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/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/init.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/hardirq.h>
23 #include <linux/scatterlist.h>
24 #include <linux/blk-mq.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_dh.h>
37 #include <trace/events/scsi.h>
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
43 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
44 #define SG_MEMPOOL_SIZE 2
46 struct scsi_host_sg_pool {
49 struct kmem_cache *slab;
53 #define SP(x) { .size = x, "sgpool-" __stringify(x) }
54 #if (SCSI_MAX_SG_SEGMENTS < 32)
55 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
57 static struct scsi_host_sg_pool scsi_sg_pools[] = {
60 #if (SCSI_MAX_SG_SEGMENTS > 32)
62 #if (SCSI_MAX_SG_SEGMENTS > 64)
64 #if (SCSI_MAX_SG_SEGMENTS > 128)
66 #if (SCSI_MAX_SG_SEGMENTS > 256)
67 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
72 SP(SCSI_MAX_SG_SEGMENTS)
76 struct kmem_cache *scsi_sdb_cache;
79 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
80 * not change behaviour from the previous unplug mechanism, experimentation
81 * may prove this needs changing.
83 #define SCSI_QUEUE_DELAY 3
86 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
88 struct Scsi_Host *host = cmd->device->host;
89 struct scsi_device *device = cmd->device;
90 struct scsi_target *starget = scsi_target(device);
93 * Set the appropriate busy bit for the device/host.
95 * If the host/device isn't busy, assume that something actually
96 * completed, and that we should be able to queue a command now.
98 * Note that the prior mid-layer assumption that any host could
99 * always queue at least one command is now broken. The mid-layer
100 * will implement a user specifiable stall (see
101 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
102 * if a command is requeued with no other commands outstanding
103 * either for the device or for the host.
106 case SCSI_MLQUEUE_HOST_BUSY:
107 atomic_set(&host->host_blocked, host->max_host_blocked);
109 case SCSI_MLQUEUE_DEVICE_BUSY:
110 case SCSI_MLQUEUE_EH_RETRY:
111 atomic_set(&device->device_blocked,
112 device->max_device_blocked);
114 case SCSI_MLQUEUE_TARGET_BUSY:
115 atomic_set(&starget->target_blocked,
116 starget->max_target_blocked);
121 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
123 struct scsi_device *sdev = cmd->device;
124 struct request_queue *q = cmd->request->q;
126 blk_mq_requeue_request(cmd->request);
127 blk_mq_kick_requeue_list(q);
128 put_device(&sdev->sdev_gendev);
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
143 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
145 struct scsi_device *device = cmd->device;
146 struct request_queue *q = device->request_queue;
149 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
150 "Inserting command %p into mlqueue\n", cmd));
152 scsi_set_blocked(cmd, reason);
155 * Decrement the counters, since these commands are no longer
156 * active on the host/device.
159 scsi_device_unbusy(device);
162 * Requeue this command. It will go before all other commands
163 * that are already in the queue. Schedule requeue work under
164 * lock such that the kblockd_schedule_work() call happens
165 * before blk_cleanup_queue() finishes.
169 scsi_mq_requeue_cmd(cmd);
172 spin_lock_irqsave(q->queue_lock, flags);
173 blk_requeue_request(q, cmd->request);
174 kblockd_schedule_work(&device->requeue_work);
175 spin_unlock_irqrestore(q->queue_lock, flags);
179 * Function: scsi_queue_insert()
181 * Purpose: Insert a command in the midlevel queue.
183 * Arguments: cmd - command that we are adding to queue.
184 * reason - why we are inserting command to queue.
186 * Lock status: Assumed that lock is not held upon entry.
190 * Notes: We do this for one of two cases. Either the host is busy
191 * and it cannot accept any more commands for the time being,
192 * or the device returned QUEUE_FULL and can accept no more
194 * Notes: This could be called either from an interrupt context or a
195 * normal process context.
197 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
199 __scsi_queue_insert(cmd, reason, 1);
202 * scsi_execute - insert request and wait for the result
205 * @data_direction: data direction
206 * @buffer: data buffer
207 * @bufflen: len of buffer
208 * @sense: optional sense buffer
209 * @timeout: request timeout in seconds
210 * @retries: number of times to retry request
211 * @flags: or into request flags;
212 * @resid: optional residual length
214 * returns the req->errors value which is the scsi_cmnd result
217 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
218 int data_direction, void *buffer, unsigned bufflen,
219 unsigned char *sense, int timeout, int retries, u64 flags,
223 int write = (data_direction == DMA_TO_DEVICE);
224 int ret = DRIVER_ERROR << 24;
226 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
229 blk_rq_set_block_pc(req);
231 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
232 buffer, bufflen, __GFP_RECLAIM))
235 req->cmd_len = COMMAND_SIZE(cmd[0]);
236 memcpy(req->cmd, cmd, req->cmd_len);
239 req->retries = retries;
240 req->timeout = timeout;
241 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
244 * head injection *required* here otherwise quiesce won't work
246 blk_execute_rq(req->q, NULL, req, 1);
249 * Some devices (USB mass-storage in particular) may transfer
250 * garbage data together with a residue indicating that the data
251 * is invalid. Prevent the garbage from being misinterpreted
252 * and prevent security leaks by zeroing out the excess data.
254 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
255 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
258 *resid = req->resid_len;
261 blk_put_request(req);
265 EXPORT_SYMBOL(scsi_execute);
267 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
268 int data_direction, void *buffer, unsigned bufflen,
269 struct scsi_sense_hdr *sshdr, int timeout, int retries,
270 int *resid, u64 flags)
276 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
278 return DRIVER_ERROR << 24;
280 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
281 sense, timeout, retries, flags, resid);
283 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
288 EXPORT_SYMBOL(scsi_execute_req_flags);
291 * Function: scsi_init_cmd_errh()
293 * Purpose: Initialize cmd fields related to error handling.
295 * Arguments: cmd - command that is ready to be queued.
297 * Notes: This function has the job of initializing a number of
298 * fields related to error handling. Typically this will
299 * be called once for each command, as required.
301 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
303 cmd->serial_number = 0;
304 scsi_set_resid(cmd, 0);
305 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
306 if (cmd->cmd_len == 0)
307 cmd->cmd_len = scsi_command_size(cmd->cmnd);
310 void scsi_device_unbusy(struct scsi_device *sdev)
312 struct Scsi_Host *shost = sdev->host;
313 struct scsi_target *starget = scsi_target(sdev);
316 atomic_dec(&shost->host_busy);
317 if (starget->can_queue > 0)
318 atomic_dec(&starget->target_busy);
320 if (unlikely(scsi_host_in_recovery(shost) &&
321 (shost->host_failed || shost->host_eh_scheduled))) {
322 spin_lock_irqsave(shost->host_lock, flags);
323 scsi_eh_wakeup(shost);
324 spin_unlock_irqrestore(shost->host_lock, flags);
327 atomic_dec(&sdev->device_busy);
330 static void scsi_kick_queue(struct request_queue *q)
333 blk_mq_start_hw_queues(q);
339 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
340 * and call blk_run_queue for all the scsi_devices on the target -
341 * including current_sdev first.
343 * Called with *no* scsi locks held.
345 static void scsi_single_lun_run(struct scsi_device *current_sdev)
347 struct Scsi_Host *shost = current_sdev->host;
348 struct scsi_device *sdev, *tmp;
349 struct scsi_target *starget = scsi_target(current_sdev);
352 spin_lock_irqsave(shost->host_lock, flags);
353 starget->starget_sdev_user = NULL;
354 spin_unlock_irqrestore(shost->host_lock, flags);
357 * Call blk_run_queue for all LUNs on the target, starting with
358 * current_sdev. We race with others (to set starget_sdev_user),
359 * but in most cases, we will be first. Ideally, each LU on the
360 * target would get some limited time or requests on the target.
362 scsi_kick_queue(current_sdev->request_queue);
364 spin_lock_irqsave(shost->host_lock, flags);
365 if (starget->starget_sdev_user)
367 list_for_each_entry_safe(sdev, tmp, &starget->devices,
368 same_target_siblings) {
369 if (sdev == current_sdev)
371 if (scsi_device_get(sdev))
374 spin_unlock_irqrestore(shost->host_lock, flags);
375 scsi_kick_queue(sdev->request_queue);
376 spin_lock_irqsave(shost->host_lock, flags);
378 scsi_device_put(sdev);
381 spin_unlock_irqrestore(shost->host_lock, flags);
384 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
386 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
388 if (atomic_read(&sdev->device_blocked) > 0)
393 static inline bool scsi_target_is_busy(struct scsi_target *starget)
395 if (starget->can_queue > 0) {
396 if (atomic_read(&starget->target_busy) >= starget->can_queue)
398 if (atomic_read(&starget->target_blocked) > 0)
404 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
406 if (shost->can_queue > 0 &&
407 atomic_read(&shost->host_busy) >= shost->can_queue)
409 if (atomic_read(&shost->host_blocked) > 0)
411 if (shost->host_self_blocked)
416 static void scsi_starved_list_run(struct Scsi_Host *shost)
418 LIST_HEAD(starved_list);
419 struct scsi_device *sdev;
422 spin_lock_irqsave(shost->host_lock, flags);
423 list_splice_init(&shost->starved_list, &starved_list);
425 while (!list_empty(&starved_list)) {
426 struct request_queue *slq;
429 * As long as shost is accepting commands and we have
430 * starved queues, call blk_run_queue. scsi_request_fn
431 * drops the queue_lock and can add us back to the
434 * host_lock protects the starved_list and starved_entry.
435 * scsi_request_fn must get the host_lock before checking
436 * or modifying starved_list or starved_entry.
438 if (scsi_host_is_busy(shost))
441 sdev = list_entry(starved_list.next,
442 struct scsi_device, starved_entry);
443 list_del_init(&sdev->starved_entry);
444 if (scsi_target_is_busy(scsi_target(sdev))) {
445 list_move_tail(&sdev->starved_entry,
446 &shost->starved_list);
451 * Once we drop the host lock, a racing scsi_remove_device()
452 * call may remove the sdev from the starved list and destroy
453 * it and the queue. Mitigate by taking a reference to the
454 * queue and never touching the sdev again after we drop the
455 * host lock. Note: if __scsi_remove_device() invokes
456 * blk_cleanup_queue() before the queue is run from this
457 * function then blk_run_queue() will return immediately since
458 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
460 slq = sdev->request_queue;
461 if (!blk_get_queue(slq))
463 spin_unlock_irqrestore(shost->host_lock, flags);
465 scsi_kick_queue(slq);
468 spin_lock_irqsave(shost->host_lock, flags);
470 /* put any unprocessed entries back */
471 list_splice(&starved_list, &shost->starved_list);
472 spin_unlock_irqrestore(shost->host_lock, flags);
476 * Function: scsi_run_queue()
478 * Purpose: Select a proper request queue to serve next
480 * Arguments: q - last request's queue
484 * Notes: The previous command was completely finished, start
485 * a new one if possible.
487 static void scsi_run_queue(struct request_queue *q)
489 struct scsi_device *sdev = q->queuedata;
491 if (scsi_target(sdev)->single_lun)
492 scsi_single_lun_run(sdev);
493 if (!list_empty(&sdev->host->starved_list))
494 scsi_starved_list_run(sdev->host);
497 blk_mq_start_stopped_hw_queues(q, false);
502 void scsi_requeue_run_queue(struct work_struct *work)
504 struct scsi_device *sdev;
505 struct request_queue *q;
507 sdev = container_of(work, struct scsi_device, requeue_work);
508 q = sdev->request_queue;
513 * Function: scsi_requeue_command()
515 * Purpose: Handle post-processing of completed commands.
517 * Arguments: q - queue to operate on
518 * cmd - command that may need to be requeued.
522 * Notes: After command completion, there may be blocks left
523 * over which weren't finished by the previous command
524 * this can be for a number of reasons - the main one is
525 * I/O errors in the middle of the request, in which case
526 * we need to request the blocks that come after the bad
528 * Notes: Upon return, cmd is a stale pointer.
530 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
532 struct scsi_device *sdev = cmd->device;
533 struct request *req = cmd->request;
536 spin_lock_irqsave(q->queue_lock, flags);
537 blk_unprep_request(req);
539 scsi_put_command(cmd);
540 blk_requeue_request(q, req);
541 spin_unlock_irqrestore(q->queue_lock, flags);
545 put_device(&sdev->sdev_gendev);
548 void scsi_run_host_queues(struct Scsi_Host *shost)
550 struct scsi_device *sdev;
552 shost_for_each_device(sdev, shost)
553 scsi_run_queue(sdev->request_queue);
556 static inline unsigned int scsi_sgtable_index(unsigned short nents)
560 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
565 index = get_count_order(nents) - 3;
570 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
572 struct scsi_host_sg_pool *sgp;
574 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
575 mempool_free(sgl, sgp->pool);
578 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
580 struct scsi_host_sg_pool *sgp;
582 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
583 return mempool_alloc(sgp->pool, gfp_mask);
586 static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
588 if (mq && sdb->table.orig_nents <= SCSI_MAX_SG_SEGMENTS)
590 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
593 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, bool mq)
595 struct scatterlist *first_chunk = NULL;
601 if (nents <= SCSI_MAX_SG_SEGMENTS) {
602 sdb->table.nents = sdb->table.orig_nents = nents;
603 sg_init_table(sdb->table.sgl, nents);
606 first_chunk = sdb->table.sgl;
609 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
610 first_chunk, GFP_ATOMIC, scsi_sg_alloc);
612 scsi_free_sgtable(sdb, mq);
616 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
618 if (cmd->request->cmd_type == REQ_TYPE_FS) {
619 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
621 if (drv->uninit_command)
622 drv->uninit_command(cmd);
626 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
628 if (cmd->sdb.table.nents)
629 scsi_free_sgtable(&cmd->sdb, true);
630 if (cmd->request->next_rq && cmd->request->next_rq->special)
631 scsi_free_sgtable(cmd->request->next_rq->special, true);
632 if (scsi_prot_sg_count(cmd))
633 scsi_free_sgtable(cmd->prot_sdb, true);
636 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
638 struct scsi_device *sdev = cmd->device;
639 struct Scsi_Host *shost = sdev->host;
642 scsi_mq_free_sgtables(cmd);
643 scsi_uninit_cmd(cmd);
645 if (shost->use_cmd_list) {
646 BUG_ON(list_empty(&cmd->list));
647 spin_lock_irqsave(&sdev->list_lock, flags);
648 list_del_init(&cmd->list);
649 spin_unlock_irqrestore(&sdev->list_lock, flags);
654 * Function: scsi_release_buffers()
656 * Purpose: Free resources allocate for a scsi_command.
658 * Arguments: cmd - command that we are bailing.
660 * Lock status: Assumed that no lock is held upon entry.
664 * Notes: In the event that an upper level driver rejects a
665 * command, we must release resources allocated during
666 * the __init_io() function. Primarily this would involve
667 * the scatter-gather table.
669 static void scsi_release_buffers(struct scsi_cmnd *cmd)
671 if (cmd->sdb.table.nents)
672 scsi_free_sgtable(&cmd->sdb, false);
674 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
676 if (scsi_prot_sg_count(cmd))
677 scsi_free_sgtable(cmd->prot_sdb, false);
680 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
682 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
684 scsi_free_sgtable(bidi_sdb, false);
685 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
686 cmd->request->next_rq->special = NULL;
689 static bool scsi_end_request(struct request *req, int error,
690 unsigned int bytes, unsigned int bidi_bytes)
692 struct scsi_cmnd *cmd = req->special;
693 struct scsi_device *sdev = cmd->device;
694 struct request_queue *q = sdev->request_queue;
696 if (blk_update_request(req, error, bytes))
699 /* Bidi request must be completed as a whole */
700 if (unlikely(bidi_bytes) &&
701 blk_update_request(req->next_rq, error, bidi_bytes))
704 if (blk_queue_add_random(q))
705 add_disk_randomness(req->rq_disk);
709 * In the MQ case the command gets freed by __blk_mq_end_request,
710 * so we have to do all cleanup that depends on it earlier.
712 * We also can't kick the queues from irq context, so we
713 * will have to defer it to a workqueue.
715 scsi_mq_uninit_cmd(cmd);
717 __blk_mq_end_request(req, error);
719 if (scsi_target(sdev)->single_lun ||
720 !list_empty(&sdev->host->starved_list))
721 kblockd_schedule_work(&sdev->requeue_work);
723 blk_mq_start_stopped_hw_queues(q, true);
728 scsi_release_bidi_buffers(cmd);
730 spin_lock_irqsave(q->queue_lock, flags);
731 blk_finish_request(req, error);
732 spin_unlock_irqrestore(q->queue_lock, flags);
734 scsi_release_buffers(cmd);
736 scsi_put_command(cmd);
740 put_device(&sdev->sdev_gendev);
745 * __scsi_error_from_host_byte - translate SCSI error code into errno
746 * @cmd: SCSI command (unused)
747 * @result: scsi error code
749 * Translate SCSI error code into standard UNIX errno.
751 * -ENOLINK temporary transport failure
752 * -EREMOTEIO permanent target failure, do not retry
753 * -EBADE permanent nexus failure, retry on other path
754 * -ENOSPC No write space available
755 * -ENODATA Medium error
756 * -EIO unspecified I/O error
758 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
762 switch(host_byte(result)) {
763 case DID_TRANSPORT_FAILFAST:
766 case DID_TARGET_FAILURE:
767 set_host_byte(cmd, DID_OK);
770 case DID_NEXUS_FAILURE:
771 set_host_byte(cmd, DID_OK);
774 case DID_ALLOC_FAILURE:
775 set_host_byte(cmd, DID_OK);
778 case DID_MEDIUM_ERROR:
779 set_host_byte(cmd, DID_OK);
791 * Function: scsi_io_completion()
793 * Purpose: Completion processing for block device I/O requests.
795 * Arguments: cmd - command that is finished.
797 * Lock status: Assumed that no lock is held upon entry.
801 * Notes: We will finish off the specified number of sectors. If we
802 * are done, the command block will be released and the queue
803 * function will be goosed. If we are not done then we have to
804 * figure out what to do next:
806 * a) We can call scsi_requeue_command(). The request
807 * will be unprepared and put back on the queue. Then
808 * a new command will be created for it. This should
809 * be used if we made forward progress, or if we want
810 * to switch from READ(10) to READ(6) for example.
812 * b) We can call __scsi_queue_insert(). The request will
813 * be put back on the queue and retried using the same
814 * command as before, possibly after a delay.
816 * c) We can call scsi_end_request() with -EIO to fail
817 * the remainder of the request.
819 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
821 int result = cmd->result;
822 struct request_queue *q = cmd->device->request_queue;
823 struct request *req = cmd->request;
825 struct scsi_sense_hdr sshdr;
826 bool sense_valid = false;
827 int sense_deferred = 0, level = 0;
828 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
829 ACTION_DELAYED_RETRY} action;
830 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
833 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
835 sense_deferred = scsi_sense_is_deferred(&sshdr);
838 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
840 if (sense_valid && req->sense) {
842 * SG_IO wants current and deferred errors
844 int len = 8 + cmd->sense_buffer[7];
846 if (len > SCSI_SENSE_BUFFERSIZE)
847 len = SCSI_SENSE_BUFFERSIZE;
848 memcpy(req->sense, cmd->sense_buffer, len);
849 req->sense_len = len;
852 error = __scsi_error_from_host_byte(cmd, result);
855 * __scsi_error_from_host_byte may have reset the host_byte
857 req->errors = cmd->result;
859 req->resid_len = scsi_get_resid(cmd);
861 if (scsi_bidi_cmnd(cmd)) {
863 * Bidi commands Must be complete as a whole,
864 * both sides at once.
866 req->next_rq->resid_len = scsi_in(cmd)->resid;
867 if (scsi_end_request(req, 0, blk_rq_bytes(req),
868 blk_rq_bytes(req->next_rq)))
872 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
874 * Certain non BLOCK_PC requests are commands that don't
875 * actually transfer anything (FLUSH), so cannot use
876 * good_bytes != blk_rq_bytes(req) as the signal for an error.
877 * This sets the error explicitly for the problem case.
879 error = __scsi_error_from_host_byte(cmd, result);
882 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
883 BUG_ON(blk_bidi_rq(req));
886 * Next deal with any sectors which we were able to correctly
889 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
890 "%u sectors total, %d bytes done.\n",
891 blk_rq_sectors(req), good_bytes));
894 * Recovered errors need reporting, but they're always treated
895 * as success, so fiddle the result code here. For BLOCK_PC
896 * we already took a copy of the original into rq->errors which
897 * is what gets returned to the user
899 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
900 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
901 * print since caller wants ATA registers. Only occurs on
902 * SCSI ATA PASS_THROUGH commands when CK_COND=1
904 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
906 else if (!(req->cmd_flags & REQ_QUIET))
907 scsi_print_sense(cmd);
909 /* BLOCK_PC may have set error */
914 * If we finished all bytes in the request we are done now.
916 if (!scsi_end_request(req, error, good_bytes, 0))
920 * Kill remainder if no retrys.
922 if (error && scsi_noretry_cmd(cmd)) {
923 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
929 * If there had been no error, but we have leftover bytes in the
930 * requeues just queue the command up again.
935 error = __scsi_error_from_host_byte(cmd, result);
937 if (host_byte(result) == DID_RESET) {
938 /* Third party bus reset or reset for error recovery
939 * reasons. Just retry the command and see what
942 action = ACTION_RETRY;
943 } else if (sense_valid && !sense_deferred) {
944 switch (sshdr.sense_key) {
946 if (cmd->device->removable) {
947 /* Detected disc change. Set a bit
948 * and quietly refuse further access.
950 cmd->device->changed = 1;
951 action = ACTION_FAIL;
953 /* Must have been a power glitch, or a
954 * bus reset. Could not have been a
955 * media change, so we just retry the
956 * command and see what happens.
958 action = ACTION_RETRY;
961 case ILLEGAL_REQUEST:
962 /* If we had an ILLEGAL REQUEST returned, then
963 * we may have performed an unsupported
964 * command. The only thing this should be
965 * would be a ten byte read where only a six
966 * byte read was supported. Also, on a system
967 * where READ CAPACITY failed, we may have
968 * read past the end of the disk.
970 if ((cmd->device->use_10_for_rw &&
971 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
972 (cmd->cmnd[0] == READ_10 ||
973 cmd->cmnd[0] == WRITE_10)) {
974 /* This will issue a new 6-byte command. */
975 cmd->device->use_10_for_rw = 0;
976 action = ACTION_REPREP;
977 } else if (sshdr.asc == 0x10) /* DIX */ {
978 action = ACTION_FAIL;
980 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
981 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
982 action = ACTION_FAIL;
985 action = ACTION_FAIL;
987 case ABORTED_COMMAND:
988 action = ACTION_FAIL;
989 if (sshdr.asc == 0x10) /* DIF */
993 /* If the device is in the process of becoming
994 * ready, or has a temporary blockage, retry.
996 if (sshdr.asc == 0x04) {
997 switch (sshdr.ascq) {
998 case 0x01: /* becoming ready */
999 case 0x04: /* format in progress */
1000 case 0x05: /* rebuild in progress */
1001 case 0x06: /* recalculation in progress */
1002 case 0x07: /* operation in progress */
1003 case 0x08: /* Long write in progress */
1004 case 0x09: /* self test in progress */
1005 case 0x14: /* space allocation in progress */
1006 action = ACTION_DELAYED_RETRY;
1009 action = ACTION_FAIL;
1013 action = ACTION_FAIL;
1015 case VOLUME_OVERFLOW:
1016 /* See SSC3rXX or current. */
1017 action = ACTION_FAIL;
1020 action = ACTION_FAIL;
1024 action = ACTION_FAIL;
1026 if (action != ACTION_FAIL &&
1027 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1028 action = ACTION_FAIL;
1032 /* Give up and fail the remainder of the request */
1033 if (!(req->cmd_flags & REQ_QUIET)) {
1034 static DEFINE_RATELIMIT_STATE(_rs,
1035 DEFAULT_RATELIMIT_INTERVAL,
1036 DEFAULT_RATELIMIT_BURST);
1038 if (unlikely(scsi_logging_level))
1039 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1040 SCSI_LOG_MLCOMPLETE_BITS);
1043 * if logging is enabled the failure will be printed
1044 * in scsi_log_completion(), so avoid duplicate messages
1046 if (!level && __ratelimit(&_rs)) {
1047 scsi_print_result(cmd, NULL, FAILED);
1048 if (driver_byte(result) & DRIVER_SENSE)
1049 scsi_print_sense(cmd);
1050 scsi_print_command(cmd);
1053 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1058 /* Unprep the request and put it back at the head of the queue.
1059 * A new command will be prepared and issued.
1062 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1063 scsi_mq_uninit_cmd(cmd);
1064 scsi_mq_requeue_cmd(cmd);
1066 scsi_release_buffers(cmd);
1067 scsi_requeue_command(q, cmd);
1071 /* Retry the same command immediately */
1072 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1074 case ACTION_DELAYED_RETRY:
1075 /* Retry the same command after a delay */
1076 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1081 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1086 * If sg table allocation fails, requeue request later.
1088 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1089 req->mq_ctx != NULL)))
1090 return BLKPREP_DEFER;
1093 * Next, walk the list, and fill in the addresses and sizes of
1096 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1097 BUG_ON(count > sdb->table.nents);
1098 sdb->table.nents = count;
1099 sdb->length = blk_rq_bytes(req);
1104 * Function: scsi_init_io()
1106 * Purpose: SCSI I/O initialize function.
1108 * Arguments: cmd - Command descriptor we wish to initialize
1110 * Returns: 0 on success
1111 * BLKPREP_DEFER if the failure is retryable
1112 * BLKPREP_KILL if the failure is fatal
1114 int scsi_init_io(struct scsi_cmnd *cmd)
1116 struct scsi_device *sdev = cmd->device;
1117 struct request *rq = cmd->request;
1118 bool is_mq = (rq->mq_ctx != NULL);
1121 BUG_ON(!rq->nr_phys_segments);
1123 error = scsi_init_sgtable(rq, &cmd->sdb);
1127 if (blk_bidi_rq(rq)) {
1128 if (!rq->q->mq_ops) {
1129 struct scsi_data_buffer *bidi_sdb =
1130 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1132 error = BLKPREP_DEFER;
1136 rq->next_rq->special = bidi_sdb;
1139 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1144 if (blk_integrity_rq(rq)) {
1145 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1148 if (prot_sdb == NULL) {
1150 * This can happen if someone (e.g. multipath)
1151 * queues a command to a device on an adapter
1152 * that does not support DIX.
1155 error = BLKPREP_KILL;
1159 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1161 if (scsi_alloc_sgtable(prot_sdb, ivecs, is_mq)) {
1162 error = BLKPREP_DEFER;
1166 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1167 prot_sdb->table.sgl);
1168 BUG_ON(unlikely(count > ivecs));
1169 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1171 cmd->prot_sdb = prot_sdb;
1172 cmd->prot_sdb->table.nents = count;
1178 scsi_mq_free_sgtables(cmd);
1180 scsi_release_buffers(cmd);
1181 cmd->request->special = NULL;
1182 scsi_put_command(cmd);
1183 put_device(&sdev->sdev_gendev);
1187 EXPORT_SYMBOL(scsi_init_io);
1189 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1190 struct request *req)
1192 struct scsi_cmnd *cmd;
1194 if (!req->special) {
1195 /* Bail if we can't get a reference to the device */
1196 if (!get_device(&sdev->sdev_gendev))
1199 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1200 if (unlikely(!cmd)) {
1201 put_device(&sdev->sdev_gendev);
1209 /* pull a tag out of the request if we have one */
1210 cmd->tag = req->tag;
1213 cmd->cmnd = req->cmd;
1214 cmd->prot_op = SCSI_PROT_NORMAL;
1219 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1221 struct scsi_cmnd *cmd = req->special;
1224 * BLOCK_PC requests may transfer data, in which case they must
1225 * a bio attached to them. Or they might contain a SCSI command
1226 * that does not transfer data, in which case they may optionally
1227 * submit a request without an attached bio.
1230 int ret = scsi_init_io(cmd);
1234 BUG_ON(blk_rq_bytes(req));
1236 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1239 cmd->cmd_len = req->cmd_len;
1240 cmd->transfersize = blk_rq_bytes(req);
1241 cmd->allowed = req->retries;
1246 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1247 * that still need to be translated to SCSI CDBs from the ULD.
1249 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1251 struct scsi_cmnd *cmd = req->special;
1253 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1254 int ret = sdev->handler->prep_fn(sdev, req);
1255 if (ret != BLKPREP_OK)
1259 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1260 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1263 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1265 struct scsi_cmnd *cmd = req->special;
1267 if (!blk_rq_bytes(req))
1268 cmd->sc_data_direction = DMA_NONE;
1269 else if (rq_data_dir(req) == WRITE)
1270 cmd->sc_data_direction = DMA_TO_DEVICE;
1272 cmd->sc_data_direction = DMA_FROM_DEVICE;
1274 switch (req->cmd_type) {
1276 return scsi_setup_fs_cmnd(sdev, req);
1277 case REQ_TYPE_BLOCK_PC:
1278 return scsi_setup_blk_pc_cmnd(sdev, req);
1280 return BLKPREP_KILL;
1285 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1287 int ret = BLKPREP_OK;
1290 * If the device is not in running state we will reject some
1293 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1294 switch (sdev->sdev_state) {
1296 case SDEV_TRANSPORT_OFFLINE:
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1302 sdev_printk(KERN_ERR, sdev,
1303 "rejecting I/O to offline device\n");
1308 * If the device is fully deleted, we refuse to
1309 * process any commands as well.
1311 sdev_printk(KERN_ERR, sdev,
1312 "rejecting I/O to dead device\n");
1316 case SDEV_CREATED_BLOCK:
1317 ret = BLKPREP_DEFER;
1321 * If the devices is blocked we defer normal commands.
1323 if (!(req->cmd_flags & REQ_PREEMPT))
1324 ret = BLKPREP_DEFER;
1328 * For any other not fully online state we only allow
1329 * special commands. In particular any user initiated
1330 * command is not allowed.
1332 if (!(req->cmd_flags & REQ_PREEMPT))
1341 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1343 struct scsi_device *sdev = q->queuedata;
1347 case BLKPREP_INVALID:
1348 req->errors = DID_NO_CONNECT << 16;
1349 /* release the command and kill it */
1351 struct scsi_cmnd *cmd = req->special;
1352 scsi_release_buffers(cmd);
1353 scsi_put_command(cmd);
1354 put_device(&sdev->sdev_gendev);
1355 req->special = NULL;
1360 * If we defer, the blk_peek_request() returns NULL, but the
1361 * queue must be restarted, so we schedule a callback to happen
1364 if (atomic_read(&sdev->device_busy) == 0)
1365 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1368 req->cmd_flags |= REQ_DONTPREP;
1374 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1376 struct scsi_device *sdev = q->queuedata;
1377 struct scsi_cmnd *cmd;
1380 ret = scsi_prep_state_check(sdev, req);
1381 if (ret != BLKPREP_OK)
1384 cmd = scsi_get_cmd_from_req(sdev, req);
1385 if (unlikely(!cmd)) {
1386 ret = BLKPREP_DEFER;
1390 ret = scsi_setup_cmnd(sdev, req);
1392 return scsi_prep_return(q, req, ret);
1395 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1397 scsi_uninit_cmd(req->special);
1401 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1404 * Called with the queue_lock held.
1406 static inline int scsi_dev_queue_ready(struct request_queue *q,
1407 struct scsi_device *sdev)
1411 busy = atomic_inc_return(&sdev->device_busy) - 1;
1412 if (atomic_read(&sdev->device_blocked)) {
1417 * unblock after device_blocked iterates to zero
1419 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1421 * For the MQ case we take care of this in the caller.
1424 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1427 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1428 "unblocking device at zero depth\n"));
1431 if (busy >= sdev->queue_depth)
1436 atomic_dec(&sdev->device_busy);
1441 * scsi_target_queue_ready: checks if there we can send commands to target
1442 * @sdev: scsi device on starget to check.
1444 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1445 struct scsi_device *sdev)
1447 struct scsi_target *starget = scsi_target(sdev);
1450 if (starget->single_lun) {
1451 spin_lock_irq(shost->host_lock);
1452 if (starget->starget_sdev_user &&
1453 starget->starget_sdev_user != sdev) {
1454 spin_unlock_irq(shost->host_lock);
1457 starget->starget_sdev_user = sdev;
1458 spin_unlock_irq(shost->host_lock);
1461 if (starget->can_queue <= 0)
1464 busy = atomic_inc_return(&starget->target_busy) - 1;
1465 if (atomic_read(&starget->target_blocked) > 0) {
1470 * unblock after target_blocked iterates to zero
1472 if (atomic_dec_return(&starget->target_blocked) > 0)
1475 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1476 "unblocking target at zero depth\n"));
1479 if (busy >= starget->can_queue)
1485 spin_lock_irq(shost->host_lock);
1486 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1487 spin_unlock_irq(shost->host_lock);
1489 if (starget->can_queue > 0)
1490 atomic_dec(&starget->target_busy);
1495 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1496 * return 0. We must end up running the queue again whenever 0 is
1497 * returned, else IO can hang.
1499 static inline int scsi_host_queue_ready(struct request_queue *q,
1500 struct Scsi_Host *shost,
1501 struct scsi_device *sdev)
1505 if (scsi_host_in_recovery(shost))
1508 busy = atomic_inc_return(&shost->host_busy) - 1;
1509 if (atomic_read(&shost->host_blocked) > 0) {
1514 * unblock after host_blocked iterates to zero
1516 if (atomic_dec_return(&shost->host_blocked) > 0)
1520 shost_printk(KERN_INFO, shost,
1521 "unblocking host at zero depth\n"));
1524 if (shost->can_queue > 0 && busy >= shost->can_queue)
1526 if (shost->host_self_blocked)
1529 /* We're OK to process the command, so we can't be starved */
1530 if (!list_empty(&sdev->starved_entry)) {
1531 spin_lock_irq(shost->host_lock);
1532 if (!list_empty(&sdev->starved_entry))
1533 list_del_init(&sdev->starved_entry);
1534 spin_unlock_irq(shost->host_lock);
1540 spin_lock_irq(shost->host_lock);
1541 if (list_empty(&sdev->starved_entry))
1542 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1543 spin_unlock_irq(shost->host_lock);
1545 atomic_dec(&shost->host_busy);
1550 * Busy state exporting function for request stacking drivers.
1552 * For efficiency, no lock is taken to check the busy state of
1553 * shost/starget/sdev, since the returned value is not guaranteed and
1554 * may be changed after request stacking drivers call the function,
1555 * regardless of taking lock or not.
1557 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1558 * needs to return 'not busy'. Otherwise, request stacking drivers
1559 * may hold requests forever.
1561 static int scsi_lld_busy(struct request_queue *q)
1563 struct scsi_device *sdev = q->queuedata;
1564 struct Scsi_Host *shost;
1566 if (blk_queue_dying(q))
1572 * Ignore host/starget busy state.
1573 * Since block layer does not have a concept of fairness across
1574 * multiple queues, congestion of host/starget needs to be handled
1577 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1584 * Kill a request for a dead device
1586 static void scsi_kill_request(struct request *req, struct request_queue *q)
1588 struct scsi_cmnd *cmd = req->special;
1589 struct scsi_device *sdev;
1590 struct scsi_target *starget;
1591 struct Scsi_Host *shost;
1593 blk_start_request(req);
1595 scmd_printk(KERN_INFO, cmd, "killing request\n");
1598 starget = scsi_target(sdev);
1600 scsi_init_cmd_errh(cmd);
1601 cmd->result = DID_NO_CONNECT << 16;
1602 atomic_inc(&cmd->device->iorequest_cnt);
1605 * SCSI request completion path will do scsi_device_unbusy(),
1606 * bump busy counts. To bump the counters, we need to dance
1607 * with the locks as normal issue path does.
1609 atomic_inc(&sdev->device_busy);
1610 atomic_inc(&shost->host_busy);
1611 if (starget->can_queue > 0)
1612 atomic_inc(&starget->target_busy);
1614 blk_complete_request(req);
1617 static void scsi_softirq_done(struct request *rq)
1619 struct scsi_cmnd *cmd = rq->special;
1620 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1623 INIT_LIST_HEAD(&cmd->eh_entry);
1625 atomic_inc(&cmd->device->iodone_cnt);
1627 atomic_inc(&cmd->device->ioerr_cnt);
1629 disposition = scsi_decide_disposition(cmd);
1630 if (disposition != SUCCESS &&
1631 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1632 sdev_printk(KERN_ERR, cmd->device,
1633 "timing out command, waited %lus\n",
1635 disposition = SUCCESS;
1638 scsi_log_completion(cmd, disposition);
1640 switch (disposition) {
1642 scsi_finish_command(cmd);
1645 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1647 case ADD_TO_MLQUEUE:
1648 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1651 if (!scsi_eh_scmd_add(cmd, 0))
1652 scsi_finish_command(cmd);
1657 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1658 * @cmd: command block we are dispatching.
1660 * Return: nonzero return request was rejected and device's queue needs to be
1663 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1665 struct Scsi_Host *host = cmd->device->host;
1668 atomic_inc(&cmd->device->iorequest_cnt);
1670 /* check if the device is still usable */
1671 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1672 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1673 * returns an immediate error upwards, and signals
1674 * that the device is no longer present */
1675 cmd->result = DID_NO_CONNECT << 16;
1679 /* Check to see if the scsi lld made this device blocked. */
1680 if (unlikely(scsi_device_blocked(cmd->device))) {
1682 * in blocked state, the command is just put back on
1683 * the device queue. The suspend state has already
1684 * blocked the queue so future requests should not
1685 * occur until the device transitions out of the
1688 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1689 "queuecommand : device blocked\n"));
1690 return SCSI_MLQUEUE_DEVICE_BUSY;
1693 /* Store the LUN value in cmnd, if needed. */
1694 if (cmd->device->lun_in_cdb)
1695 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1696 (cmd->device->lun << 5 & 0xe0);
1701 * Before we queue this command, check if the command
1702 * length exceeds what the host adapter can handle.
1704 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1705 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1706 "queuecommand : command too long. "
1707 "cdb_size=%d host->max_cmd_len=%d\n",
1708 cmd->cmd_len, cmd->device->host->max_cmd_len));
1709 cmd->result = (DID_ABORT << 16);
1713 if (unlikely(host->shost_state == SHOST_DEL)) {
1714 cmd->result = (DID_NO_CONNECT << 16);
1719 trace_scsi_dispatch_cmd_start(cmd);
1720 rtn = host->hostt->queuecommand(host, cmd);
1722 trace_scsi_dispatch_cmd_error(cmd, rtn);
1723 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1724 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1725 rtn = SCSI_MLQUEUE_HOST_BUSY;
1727 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1728 "queuecommand : request rejected\n"));
1733 cmd->scsi_done(cmd);
1738 * scsi_done - Invoke completion on finished SCSI command.
1739 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1740 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1742 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1743 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1744 * calls blk_complete_request() for further processing.
1746 * This function is interrupt context safe.
1748 static void scsi_done(struct scsi_cmnd *cmd)
1750 trace_scsi_dispatch_cmd_done(cmd);
1751 blk_complete_request(cmd->request);
1755 * Function: scsi_request_fn()
1757 * Purpose: Main strategy routine for SCSI.
1759 * Arguments: q - Pointer to actual queue.
1763 * Lock status: IO request lock assumed to be held when called.
1765 static void scsi_request_fn(struct request_queue *q)
1766 __releases(q->queue_lock)
1767 __acquires(q->queue_lock)
1769 struct scsi_device *sdev = q->queuedata;
1770 struct Scsi_Host *shost;
1771 struct scsi_cmnd *cmd;
1772 struct request *req;
1775 * To start with, we keep looping until the queue is empty, or until
1776 * the host is no longer able to accept any more requests.
1782 * get next queueable request. We do this early to make sure
1783 * that the request is fully prepared even if we cannot
1786 req = blk_peek_request(q);
1790 if (unlikely(!scsi_device_online(sdev))) {
1791 sdev_printk(KERN_ERR, sdev,
1792 "rejecting I/O to offline device\n");
1793 scsi_kill_request(req, q);
1797 if (!scsi_dev_queue_ready(q, sdev))
1801 * Remove the request from the request list.
1803 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1804 blk_start_request(req);
1806 spin_unlock_irq(q->queue_lock);
1808 if (unlikely(cmd == NULL)) {
1809 printk(KERN_CRIT "impossible request in %s.\n"
1810 "please mail a stack trace to "
1811 "linux-scsi@vger.kernel.org\n",
1813 blk_dump_rq_flags(req, "foo");
1818 * We hit this when the driver is using a host wide
1819 * tag map. For device level tag maps the queue_depth check
1820 * in the device ready fn would prevent us from trying
1821 * to allocate a tag. Since the map is a shared host resource
1822 * we add the dev to the starved list so it eventually gets
1823 * a run when a tag is freed.
1825 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1826 spin_lock_irq(shost->host_lock);
1827 if (list_empty(&sdev->starved_entry))
1828 list_add_tail(&sdev->starved_entry,
1829 &shost->starved_list);
1830 spin_unlock_irq(shost->host_lock);
1834 if (!scsi_target_queue_ready(shost, sdev))
1837 if (!scsi_host_queue_ready(q, shost, sdev))
1838 goto host_not_ready;
1840 if (sdev->simple_tags)
1841 cmd->flags |= SCMD_TAGGED;
1843 cmd->flags &= ~SCMD_TAGGED;
1846 * Finally, initialize any error handling parameters, and set up
1847 * the timers for timeouts.
1849 scsi_init_cmd_errh(cmd);
1852 * Dispatch the command to the low-level driver.
1854 cmd->scsi_done = scsi_done;
1855 rtn = scsi_dispatch_cmd(cmd);
1857 scsi_queue_insert(cmd, rtn);
1858 spin_lock_irq(q->queue_lock);
1861 spin_lock_irq(q->queue_lock);
1867 if (scsi_target(sdev)->can_queue > 0)
1868 atomic_dec(&scsi_target(sdev)->target_busy);
1871 * lock q, handle tag, requeue req, and decrement device_busy. We
1872 * must return with queue_lock held.
1874 * Decrementing device_busy without checking it is OK, as all such
1875 * cases (host limits or settings) should run the queue at some
1878 spin_lock_irq(q->queue_lock);
1879 blk_requeue_request(q, req);
1880 atomic_dec(&sdev->device_busy);
1882 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1883 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1886 static inline int prep_to_mq(int ret)
1892 return BLK_MQ_RQ_QUEUE_BUSY;
1894 return BLK_MQ_RQ_QUEUE_ERROR;
1898 static int scsi_mq_prep_fn(struct request *req)
1900 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1901 struct scsi_device *sdev = req->q->queuedata;
1902 struct Scsi_Host *shost = sdev->host;
1903 unsigned char *sense_buf = cmd->sense_buffer;
1904 struct scatterlist *sg;
1906 memset(cmd, 0, sizeof(struct scsi_cmnd));
1912 cmd->sense_buffer = sense_buf;
1914 cmd->tag = req->tag;
1916 cmd->cmnd = req->cmd;
1917 cmd->prot_op = SCSI_PROT_NORMAL;
1919 INIT_LIST_HEAD(&cmd->list);
1920 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1921 cmd->jiffies_at_alloc = jiffies;
1923 if (shost->use_cmd_list) {
1924 spin_lock_irq(&sdev->list_lock);
1925 list_add_tail(&cmd->list, &sdev->cmd_list);
1926 spin_unlock_irq(&sdev->list_lock);
1929 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1930 cmd->sdb.table.sgl = sg;
1932 if (scsi_host_get_prot(shost)) {
1933 cmd->prot_sdb = (void *)sg +
1935 shost->sg_tablesize, SCSI_MAX_SG_SEGMENTS) *
1936 sizeof(struct scatterlist);
1937 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1939 cmd->prot_sdb->table.sgl =
1940 (struct scatterlist *)(cmd->prot_sdb + 1);
1943 if (blk_bidi_rq(req)) {
1944 struct request *next_rq = req->next_rq;
1945 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1947 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1948 bidi_sdb->table.sgl =
1949 (struct scatterlist *)(bidi_sdb + 1);
1951 next_rq->special = bidi_sdb;
1954 blk_mq_start_request(req);
1956 return scsi_setup_cmnd(sdev, req);
1959 static void scsi_mq_done(struct scsi_cmnd *cmd)
1961 trace_scsi_dispatch_cmd_done(cmd);
1962 blk_mq_complete_request(cmd->request, cmd->request->errors);
1965 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1966 const struct blk_mq_queue_data *bd)
1968 struct request *req = bd->rq;
1969 struct request_queue *q = req->q;
1970 struct scsi_device *sdev = q->queuedata;
1971 struct Scsi_Host *shost = sdev->host;
1972 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1976 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1980 ret = BLK_MQ_RQ_QUEUE_BUSY;
1981 if (!get_device(&sdev->sdev_gendev))
1984 if (!scsi_dev_queue_ready(q, sdev))
1985 goto out_put_device;
1986 if (!scsi_target_queue_ready(shost, sdev))
1987 goto out_dec_device_busy;
1988 if (!scsi_host_queue_ready(q, shost, sdev))
1989 goto out_dec_target_busy;
1992 if (!(req->cmd_flags & REQ_DONTPREP)) {
1993 ret = prep_to_mq(scsi_mq_prep_fn(req));
1995 goto out_dec_host_busy;
1996 req->cmd_flags |= REQ_DONTPREP;
1998 blk_mq_start_request(req);
2001 if (sdev->simple_tags)
2002 cmd->flags |= SCMD_TAGGED;
2004 cmd->flags &= ~SCMD_TAGGED;
2006 scsi_init_cmd_errh(cmd);
2007 cmd->scsi_done = scsi_mq_done;
2009 reason = scsi_dispatch_cmd(cmd);
2011 scsi_set_blocked(cmd, reason);
2012 ret = BLK_MQ_RQ_QUEUE_BUSY;
2013 goto out_dec_host_busy;
2016 return BLK_MQ_RQ_QUEUE_OK;
2019 atomic_dec(&shost->host_busy);
2020 out_dec_target_busy:
2021 if (scsi_target(sdev)->can_queue > 0)
2022 atomic_dec(&scsi_target(sdev)->target_busy);
2023 out_dec_device_busy:
2024 atomic_dec(&sdev->device_busy);
2026 put_device(&sdev->sdev_gendev);
2029 case BLK_MQ_RQ_QUEUE_BUSY:
2030 blk_mq_stop_hw_queue(hctx);
2031 if (atomic_read(&sdev->device_busy) == 0 &&
2032 !scsi_device_blocked(sdev))
2033 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2035 case BLK_MQ_RQ_QUEUE_ERROR:
2037 * Make sure to release all allocated ressources when
2038 * we hit an error, as we will never see this command
2041 if (req->cmd_flags & REQ_DONTPREP)
2042 scsi_mq_uninit_cmd(cmd);
2050 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2054 return BLK_EH_RESET_TIMER;
2055 return scsi_times_out(req);
2058 static int scsi_init_request(void *data, struct request *rq,
2059 unsigned int hctx_idx, unsigned int request_idx,
2060 unsigned int numa_node)
2062 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2064 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
2066 if (!cmd->sense_buffer)
2071 static void scsi_exit_request(void *data, struct request *rq,
2072 unsigned int hctx_idx, unsigned int request_idx)
2074 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2076 kfree(cmd->sense_buffer);
2079 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2081 struct device *host_dev;
2082 u64 bounce_limit = 0xffffffff;
2084 if (shost->unchecked_isa_dma)
2085 return BLK_BOUNCE_ISA;
2087 * Platforms with virtual-DMA translation
2088 * hardware have no practical limit.
2090 if (!PCI_DMA_BUS_IS_PHYS)
2091 return BLK_BOUNCE_ANY;
2093 host_dev = scsi_get_device(shost);
2094 if (host_dev && host_dev->dma_mask)
2095 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2097 return bounce_limit;
2100 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2102 struct device *dev = shost->dma_dev;
2105 * this limit is imposed by hardware restrictions
2107 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2108 SCSI_MAX_SG_CHAIN_SEGMENTS));
2110 if (scsi_host_prot_dma(shost)) {
2111 shost->sg_prot_tablesize =
2112 min_not_zero(shost->sg_prot_tablesize,
2113 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2114 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2115 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2118 blk_queue_max_hw_sectors(q, shost->max_sectors);
2119 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2120 blk_queue_segment_boundary(q, shost->dma_boundary);
2121 dma_set_seg_boundary(dev, shost->dma_boundary);
2123 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2125 if (!shost->use_clustering)
2126 q->limits.cluster = 0;
2129 * set a reasonable default alignment on word boundaries: the
2130 * host and device may alter it using
2131 * blk_queue_update_dma_alignment() later.
2133 blk_queue_dma_alignment(q, 0x03);
2136 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2137 request_fn_proc *request_fn)
2139 struct request_queue *q;
2141 q = blk_init_queue(request_fn, NULL);
2144 __scsi_init_queue(shost, q);
2147 EXPORT_SYMBOL(__scsi_alloc_queue);
2149 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2151 struct request_queue *q;
2153 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2157 blk_queue_prep_rq(q, scsi_prep_fn);
2158 blk_queue_unprep_rq(q, scsi_unprep_fn);
2159 blk_queue_softirq_done(q, scsi_softirq_done);
2160 blk_queue_rq_timed_out(q, scsi_times_out);
2161 blk_queue_lld_busy(q, scsi_lld_busy);
2165 static struct blk_mq_ops scsi_mq_ops = {
2166 .map_queue = blk_mq_map_queue,
2167 .queue_rq = scsi_queue_rq,
2168 .complete = scsi_softirq_done,
2169 .timeout = scsi_timeout,
2170 .init_request = scsi_init_request,
2171 .exit_request = scsi_exit_request,
2174 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2176 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2177 if (IS_ERR(sdev->request_queue))
2180 sdev->request_queue->queuedata = sdev;
2181 __scsi_init_queue(sdev->host, sdev->request_queue);
2182 return sdev->request_queue;
2185 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2187 unsigned int cmd_size, sgl_size, tbl_size;
2189 tbl_size = shost->sg_tablesize;
2190 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2191 tbl_size = SCSI_MAX_SG_SEGMENTS;
2192 sgl_size = tbl_size * sizeof(struct scatterlist);
2193 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2194 if (scsi_host_get_prot(shost))
2195 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2197 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2198 shost->tag_set.ops = &scsi_mq_ops;
2199 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2200 shost->tag_set.queue_depth = shost->can_queue;
2201 shost->tag_set.cmd_size = cmd_size;
2202 shost->tag_set.numa_node = NUMA_NO_NODE;
2203 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2204 shost->tag_set.flags |=
2205 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2206 shost->tag_set.driver_data = shost;
2208 return blk_mq_alloc_tag_set(&shost->tag_set);
2211 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2213 blk_mq_free_tag_set(&shost->tag_set);
2217 * Function: scsi_block_requests()
2219 * Purpose: Utility function used by low-level drivers to prevent further
2220 * commands from being queued to the device.
2222 * Arguments: shost - Host in question
2226 * Lock status: No locks are assumed held.
2228 * Notes: There is no timer nor any other means by which the requests
2229 * get unblocked other than the low-level driver calling
2230 * scsi_unblock_requests().
2232 void scsi_block_requests(struct Scsi_Host *shost)
2234 shost->host_self_blocked = 1;
2236 EXPORT_SYMBOL(scsi_block_requests);
2239 * Function: scsi_unblock_requests()
2241 * Purpose: Utility function used by low-level drivers to allow further
2242 * commands from being queued to the device.
2244 * Arguments: shost - Host in question
2248 * Lock status: No locks are assumed held.
2250 * Notes: There is no timer nor any other means by which the requests
2251 * get unblocked other than the low-level driver calling
2252 * scsi_unblock_requests().
2254 * This is done as an API function so that changes to the
2255 * internals of the scsi mid-layer won't require wholesale
2256 * changes to drivers that use this feature.
2258 void scsi_unblock_requests(struct Scsi_Host *shost)
2260 shost->host_self_blocked = 0;
2261 scsi_run_host_queues(shost);
2263 EXPORT_SYMBOL(scsi_unblock_requests);
2265 int __init scsi_init_queue(void)
2269 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2270 sizeof(struct scsi_data_buffer),
2272 if (!scsi_sdb_cache) {
2273 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2277 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2278 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2279 int size = sgp->size * sizeof(struct scatterlist);
2281 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2282 SLAB_HWCACHE_ALIGN, NULL);
2284 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2289 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2292 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2301 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2302 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2304 mempool_destroy(sgp->pool);
2306 kmem_cache_destroy(sgp->slab);
2308 kmem_cache_destroy(scsi_sdb_cache);
2313 void scsi_exit_queue(void)
2317 kmem_cache_destroy(scsi_sdb_cache);
2319 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2320 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2321 mempool_destroy(sgp->pool);
2322 kmem_cache_destroy(sgp->slab);
2327 * scsi_mode_select - issue a mode select
2328 * @sdev: SCSI device to be queried
2329 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2330 * @sp: Save page bit (0 == don't save, 1 == save)
2331 * @modepage: mode page being requested
2332 * @buffer: request buffer (may not be smaller than eight bytes)
2333 * @len: length of request buffer.
2334 * @timeout: command timeout
2335 * @retries: number of retries before failing
2336 * @data: returns a structure abstracting the mode header data
2337 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2338 * must be SCSI_SENSE_BUFFERSIZE big.
2340 * Returns zero if successful; negative error number or scsi
2345 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2346 unsigned char *buffer, int len, int timeout, int retries,
2347 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2349 unsigned char cmd[10];
2350 unsigned char *real_buffer;
2353 memset(cmd, 0, sizeof(cmd));
2354 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2356 if (sdev->use_10_for_ms) {
2359 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2362 memcpy(real_buffer + 8, buffer, len);
2366 real_buffer[2] = data->medium_type;
2367 real_buffer[3] = data->device_specific;
2368 real_buffer[4] = data->longlba ? 0x01 : 0;
2370 real_buffer[6] = data->block_descriptor_length >> 8;
2371 real_buffer[7] = data->block_descriptor_length;
2373 cmd[0] = MODE_SELECT_10;
2377 if (len > 255 || data->block_descriptor_length > 255 ||
2381 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2384 memcpy(real_buffer + 4, buffer, len);
2387 real_buffer[1] = data->medium_type;
2388 real_buffer[2] = data->device_specific;
2389 real_buffer[3] = data->block_descriptor_length;
2392 cmd[0] = MODE_SELECT;
2396 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2397 sshdr, timeout, retries, NULL);
2401 EXPORT_SYMBOL_GPL(scsi_mode_select);
2404 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2405 * @sdev: SCSI device to be queried
2406 * @dbd: set if mode sense will allow block descriptors to be returned
2407 * @modepage: mode page being requested
2408 * @buffer: request buffer (may not be smaller than eight bytes)
2409 * @len: length of request buffer.
2410 * @timeout: command timeout
2411 * @retries: number of retries before failing
2412 * @data: returns a structure abstracting the mode header data
2413 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2414 * must be SCSI_SENSE_BUFFERSIZE big.
2416 * Returns zero if unsuccessful, or the header offset (either 4
2417 * or 8 depending on whether a six or ten byte command was
2418 * issued) if successful.
2421 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2422 unsigned char *buffer, int len, int timeout, int retries,
2423 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2425 unsigned char cmd[12];
2428 int result, retry_count = retries;
2429 struct scsi_sense_hdr my_sshdr;
2431 memset(data, 0, sizeof(*data));
2432 memset(&cmd[0], 0, 12);
2433 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2436 /* caller might not be interested in sense, but we need it */
2441 use_10_for_ms = sdev->use_10_for_ms;
2443 if (use_10_for_ms) {
2447 cmd[0] = MODE_SENSE_10;
2454 cmd[0] = MODE_SENSE;
2459 memset(buffer, 0, len);
2461 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2462 sshdr, timeout, retries, NULL);
2464 /* This code looks awful: what it's doing is making sure an
2465 * ILLEGAL REQUEST sense return identifies the actual command
2466 * byte as the problem. MODE_SENSE commands can return
2467 * ILLEGAL REQUEST if the code page isn't supported */
2469 if (use_10_for_ms && !scsi_status_is_good(result) &&
2470 (driver_byte(result) & DRIVER_SENSE)) {
2471 if (scsi_sense_valid(sshdr)) {
2472 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2473 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2475 * Invalid command operation code
2477 sdev->use_10_for_ms = 0;
2483 if(scsi_status_is_good(result)) {
2484 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2485 (modepage == 6 || modepage == 8))) {
2486 /* Initio breakage? */
2489 data->medium_type = 0;
2490 data->device_specific = 0;
2492 data->block_descriptor_length = 0;
2493 } else if(use_10_for_ms) {
2494 data->length = buffer[0]*256 + buffer[1] + 2;
2495 data->medium_type = buffer[2];
2496 data->device_specific = buffer[3];
2497 data->longlba = buffer[4] & 0x01;
2498 data->block_descriptor_length = buffer[6]*256
2501 data->length = buffer[0] + 1;
2502 data->medium_type = buffer[1];
2503 data->device_specific = buffer[2];
2504 data->block_descriptor_length = buffer[3];
2506 data->header_length = header_length;
2507 } else if ((status_byte(result) == CHECK_CONDITION) &&
2508 scsi_sense_valid(sshdr) &&
2509 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2516 EXPORT_SYMBOL(scsi_mode_sense);
2519 * scsi_test_unit_ready - test if unit is ready
2520 * @sdev: scsi device to change the state of.
2521 * @timeout: command timeout
2522 * @retries: number of retries before failing
2523 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2524 * returning sense. Make sure that this is cleared before passing
2527 * Returns zero if unsuccessful or an error if TUR failed. For
2528 * removable media, UNIT_ATTENTION sets ->changed flag.
2531 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2532 struct scsi_sense_hdr *sshdr_external)
2535 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2537 struct scsi_sense_hdr *sshdr;
2540 if (!sshdr_external)
2541 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2543 sshdr = sshdr_external;
2545 /* try to eat the UNIT_ATTENTION if there are enough retries */
2547 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2548 timeout, retries, NULL);
2549 if (sdev->removable && scsi_sense_valid(sshdr) &&
2550 sshdr->sense_key == UNIT_ATTENTION)
2552 } while (scsi_sense_valid(sshdr) &&
2553 sshdr->sense_key == UNIT_ATTENTION && --retries);
2555 if (!sshdr_external)
2559 EXPORT_SYMBOL(scsi_test_unit_ready);
2562 * scsi_device_set_state - Take the given device through the device state model.
2563 * @sdev: scsi device to change the state of.
2564 * @state: state to change to.
2566 * Returns zero if unsuccessful or an error if the requested
2567 * transition is illegal.
2570 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2572 enum scsi_device_state oldstate = sdev->sdev_state;
2574 if (state == oldstate)
2580 case SDEV_CREATED_BLOCK:
2591 case SDEV_TRANSPORT_OFFLINE:
2604 case SDEV_TRANSPORT_OFFLINE:
2612 case SDEV_TRANSPORT_OFFLINE:
2627 case SDEV_CREATED_BLOCK:
2634 case SDEV_CREATED_BLOCK:
2649 case SDEV_TRANSPORT_OFFLINE:
2662 case SDEV_TRANSPORT_OFFLINE:
2664 case SDEV_CREATED_BLOCK:
2672 sdev->sdev_state = state;
2676 SCSI_LOG_ERROR_RECOVERY(1,
2677 sdev_printk(KERN_ERR, sdev,
2678 "Illegal state transition %s->%s",
2679 scsi_device_state_name(oldstate),
2680 scsi_device_state_name(state))
2684 EXPORT_SYMBOL(scsi_device_set_state);
2687 * sdev_evt_emit - emit a single SCSI device uevent
2688 * @sdev: associated SCSI device
2689 * @evt: event to emit
2691 * Send a single uevent (scsi_event) to the associated scsi_device.
2693 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2698 switch (evt->evt_type) {
2699 case SDEV_EVT_MEDIA_CHANGE:
2700 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2702 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2703 scsi_rescan_device(&sdev->sdev_gendev);
2704 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2706 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2707 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2709 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2710 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2712 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2713 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2715 case SDEV_EVT_LUN_CHANGE_REPORTED:
2716 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2718 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2719 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2728 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2732 * sdev_evt_thread - send a uevent for each scsi event
2733 * @work: work struct for scsi_device
2735 * Dispatch queued events to their associated scsi_device kobjects
2738 void scsi_evt_thread(struct work_struct *work)
2740 struct scsi_device *sdev;
2741 enum scsi_device_event evt_type;
2742 LIST_HEAD(event_list);
2744 sdev = container_of(work, struct scsi_device, event_work);
2746 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2747 if (test_and_clear_bit(evt_type, sdev->pending_events))
2748 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2751 struct scsi_event *evt;
2752 struct list_head *this, *tmp;
2753 unsigned long flags;
2755 spin_lock_irqsave(&sdev->list_lock, flags);
2756 list_splice_init(&sdev->event_list, &event_list);
2757 spin_unlock_irqrestore(&sdev->list_lock, flags);
2759 if (list_empty(&event_list))
2762 list_for_each_safe(this, tmp, &event_list) {
2763 evt = list_entry(this, struct scsi_event, node);
2764 list_del(&evt->node);
2765 scsi_evt_emit(sdev, evt);
2772 * sdev_evt_send - send asserted event to uevent thread
2773 * @sdev: scsi_device event occurred on
2774 * @evt: event to send
2776 * Assert scsi device event asynchronously.
2778 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2780 unsigned long flags;
2783 /* FIXME: currently this check eliminates all media change events
2784 * for polled devices. Need to update to discriminate between AN
2785 * and polled events */
2786 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2792 spin_lock_irqsave(&sdev->list_lock, flags);
2793 list_add_tail(&evt->node, &sdev->event_list);
2794 schedule_work(&sdev->event_work);
2795 spin_unlock_irqrestore(&sdev->list_lock, flags);
2797 EXPORT_SYMBOL_GPL(sdev_evt_send);
2800 * sdev_evt_alloc - allocate a new scsi event
2801 * @evt_type: type of event to allocate
2802 * @gfpflags: GFP flags for allocation
2804 * Allocates and returns a new scsi_event.
2806 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2809 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2813 evt->evt_type = evt_type;
2814 INIT_LIST_HEAD(&evt->node);
2816 /* evt_type-specific initialization, if any */
2818 case SDEV_EVT_MEDIA_CHANGE:
2819 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2820 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2821 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2822 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2823 case SDEV_EVT_LUN_CHANGE_REPORTED:
2824 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2832 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2835 * sdev_evt_send_simple - send asserted event to uevent thread
2836 * @sdev: scsi_device event occurred on
2837 * @evt_type: type of event to send
2838 * @gfpflags: GFP flags for allocation
2840 * Assert scsi device event asynchronously, given an event type.
2842 void sdev_evt_send_simple(struct scsi_device *sdev,
2843 enum scsi_device_event evt_type, gfp_t gfpflags)
2845 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2847 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2852 sdev_evt_send(sdev, evt);
2854 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
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
2939 * Block request made by scsi lld's to temporarily stop all
2940 * scsi commands on the specified device. Called from interrupt
2941 * or normal process context.
2943 * Returns zero if successful or error if not
2946 * This routine transitions the device to the SDEV_BLOCK state
2947 * (which must be a legal transition). When the device is in this
2948 * state, all commands are deferred until the scsi lld reenables
2949 * the device with scsi_device_unblock or device_block_tmo fires.
2952 scsi_internal_device_block(struct scsi_device *sdev)
2954 struct request_queue *q = sdev->request_queue;
2955 unsigned long flags;
2958 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2960 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2967 * The device has transitioned to SDEV_BLOCK. Stop the
2968 * block layer from calling the midlayer with this device's
2972 blk_mq_stop_hw_queues(q);
2974 spin_lock_irqsave(q->queue_lock, flags);
2976 spin_unlock_irqrestore(q->queue_lock, flags);
2981 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2984 * scsi_internal_device_unblock - resume a device after a block request
2985 * @sdev: device to resume
2986 * @new_state: state to set devices to after unblocking
2988 * Called by scsi lld's or the midlayer to restart the device queue
2989 * for the previously suspended scsi device. Called from interrupt or
2990 * normal process context.
2992 * Returns zero if successful or error if not.
2995 * This routine transitions the device to the SDEV_RUNNING state
2996 * or to one of the offline states (which must be a legal transition)
2997 * allowing the midlayer to goose the queue for this device.
3000 scsi_internal_device_unblock(struct scsi_device *sdev,
3001 enum scsi_device_state new_state)
3003 struct request_queue *q = sdev->request_queue;
3004 unsigned long flags;
3007 * Try to transition the scsi device to SDEV_RUNNING or one of the
3008 * offlined states and goose the device queue if successful.
3010 if ((sdev->sdev_state == SDEV_BLOCK) ||
3011 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3012 sdev->sdev_state = new_state;
3013 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3014 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3015 new_state == SDEV_OFFLINE)
3016 sdev->sdev_state = new_state;
3018 sdev->sdev_state = SDEV_CREATED;
3019 } else if (sdev->sdev_state != SDEV_CANCEL &&
3020 sdev->sdev_state != SDEV_OFFLINE)
3024 blk_mq_start_stopped_hw_queues(q, false);
3026 spin_lock_irqsave(q->queue_lock, flags);
3028 spin_unlock_irqrestore(q->queue_lock, flags);
3033 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3036 device_block(struct scsi_device *sdev, void *data)
3038 scsi_internal_device_block(sdev);
3042 target_block(struct device *dev, void *data)
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), NULL,
3051 scsi_target_block(struct device *dev)
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), NULL,
3057 device_for_each_child(dev, NULL, target_block);
3059 EXPORT_SYMBOL_GPL(scsi_target_block);
3062 device_unblock(struct scsi_device *sdev, void *data)
3064 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3068 target_unblock(struct device *dev, void *data)
3070 if (scsi_is_target_device(dev))
3071 starget_for_each_device(to_scsi_target(dev), data,
3077 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3079 if (scsi_is_target_device(dev))
3080 starget_for_each_device(to_scsi_target(dev), &new_state,
3083 device_for_each_child(dev, &new_state, target_unblock);
3085 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3088 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3089 * @sgl: scatter-gather list
3090 * @sg_count: number of segments in sg
3091 * @offset: offset in bytes into sg, on return offset into the mapped area
3092 * @len: bytes to map, on return number of bytes mapped
3094 * Returns virtual address of the start of the mapped page
3096 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3097 size_t *offset, size_t *len)
3100 size_t sg_len = 0, len_complete = 0;
3101 struct scatterlist *sg;
3104 WARN_ON(!irqs_disabled());
3106 for_each_sg(sgl, sg, sg_count, i) {
3107 len_complete = sg_len; /* Complete sg-entries */
3108 sg_len += sg->length;
3109 if (sg_len > *offset)
3113 if (unlikely(i == sg_count)) {
3114 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3116 __func__, sg_len, *offset, sg_count);
3121 /* Offset starting from the beginning of first page in this sg-entry */
3122 *offset = *offset - len_complete + sg->offset;
3124 /* Assumption: contiguous pages can be accessed as "page + i" */
3125 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3126 *offset &= ~PAGE_MASK;
3128 /* Bytes in this sg-entry from *offset to the end of the page */
3129 sg_len = PAGE_SIZE - *offset;
3133 return kmap_atomic(page);
3135 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3138 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3139 * @virt: virtual address to be unmapped
3141 void scsi_kunmap_atomic_sg(void *virt)
3143 kunmap_atomic(virt);
3145 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3147 void sdev_disable_disk_events(struct scsi_device *sdev)
3149 atomic_inc(&sdev->disk_events_disable_depth);
3151 EXPORT_SYMBOL(sdev_disable_disk_events);
3153 void sdev_enable_disk_events(struct scsi_device *sdev)
3155 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3157 atomic_dec(&sdev->disk_events_disable_depth);
3159 EXPORT_SYMBOL(sdev_enable_disk_events);
3162 * scsi_vpd_lun_id - return a unique device identification
3163 * @sdev: SCSI device
3164 * @id: buffer for the identification
3165 * @id_len: length of the buffer
3167 * Copies a unique device identification into @id based
3168 * on the information in the VPD page 0x83 of the device.
3169 * The string will be formatted as a SCSI name string.
3171 * Returns the length of the identification or error on failure.
3172 * If the identifier is longer than the supplied buffer the actual
3173 * identifier length is returned and the buffer is not zero-padded.
3175 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3177 u8 cur_id_type = 0xff;
3179 unsigned char *d, *cur_id_str;
3180 unsigned char __rcu *vpd_pg83;
3181 int id_size = -EINVAL;
3184 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3191 * Look for the correct descriptor.
3192 * Order of preference for lun descriptor:
3193 * - SCSI name string
3194 * - NAA IEEE Registered Extended
3195 * - EUI-64 based 16-byte
3196 * - EUI-64 based 12-byte
3197 * - NAA IEEE Registered
3198 * - NAA IEEE Extended
3199 * as longer descriptors reduce the likelyhood
3200 * of identification clashes.
3203 /* The id string must be at least 20 bytes + terminating NULL byte */
3209 memset(id, 0, id_len);
3211 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3212 /* Skip designators not referring to the LUN */
3213 if ((d[1] & 0x30) != 0x00)
3216 switch (d[1] & 0xf) {
3219 if (cur_id_size > d[3])
3221 /* Prefer NAA IEEE Registered Extended */
3222 if (cur_id_type == 0x3 &&
3223 cur_id_size == d[3])
3227 cur_id_type = d[1] & 0xf;
3228 switch (cur_id_size) {
3230 id_size = snprintf(id, id_len,
3235 id_size = snprintf(id, id_len,
3240 id_size = snprintf(id, id_len,
3251 if (cur_id_size > d[3])
3255 cur_id_type = d[1] & 0xf;
3256 switch (cur_id_size) {
3258 id_size = snprintf(id, id_len,
3263 id_size = snprintf(id, id_len,
3273 /* SCSI name string */
3274 if (cur_id_size + 4 > d[3])
3276 /* Prefer others for truncated descriptor */
3277 if (cur_id_size && d[3] > id_len)
3279 cur_id_size = id_size = d[3];
3281 cur_id_type = d[1] & 0xf;
3282 if (cur_id_size >= id_len)
3283 cur_id_size = id_len - 1;
3284 memcpy(id, cur_id_str, cur_id_size);
3285 /* Decrease priority for truncated descriptor */
3286 if (cur_id_size != id_size)
3299 EXPORT_SYMBOL(scsi_vpd_lun_id);
3302 * scsi_vpd_tpg_id - return a target port group identifier
3303 * @sdev: SCSI device
3305 * Returns the Target Port Group identifier from the information
3306 * froom VPD page 0x83 of the device.
3308 * Returns the identifier or error on failure.
3310 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3313 unsigned char __rcu *vpd_pg83;
3314 int group_id = -EAGAIN, rel_port = -1;
3317 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3323 d = sdev->vpd_pg83 + 4;
3324 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3325 switch (d[1] & 0xf) {
3327 /* Relative target port */
3328 rel_port = get_unaligned_be16(&d[6]);
3331 /* Target port group */
3332 group_id = get_unaligned_be16(&d[6]);
3341 if (group_id >= 0 && rel_id && rel_port != -1)
3346 EXPORT_SYMBOL(scsi_vpd_tpg_id);