_device *device = cmd->device;
106 struct scsi_target *starget = scsi_target(device);
109 * Set the appropriate busy bit for the device/host.
111 * If the host/device isn't busy, assume that something actually
112 * completed, and that we should be able to queue a command now.
114 * Note that the prior mid-layer assumption that any host could
115 * always queue at least one command is now broken. The mid-layer
116 * will implement a user specifiable stall (see
117 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
118 * if a command is requeued with no other commands outstanding
119 * either for the device or for the host.
122 case SCSI_MLQUEUE_HOST_BUSY:
123 atomic_set(&host->host_blocked, host->max_host_blocked);
125 case SCSI_MLQUEUE_DEVICE_BUSY:
126 case SCSI_MLQUEUE_EH_RETRY:
127 atomic_set(&device->device_blocked,
128 device->max_device_blocked);
130 case SCSI_MLQUEUE_TARGET_BUSY:
131 atomic_set(&starget->target_blocked,
132 starget->max_target_blocked);
137 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
139 struct scsi_device *sdev = cmd->device;
141 blk_mq_requeue_request(cmd->request, true);
142 put_device(&sdev->sdev_gendev);
146 * __scsi_queue_insert - private queue insertion
147 * @cmd: The SCSI command being requeued
148 * @reason: The reason for the requeue
149 * @unbusy: Whether the queue should be unbusied
151 * This is a private queue insertion. The public interface
152 * scsi_queue_insert() always assumes the queue should be unbusied
153 * because it's always called before the completion. This function is
154 * for a requeue after completion, which should only occur in this
157 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
159 struct scsi_device *device = cmd->device;
160 struct request_queue *q = device->request_queue;
163 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
164 "Inserting command %p into mlqueue\n", cmd));
166 scsi_set_blocked(cmd, reason);
169 * Decrement the counters, since these commands are no longer
170 * active on the host/device.
173 scsi_device_unbusy(device);
176 * Requeue this command. It will go before all other commands
177 * that are already in the queue. Schedule requeue work under
178 * lock such that the kblockd_schedule_work() call happens
179 * before blk_cleanup_queue() finishes.
183 scsi_mq_requeue_cmd(cmd);
186 spin_lock_irqsave(q->queue_lock, flags);
187 blk_requeue_request(q, cmd->request);
188 kblockd_schedule_work(&device->requeue_work);
189 spin_unlock_irqrestore(q->queue_lock, flags);
193 * Function: scsi_queue_insert()
195 * Purpose: Insert a command in the midlevel queue.
197 * Arguments: cmd - command that we are adding to queue.
198 * reason - why we are inserting command to queue.
200 * Lock status: Assumed that lock is not held upon entry.
204 * Notes: We do this for one of two cases. Either the host is busy
205 * and it cannot accept any more commands for the time being,
206 * or the device returned QUEUE_FULL and can accept no more
208 * Notes: This could be called either from an interrupt context or a
209 * normal process context.
211 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
213 __scsi_queue_insert(cmd, reason, 1);
218 * scsi_execute - insert request and wait for the result
221 * @data_direction: data direction
222 * @buffer: data buffer
223 * @bufflen: len of buffer
224 * @sense: optional sense buffer
225 * @sshdr: optional decoded sense header
226 * @timeout: request timeout in seconds
227 * @retries: number of times to retry request
228 * @flags: flags for ->cmd_flags
229 * @rq_flags: flags for ->rq_flags
230 * @resid: optional residual length
232 * returns the req->errors value which is the scsi_cmnd result
235 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
236 int data_direction, void *buffer, unsigned bufflen,
237 unsigned char *sense, struct scsi_sense_hdr *sshdr,
238 int timeout, int retries, u64 flags, req_flags_t rq_flags,
242 struct scsi_request *rq;
243 int ret = DRIVER_ERROR << 24;
245 req = blk_get_request(sdev->request_queue,
246 data_direction == DMA_TO_DEVICE ?
247 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, __GFP_RECLAIM);
253 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
254 buffer, bufflen, __GFP_RECLAIM))
257 rq->cmd_len = COMMAND_SIZE(cmd[0]);
258 memcpy(rq->cmd, cmd, rq->cmd_len);
259 req->retries = retries;
260 req->timeout = timeout;
261 req->cmd_flags |= flags;
262 req->rq_flags |= rq_flags | RQF_QUIET | RQF_PREEMPT;
265 * head injection *required* here otherwise quiesce won't work
267 blk_execute_rq(req->q, NULL, req, 1);
270 * Some devices (USB mass-storage in particular) may transfer
271 * garbage data together with a residue indicating that the data
272 * is invalid. Prevent the garbage from being misinterpreted
273 * and prevent security leaks by zeroing out the excess data.
275 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
276 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
279 *resid = rq->resid_len;
280 if (sense && rq->sense_len)
281 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
283 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
286 blk_put_request(req);
290 EXPORT_SYMBOL(scsi_execute);
293 * Function: scsi_init_cmd_errh()
295 * Purpose: Initialize cmd fields related to error handling.
297 * Arguments: cmd - command that is ready to be queued.
299 * Notes: This function has the job of initializing a number of
300 * fields related to error handling. Typically this will
301 * be called once for each command, as required.
303 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
305 cmd->serial_number = 0;
306 scsi_set_resid(cmd, 0);
307 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
308 if (cmd->cmd_len == 0)
309 cmd->cmd_len = scsi_command_size(cmd->cmnd);
312 void scsi_device_unbusy(struct scsi_device *sdev)
314 struct Scsi_Host *shost = sdev->host;
315 struct scsi_target *starget = scsi_target(sdev);
318 atomic_dec(&shost->host_busy);
319 if (starget->can_queue > 0)
320 atomic_dec(&starget->target_busy);
322 if (unlikely(scsi_host_in_recovery(shost) &&
323 (shost->host_failed || shost->host_eh_scheduled))) {
324 spin_lock_irqsave(shost->host_lock, flags);
325 scsi_eh_wakeup(shost);
326 spin_unlock_irqrestore(shost->host_lock, flags);
329 atomic_dec(&sdev->device_busy);
332 static void scsi_kick_queue(struct request_queue *q)
335 blk_mq_start_hw_queues(q);
341 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
342 * and call blk_run_queue for all the scsi_devices on the target -
343 * including current_sdev first.
345 * Called with *no* scsi locks held.
347 static void scsi_single_lun_run(struct scsi_device *current_sdev)
349 struct Scsi_Host *shost = current_sdev->host;
350 struct scsi_device *sdev, *tmp;
351 struct scsi_target *starget = scsi_target(current_sdev);
354 spin_lock_irqsave(shost->host_lock, flags);
355 starget->starget_sdev_user = NULL;
356 spin_unlock_irqrestore(shost->host_lock, flags);
359 * Call blk_run_queue for all LUNs on the target, starting with
360 * current_sdev. We race with others (to set starget_sdev_user),
361 * but in most cases, we will be first. Ideally, each LU on the
362 * target would get some limited time or requests on the target.
364 scsi_kick_queue(current_sdev->request_queue);
366 spin_lock_irqsave(shost->host_lock, flags);
367 if (starget->starget_sdev_user)
369 list_for_each_entry_safe(sdev, tmp, &starget->devices,
370 same_target_siblings) {
371 if (sdev == current_sdev)
373 if (scsi_device_get(sdev))
376 spin_unlock_irqrestore(shost->host_lock, flags);
377 scsi_kick_queue(sdev->request_queue);
378 spin_lock_irqsave(shost->host_lock, flags);
380 scsi_device_put(sdev);
383 spin_unlock_irqrestore(shost->host_lock, flags);
386 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
388 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
390 if (atomic_read(&sdev->device_blocked) > 0)
395 static inline bool scsi_target_is_busy(struct scsi_target *starget)
397 if (starget->can_queue > 0) {
398 if (atomic_read(&starget->target_busy) >= starget->can_queue)
400 if (atomic_read(&starget->target_blocked) > 0)
406 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
408 if (shost->can_queue > 0 &&
409 atomic_read(&shost->host_busy) >= shost->can_queue)
411 if (atomic_read(&shost->host_blocked) > 0)
413 if (shost->host_self_blocked)
418 static void scsi_starved_list_run(struct Scsi_Host *shost)
420 LIST_HEAD(starved_list);
421 struct scsi_device *sdev;
424 spin_lock_irqsave(shost->host_lock, flags);
425 list_splice_init(&shost->starved_list, &starved_list);
427 while (!list_empty(&starved_list)) {
428 struct request_queue *slq;
431 * As long as shost is accepting commands and we have
432 * starved queues, call blk_run_queue. scsi_request_fn
433 * drops the queue_lock and can add us back to the
436 * host_lock protects the starved_list and starved_entry.
437 * scsi_request_fn must get the host_lock before checking
438 * or modifying starved_list or starved_entry.
440 if (scsi_host_is_busy(shost))
443 sdev = list_entry(starved_list.next,
444 struct scsi_device, starved_entry);
445 list_del_init(&sdev->starved_entry);
446 if (scsi_target_is_busy(scsi_target(sdev))) {
447 list_move_tail(&sdev->starved_entry,
448 &shost->starved_list);
453 * Once we drop the host lock, a racing scsi_remove_device()
454 * call may remove the sdev from the starved list and destroy
455 * it and the queue. Mitigate by taking a reference to the
456 * queue and never touching the sdev again after we drop the
457 * host lock. Note: if __scsi_remove_device() invokes
458 * blk_cleanup_queue() before the queue is run from this
459 * function then blk_run_queue() will return immediately since
460 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
462 slq = sdev->request_queue;
463 if (!blk_get_queue(slq))
465 spin_unlock_irqrestore(shost->host_lock, flags);
467 scsi_kick_queue(slq);
470 spin_lock_irqsave(shost->host_lock, flags);
472 /* put any unprocessed entries back */
473 list_splice(&starved_list, &shost->starved_list);
474 spin_unlock_irqrestore(shost->host_lock, flags);
478 * Function: scsi_run_queue()
480 * Purpose: Select a proper request queue to serve next
482 * Arguments: q - last request's queue
486 * Notes: The previous command was completely finished, start
487 * a new one if possible.
489 static void scsi_run_queue(struct request_queue *q)
491 struct scsi_device *sdev = q->queuedata;
493 if (scsi_target(sdev)->single_lun)
494 scsi_single_lun_run(sdev);
495 if (!list_empty(&sdev->host->starved_list))
496 scsi_starved_list_run(sdev->host);
499 blk_mq_start_stopped_hw_queues(q, false);
504 void scsi_requeue_run_queue(struct work_struct *work)
506 struct scsi_device *sdev;
507 struct request_queue *q;
509 sdev = container_of(work, struct scsi_device, requeue_work);
510 q = sdev->request_queue;
515 * Function: scsi_requeue_command()
517 * Purpose: Handle post-processing of completed commands.
519 * Arguments: q - queue to operate on
520 * cmd - command that may need to be requeued.
524 * Notes: After command completion, there may be blocks left
525 * over which weren't finished by the previous command
526 * this can be for a number of reasons - the main one is
527 * I/O errors in the middle of the request, in which case
528 * we need to request the blocks that come after the bad
530 * Notes: Upon return, cmd is a stale pointer.
532 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
534 struct scsi_device *sdev = cmd->device;
535 struct request *req = cmd->request;
538 spin_lock_irqsave(q->queue_lock, flags);
539 blk_unprep_request(req);
541 scsi_put_command(cmd);
542 blk_requeue_request(q, req);
543 spin_unlock_irqrestore(q->queue_lock, flags);
547 put_device(&sdev->sdev_gendev);
550 void scsi_run_host_queues(struct Scsi_Host *shost)
552 struct scsi_device *sdev;
554 shost_for_each_device(sdev, shost)
555 scsi_run_queue(sdev->request_queue);
558 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
560 if (!blk_rq_is_passthrough(cmd->request)) {
561 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
563 if (drv->uninit_command)
564 drv->uninit_command(cmd);
568 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
570 struct scsi_data_buffer *sdb;
572 if (cmd->sdb.table.nents)
573 sg_free_table_chained(&cmd->sdb.table, true);
574 if (cmd->request->next_rq) {
575 sdb = cmd->request->next_rq->special;
577 sg_free_table_chained(&sdb->table, true);
579 if (scsi_prot_sg_count(cmd))
580 sg_free_table_chained(&cmd->prot_sdb->table, true);
583 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
585 struct scsi_device *sdev = cmd->device;
586 struct Scsi_Host *shost = sdev->host;
589 scsi_mq_free_sgtables(cmd);
590 scsi_uninit_cmd(cmd);
592 if (shost->use_cmd_list) {
593 BUG_ON(list_empty(&cmd->list));
594 spin_lock_irqsave(&sdev->list_lock, flags);
595 list_del_init(&cmd->list);
596 spin_unlock_irqrestore(&sdev->list_lock, flags);
601 * Function: scsi_release_buffers()
603 * Purpose: Free resources allocate for a scsi_command.
605 * Arguments: cmd - command that we are bailing.
607 * Lock status: Assumed that no lock is held upon entry.
611 * Notes: In the event that an upper level driver rejects a
612 * command, we must release resources allocated during
613 * the __init_io() function. Primarily this would involve
614 * the scatter-gather table.
616 static void scsi_release_buffers(struct scsi_cmnd *cmd)
618 if (cmd->sdb.table.nents)
619 sg_free_table_chained(&cmd->sdb.table, false);
621 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
623 if (scsi_prot_sg_count(cmd))
624 sg_free_table_chained(&cmd->prot_sdb->table, false);
627 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
629 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
631 sg_free_table_chained(&bidi_sdb->table, false);
632 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
633 cmd->request->next_rq->special = NULL;
636 static bool scsi_end_request(struct request *req, int error,
637 unsigned int bytes, unsigned int bidi_bytes)
639 struct scsi_cmnd *cmd = req->special;
640 struct scsi_device *sdev = cmd->device;
641 struct request_queue *q = sdev->request_queue;
643 if (blk_update_request(req, error, bytes))
646 /* Bidi request must be completed as a whole */
647 if (unlikely(bidi_bytes) &&
648 blk_update_request(req->next_rq, error, bidi_bytes))
651 if (blk_queue_add_random(q))
652 add_disk_randomness(req->rq_disk);
656 * In the MQ case the command gets freed by __blk_mq_end_request,
657 * so we have to do all cleanup that depends on it earlier.
659 * We also can't kick the queues from irq context, so we
660 * will have to defer it to a workqueue.
662 scsi_mq_uninit_cmd(cmd);
664 __blk_mq_end_request(req, error);
666 if (scsi_target(sdev)->single_lun ||
667 !list_empty(&sdev->host->starved_list))
668 kblockd_schedule_work(&sdev->requeue_work);
670 blk_mq_start_stopped_hw_queues(q, true);
675 scsi_release_bidi_buffers(cmd);
676 scsi_release_buffers(cmd);
677 scsi_put_command(cmd);
679 spin_lock_irqsave(q->queue_lock, flags);
680 blk_finish_request(req, error);
681 spin_unlock_irqrestore(q->queue_lock, flags);
686 put_device(&sdev->sdev_gendev);
691 * __scsi_error_from_host_byte - translate SCSI error code into errno
692 * @cmd: SCSI command (unused)
693 * @result: scsi error code
695 * Translate SCSI error code into standard UNIX errno.
697 * -ENOLINK temporary transport failure
698 * -EREMOTEIO permanent target failure, do not retry
699 * -EBADE permanent nexus failure, retry on other path
700 * -ENOSPC No write space available
701 * -ENODATA Medium error
702 * -EIO unspecified I/O error
704 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
708 switch(host_byte(result)) {
709 case DID_TRANSPORT_FAILFAST:
712 case DID_TARGET_FAILURE:
713 set_host_byte(cmd, DID_OK);
716 case DID_NEXUS_FAILURE:
717 set_host_byte(cmd, DID_OK);
720 case DID_ALLOC_FAILURE:
721 set_host_byte(cmd, DID_OK);
724 case DID_MEDIUM_ERROR:
725 set_host_byte(cmd, DID_OK);
737 * Function: scsi_io_completion()
739 * Purpose: Completion processing for block device I/O requests.
741 * Arguments: cmd - command that is finished.
743 * Lock status: Assumed that no lock is held upon entry.
747 * Notes: We will finish off the specified number of sectors. If we
748 * are done, the command block will be released and the queue
749 * function will be goosed. If we are not done then we have to
750 * figure out what to do next:
752 * a) We can call scsi_requeue_command(). The request
753 * will be unprepared and put back on the queue. Then
754 * a new command will be created for it. This should
755 * be used if we made forward progress, or if we want
756 * to switch from READ(10) to READ(6) for example.
758 * b) We can call __scsi_queue_insert(). The request will
759 * be put back on the queue and retried using the same
760 * command as before, possibly after a delay.
762 * c) We can call scsi_end_request() with -EIO to fail
763 * the remainder of the request.
765 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
767 int result = cmd->result;
768 struct request_queue *q = cmd->device->request_queue;
769 struct request *req = cmd->request;
771 struct scsi_sense_hdr sshdr;
772 bool sense_valid = false;
773 int sense_deferred = 0, level = 0;
774 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
775 ACTION_DELAYED_RETRY} action;
776 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
779 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
781 sense_deferred = scsi_sense_is_deferred(&sshdr);
784 if (blk_rq_is_passthrough(req)) {
788 * SG_IO wants current and deferred errors
790 scsi_req(req)->sense_len =
791 min(8 + cmd->sense_buffer[7],
792 SCSI_SENSE_BUFFERSIZE);
795 error = __scsi_error_from_host_byte(cmd, result);
798 * __scsi_error_from_host_byte may have reset the host_byte
800 req->errors = cmd->result;
802 scsi_req(req)->resid_len = scsi_get_resid(cmd);
804 if (scsi_bidi_cmnd(cmd)) {
806 * Bidi commands Must be complete as a whole,
807 * both sides at once.
809 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
810 if (scsi_end_request(req, 0, blk_rq_bytes(req),
811 blk_rq_bytes(req->next_rq)))
815 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
817 * Flush commands do not transfers any data, and thus cannot use
818 * good_bytes != blk_rq_bytes(req) as the signal for an error.
819 * This sets the error explicitly for the problem case.
821 error = __scsi_error_from_host_byte(cmd, result);
824 /* no bidi support for !blk_rq_is_passthrough yet */
825 BUG_ON(blk_bidi_rq(req));
828 * Next deal with any sectors which we were able to correctly
831 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
832 "%u sectors total, %d bytes done.\n",
833 blk_rq_sectors(req), good_bytes));
836 * Recovered errors need reporting, but they're always treated as
837 * success, so fiddle the result code here. For passthrough requests
838 * we already took a copy of the original into rq->errors which
839 * is what gets returned to the user
841 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
842 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
843 * print since caller wants ATA registers. Only occurs on
844 * SCSI ATA PASS_THROUGH commands when CK_COND=1
846 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
848 else if (!(req->rq_flags & RQF_QUIET))
849 scsi_print_sense(cmd);
851 /* for passthrough error may be set */
856 * special case: failed zero length commands always need to
857 * drop down into the retry code. Otherwise, if we finished
858 * all bytes in the request we are done now.
860 if (!(blk_rq_bytes(req) == 0 && error) &&
861 !scsi_end_request(req, error, good_bytes, 0))
865 * Kill remainder if no retrys.
867 if (error && scsi_noretry_cmd(cmd)) {
868 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
874 * If there had been no error, but we have leftover bytes in the
875 * requeues just queue the command up again.
880 error = __scsi_error_from_host_byte(cmd, result);
882 if (host_byte(result) == DID_RESET) {
883 /* Third party bus reset or reset for error recovery
884 * reasons. Just retry the command and see what
887 action = ACTION_RETRY;
888 } else if (sense_valid && !sense_deferred) {
889 switch (sshdr.sense_key) {
891 if (cmd->device->removable) {
892 /* Detected disc change. Set a bit
893 * and quietly refuse further access.
895 cmd->device->changed = 1;
896 action = ACTION_FAIL;
898 /* Must have been a power glitch, or a
899 * bus reset. Could not have been a
900 * media change, so we just retry the
901 * command and see what happens.
903 action = ACTION_RETRY;
906 case ILLEGAL_REQUEST:
907 /* If we had an ILLEGAL REQUEST returned, then
908 * we may have performed an unsupported
909 * command. The only thing this should be
910 * would be a ten byte read where only a six
911 * byte read was supported. Also, on a system
912 * where READ CAPACITY failed, we may have
913 * read past the end of the disk.
915 if ((cmd->device->use_10_for_rw &&
916 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
917 (cmd->cmnd[0] == READ_10 ||
918 cmd->cmnd[0] == WRITE_10)) {
919 /* This will issue a new 6-byte command. */
920 cmd->device->use_10_for_rw = 0;
921 action = ACTION_REPREP;
922 } else if (sshdr.asc == 0x10) /* DIX */ {
923 action = ACTION_FAIL;
925 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
926 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
927 action = ACTION_FAIL;
930 action = ACTION_FAIL;
932 case ABORTED_COMMAND:
933 action = ACTION_FAIL;
934 if (sshdr.asc == 0x10) /* DIF */
938 /* If the device is in the process of becoming
939 * ready, or has a temporary blockage, retry.
941 if (sshdr.asc == 0x04) {
942 switch (sshdr.ascq) {
943 case 0x01: /* becoming ready */
944 case 0x04: /* format in progress */
945 case 0x05: /* rebuild in progress */
946 case 0x06: /* recalculation in progress */
947 case 0x07: /* operation in progress */
948 case 0x08: /* Long write in progress */
949 case 0x09: /* self test in progress */
950 case 0x14: /* space allocation in progress */
951 action = ACTION_DELAYED_RETRY;
954 action = ACTION_FAIL;
958 action = ACTION_FAIL;
960 case VOLUME_OVERFLOW:
961 /* See SSC3rXX or current. */
962 action = ACTION_FAIL;
965 action = ACTION_FAIL;
969 action = ACTION_FAIL;
971 if (action != ACTION_FAIL &&
972 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
973 action = ACTION_FAIL;
977 /* Give up and fail the remainder of the request */
978 if (!(req->rq_flags & RQF_QUIET)) {
979 static DEFINE_RATELIMIT_STATE(_rs,
980 DEFAULT_RATELIMIT_INTERVAL,
981 DEFAULT_RATELIMIT_BURST);
983 if (unlikely(scsi_logging_level))
984 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
985 SCSI_LOG_MLCOMPLETE_BITS);
988 * if logging is enabled the failure will be printed
989 * in scsi_log_completion(), so avoid duplicate messages
991 if (!level && __ratelimit(&_rs)) {
992 scsi_print_result(cmd, NULL, FAILED);
993 if (driver_byte(result) & DRIVER_SENSE)
994 scsi_print_sense(cmd);
995 scsi_print_command(cmd);
998 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1003 /* Unprep the request and put it back at the head of the queue.
1004 * A new command will be prepared and issued.
1007 cmd->request->rq_flags &= ~RQF_DONTPREP;
1008 scsi_mq_uninit_cmd(cmd);
1009 scsi_mq_requeue_cmd(cmd);
1011 scsi_release_buffers(cmd);
1012 scsi_requeue_command(q, cmd);
1016 /* Retry the same command immediately */
1017 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1019 case ACTION_DELAYED_RETRY:
1020 /* Retry the same command after a delay */
1021 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1026 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1031 * If sg table allocation fails, requeue request later.
1033 if (unlikely(sg_alloc_table_chained(&sdb->table,
1034 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1035 return BLKPREP_DEFER;
1038 * Next, walk the list, and fill in the addresses and sizes of
1041 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1042 BUG_ON(count > sdb->table.nents);
1043 sdb->table.nents = count;
1044 sdb->length = blk_rq_payload_bytes(req);
1049 * Function: scsi_init_io()
1051 * Purpose: SCSI I/O initialize function.
1053 * Arguments: cmd - Command descriptor we wish to initialize
1055 * Returns: 0 on success
1056 * BLKPREP_DEFER if the failure is retryable
1057 * BLKPREP_KILL if the failure is fatal
1059 int scsi_init_io(struct scsi_cmnd *cmd)
1061 struct scsi_device *sdev = cmd->device;
1062 struct request *rq = cmd->request;
1063 bool is_mq = (rq->mq_ctx != NULL);
1066 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1069 error = scsi_init_sgtable(rq, &cmd->sdb);
1073 if (blk_bidi_rq(rq)) {
1074 if (!rq->q->mq_ops) {
1075 struct scsi_data_buffer *bidi_sdb =
1076 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1078 error = BLKPREP_DEFER;
1082 rq->next_rq->special = bidi_sdb;
1085 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1090 if (blk_integrity_rq(rq)) {
1091 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1094 if (prot_sdb == NULL) {
1096 * This can happen if someone (e.g. multipath)
1097 * queues a command to a device on an adapter
1098 * that does not support DIX.
1101 error = BLKPREP_KILL;
1105 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1107 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1108 prot_sdb->table.sgl)) {
1109 error = BLKPREP_DEFER;
1113 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1114 prot_sdb->table.sgl);
1115 BUG_ON(unlikely(count > ivecs));
1116 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1118 cmd->prot_sdb = prot_sdb;
1119 cmd->prot_sdb->table.nents = count;
1125 scsi_mq_free_sgtables(cmd);
1127 scsi_release_buffers(cmd);
1128 cmd->request->special = NULL;
1129 scsi_put_command(cmd);
1130 put_device(&sdev->sdev_gendev);
1134 EXPORT_SYMBOL(scsi_init_io);
1136 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1138 void *buf = cmd->sense_buffer;
1139 void *prot = cmd->prot_sdb;
1140 unsigned long flags;
1142 /* zero out the cmd, except for the embedded scsi_request */
1143 memset((char *)cmd + sizeof(cmd->req), 0,
1144 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1147 cmd->sense_buffer = buf;
1148 cmd->prot_sdb = prot;
1149 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1150 cmd->jiffies_at_alloc = jiffies;
1152 spin_lock_irqsave(&dev->list_lock, flags);
1153 list_add_tail(&cmd->list, &dev->cmd_list);
1154 spin_unlock_irqrestore(&dev->list_lock, flags);
1157 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1159 struct scsi_cmnd *cmd = req->special;
1162 * Passthrough requests may transfer data, in which case they must
1163 * a bio attached to them. Or they might contain a SCSI command
1164 * that does not transfer data, in which case they may optionally
1165 * submit a request without an attached bio.
1168 int ret = scsi_init_io(cmd);
1172 BUG_ON(blk_rq_bytes(req));
1174 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1177 cmd->cmd_len = scsi_req(req)->cmd_len;
1178 cmd->cmnd = scsi_req(req)->cmd;
1179 cmd->transfersize = blk_rq_bytes(req);
1180 cmd->allowed = req->retries;
1185 * Setup a normal block command. These are simple request from filesystems
1186 * that still need to be translated to SCSI CDBs from the ULD.
1188 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1190 struct scsi_cmnd *cmd = req->special;
1192 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1193 int ret = sdev->handler->prep_fn(sdev, req);
1194 if (ret != BLKPREP_OK)
1198 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1199 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1200 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1203 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1205 struct scsi_cmnd *cmd = req->special;
1207 if (!blk_rq_bytes(req))
1208 cmd->sc_data_direction = DMA_NONE;
1209 else if (rq_data_dir(req) == WRITE)
1210 cmd->sc_data_direction = DMA_TO_DEVICE;
1212 cmd->sc_data_direction = DMA_FROM_DEVICE;
1214 if (blk_rq_is_scsi(req))
1215 return scsi_setup_scsi_cmnd(sdev, req);
1217 return scsi_setup_fs_cmnd(sdev, req);
1221 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1223 int ret = BLKPREP_OK;
1226 * If the device is not in running state we will reject some
1229 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1230 switch (sdev->sdev_state) {
1232 case SDEV_TRANSPORT_OFFLINE:
1234 * If the device is offline we refuse to process any
1235 * commands. The device must be brought online
1236 * before trying any recovery commands.
1238 sdev_printk(KERN_ERR, sdev,
1239 "rejecting I/O to offline device\n");
1244 * If the device is fully deleted, we refuse to
1245 * process any commands as well.
1247 sdev_printk(KERN_ERR, sdev,
1248 "rejecting I/O to dead device\n");
1252 case SDEV_CREATED_BLOCK:
1253 ret = BLKPREP_DEFER;
1257 * If the devices is blocked we defer normal commands.
1259 if (!(req->rq_flags & RQF_PREEMPT))
1260 ret = BLKPREP_DEFER;
1264 * For any other not fully online state we only allow
1265 * special commands. In particular any user initiated
1266 * command is not allowed.
1268 if (!(req->rq_flags & RQF_PREEMPT))
1277 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1279 struct scsi_device *sdev = q->queuedata;
1283 case BLKPREP_INVALID:
1284 req->errors = DID_NO_CONNECT << 16;
1285 /* release the command and kill it */
1287 struct scsi_cmnd *cmd = req->special;
1288 scsi_release_buffers(cmd);
1289 scsi_put_command(cmd);
1290 put_device(&sdev->sdev_gendev);
1291 req->special = NULL;
1296 * If we defer, the blk_peek_request() returns NULL, but the
1297 * queue must be restarted, so we schedule a callback to happen
1300 if (atomic_read(&sdev->device_busy) == 0)
1301 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1304 req->rq_flags |= RQF_DONTPREP;
1310 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1312 struct scsi_device *sdev = q->queuedata;
1313 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1316 ret = scsi_prep_state_check(sdev, req);
1317 if (ret != BLKPREP_OK)
1320 if (!req->special) {
1321 /* Bail if we can't get a reference to the device */
1322 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1323 ret = BLKPREP_DEFER;
1327 scsi_init_command(sdev, cmd);
1331 cmd->tag = req->tag;
1333 cmd->prot_op = SCSI_PROT_NORMAL;
1335 ret = scsi_setup_cmnd(sdev, req);
1337 return scsi_prep_return(q, req, ret);
1340 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1342 scsi_uninit_cmd(req->special);
1346 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1349 * Called with the queue_lock held.
1351 static inline int scsi_dev_queue_ready(struct request_queue *q,
1352 struct scsi_device *sdev)
1356 busy = atomic_inc_return(&sdev->device_busy) - 1;
1357 if (atomic_read(&sdev->device_blocked)) {
1362 * unblock after device_blocked iterates to zero
1364 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1366 * For the MQ case we take care of this in the caller.
1369 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1372 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1373 "unblocking device at zero depth\n"));
1376 if (busy >= sdev->queue_depth)
1381 atomic_dec(&sdev->device_busy);
1386 * scsi_target_queue_ready: checks if there we can send commands to target
1387 * @sdev: scsi device on starget to check.
1389 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1390 struct scsi_device *sdev)
1392 struct scsi_target *starget = scsi_target(sdev);
1395 if (starget->single_lun) {
1396 spin_lock_irq(shost->host_lock);
1397 if (starget->starget_sdev_user &&
1398 starget->starget_sdev_user != sdev) {
1399 spin_unlock_irq(shost->host_lock);
1402 starget->starget_sdev_user = sdev;
1403 spin_unlock_irq(shost->host_lock);
1406 if (starget->can_queue <= 0)
1409 busy = atomic_inc_return(&starget->target_busy) - 1;
1410 if (atomic_read(&starget->target_blocked) > 0) {
1415 * unblock after target_blocked iterates to zero
1417 if (atomic_dec_return(&starget->target_blocked) > 0)
1420 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1421 "unblocking target at zero depth\n"));
1424 if (busy >= starget->can_queue)
1430 spin_lock_irq(shost->host_lock);
1431 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1432 spin_unlock_irq(shost->host_lock);
1434 if (starget->can_queue > 0)
1435 atomic_dec(&starget->target_busy);
1440 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1441 * return 0. We must end up running the queue again whenever 0 is
1442 * returned, else IO can hang.
1444 static inline int scsi_host_queue_ready(struct request_queue *q,
1445 struct Scsi_Host *shost,
1446 struct scsi_device *sdev)
1450 if (scsi_host_in_recovery(shost))
1453 busy = atomic_inc_return(&shost->host_busy) - 1;
1454 if (atomic_read(&shost->host_blocked) > 0) {
1459 * unblock after host_blocked iterates to zero
1461 if (atomic_dec_return(&shost->host_blocked) > 0)
1465 shost_printk(KERN_INFO, shost,
1466 "unblocking host at zero depth\n"));
1469 if (shost->can_queue > 0 && busy >= shost->can_queue)
1471 if (shost->host_self_blocked)
1474 /* We're OK to process the command, so we can't be starved */
1475 if (!list_empty(&sdev->starved_entry)) {
1476 spin_lock_irq(shost->host_lock);
1477 if (!list_empty(&sdev->starved_entry))
1478 list_del_init(&sdev->starved_entry);
1479 spin_unlock_irq(shost->host_lock);
1485 spin_lock_irq(shost->host_lock);
1486 if (list_empty(&sdev->starved_entry))
1487 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1488 spin_unlock_irq(shost->host_lock);
1490 atomic_dec(&shost->host_busy);
1495 * Busy state exporting function for request stacking drivers.
1497 * For efficiency, no lock is taken to check the busy state of
1498 * shost/starget/sdev, since the returned value is not guaranteed and
1499 * may be changed after request stacking drivers call the function,
1500 * regardless of taking lock or not.
1502 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1503 * needs to return 'not busy'. Otherwise, request stacking drivers
1504 * may hold requests forever.
1506 static int scsi_lld_busy(struct request_queue *q)
1508 struct scsi_device *sdev = q->queuedata;
1509 struct Scsi_Host *shost;
1511 if (blk_queue_dying(q))
1517 * Ignore host/starget busy state.
1518 * Since block layer does not have a concept of fairness across
1519 * multiple queues, congestion of host/starget needs to be handled
1522 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1529 * Kill a request for a dead device
1531 static void scsi_kill_request(struct request *req, struct request_queue *q)
1533 struct scsi_cmnd *cmd = req->special;
1534 struct scsi_device *sdev;
1535 struct scsi_target *starget;
1536 struct Scsi_Host *shost;
1538 blk_start_request(req);
1540 scmd_printk(KERN_INFO, cmd, "killing request\n");
1543 starget = scsi_target(sdev);
1545 scsi_init_cmd_errh(cmd);
1546 cmd->result = DID_NO_CONNECT << 16;
1547 atomic_inc(&cmd->device->iorequest_cnt);
1550 * SCSI request completion path will do scsi_device_unbusy(),
1551 * bump busy counts. To bump the counters, we need to dance
1552 * with the locks as normal issue path does.
1554 atomic_inc(&sdev->device_busy);
1555 atomic_inc(&shost->host_busy);
1556 if (starget->can_queue > 0)
1557 atomic_inc(&starget->target_busy);
1559 blk_complete_request(req);
1562 static void scsi_softirq_done(struct request *rq)
1564 struct scsi_cmnd *cmd = rq->special;
1565 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1568 INIT_LIST_HEAD(&cmd->eh_entry);
1570 atomic_inc(&cmd->device->iodone_cnt);
1572 atomic_inc(&cmd->device->ioerr_cnt);
1574 disposition = scsi_decide_disposition(cmd);
1575 if (disposition != SUCCESS &&
1576 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1577 sdev_printk(KERN_ERR, cmd->device,
1578 "timing out command, waited %lus\n",
1580 disposition = SUCCESS;
1583 scsi_log_completion(cmd, disposition);
1585 switch (disposition) {
1587 scsi_finish_command(cmd);
1590 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1592 case ADD_TO_MLQUEUE:
1593 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1596 if (!scsi_eh_scmd_add(cmd, 0))
1597 scsi_finish_command(cmd);
1602 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1603 * @cmd: command block we are dispatching.
1605 * Return: nonzero return request was rejected and device's queue needs to be
1608 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1610 struct Scsi_Host *host = cmd->device->host;
1613 atomic_inc(&cmd->device->iorequest_cnt);
1615 /* check if the device is still usable */
1616 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1617 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1618 * returns an immediate error upwards, and signals
1619 * that the device is no longer present */
1620 cmd->result = DID_NO_CONNECT << 16;
1624 /* Check to see if the scsi lld made this device blocked. */
1625 if (unlikely(scsi_device_blocked(cmd->device))) {
1627 * in blocked state, the command is just put back on
1628 * the device queue. The suspend state has already
1629 * blocked the queue so future requests should not
1630 * occur until the device transitions out of the
1633 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1634 "queuecommand : device blocked\n"));
1635 return SCSI_MLQUEUE_DEVICE_BUSY;
1638 /* Store the LUN value in cmnd, if needed. */
1639 if (cmd->device->lun_in_cdb)
1640 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1641 (cmd->device->lun << 5 & 0xe0);
1646 * Before we queue this command, check if the command
1647 * length exceeds what the host adapter can handle.
1649 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1650 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1651 "queuecommand : command too long. "
1652 "cdb_size=%d host->max_cmd_len=%d\n",
1653 cmd->cmd_len, cmd->device->host->max_cmd_len));
1654 cmd->result = (DID_ABORT << 16);
1658 if (unlikely(host->shost_state == SHOST_DEL)) {
1659 cmd->result = (DID_NO_CONNECT << 16);
1664 trace_scsi_dispatch_cmd_start(cmd);
1665 rtn = host->hostt->queuecommand(host, cmd);
1667 trace_scsi_dispatch_cmd_error(cmd, rtn);
1668 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1669 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1670 rtn = SCSI_MLQUEUE_HOST_BUSY;
1672 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1673 "queuecommand : request rejected\n"));
1678 cmd->scsi_done(cmd);
1683 * scsi_done - Invoke completion on finished SCSI command.
1684 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1685 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1687 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1688 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1689 * calls blk_complete_request() for further processing.
1691 * This function is interrupt context safe.
1693 static void scsi_done(struct scsi_cmnd *cmd)
1695 trace_scsi_dispatch_cmd_done(cmd);
1696 blk_complete_request(cmd->request);
1700 * Function: scsi_request_fn()
1702 * Purpose: Main strategy routine for SCSI.
1704 * Arguments: q - Pointer to actual queue.
1708 * Lock status: IO request lock assumed to be held when called.
1710 static void scsi_request_fn(struct request_queue *q)
1711 __releases(q->queue_lock)
1712 __acquires(q->queue_lock)
1714 struct scsi_device *sdev = q->queuedata;
1715 struct Scsi_Host *shost;
1716 struct scsi_cmnd *cmd;
1717 struct request *req;
1720 * To start with, we keep looping until the queue is empty, or until
1721 * the host is no longer able to accept any more requests.
1727 * get next queueable request. We do this early to make sure
1728 * that the request is fully prepared even if we cannot
1731 req = blk_peek_request(q);
1735 if (unlikely(!scsi_device_online(sdev))) {
1736 sdev_printk(KERN_ERR, sdev,
1737 "rejecting I/O to offline device\n");
1738 scsi_kill_request(req, q);
1742 if (!scsi_dev_queue_ready(q, sdev))
1746 * Remove the request from the request list.
1748 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1749 blk_start_request(req);
1751 spin_unlock_irq(q->queue_lock);
1753 if (unlikely(cmd == NULL)) {
1754 printk(KERN_CRIT "impossible request in %s.\n"
1755 "please mail a stack trace to "
1756 "linux-scsi@vger.kernel.org\n",
1758 blk_dump_rq_flags(req, "foo");
1763 * We hit this when the driver is using a host wide
1764 * tag map. For device level tag maps the queue_depth check
1765 * in the device ready fn would prevent us from trying
1766 * to allocate a tag. Since the map is a shared host resource
1767 * we add the dev to the starved list so it eventually gets
1768 * a run when a tag is freed.
1770 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1771 spin_lock_irq(shost->host_lock);
1772 if (list_empty(&sdev->starved_entry))
1773 list_add_tail(&sdev->starved_entry,
1774 &shost->starved_list);
1775 spin_unlock_irq(shost->host_lock);
1779 if (!scsi_target_queue_ready(shost, sdev))
1782 if (!scsi_host_queue_ready(q, shost, sdev))
1783 goto host_not_ready;
1785 if (sdev->simple_tags)
1786 cmd->flags |= SCMD_TAGGED;
1788 cmd->flags &= ~SCMD_TAGGED;
1791 * Finally, initialize any error handling parameters, and set up
1792 * the timers for timeouts.
1794 scsi_init_cmd_errh(cmd);
1797 * Dispatch the command to the low-level driver.
1799 cmd->scsi_done = scsi_done;
1800 rtn = scsi_dispatch_cmd(cmd);
1802 scsi_queue_insert(cmd, rtn);
1803 spin_lock_irq(q->queue_lock);
1806 spin_lock_irq(q->queue_lock);
1812 if (scsi_target(sdev)->can_queue > 0)
1813 atomic_dec(&scsi_target(sdev)->target_busy);
1816 * lock q, handle tag, requeue req, and decrement device_busy. We
1817 * must return with queue_lock held.
1819 * Decrementing device_busy without checking it is OK, as all such
1820 * cases (host limits or settings) should run the queue at some
1823 spin_lock_irq(q->queue_lock);
1824 blk_requeue_request(q, req);
1825 atomic_dec(&sdev->device_busy);
1827 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1828 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1831 static inline int prep_to_mq(int ret)
1835 return BLK_MQ_RQ_QUEUE_OK;
1837 return BLK_MQ_RQ_QUEUE_BUSY;
1839 return BLK_MQ_RQ_QUEUE_ERROR;
1843 static int scsi_mq_prep_fn(struct request *req)
1845 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1846 struct scsi_device *sdev = req->q->queuedata;
1847 struct Scsi_Host *shost = sdev->host;
1848 unsigned char *sense_buf = cmd->sense_buffer;
1849 struct scatterlist *sg;
1851 /* zero out the cmd, except for the embedded scsi_request */
1852 memset((char *)cmd + sizeof(cmd->req), 0,
1853 sizeof(*cmd) - sizeof(cmd->req));
1859 cmd->sense_buffer = sense_buf;
1861 cmd->tag = req->tag;
1863 cmd->prot_op = SCSI_PROT_NORMAL;
1865 INIT_LIST_HEAD(&cmd->list);
1866 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1867 cmd->jiffies_at_alloc = jiffies;
1869 if (shost->use_cmd_list) {
1870 spin_lock_irq(&sdev->list_lock);
1871 list_add_tail(&cmd->list, &sdev->cmd_list);
1872 spin_unlock_irq(&sdev->list_lock);
1875 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1876 cmd->sdb.table.sgl = sg;
1878 if (scsi_host_get_prot(shost)) {
1879 cmd->prot_sdb = (void *)sg +
1881 shost->sg_tablesize, SG_CHUNK_SIZE) *
1882 sizeof(struct scatterlist);
1883 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1885 cmd->prot_sdb->table.sgl =
1886 (struct scatterlist *)(cmd->prot_sdb + 1);
1889 if (blk_bidi_rq(req)) {
1890 struct request *next_rq = req->next_rq;
1891 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1893 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1894 bidi_sdb->table.sgl =
1895 (struct scatterlist *)(bidi_sdb + 1);
1897 next_rq->special = bidi_sdb;
1900 blk_mq_start_request(req);
1902 return scsi_setup_cmnd(sdev, req);
1905 static void scsi_mq_done(struct scsi_cmnd *cmd)
1907 trace_scsi_dispatch_cmd_done(cmd);
1908 blk_mq_complete_request(cmd->request, cmd->request->errors);
1911 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1912 const struct blk_mq_queue_data *bd)
1914 struct request *req = bd->rq;
1915 struct request_queue *q = req->q;
1916 struct scsi_device *sdev = q->queuedata;
1917 struct Scsi_Host *shost = sdev->host;
1918 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1922 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1923 if (ret != BLK_MQ_RQ_QUEUE_OK)
1926 ret = BLK_MQ_RQ_QUEUE_BUSY;
1927 if (!get_device(&sdev->sdev_gendev))
1930 if (!scsi_dev_queue_ready(q, sdev))
1931 goto out_put_device;
1932 if (!scsi_target_queue_ready(shost, sdev))
1933 goto out_dec_device_busy;
1934 if (!scsi_host_queue_ready(q, shost, sdev))
1935 goto out_dec_target_busy;
1937 if (!(req->rq_flags & RQF_DONTPREP)) {
1938 ret = prep_to_mq(scsi_mq_prep_fn(req));
1939 if (ret != BLK_MQ_RQ_QUEUE_OK)
1940 goto out_dec_host_busy;
1941 req->rq_flags |= RQF_DONTPREP;
1943 blk_mq_start_request(req);
1946 if (sdev->simple_tags)
1947 cmd->flags |= SCMD_TAGGED;
1949 cmd->flags &= ~SCMD_TAGGED;
1951 scsi_init_cmd_errh(cmd);
1952 cmd->scsi_done = scsi_mq_done;
1954 reason = scsi_dispatch_cmd(cmd);
1956 scsi_set_blocked(cmd, reason);
1957 ret = BLK_MQ_RQ_QUEUE_BUSY;
1958 goto out_dec_host_busy;
1961 return BLK_MQ_RQ_QUEUE_OK;
1964 atomic_dec(&shost->host_busy);
1965 out_dec_target_busy:
1966 if (scsi_target(sdev)->can_queue > 0)
1967 atomic_dec(&scsi_target(sdev)->target_busy);
1968 out_dec_device_busy:
1969 atomic_dec(&sdev->device_busy);
1971 put_device(&sdev->sdev_gendev);
1974 case BLK_MQ_RQ_QUEUE_BUSY:
1975 if (atomic_read(&sdev->device_busy) == 0 &&
1976 !scsi_device_blocked(sdev))
1977 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
1979 case BLK_MQ_RQ_QUEUE_ERROR:
1981 * Make sure to release all allocated ressources when
1982 * we hit an error, as we will never see this command
1985 if (req->rq_flags & RQF_DONTPREP)
1986 scsi_mq_uninit_cmd(cmd);
1994 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1998 return BLK_EH_RESET_TIMER;
1999 return scsi_times_out(req);
2002 static int scsi_init_request(void *data, struct request *rq,
2003 unsigned int hctx_idx, unsigned int request_idx,
2004 unsigned int numa_node)
2006 struct Scsi_Host *shost = data;
2007 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2010 scsi_alloc_sense_buffer(shost, GFP_KERNEL, numa_node);
2011 if (!cmd->sense_buffer)
2013 cmd->req.sense = cmd->sense_buffer;
2017 static void scsi_exit_request(void *data, struct request *rq,
2018 unsigned int hctx_idx, unsigned int request_idx)
2020 struct Scsi_Host *shost = data;
2021 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2023 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2026 static int scsi_map_queues(struct blk_mq_tag_set *set)
2028 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2030 if (shost->hostt->map_queues)
2031 return shost->hostt->map_queues(shost);
2032 return blk_mq_map_queues(set);
2035 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2037 struct device *host_dev;
2038 u64 bounce_limit = 0xffffffff;
2040 if (shost->unchecked_isa_dma)
2041 return BLK_BOUNCE_ISA;
2043 * Platforms with virtual-DMA translation
2044 * hardware have no practical limit.
2046 if (!PCI_DMA_BUS_IS_PHYS)
2047 return BLK_BOUNCE_ANY;
2049 host_dev = scsi_get_device(shost);
2050 if (host_dev && host_dev->dma_mask)
2051 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2053 return bounce_limit;
2056 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2058 struct device *dev = shost->dma_dev;
2061 * this limit is imposed by hardware restrictions
2063 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2066 if (scsi_host_prot_dma(shost)) {
2067 shost->sg_prot_tablesize =
2068 min_not_zero(shost->sg_prot_tablesize,
2069 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2070 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2071 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2074 blk_queue_max_hw_sectors(q, shost->max_sectors);
2075 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2076 blk_queue_segment_boundary(q, shost->dma_boundary);
2077 dma_set_seg_boundary(dev, shost->dma_boundary);
2079 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2081 if (!shost->use_clustering)
2082 q->limits.cluster = 0;
2085 * set a reasonable default alignment on word boundaries: the
2086 * host and device may alter it using
2087 * blk_queue_update_dma_alignment() later.
2089 blk_queue_dma_alignment(q, 0x03);
2091 EXPORT_SYMBOL_GPL(__scsi_init_queue);
2093 static int scsi_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp)
2095 struct Scsi_Host *shost = q->rq_alloc_data;
2096 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2098 memset(cmd, 0, sizeof(*cmd));
2100 cmd->sense_buffer = scsi_alloc_sense_buffer(shost, gfp, NUMA_NO_NODE);
2101 if (!cmd->sense_buffer)
2103 cmd->req.sense = cmd->sense_buffer;
2105 if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2106 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2108 goto fail_free_sense;
2114 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2119 static void scsi_exit_rq(struct request_queue *q, struct request *rq)
2121 struct Scsi_Host *shost = q->rq_alloc_data;
2122 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2125 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2126 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2129 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2131 struct Scsi_Host *shost = sdev->host;
2132 struct request_queue *q;
2134 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE);
2137 q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2138 q->rq_alloc_data = shost;
2139 q->request_fn = scsi_request_fn;
2140 q->init_rq_fn = scsi_init_rq;
2141 q->exit_rq_fn = scsi_exit_rq;
2143 if (blk_init_allocated_queue(q) < 0) {
2144 blk_cleanup_queue(q);
2148 __scsi_init_queue(shost, q);
2149 blk_queue_prep_rq(q, scsi_prep_fn);
2150 blk_queue_unprep_rq(q, scsi_unprep_fn);
2151 blk_queue_softirq_done(q, scsi_softirq_done);
2152 blk_queue_rq_timed_out(q, scsi_times_out);
2153 blk_queue_lld_busy(q, scsi_lld_busy);
2157 static struct blk_mq_ops scsi_mq_ops = {
2158 .queue_rq = scsi_queue_rq,
2159 .complete = scsi_softirq_done,
2160 .timeout = scsi_timeout,
2161 .init_request = scsi_init_request,
2162 .exit_request = scsi_exit_request,
2163 .map_queues = scsi_map_queues,
2166 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2168 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2169 if (IS_ERR(sdev->request_queue))
2172 sdev->request_queue->queuedata = sdev;
2173 __scsi_init_queue(sdev->host, sdev->request_queue);
2174 return sdev->request_queue;
2177 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2179 unsigned int cmd_size, sgl_size, tbl_size;
2181 tbl_size = shost->sg_tablesize;
2182 if (tbl_size > SG_CHUNK_SIZE)
2183 tbl_size = SG_CHUNK_SIZE;
2184 sgl_size = tbl_size * sizeof(struct scatterlist);
2185 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2186 if (scsi_host_get_prot(shost))
2187 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2189 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2190 shost->tag_set.ops = &scsi_mq_ops;
2191 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2192 shost->tag_set.queue_depth = shost->can_queue;
2193 shost->tag_set.cmd_size = cmd_size;
2194 shost->tag_set.numa_node = NUMA_NO_NODE;
2195 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2196 shost->tag_set.flags |=
2197 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2198 shost->tag_set.driver_data = shost;
2200 return blk_mq_alloc_tag_set(&shost->tag_set);
2203 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2205 blk_mq_free_tag_set(&shost->tag_set);
2209 * scsi_device_from_queue - return sdev associated with a request_queue
2210 * @q: The request queue to return the sdev from
2212 * Return the sdev associated with a request queue or NULL if the
2213 * request_queue does not reference a SCSI device.
2215 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2217 struct scsi_device *sdev = NULL;
2220 if (q->mq_ops == &scsi_mq_ops)
2221 sdev = q->queuedata;
2222 } else if (q->request_fn == scsi_request_fn)
2223 sdev = q->queuedata;
2224 if (!sdev || !get_device(&sdev->sdev_gendev))
2229 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2232 * Function: scsi_block_requests()
2234 * Purpose: Utility function used by low-level drivers to prevent further
2235 * commands from being queued to the device.
2237 * Arguments: shost - Host in question
2241 * Lock status: No locks are assumed held.
2243 * Notes: There is no timer nor any other means by which the requests
2244 * get unblocked other than the low-level driver calling
2245 * scsi_unblock_requests().
2247 void scsi_block_requests(struct Scsi_Host *shost)
2249 shost->host_self_blocked = 1;
2251 EXPORT_SYMBOL(scsi_block_requests);
2254 * Function: scsi_unblock_requests()
2256 * Purpose: Utility function used by low-level drivers to allow further
2257 * commands from being queued to the device.
2259 * Arguments: shost - Host in question
2263 * Lock status: No locks are assumed held.
2265 * Notes: There is no timer nor any other means by which the requests
2266 * get unblocked other than the low-level driver calling
2267 * scsi_unblock_requests().
2269 * This is done as an API function so that changes to the
2270 * internals of the scsi mid-layer won't require wholesale
2271 * changes to drivers that use this feature.
2273 void scsi_unblock_requests(struct Scsi_Host *shost)
2275 shost->host_self_blocked = 0;
2276 scsi_run_host_queues(shost);
2278 EXPORT_SYMBOL(scsi_unblock_requests);
2280 int __init scsi_init_queue(void)
2282 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2283 sizeof(struct scsi_data_buffer),
2285 if (!scsi_sdb_cache) {
2286 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2293 void scsi_exit_queue(void)
2295 kmem_cache_destroy(scsi_sense_cache);
2296 kmem_cache_destroy(scsi_sense_isadma_cache);
2297 kmem_cache_destroy(scsi_sdb_cache);
2301 * scsi_mode_select - issue a mode select
2302 * @sdev: SCSI device to be queried
2303 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2304 * @sp: Save page bit (0 == don't save, 1 == save)
2305 * @modepage: mode page being requested
2306 * @buffer: request buffer (may not be smaller than eight bytes)
2307 * @len: length of request buffer.
2308 * @timeout: command timeout
2309 * @retries: number of retries before failing
2310 * @data: returns a structure abstracting the mode header data
2311 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2312 * must be SCSI_SENSE_BUFFERSIZE big.
2314 * Returns zero if successful; negative error number or scsi
2319 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2320 unsigned char *buffer, int len, int timeout, int retries,
2321 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2323 unsigned char cmd[10];
2324 unsigned char *real_buffer;
2327 memset(cmd, 0, sizeof(cmd));
2328 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2330 if (sdev->use_10_for_ms) {
2333 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2336 memcpy(real_buffer + 8, buffer, len);
2340 real_buffer[2] = data->medium_type;
2341 real_buffer[3] = data->device_specific;
2342 real_buffer[4] = data->longlba ? 0x01 : 0;
2344 real_buffer[6] = data->block_descriptor_length >> 8;
2345 real_buffer[7] = data->block_descriptor_length;
2347 cmd[0] = MODE_SELECT_10;
2351 if (len > 255 || data->block_descriptor_length > 255 ||
2355 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2358 memcpy(real_buffer + 4, buffer, len);
2361 real_buffer[1] = data->medium_type;
2362 real_buffer[2] = data->device_specific;
2363 real_buffer[3] = data->block_descriptor_length;
2366 cmd[0] = MODE_SELECT;
2370 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2371 sshdr, timeout, retries, NULL);
2375 EXPORT_SYMBOL_GPL(scsi_mode_select);
2378 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2379 * @sdev: SCSI device to be queried
2380 * @dbd: set if mode sense will allow block descriptors to be returned
2381 * @modepage: mode page being requested
2382 * @buffer: request buffer (may not be smaller than eight bytes)
2383 * @len: length of request buffer.
2384 * @timeout: command timeout
2385 * @retries: number of retries before failing
2386 * @data: returns a structure abstracting the mode header data
2387 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2388 * must be SCSI_SENSE_BUFFERSIZE big.
2390 * Returns zero if unsuccessful, or the header offset (either 4
2391 * or 8 depending on whether a six or ten byte command was
2392 * issued) if successful.
2395 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2396 unsigned char *buffer, int len, int timeout, int retries,
2397 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2399 unsigned char cmd[12];
2402 int result, retry_count = retries;
2403 struct scsi_sense_hdr my_sshdr;
2405 memset(data, 0, sizeof(*data));
2406 memset(&cmd[0], 0, 12);
2407 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2410 /* caller might not be interested in sense, but we need it */
2415 use_10_for_ms = sdev->use_10_for_ms;
2417 if (use_10_for_ms) {
2421 cmd[0] = MODE_SENSE_10;
2428 cmd[0] = MODE_SENSE;
2433 memset(buffer, 0, len);
2435 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2436 sshdr, timeout, retries, NULL);
2438 /* This code looks awful: what it's doing is making sure an
2439 * ILLEGAL REQUEST sense return identifies the actual command
2440 * byte as the problem. MODE_SENSE commands can return
2441 * ILLEGAL REQUEST if the code page isn't supported */
2443 if (use_10_for_ms && !scsi_status_is_good(result) &&
2444 (driver_byte(result) & DRIVER_SENSE)) {
2445 if (scsi_sense_valid(sshdr)) {
2446 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2447 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2449 * Invalid command operation code
2451 sdev->use_10_for_ms = 0;
2457 if(scsi_status_is_good(result)) {
2458 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2459 (modepage == 6 || modepage == 8))) {
2460 /* Initio breakage? */
2463 data->medium_type = 0;
2464 data->device_specific = 0;
2466 data->block_descriptor_length = 0;
2467 } else if(use_10_for_ms) {
2468 data->length = buffer[0]*256 + buffer[1] + 2;
2469 data->medium_type = buffer[2];
2470 data->device_specific = buffer[3];
2471 data->longlba = buffer[4] & 0x01;
2472 data->block_descriptor_length = buffer[6]*256
2475 data->length = buffer[0] + 1;
2476 data->medium_type = buffer[1];
2477 data->device_specific = buffer[2];
2478 data->block_descriptor_length = buffer[3];
2480 data->header_length = header_length;
2481 } else if ((status_byte(result) == CHECK_CONDITION) &&
2482 scsi_sense_valid(sshdr) &&
2483 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2490 EXPORT_SYMBOL(scsi_mode_sense);
2493 * scsi_test_unit_ready - test if unit is ready
2494 * @sdev: scsi device to change the state of.
2495 * @timeout: command timeout
2496 * @retries: number of retries before failing
2497 * @sshdr: outpout pointer for decoded sense information.
2499 * Returns zero if unsuccessful or an error if TUR failed. For
2500 * removable media, UNIT_ATTENTION sets ->changed flag.
2503 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2504 struct scsi_sense_hdr *sshdr)
2507 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2511 /* try to eat the UNIT_ATTENTION if there are enough retries */
2513 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2514 timeout, retries, NULL);
2515 if (sdev->removable && scsi_sense_valid(sshdr) &&
2516 sshdr->sense_key == UNIT_ATTENTION)
2518 } while (scsi_sense_valid(sshdr) &&
2519 sshdr->sense_key == UNIT_ATTENTION && --retries);
2523 EXPORT_SYMBOL(scsi_test_unit_ready);
2526 * scsi_device_set_state - Take the given device through the device state model.
2527 * @sdev: scsi device to change the state of.
2528 * @state: state to change to.
2530 * Returns zero if unsuccessful or an error if the requested
2531 * transition is illegal.
2534 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2536 enum scsi_device_state oldstate = sdev->sdev_state;
2538 if (state == oldstate)
2544 case SDEV_CREATED_BLOCK:
2555 case SDEV_TRANSPORT_OFFLINE:
2568 case SDEV_TRANSPORT_OFFLINE:
2576 case SDEV_TRANSPORT_OFFLINE:
2591 case SDEV_CREATED_BLOCK:
2598 case SDEV_CREATED_BLOCK:
2613 case SDEV_TRANSPORT_OFFLINE:
2626 case SDEV_TRANSPORT_OFFLINE:
2628 case SDEV_CREATED_BLOCK:
2636 sdev->sdev_state = state;
2640 SCSI_LOG_ERROR_RECOVERY(1,
2641 sdev_printk(KERN_ERR, sdev,
2642 "Illegal state transition %s->%s",
2643 scsi_device_state_name(oldstate),
2644 scsi_device_state_name(state))
2648 EXPORT_SYMBOL(scsi_device_set_state);
2651 * sdev_evt_emit - emit a single SCSI device uevent
2652 * @sdev: associated SCSI device
2653 * @evt: event to emit
2655 * Send a single uevent (scsi_event) to the associated scsi_device.
2657 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2662 switch (evt->evt_type) {
2663 case SDEV_EVT_MEDIA_CHANGE:
2664 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2666 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2667 scsi_rescan_device(&sdev->sdev_gendev);
2668 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2670 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2671 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2673 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2674 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2676 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2677 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2679 case SDEV_EVT_LUN_CHANGE_REPORTED:
2680 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2682 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2683 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2692 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2696 * sdev_evt_thread - send a uevent for each scsi event
2697 * @work: work struct for scsi_device
2699 * Dispatch queued events to their associated scsi_device kobjects
2702 void scsi_evt_thread(struct work_struct *work)
2704 struct scsi_device *sdev;
2705 enum scsi_device_event evt_type;
2706 LIST_HEAD(event_list);
2708 sdev = container_of(work, struct scsi_device, event_work);
2710 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2711 if (test_and_clear_bit(evt_type, sdev->pending_events))
2712 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2715 struct scsi_event *evt;
2716 struct list_head *this, *tmp;
2717 unsigned long flags;
2719 spin_lock_irqsave(&sdev->list_lock, flags);
2720 list_splice_init(&sdev->event_list, &event_list);
2721 spin_unlock_irqrestore(&sdev->list_lock, flags);
2723 if (list_empty(&event_list))
2726 list_for_each_safe(this, tmp, &event_list) {
2727 evt = list_entry(this, struct scsi_event, node);
2728 list_del(&evt->node);
2729 scsi_evt_emit(sdev, evt);
2736 * sdev_evt_send - send asserted event to uevent thread
2737 * @sdev: scsi_device event occurred on
2738 * @evt: event to send
2740 * Assert scsi device event asynchronously.
2742 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2744 unsigned long flags;
2747 /* FIXME: currently this check eliminates all media change events
2748 * for polled devices. Need to update to discriminate between AN
2749 * and polled events */
2750 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2756 spin_lock_irqsave(&sdev->list_lock, flags);
2757 list_add_tail(&evt->node, &sdev->event_list);
2758 schedule_work(&sdev->event_work);
2759 spin_unlock_irqrestore(&sdev->list_lock, flags);
2761 EXPORT_SYMBOL_GPL(sdev_evt_send);
2764 * sdev_evt_alloc - allocate a new scsi event
2765 * @evt_type: type of event to allocate
2766 * @gfpflags: GFP flags for allocation
2768 * Allocates and returns a new scsi_event.
2770 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2773 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2777 evt->evt_type = evt_type;
2778 INIT_LIST_HEAD(&evt->node);
2780 /* evt_type-specific initialization, if any */
2782 case SDEV_EVT_MEDIA_CHANGE:
2783 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2784 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2785 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2786 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2787 case SDEV_EVT_LUN_CHANGE_REPORTED:
2788 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2796 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2799 * sdev_evt_send_simple - send asserted event to uevent thread
2800 * @sdev: scsi_device event occurred on
2801 * @evt_type: type of event to send
2802 * @gfpflags: GFP flags for allocation
2804 * Assert scsi device event asynchronously, given an event type.
2806 void sdev_evt_send_simple(struct scsi_device *sdev,
2807 enum scsi_device_event evt_type, gfp_t gfpflags)
2809 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2811 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2816 sdev_evt_send(sdev, evt);
2818 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2821 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2822 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2824 static int scsi_request_fn_active(struct scsi_device *sdev)
2826 struct request_queue *q = sdev->request_queue;
2827 int request_fn_active;
2829 WARN_ON_ONCE(sdev->host->use_blk_mq);
2831 spin_lock_irq(q->queue_lock);
2832 request_fn_active = q->request_fn_active;
2833 spin_unlock_irq(q->queue_lock);
2835 return request_fn_active;
2839 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2840 * @sdev: SCSI device pointer.
2842 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2843 * invoked from scsi_request_fn() have finished.
2845 static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2847 WARN_ON_ONCE(sdev->host->use_blk_mq);
2849 while (scsi_request_fn_active(sdev))
2854 * scsi_device_quiesce - Block user issued commands.
2855 * @sdev: scsi device to quiesce.
2857 * This works by trying to transition to the SDEV_QUIESCE state
2858 * (which must be a legal transition). When the device is in this
2859 * state, only special requests will be accepted, all others will
2860 * be deferred. Since special requests may also be requeued requests,
2861 * a successful return doesn't guarantee the device will be
2862 * totally quiescent.
2864 * Must be called with user context, may sleep.
2866 * Returns zero if unsuccessful or an error if not.
2869 scsi_device_quiesce(struct scsi_device *sdev)
2871 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2875 scsi_run_queue(sdev->request_queue);
2876 while (atomic_read(&sdev->device_busy)) {
2877 msleep_interruptible(200);
2878 scsi_run_queue(sdev->request_queue);
2882 EXPORT_SYMBOL(scsi_device_quiesce);
2885 * scsi_device_resume - Restart user issued commands to a quiesced device.
2886 * @sdev: scsi device to resume.
2888 * Moves the device from quiesced back to running and restarts the
2891 * Must be called with user context, may sleep.
2893 void scsi_device_resume(struct scsi_device *sdev)
2895 /* check if the device state was mutated prior to resume, and if
2896 * so assume the state is being managed elsewhere (for example
2897 * device deleted during suspend)
2899 if (sdev->sdev_state != SDEV_QUIESCE ||
2900 scsi_device_set_state(sdev, SDEV_RUNNING))
2902 scsi_run_queue(sdev->request_queue);
2904 EXPORT_SYMBOL(scsi_device_resume);
2907 device_quiesce_fn(struct scsi_device *sdev, void *data)
2909 scsi_device_quiesce(sdev);
2913 scsi_target_quiesce(struct scsi_target *starget)
2915 starget_for_each_device(starget, NULL, device_quiesce_fn);
2917 EXPORT_SYMBOL(scsi_target_quiesce);
2920 device_resume_fn(struct scsi_device *sdev, void *data)
2922 scsi_device_resume(sdev);
2926 scsi_target_resume(struct scsi_target *starget)
2928 starget_for_each_device(starget, NULL, device_resume_fn);
2930 EXPORT_SYMBOL(scsi_target_resume);
2933 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2934 * @sdev: device to block
2935 * @wait: Whether or not to wait until ongoing .queuecommand() /
2936 * .queue_rq() calls have finished.
2938 * Block request made by scsi lld's to temporarily stop all
2939 * scsi commands on the specified device. May sleep.
2941 * Returns zero if successful or error if not
2944 * This routine transitions the device to the SDEV_BLOCK state
2945 * (which must be a legal transition). When the device is in this
2946 * state, all commands are deferred until the scsi lld reenables
2947 * the device with scsi_device_unblock or device_block_tmo fires.
2949 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2950 * scsi_internal_device_block() has blocked a SCSI device and also
2951 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2954 scsi_internal_device_block(struct scsi_device *sdev, bool wait)
2956 struct request_queue *q = sdev->request_queue;
2957 unsigned long flags;
2960 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2962 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2969 * The device has transitioned to SDEV_BLOCK. Stop the
2970 * block layer from calling the midlayer with this device's
2975 blk_mq_quiesce_queue(q);
2977 blk_mq_stop_hw_queues(q);
2979 spin_lock_irqsave(q->queue_lock, flags);
2981 spin_unlock_irqrestore(q->queue_lock, flags);
2983 scsi_wait_for_queuecommand(sdev);
2988 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2991 * scsi_internal_device_unblock - resume a device after a block request
2992 * @sdev: device to resume
2993 * @new_state: state to set devices to after unblocking
2995 * Called by scsi lld's or the midlayer to restart the device queue
2996 * for the previously suspended scsi device. Called from interrupt or
2997 * normal process context.
2999 * Returns zero if successful or error if not.
3002 * This routine transitions the device to the SDEV_RUNNING state
3003 * or to one of the offline states (which must be a legal transition)
3004 * allowing the midlayer to goose the queue for this device.
3007 scsi_internal_device_unblock(struct scsi_device *sdev,
3008 enum scsi_device_state new_state)
3010 struct request_queue *q = sdev->request_queue;
3011 unsigned long flags;
3014 * Try to transition the scsi device to SDEV_RUNNING or one of the
3015 * offlined states and goose the device queue if successful.
3017 if ((sdev->sdev_state == SDEV_BLOCK) ||
3018 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3019 sdev->sdev_state = new_state;
3020 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3021 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3022 new_state == SDEV_OFFLINE)
3023 sdev->sdev_state = new_state;
3025 sdev->sdev_state = SDEV_CREATED;
3026 } else if (sdev->sdev_state != SDEV_CANCEL &&
3027 sdev->sdev_state != SDEV_OFFLINE)
3031 blk_mq_start_stopped_hw_queues(q, false);
3033 spin_lock_irqsave(q->queue_lock, flags);
3035 spin_unlock_irqrestore(q->queue_lock, flags);
3040 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3043 device_block(struct scsi_device *sdev, void *data)
3045 scsi_internal_device_block(sdev, true);
3049 target_block(struct device *dev, void *data)
3051 if (scsi_is_target_device(dev))
3052 starget_for_each_device(to_scsi_target(dev), NULL,
3058 scsi_target_block(struct device *dev)
3060 if (scsi_is_target_device(dev))
3061 starget_for_each_device(to_scsi_target(dev), NULL,
3064 device_for_each_child(dev, NULL, target_block);
3066 EXPORT_SYMBOL_GPL(scsi_target_block);
3069 device_unblock(struct scsi_device *sdev, void *data)
3071 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3075 target_unblock(struct device *dev, void *data)
3077 if (scsi_is_target_device(dev))
3078 starget_for_each_device(to_scsi_target(dev), data,
3084 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3086 if (scsi_is_target_device(dev))
3087 starget_for_each_device(to_scsi_target(dev), &new_state,
3090 device_for_each_child(dev, &new_state, target_unblock);
3092 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3095 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3096 * @sgl: scatter-gather list
3097 * @sg_count: number of segments in sg
3098 * @offset: offset in bytes into sg, on return offset into the mapped area
3099 * @len: bytes to map, on return number of bytes mapped
3101 * Returns virtual address of the start of the mapped page
3103 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3104 size_t *offset, size_t *len)
3107 size_t sg_len = 0, len_complete = 0;
3108 struct scatterlist *sg;
3111 WARN_ON(!irqs_disabled());
3113 for_each_sg(sgl, sg, sg_count, i) {
3114 len_complete = sg_len; /* Complete sg-entries */
3115 sg_len += sg->length;
3116 if (sg_len > *offset)
3120 if (unlikely(i == sg_count)) {
3121 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3123 __func__, sg_len, *offset, sg_count);
3128 /* Offset starting from the beginning of first page in this sg-entry */
3129 *offset = *offset - len_complete + sg->offset;
3131 /* Assumption: contiguous pages can be accessed as "page + i" */
3132 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3133 *offset &= ~PAGE_MASK;
3135 /* Bytes in this sg-entry from *offset to the end of the page */
3136 sg_len = PAGE_SIZE - *offset;
3140 return kmap_atomic(page);
3142 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3145 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3146 * @virt: virtual address to be unmapped
3148 void scsi_kunmap_atomic_sg(void *virt)
3150 kunmap_atomic(virt);
3152 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3154 void sdev_disable_disk_events(struct scsi_device *sdev)
3156 atomic_inc(&sdev->disk_events_disable_depth);
3158 EXPORT_SYMBOL(sdev_disable_disk_events);
3160 void sdev_enable_disk_events(struct scsi_device *sdev)
3162 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3164 atomic_dec(&sdev->disk_events_disable_depth);
3166 EXPORT_SYMBOL(sdev_enable_disk_events);
3169 * scsi_vpd_lun_id - return a unique device identification
3170 * @sdev: SCSI device
3171 * @id: buffer for the identification
3172 * @id_len: length of the buffer
3174 * Copies a unique device identification into @id based
3175 * on the information in the VPD page 0x83 of the device.
3176 * The string will be formatted as a SCSI name string.
3178 * Returns the length of the identification or error on failure.
3179 * If the identifier is longer than the supplied buffer the actual
3180 * identifier length is returned and the buffer is not zero-padded.
3182 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3184 u8 cur_id_type = 0xff;
3186 unsigned char *d, *cur_id_str;
3187 unsigned char __rcu *vpd_pg83;
3188 int id_size = -EINVAL;
3191 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3198 * Look for the correct descriptor.
3199 * Order of preference for lun descriptor:
3200 * - SCSI name string
3201 * - NAA IEEE Registered Extended
3202 * - EUI-64 based 16-byte
3203 * - EUI-64 based 12-byte
3204 * - NAA IEEE Registered
3205 * - NAA IEEE Extended
3207 * as longer descriptors reduce the likelyhood
3208 * of identification clashes.
3211 /* The id string must be at least 20 bytes + terminating NULL byte */
3217 memset(id, 0, id_len);
3219 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3220 /* Skip designators not referring to the LUN */
3221 if ((d[1] & 0x30) != 0x00)
3224 switch (d[1] & 0xf) {
3227 if (cur_id_size > d[3])
3229 /* Prefer anything */
3230 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3233 if (cur_id_size + 4 > id_len)
3234 cur_id_size = id_len - 4;
3236 cur_id_type = d[1] & 0xf;
3237 id_size = snprintf(id, id_len, "t10.%*pE",
3238 cur_id_size, cur_id_str);
3242 if (cur_id_size > d[3])
3244 /* Prefer NAA IEEE Registered Extended */
3245 if (cur_id_type == 0x3 &&
3246 cur_id_size == d[3])
3250 cur_id_type = d[1] & 0xf;
3251 switch (cur_id_size) {
3253 id_size = snprintf(id, id_len,
3258 id_size = snprintf(id, id_len,
3263 id_size = snprintf(id, id_len,
3274 if (cur_id_size > d[3])
3278 cur_id_type = d[1] & 0xf;
3279 switch (cur_id_size) {
3281 id_size = snprintf(id, id_len,
3286 id_size = snprintf(id, id_len,
3296 /* SCSI name string */
3297 if (cur_id_size + 4 > d[3])
3299 /* Prefer others for truncated descriptor */
3300 if (cur_id_size && d[3] > id_len)
3302 cur_id_size = id_size = d[3];
3304 cur_id_type = d[1] & 0xf;
3305 if (cur_id_size >= id_len)
3306 cur_id_size = id_len - 1;
3307 memcpy(id, cur_id_str, cur_id_size);
3308 /* Decrease priority for truncated descriptor */
3309 if (cur_id_size != id_size)
3322 EXPORT_SYMBOL(scsi_vpd_lun_id);
3325 * scsi_vpd_tpg_id - return a target port group identifier
3326 * @sdev: SCSI device
3328 * Returns the Target Port Group identifier from the information
3329 * froom VPD page 0x83 of the device.
3331 * Returns the identifier or error on failure.
3333 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3336 unsigned char __rcu *vpd_pg83;
3337 int group_id = -EAGAIN, rel_port = -1;
3340 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3346 d = sdev->vpd_pg83 + 4;
3347 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3348 switch (d[1] & 0xf) {
3350 /* Relative target port */
3351 rel_port = get_unaligned_be16(&d[6]);
3354 /* Target port group */
3355 group_id = get_unaligned_be16(&d[6]);
3364 if (group_id >= 0 && rel_id && rel_port != -1)
3369 EXPORT_SYMBOL(scsi_vpd_tpg_id);