2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
34 #include <scsi/scsi_dh.h>
36 #include <trace/events/scsi.h>
38 #include "scsi_debugfs.h"
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
42 static struct kmem_cache *scsi_sdb_cache;
43 static struct kmem_cache *scsi_sense_cache;
44 static struct kmem_cache *scsi_sense_isadma_cache;
45 static DEFINE_MUTEX(scsi_sense_cache_mutex);
47 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
49 static inline struct kmem_cache *
50 scsi_select_sense_cache(bool unchecked_isa_dma)
52 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
55 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56 unsigned char *sense_buffer)
58 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
62 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
63 gfp_t gfp_mask, int numa_node)
65 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
69 int scsi_init_sense_cache(struct Scsi_Host *shost)
71 struct kmem_cache *cache;
74 mutex_lock(&scsi_sense_cache_mutex);
75 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
79 if (shost->unchecked_isa_dma) {
80 scsi_sense_isadma_cache =
81 kmem_cache_create("scsi_sense_cache(DMA)",
82 SCSI_SENSE_BUFFERSIZE, 0,
83 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
84 if (!scsi_sense_isadma_cache)
88 kmem_cache_create_usercopy("scsi_sense_cache",
89 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
90 0, SCSI_SENSE_BUFFERSIZE, NULL);
91 if (!scsi_sense_cache)
95 mutex_unlock(&scsi_sense_cache_mutex);
100 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
101 * not change behaviour from the previous unplug mechanism, experimentation
102 * may prove this needs changing.
104 #define SCSI_QUEUE_DELAY 3
107 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
109 struct Scsi_Host *host = cmd->device->host;
110 struct scsi_device *device = cmd->device;
111 struct scsi_target *starget = scsi_target(device);
114 * Set the appropriate busy bit for the device/host.
116 * If the host/device isn't busy, assume that something actually
117 * completed, and that we should be able to queue a command now.
119 * Note that the prior mid-layer assumption that any host could
120 * always queue at least one command is now broken. The mid-layer
121 * will implement a user specifiable stall (see
122 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
123 * if a command is requeued with no other commands outstanding
124 * either for the device or for the host.
127 case SCSI_MLQUEUE_HOST_BUSY:
128 atomic_set(&host->host_blocked, host->max_host_blocked);
130 case SCSI_MLQUEUE_DEVICE_BUSY:
131 case SCSI_MLQUEUE_EH_RETRY:
132 atomic_set(&device->device_blocked,
133 device->max_device_blocked);
135 case SCSI_MLQUEUE_TARGET_BUSY:
136 atomic_set(&starget->target_blocked,
137 starget->max_target_blocked);
142 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
144 struct scsi_device *sdev = cmd->device;
146 if (cmd->request->rq_flags & RQF_DONTPREP) {
147 cmd->request->rq_flags &= ~RQF_DONTPREP;
148 scsi_mq_uninit_cmd(cmd);
152 blk_mq_requeue_request(cmd->request, true);
153 put_device(&sdev->sdev_gendev);
157 * __scsi_queue_insert - private queue insertion
158 * @cmd: The SCSI command being requeued
159 * @reason: The reason for the requeue
160 * @unbusy: Whether the queue should be unbusied
162 * This is a private queue insertion. The public interface
163 * scsi_queue_insert() always assumes the queue should be unbusied
164 * because it's always called before the completion. This function is
165 * for a requeue after completion, which should only occur in this
168 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
170 struct scsi_device *device = cmd->device;
171 struct request_queue *q = device->request_queue;
174 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
175 "Inserting command %p into mlqueue\n", cmd));
177 scsi_set_blocked(cmd, reason);
180 * Decrement the counters, since these commands are no longer
181 * active on the host/device.
184 scsi_device_unbusy(device);
187 * Requeue this command. It will go before all other commands
188 * that are already in the queue. Schedule requeue work under
189 * lock such that the kblockd_schedule_work() call happens
190 * before blk_cleanup_queue() finishes.
195 * Before a SCSI command is dispatched,
196 * get_device(&sdev->sdev_gendev) is called and the host,
197 * target and device busy counters are increased. Since
198 * requeuing a request causes these actions to be repeated and
199 * since scsi_device_unbusy() has already been called,
200 * put_device(&device->sdev_gendev) must still be called. Call
201 * put_device() after blk_mq_requeue_request() to avoid that
202 * removal of the SCSI device can start before requeueing has
205 blk_mq_requeue_request(cmd->request, true);
206 put_device(&device->sdev_gendev);
209 spin_lock_irqsave(q->queue_lock, flags);
210 blk_requeue_request(q, cmd->request);
211 kblockd_schedule_work(&device->requeue_work);
212 spin_unlock_irqrestore(q->queue_lock, flags);
216 * Function: scsi_queue_insert()
218 * Purpose: Insert a command in the midlevel queue.
220 * Arguments: cmd - command that we are adding to queue.
221 * reason - why we are inserting command to queue.
223 * Lock status: Assumed that lock is not held upon entry.
227 * Notes: We do this for one of two cases. Either the host is busy
228 * and it cannot accept any more commands for the time being,
229 * or the device returned QUEUE_FULL and can accept no more
231 * Notes: This could be called either from an interrupt context or a
232 * normal process context.
234 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
236 __scsi_queue_insert(cmd, reason, true);
241 * __scsi_execute - insert request and wait for the result
244 * @data_direction: data direction
245 * @buffer: data buffer
246 * @bufflen: len of buffer
247 * @sense: optional sense buffer
248 * @sshdr: optional decoded sense header
249 * @timeout: request timeout in seconds
250 * @retries: number of times to retry request
251 * @flags: flags for ->cmd_flags
252 * @rq_flags: flags for ->rq_flags
253 * @resid: optional residual length
255 * Returns the scsi_cmnd result field if a command was executed, or a negative
256 * Linux error code if we didn't get that far.
258 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
259 int data_direction, void *buffer, unsigned bufflen,
260 unsigned char *sense, struct scsi_sense_hdr *sshdr,
261 int timeout, int retries, u64 flags, req_flags_t rq_flags,
265 struct scsi_request *rq;
266 int ret = DRIVER_ERROR << 24;
268 req = blk_get_request(sdev->request_queue,
269 data_direction == DMA_TO_DEVICE ?
270 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
275 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
276 buffer, bufflen, GFP_NOIO))
279 rq->cmd_len = COMMAND_SIZE(cmd[0]);
280 memcpy(rq->cmd, cmd, rq->cmd_len);
281 rq->retries = retries;
282 req->timeout = timeout;
283 req->cmd_flags |= flags;
284 req->rq_flags |= rq_flags | RQF_QUIET;
287 * head injection *required* here otherwise quiesce won't work
289 blk_execute_rq(req->q, NULL, req, 1);
292 * Some devices (USB mass-storage in particular) may transfer
293 * garbage data together with a residue indicating that the data
294 * is invalid. Prevent the garbage from being misinterpreted
295 * and prevent security leaks by zeroing out the excess data.
297 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
298 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
301 *resid = rq->resid_len;
302 if (sense && rq->sense_len)
303 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
305 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
308 blk_put_request(req);
312 EXPORT_SYMBOL(__scsi_execute);
315 * Function: scsi_init_cmd_errh()
317 * Purpose: Initialize cmd fields related to error handling.
319 * Arguments: cmd - command that is ready to be queued.
321 * Notes: This function has the job of initializing a number of
322 * fields related to error handling. Typically this will
323 * be called once for each command, as required.
325 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
327 cmd->serial_number = 0;
328 scsi_set_resid(cmd, 0);
329 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
330 if (cmd->cmd_len == 0)
331 cmd->cmd_len = scsi_command_size(cmd->cmnd);
335 * Decrement the host_busy counter and wake up the error handler if necessary.
336 * Avoid as follows that the error handler is not woken up if shost->host_busy
337 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
338 * with an RCU read lock in this function to ensure that this function in its
339 * entirety either finishes before scsi_eh_scmd_add() increases the
340 * host_failed counter or that it notices the shost state change made by
341 * scsi_eh_scmd_add().
343 static void scsi_dec_host_busy(struct Scsi_Host *shost)
348 atomic_dec(&shost->host_busy);
349 if (unlikely(scsi_host_in_recovery(shost))) {
350 spin_lock_irqsave(shost->host_lock, flags);
351 if (shost->host_failed || shost->host_eh_scheduled)
352 scsi_eh_wakeup(shost);
353 spin_unlock_irqrestore(shost->host_lock, flags);
358 void scsi_device_unbusy(struct scsi_device *sdev)
360 struct Scsi_Host *shost = sdev->host;
361 struct scsi_target *starget = scsi_target(sdev);
363 scsi_dec_host_busy(shost);
365 if (starget->can_queue > 0)
366 atomic_dec(&starget->target_busy);
368 atomic_dec(&sdev->device_busy);
371 static void scsi_kick_queue(struct request_queue *q)
374 blk_mq_run_hw_queues(q, false);
380 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
381 * and call blk_run_queue for all the scsi_devices on the target -
382 * including current_sdev first.
384 * Called with *no* scsi locks held.
386 static void scsi_single_lun_run(struct scsi_device *current_sdev)
388 struct Scsi_Host *shost = current_sdev->host;
389 struct scsi_device *sdev, *tmp;
390 struct scsi_target *starget = scsi_target(current_sdev);
393 spin_lock_irqsave(shost->host_lock, flags);
394 starget->starget_sdev_user = NULL;
395 spin_unlock_irqrestore(shost->host_lock, flags);
398 * Call blk_run_queue for all LUNs on the target, starting with
399 * current_sdev. We race with others (to set starget_sdev_user),
400 * but in most cases, we will be first. Ideally, each LU on the
401 * target would get some limited time or requests on the target.
403 scsi_kick_queue(current_sdev->request_queue);
405 spin_lock_irqsave(shost->host_lock, flags);
406 if (starget->starget_sdev_user)
408 list_for_each_entry_safe(sdev, tmp, &starget->devices,
409 same_target_siblings) {
410 if (sdev == current_sdev)
412 if (scsi_device_get(sdev))
415 spin_unlock_irqrestore(shost->host_lock, flags);
416 scsi_kick_queue(sdev->request_queue);
417 spin_lock_irqsave(shost->host_lock, flags);
419 scsi_device_put(sdev);
422 spin_unlock_irqrestore(shost->host_lock, flags);
425 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
427 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
429 if (atomic_read(&sdev->device_blocked) > 0)
434 static inline bool scsi_target_is_busy(struct scsi_target *starget)
436 if (starget->can_queue > 0) {
437 if (atomic_read(&starget->target_busy) >= starget->can_queue)
439 if (atomic_read(&starget->target_blocked) > 0)
445 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
447 if (shost->can_queue > 0 &&
448 atomic_read(&shost->host_busy) >= shost->can_queue)
450 if (atomic_read(&shost->host_blocked) > 0)
452 if (shost->host_self_blocked)
457 static void scsi_starved_list_run(struct Scsi_Host *shost)
459 LIST_HEAD(starved_list);
460 struct scsi_device *sdev;
463 spin_lock_irqsave(shost->host_lock, flags);
464 list_splice_init(&shost->starved_list, &starved_list);
466 while (!list_empty(&starved_list)) {
467 struct request_queue *slq;
470 * As long as shost is accepting commands and we have
471 * starved queues, call blk_run_queue. scsi_request_fn
472 * drops the queue_lock and can add us back to the
475 * host_lock protects the starved_list and starved_entry.
476 * scsi_request_fn must get the host_lock before checking
477 * or modifying starved_list or starved_entry.
479 if (scsi_host_is_busy(shost))
482 sdev = list_entry(starved_list.next,
483 struct scsi_device, starved_entry);
484 list_del_init(&sdev->starved_entry);
485 if (scsi_target_is_busy(scsi_target(sdev))) {
486 list_move_tail(&sdev->starved_entry,
487 &shost->starved_list);
492 * Once we drop the host lock, a racing scsi_remove_device()
493 * call may remove the sdev from the starved list and destroy
494 * it and the queue. Mitigate by taking a reference to the
495 * queue and never touching the sdev again after we drop the
496 * host lock. Note: if __scsi_remove_device() invokes
497 * blk_cleanup_queue() before the queue is run from this
498 * function then blk_run_queue() will return immediately since
499 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
501 slq = sdev->request_queue;
502 if (!blk_get_queue(slq))
504 spin_unlock_irqrestore(shost->host_lock, flags);
506 scsi_kick_queue(slq);
509 spin_lock_irqsave(shost->host_lock, flags);
511 /* put any unprocessed entries back */
512 list_splice(&starved_list, &shost->starved_list);
513 spin_unlock_irqrestore(shost->host_lock, flags);
517 * Function: scsi_run_queue()
519 * Purpose: Select a proper request queue to serve next
521 * Arguments: q - last request's queue
525 * Notes: The previous command was completely finished, start
526 * a new one if possible.
528 static void scsi_run_queue(struct request_queue *q)
530 struct scsi_device *sdev = q->queuedata;
532 if (scsi_target(sdev)->single_lun)
533 scsi_single_lun_run(sdev);
534 if (!list_empty(&sdev->host->starved_list))
535 scsi_starved_list_run(sdev->host);
538 blk_mq_run_hw_queues(q, false);
543 void scsi_requeue_run_queue(struct work_struct *work)
545 struct scsi_device *sdev;
546 struct request_queue *q;
548 sdev = container_of(work, struct scsi_device, requeue_work);
549 q = sdev->request_queue;
554 * Function: scsi_requeue_command()
556 * Purpose: Handle post-processing of completed commands.
558 * Arguments: q - queue to operate on
559 * cmd - command that may need to be requeued.
563 * Notes: After command completion, there may be blocks left
564 * over which weren't finished by the previous command
565 * this can be for a number of reasons - the main one is
566 * I/O errors in the middle of the request, in which case
567 * we need to request the blocks that come after the bad
569 * Notes: Upon return, cmd is a stale pointer.
571 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
573 struct scsi_device *sdev = cmd->device;
574 struct request *req = cmd->request;
577 spin_lock_irqsave(q->queue_lock, flags);
578 blk_unprep_request(req);
580 scsi_put_command(cmd);
581 blk_requeue_request(q, req);
582 spin_unlock_irqrestore(q->queue_lock, flags);
586 put_device(&sdev->sdev_gendev);
589 void scsi_run_host_queues(struct Scsi_Host *shost)
591 struct scsi_device *sdev;
593 shost_for_each_device(sdev, shost)
594 scsi_run_queue(sdev->request_queue);
597 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
599 if (!blk_rq_is_passthrough(cmd->request)) {
600 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
602 if (drv->uninit_command)
603 drv->uninit_command(cmd);
607 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
609 struct scsi_data_buffer *sdb;
611 if (cmd->sdb.table.nents)
612 sg_free_table_chained(&cmd->sdb.table, true);
613 if (cmd->request->next_rq) {
614 sdb = cmd->request->next_rq->special;
616 sg_free_table_chained(&sdb->table, true);
618 if (scsi_prot_sg_count(cmd))
619 sg_free_table_chained(&cmd->prot_sdb->table, true);
622 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
624 scsi_mq_free_sgtables(cmd);
625 scsi_uninit_cmd(cmd);
626 scsi_del_cmd_from_list(cmd);
630 * Function: scsi_release_buffers()
632 * Purpose: Free resources allocate for a scsi_command.
634 * Arguments: cmd - command that we are bailing.
636 * Lock status: Assumed that no lock is held upon entry.
640 * Notes: In the event that an upper level driver rejects a
641 * command, we must release resources allocated during
642 * the __init_io() function. Primarily this would involve
643 * the scatter-gather table.
645 static void scsi_release_buffers(struct scsi_cmnd *cmd)
647 if (cmd->sdb.table.nents)
648 sg_free_table_chained(&cmd->sdb.table, false);
650 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
652 if (scsi_prot_sg_count(cmd))
653 sg_free_table_chained(&cmd->prot_sdb->table, false);
656 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
658 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
660 sg_free_table_chained(&bidi_sdb->table, false);
661 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
662 cmd->request->next_rq->special = NULL;
665 /* Returns false when no more bytes to process, true if there are more */
666 static bool scsi_end_request(struct request *req, blk_status_t error,
667 unsigned int bytes, unsigned int bidi_bytes)
669 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
670 struct scsi_device *sdev = cmd->device;
671 struct request_queue *q = sdev->request_queue;
673 if (blk_update_request(req, error, bytes))
676 /* Bidi request must be completed as a whole */
677 if (unlikely(bidi_bytes) &&
678 blk_update_request(req->next_rq, error, bidi_bytes))
681 if (blk_queue_add_random(q))
682 add_disk_randomness(req->rq_disk);
684 if (!blk_rq_is_scsi(req)) {
685 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
686 cmd->flags &= ~SCMD_INITIALIZED;
687 destroy_rcu_head(&cmd->rcu);
692 * In the MQ case the command gets freed by __blk_mq_end_request,
693 * so we have to do all cleanup that depends on it earlier.
695 * We also can't kick the queues from irq context, so we
696 * will have to defer it to a workqueue.
698 scsi_mq_uninit_cmd(cmd);
701 * queue is still alive, so grab the ref for preventing it
702 * from being cleaned up during running queue.
704 percpu_ref_get(&q->q_usage_counter);
706 __blk_mq_end_request(req, error);
708 if (scsi_target(sdev)->single_lun ||
709 !list_empty(&sdev->host->starved_list))
710 kblockd_schedule_work(&sdev->requeue_work);
712 blk_mq_run_hw_queues(q, true);
714 percpu_ref_put(&q->q_usage_counter);
719 scsi_release_bidi_buffers(cmd);
720 scsi_release_buffers(cmd);
721 scsi_put_command(cmd);
723 spin_lock_irqsave(q->queue_lock, flags);
724 blk_finish_request(req, error);
725 spin_unlock_irqrestore(q->queue_lock, flags);
730 put_device(&sdev->sdev_gendev);
735 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
737 * @result: scsi error code
739 * Translate a SCSI result code into a blk_status_t value. May reset the host
740 * byte of @cmd->result.
742 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
744 switch (host_byte(result)) {
747 * Also check the other bytes than the status byte in result
748 * to handle the case when a SCSI LLD sets result to
749 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
751 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
753 return BLK_STS_IOERR;
754 case DID_TRANSPORT_FAILFAST:
755 return BLK_STS_TRANSPORT;
756 case DID_TARGET_FAILURE:
757 set_host_byte(cmd, DID_OK);
758 return BLK_STS_TARGET;
759 case DID_NEXUS_FAILURE:
760 set_host_byte(cmd, DID_OK);
761 return BLK_STS_NEXUS;
762 case DID_ALLOC_FAILURE:
763 set_host_byte(cmd, DID_OK);
764 return BLK_STS_NOSPC;
765 case DID_MEDIUM_ERROR:
766 set_host_byte(cmd, DID_OK);
767 return BLK_STS_MEDIUM;
769 return BLK_STS_IOERR;
773 /* Helper for scsi_io_completion() when "reprep" action required. */
774 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
775 struct request_queue *q)
777 /* A new command will be prepared and issued. */
779 scsi_mq_requeue_cmd(cmd);
781 /* Unprep request and put it back at head of the queue. */
782 scsi_release_buffers(cmd);
783 scsi_requeue_command(q, cmd);
787 /* Helper for scsi_io_completion() when special action required. */
788 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
790 struct request_queue *q = cmd->device->request_queue;
791 struct request *req = cmd->request;
793 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
794 ACTION_DELAYED_RETRY} action;
795 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
796 struct scsi_sense_hdr sshdr;
798 bool sense_current = true; /* false implies "deferred sense" */
799 blk_status_t blk_stat;
801 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
803 sense_current = !scsi_sense_is_deferred(&sshdr);
805 blk_stat = scsi_result_to_blk_status(cmd, result);
807 if (host_byte(result) == DID_RESET) {
808 /* Third party bus reset or reset for error recovery
809 * reasons. Just retry the command and see what
812 action = ACTION_RETRY;
813 } else if (sense_valid && sense_current) {
814 switch (sshdr.sense_key) {
816 if (cmd->device->removable) {
817 /* Detected disc change. Set a bit
818 * and quietly refuse further access.
820 cmd->device->changed = 1;
821 action = ACTION_FAIL;
823 /* Must have been a power glitch, or a
824 * bus reset. Could not have been a
825 * media change, so we just retry the
826 * command and see what happens.
828 action = ACTION_RETRY;
831 case ILLEGAL_REQUEST:
832 /* If we had an ILLEGAL REQUEST returned, then
833 * we may have performed an unsupported
834 * command. The only thing this should be
835 * would be a ten byte read where only a six
836 * byte read was supported. Also, on a system
837 * where READ CAPACITY failed, we may have
838 * read past the end of the disk.
840 if ((cmd->device->use_10_for_rw &&
841 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
842 (cmd->cmnd[0] == READ_10 ||
843 cmd->cmnd[0] == WRITE_10)) {
844 /* This will issue a new 6-byte command. */
845 cmd->device->use_10_for_rw = 0;
846 action = ACTION_REPREP;
847 } else if (sshdr.asc == 0x10) /* DIX */ {
848 action = ACTION_FAIL;
849 blk_stat = BLK_STS_PROTECTION;
850 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
851 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
852 action = ACTION_FAIL;
853 blk_stat = BLK_STS_TARGET;
855 action = ACTION_FAIL;
857 case ABORTED_COMMAND:
858 action = ACTION_FAIL;
859 if (sshdr.asc == 0x10) /* DIF */
860 blk_stat = BLK_STS_PROTECTION;
863 /* If the device is in the process of becoming
864 * ready, or has a temporary blockage, retry.
866 if (sshdr.asc == 0x04) {
867 switch (sshdr.ascq) {
868 case 0x01: /* becoming ready */
869 case 0x04: /* format in progress */
870 case 0x05: /* rebuild in progress */
871 case 0x06: /* recalculation in progress */
872 case 0x07: /* operation in progress */
873 case 0x08: /* Long write in progress */
874 case 0x09: /* self test in progress */
875 case 0x11: /* notify (enable spinup) required */
876 case 0x14: /* space allocation in progress */
877 case 0x1a: /* start stop unit in progress */
878 case 0x1b: /* sanitize in progress */
879 case 0x1d: /* configuration in progress */
880 case 0x24: /* depopulation in progress */
881 action = ACTION_DELAYED_RETRY;
884 action = ACTION_FAIL;
888 action = ACTION_FAIL;
890 case VOLUME_OVERFLOW:
891 /* See SSC3rXX or current. */
892 action = ACTION_FAIL;
895 action = ACTION_FAIL;
899 action = ACTION_FAIL;
901 if (action != ACTION_FAIL &&
902 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
903 action = ACTION_FAIL;
907 /* Give up and fail the remainder of the request */
908 if (!(req->rq_flags & RQF_QUIET)) {
909 static DEFINE_RATELIMIT_STATE(_rs,
910 DEFAULT_RATELIMIT_INTERVAL,
911 DEFAULT_RATELIMIT_BURST);
913 if (unlikely(scsi_logging_level))
915 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
916 SCSI_LOG_MLCOMPLETE_BITS);
919 * if logging is enabled the failure will be printed
920 * in scsi_log_completion(), so avoid duplicate messages
922 if (!level && __ratelimit(&_rs)) {
923 scsi_print_result(cmd, NULL, FAILED);
924 if (driver_byte(result) == DRIVER_SENSE)
925 scsi_print_sense(cmd);
926 scsi_print_command(cmd);
929 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req), 0))
933 scsi_io_completion_reprep(cmd, q);
936 /* Retry the same command immediately */
937 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
939 case ACTION_DELAYED_RETRY:
940 /* Retry the same command after a delay */
941 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
947 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
948 * new result that may suppress further error checking. Also modifies
949 * *blk_statp in some cases.
951 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
952 blk_status_t *blk_statp)
955 bool sense_current = true; /* false implies "deferred sense" */
956 struct request *req = cmd->request;
957 struct scsi_sense_hdr sshdr;
959 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
961 sense_current = !scsi_sense_is_deferred(&sshdr);
963 if (blk_rq_is_passthrough(req)) {
966 * SG_IO wants current and deferred errors
968 scsi_req(req)->sense_len =
969 min(8 + cmd->sense_buffer[7],
970 SCSI_SENSE_BUFFERSIZE);
973 *blk_statp = scsi_result_to_blk_status(cmd, result);
974 } else if (blk_rq_bytes(req) == 0 && sense_current) {
976 * Flush commands do not transfers any data, and thus cannot use
977 * good_bytes != blk_rq_bytes(req) as the signal for an error.
978 * This sets *blk_statp explicitly for the problem case.
980 *blk_statp = scsi_result_to_blk_status(cmd, result);
983 * Recovered errors need reporting, but they're always treated as
984 * success, so fiddle the result code here. For passthrough requests
985 * we already took a copy of the original into sreq->result which
986 * is what gets returned to the user
988 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
989 bool do_print = true;
991 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
992 * skip print since caller wants ATA registers. Only occurs
993 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
995 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
997 else if (req->rq_flags & RQF_QUIET)
1000 scsi_print_sense(cmd);
1002 /* for passthrough, *blk_statp may be set */
1003 *blk_statp = BLK_STS_OK;
1006 * Another corner case: the SCSI status byte is non-zero but 'good'.
1007 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1008 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1009 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1010 * intermediate statuses (both obsolete in SAM-4) as good.
1012 if (status_byte(result) && scsi_status_is_good(result)) {
1014 *blk_statp = BLK_STS_OK;
1020 * Function: scsi_io_completion()
1022 * Purpose: Completion processing for block device I/O requests.
1024 * Arguments: cmd - command that is finished.
1026 * Lock status: Assumed that no lock is held upon entry.
1030 * Notes: We will finish off the specified number of sectors. If we
1031 * are done, the command block will be released and the queue
1032 * function will be goosed. If we are not done then we have to
1033 * figure out what to do next:
1035 * a) We can call scsi_requeue_command(). The request
1036 * will be unprepared and put back on the queue. Then
1037 * a new command will be created for it. This should
1038 * be used if we made forward progress, or if we want
1039 * to switch from READ(10) to READ(6) for example.
1041 * b) We can call __scsi_queue_insert(). The request will
1042 * be put back on the queue and retried using the same
1043 * command as before, possibly after a delay.
1045 * c) We can call scsi_end_request() with blk_stat other than
1046 * BLK_STS_OK, to fail the remainder of the request.
1048 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1050 int result = cmd->result;
1051 struct request_queue *q = cmd->device->request_queue;
1052 struct request *req = cmd->request;
1053 blk_status_t blk_stat = BLK_STS_OK;
1055 if (unlikely(result)) /* a nz result may or may not be an error */
1056 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1058 if (unlikely(blk_rq_is_passthrough(req))) {
1060 * scsi_result_to_blk_status may have reset the host_byte
1062 scsi_req(req)->result = cmd->result;
1063 scsi_req(req)->resid_len = scsi_get_resid(cmd);
1065 if (unlikely(scsi_bidi_cmnd(cmd))) {
1067 * Bidi commands Must be complete as a whole,
1068 * both sides at once.
1070 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
1071 if (scsi_end_request(req, BLK_STS_OK, blk_rq_bytes(req),
1072 blk_rq_bytes(req->next_rq)))
1074 "Bidi command with remaining bytes");
1079 /* no bidi support yet, other than in pass-through */
1080 if (unlikely(blk_bidi_rq(req))) {
1081 WARN_ONCE(true, "Only support bidi command in passthrough");
1082 scmd_printk(KERN_ERR, cmd, "Killing bidi command\n");
1083 if (scsi_end_request(req, BLK_STS_IOERR, blk_rq_bytes(req),
1084 blk_rq_bytes(req->next_rq)))
1085 WARN_ONCE(true, "Bidi command with remaining bytes");
1090 * Next deal with any sectors which we were able to correctly
1093 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1094 "%u sectors total, %d bytes done.\n",
1095 blk_rq_sectors(req), good_bytes));
1098 * Next deal with any sectors which we were able to correctly
1099 * handle. Failed, zero length commands always need to drop down
1100 * to retry code. Fast path should return in this block.
1102 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1103 if (likely(!scsi_end_request(req, blk_stat, good_bytes, 0)))
1104 return; /* no bytes remaining */
1107 /* Kill remainder if no retries. */
1108 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1109 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req), 0))
1111 "Bytes remaining after failed, no-retry command");
1116 * If there had been no error, but we have leftover bytes in the
1117 * requeues just queue the command up again.
1119 if (likely(result == 0))
1120 scsi_io_completion_reprep(cmd, q);
1122 scsi_io_completion_action(cmd, result);
1125 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1130 * If sg table allocation fails, requeue request later.
1132 if (unlikely(sg_alloc_table_chained(&sdb->table,
1133 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1134 return BLKPREP_DEFER;
1137 * Next, walk the list, and fill in the addresses and sizes of
1140 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1141 BUG_ON(count > sdb->table.nents);
1142 sdb->table.nents = count;
1143 sdb->length = blk_rq_payload_bytes(req);
1148 * Function: scsi_init_io()
1150 * Purpose: SCSI I/O initialize function.
1152 * Arguments: cmd - Command descriptor we wish to initialize
1154 * Returns: 0 on success
1155 * BLKPREP_DEFER if the failure is retryable
1156 * BLKPREP_KILL if the failure is fatal
1158 int scsi_init_io(struct scsi_cmnd *cmd)
1160 struct scsi_device *sdev = cmd->device;
1161 struct request *rq = cmd->request;
1162 bool is_mq = (rq->mq_ctx != NULL);
1163 int error = BLKPREP_KILL;
1165 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1168 error = scsi_init_sgtable(rq, &cmd->sdb);
1172 if (blk_bidi_rq(rq)) {
1173 if (!rq->q->mq_ops) {
1174 struct scsi_data_buffer *bidi_sdb =
1175 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1177 error = BLKPREP_DEFER;
1181 rq->next_rq->special = bidi_sdb;
1184 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1189 if (blk_integrity_rq(rq)) {
1190 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1193 if (prot_sdb == NULL) {
1195 * This can happen if someone (e.g. multipath)
1196 * queues a command to a device on an adapter
1197 * that does not support DIX.
1200 error = BLKPREP_KILL;
1204 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1206 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1207 prot_sdb->table.sgl)) {
1208 error = BLKPREP_DEFER;
1212 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1213 prot_sdb->table.sgl);
1214 BUG_ON(unlikely(count > ivecs));
1215 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1217 cmd->prot_sdb = prot_sdb;
1218 cmd->prot_sdb->table.nents = count;
1224 scsi_mq_free_sgtables(cmd);
1226 scsi_release_buffers(cmd);
1227 cmd->request->special = NULL;
1228 scsi_put_command(cmd);
1229 put_device(&sdev->sdev_gendev);
1233 EXPORT_SYMBOL(scsi_init_io);
1236 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1237 * @rq: Request associated with the SCSI command to be initialized.
1239 * This function initializes the members of struct scsi_cmnd that must be
1240 * initialized before request processing starts and that won't be
1241 * reinitialized if a SCSI command is requeued.
1243 * Called from inside blk_get_request() for pass-through requests and from
1244 * inside scsi_init_command() for filesystem requests.
1246 static void scsi_initialize_rq(struct request *rq)
1248 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1250 scsi_req_init(&cmd->req);
1251 init_rcu_head(&cmd->rcu);
1252 cmd->jiffies_at_alloc = jiffies;
1257 * Only called when the request isn't completed by SCSI, and not freed by
1260 static void scsi_cleanup_rq(struct request *rq)
1262 if (rq->rq_flags & RQF_DONTPREP) {
1263 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1264 rq->rq_flags &= ~RQF_DONTPREP;
1268 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1269 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1271 struct scsi_device *sdev = cmd->device;
1272 struct Scsi_Host *shost = sdev->host;
1273 unsigned long flags;
1275 if (shost->use_cmd_list) {
1276 spin_lock_irqsave(&sdev->list_lock, flags);
1277 list_add_tail(&cmd->list, &sdev->cmd_list);
1278 spin_unlock_irqrestore(&sdev->list_lock, flags);
1282 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1283 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1285 struct scsi_device *sdev = cmd->device;
1286 struct Scsi_Host *shost = sdev->host;
1287 unsigned long flags;
1289 if (shost->use_cmd_list) {
1290 spin_lock_irqsave(&sdev->list_lock, flags);
1291 BUG_ON(list_empty(&cmd->list));
1292 list_del_init(&cmd->list);
1293 spin_unlock_irqrestore(&sdev->list_lock, flags);
1297 /* Called after a request has been started. */
1298 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1300 void *buf = cmd->sense_buffer;
1301 void *prot = cmd->prot_sdb;
1302 struct request *rq = blk_mq_rq_from_pdu(cmd);
1303 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1304 unsigned long jiffies_at_alloc;
1307 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1308 flags |= SCMD_INITIALIZED;
1309 scsi_initialize_rq(rq);
1312 jiffies_at_alloc = cmd->jiffies_at_alloc;
1313 retries = cmd->retries;
1314 /* zero out the cmd, except for the embedded scsi_request */
1315 memset((char *)cmd + sizeof(cmd->req), 0,
1316 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1319 cmd->sense_buffer = buf;
1320 cmd->prot_sdb = prot;
1322 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1323 cmd->jiffies_at_alloc = jiffies_at_alloc;
1324 cmd->retries = retries;
1326 scsi_add_cmd_to_list(cmd);
1329 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1331 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1334 * Passthrough requests may transfer data, in which case they must
1335 * a bio attached to them. Or they might contain a SCSI command
1336 * that does not transfer data, in which case they may optionally
1337 * submit a request without an attached bio.
1340 int ret = scsi_init_io(cmd);
1344 BUG_ON(blk_rq_bytes(req));
1346 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1349 cmd->cmd_len = scsi_req(req)->cmd_len;
1350 cmd->cmnd = scsi_req(req)->cmd;
1351 cmd->transfersize = blk_rq_bytes(req);
1352 cmd->allowed = scsi_req(req)->retries;
1357 * Setup a normal block command. These are simple request from filesystems
1358 * that still need to be translated to SCSI CDBs from the ULD.
1360 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1362 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1364 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1365 int ret = sdev->handler->prep_fn(sdev, req);
1366 if (ret != BLKPREP_OK)
1370 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1371 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1372 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1375 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1377 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1379 if (!blk_rq_bytes(req))
1380 cmd->sc_data_direction = DMA_NONE;
1381 else if (rq_data_dir(req) == WRITE)
1382 cmd->sc_data_direction = DMA_TO_DEVICE;
1384 cmd->sc_data_direction = DMA_FROM_DEVICE;
1386 if (blk_rq_is_scsi(req))
1387 return scsi_setup_scsi_cmnd(sdev, req);
1389 return scsi_setup_fs_cmnd(sdev, req);
1393 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1395 int ret = BLKPREP_OK;
1398 * If the device is not in running state we will reject some
1401 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1402 switch (sdev->sdev_state) {
1404 case SDEV_TRANSPORT_OFFLINE:
1406 * If the device is offline we refuse to process any
1407 * commands. The device must be brought online
1408 * before trying any recovery commands.
1410 sdev_printk(KERN_ERR, sdev,
1411 "rejecting I/O to offline device\n");
1416 * If the device is fully deleted, we refuse to
1417 * process any commands as well.
1419 sdev_printk(KERN_ERR, sdev,
1420 "rejecting I/O to dead device\n");
1424 case SDEV_CREATED_BLOCK:
1425 ret = BLKPREP_DEFER;
1429 * If the devices is blocked we defer normal commands.
1431 if (req && !(req->rq_flags & RQF_PREEMPT))
1432 ret = BLKPREP_DEFER;
1436 * For any other not fully online state we only allow
1437 * special commands. In particular any user initiated
1438 * command is not allowed.
1440 if (req && !(req->rq_flags & RQF_PREEMPT))
1449 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1451 struct scsi_device *sdev = q->queuedata;
1455 case BLKPREP_INVALID:
1456 scsi_req(req)->result = DID_NO_CONNECT << 16;
1457 /* release the command and kill it */
1459 struct scsi_cmnd *cmd = req->special;
1460 scsi_release_buffers(cmd);
1461 scsi_put_command(cmd);
1462 put_device(&sdev->sdev_gendev);
1463 req->special = NULL;
1468 * If we defer, the blk_peek_request() returns NULL, but the
1469 * queue must be restarted, so we schedule a callback to happen
1472 if (atomic_read(&sdev->device_busy) == 0)
1473 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1476 req->rq_flags |= RQF_DONTPREP;
1482 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1484 struct scsi_device *sdev = q->queuedata;
1485 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1488 ret = scsi_prep_state_check(sdev, req);
1489 if (ret != BLKPREP_OK)
1492 if (!req->special) {
1493 /* Bail if we can't get a reference to the device */
1494 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1495 ret = BLKPREP_DEFER;
1499 scsi_init_command(sdev, cmd);
1503 cmd->tag = req->tag;
1505 cmd->prot_op = SCSI_PROT_NORMAL;
1507 ret = scsi_setup_cmnd(sdev, req);
1509 return scsi_prep_return(q, req, ret);
1512 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1514 scsi_uninit_cmd(blk_mq_rq_to_pdu(req));
1518 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1521 * Called with the queue_lock held.
1523 static inline int scsi_dev_queue_ready(struct request_queue *q,
1524 struct scsi_device *sdev)
1528 busy = atomic_inc_return(&sdev->device_busy) - 1;
1529 if (atomic_read(&sdev->device_blocked)) {
1534 * unblock after device_blocked iterates to zero
1536 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1538 * For the MQ case we take care of this in the caller.
1541 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1544 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1545 "unblocking device at zero depth\n"));
1548 if (busy >= sdev->queue_depth)
1553 atomic_dec(&sdev->device_busy);
1558 * scsi_target_queue_ready: checks if there we can send commands to target
1559 * @sdev: scsi device on starget to check.
1561 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1562 struct scsi_device *sdev)
1564 struct scsi_target *starget = scsi_target(sdev);
1567 if (starget->single_lun) {
1568 spin_lock_irq(shost->host_lock);
1569 if (starget->starget_sdev_user &&
1570 starget->starget_sdev_user != sdev) {
1571 spin_unlock_irq(shost->host_lock);
1574 starget->starget_sdev_user = sdev;
1575 spin_unlock_irq(shost->host_lock);
1578 if (starget->can_queue <= 0)
1581 busy = atomic_inc_return(&starget->target_busy) - 1;
1582 if (atomic_read(&starget->target_blocked) > 0) {
1587 * unblock after target_blocked iterates to zero
1589 if (atomic_dec_return(&starget->target_blocked) > 0)
1592 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1593 "unblocking target at zero depth\n"));
1596 if (busy >= starget->can_queue)
1602 spin_lock_irq(shost->host_lock);
1603 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1604 spin_unlock_irq(shost->host_lock);
1606 if (starget->can_queue > 0)
1607 atomic_dec(&starget->target_busy);
1612 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1613 * return 0. We must end up running the queue again whenever 0 is
1614 * returned, else IO can hang.
1616 static inline int scsi_host_queue_ready(struct request_queue *q,
1617 struct Scsi_Host *shost,
1618 struct scsi_device *sdev)
1622 if (scsi_host_in_recovery(shost))
1625 busy = atomic_inc_return(&shost->host_busy) - 1;
1626 if (atomic_read(&shost->host_blocked) > 0) {
1631 * unblock after host_blocked iterates to zero
1633 if (atomic_dec_return(&shost->host_blocked) > 0)
1637 shost_printk(KERN_INFO, shost,
1638 "unblocking host at zero depth\n"));
1641 if (shost->can_queue > 0 && busy >= shost->can_queue)
1643 if (shost->host_self_blocked)
1646 /* We're OK to process the command, so we can't be starved */
1647 if (!list_empty(&sdev->starved_entry)) {
1648 spin_lock_irq(shost->host_lock);
1649 if (!list_empty(&sdev->starved_entry))
1650 list_del_init(&sdev->starved_entry);
1651 spin_unlock_irq(shost->host_lock);
1657 spin_lock_irq(shost->host_lock);
1658 if (list_empty(&sdev->starved_entry))
1659 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1660 spin_unlock_irq(shost->host_lock);
1662 scsi_dec_host_busy(shost);
1667 * Busy state exporting function for request stacking drivers.
1669 * For efficiency, no lock is taken to check the busy state of
1670 * shost/starget/sdev, since the returned value is not guaranteed and
1671 * may be changed after request stacking drivers call the function,
1672 * regardless of taking lock or not.
1674 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1675 * needs to return 'not busy'. Otherwise, request stacking drivers
1676 * may hold requests forever.
1678 static int scsi_lld_busy(struct request_queue *q)
1680 struct scsi_device *sdev = q->queuedata;
1681 struct Scsi_Host *shost;
1683 if (blk_queue_dying(q))
1689 * Ignore host/starget busy state.
1690 * Since block layer does not have a concept of fairness across
1691 * multiple queues, congestion of host/starget needs to be handled
1694 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1701 * Kill a request for a dead device
1703 static void scsi_kill_request(struct request *req, struct request_queue *q)
1705 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1706 struct scsi_device *sdev;
1707 struct scsi_target *starget;
1708 struct Scsi_Host *shost;
1710 blk_start_request(req);
1712 scmd_printk(KERN_INFO, cmd, "killing request\n");
1715 starget = scsi_target(sdev);
1717 scsi_init_cmd_errh(cmd);
1718 cmd->result = DID_NO_CONNECT << 16;
1719 atomic_inc(&cmd->device->iorequest_cnt);
1722 * SCSI request completion path will do scsi_device_unbusy(),
1723 * bump busy counts. To bump the counters, we need to dance
1724 * with the locks as normal issue path does.
1726 atomic_inc(&sdev->device_busy);
1727 atomic_inc(&shost->host_busy);
1728 if (starget->can_queue > 0)
1729 atomic_inc(&starget->target_busy);
1731 blk_complete_request(req);
1734 static void scsi_softirq_done(struct request *rq)
1736 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1737 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1740 INIT_LIST_HEAD(&cmd->eh_entry);
1742 atomic_inc(&cmd->device->iodone_cnt);
1744 atomic_inc(&cmd->device->ioerr_cnt);
1746 disposition = scsi_decide_disposition(cmd);
1747 if (disposition != SUCCESS &&
1748 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1749 sdev_printk(KERN_ERR, cmd->device,
1750 "timing out command, waited %lus\n",
1752 disposition = SUCCESS;
1755 scsi_log_completion(cmd, disposition);
1757 switch (disposition) {
1759 scsi_finish_command(cmd);
1762 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1764 case ADD_TO_MLQUEUE:
1765 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1768 scsi_eh_scmd_add(cmd);
1774 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1775 * @cmd: command block we are dispatching.
1777 * Return: nonzero return request was rejected and device's queue needs to be
1780 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1782 struct Scsi_Host *host = cmd->device->host;
1785 atomic_inc(&cmd->device->iorequest_cnt);
1787 /* check if the device is still usable */
1788 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1789 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1790 * returns an immediate error upwards, and signals
1791 * that the device is no longer present */
1792 cmd->result = DID_NO_CONNECT << 16;
1796 /* Check to see if the scsi lld made this device blocked. */
1797 if (unlikely(scsi_device_blocked(cmd->device))) {
1799 * in blocked state, the command is just put back on
1800 * the device queue. The suspend state has already
1801 * blocked the queue so future requests should not
1802 * occur until the device transitions out of the
1805 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1806 "queuecommand : device blocked\n"));
1807 return SCSI_MLQUEUE_DEVICE_BUSY;
1810 /* Store the LUN value in cmnd, if needed. */
1811 if (cmd->device->lun_in_cdb)
1812 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1813 (cmd->device->lun << 5 & 0xe0);
1818 * Before we queue this command, check if the command
1819 * length exceeds what the host adapter can handle.
1821 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1822 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1823 "queuecommand : command too long. "
1824 "cdb_size=%d host->max_cmd_len=%d\n",
1825 cmd->cmd_len, cmd->device->host->max_cmd_len));
1826 cmd->result = (DID_ABORT << 16);
1830 if (unlikely(host->shost_state == SHOST_DEL)) {
1831 cmd->result = (DID_NO_CONNECT << 16);
1836 trace_scsi_dispatch_cmd_start(cmd);
1837 rtn = host->hostt->queuecommand(host, cmd);
1839 trace_scsi_dispatch_cmd_error(cmd, rtn);
1840 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1841 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1842 rtn = SCSI_MLQUEUE_HOST_BUSY;
1844 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1845 "queuecommand : request rejected\n"));
1850 cmd->scsi_done(cmd);
1855 * scsi_done - Invoke completion on finished SCSI command.
1856 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1857 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1859 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1860 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1861 * calls blk_complete_request() for further processing.
1863 * This function is interrupt context safe.
1865 static void scsi_done(struct scsi_cmnd *cmd)
1867 trace_scsi_dispatch_cmd_done(cmd);
1868 blk_complete_request(cmd->request);
1872 * Function: scsi_request_fn()
1874 * Purpose: Main strategy routine for SCSI.
1876 * Arguments: q - Pointer to actual queue.
1880 * Lock status: request queue lock assumed to be held when called.
1882 * Note: See sd_zbc.c sd_zbc_write_lock_zone() for write order
1883 * protection for ZBC disks.
1885 static void scsi_request_fn(struct request_queue *q)
1886 __releases(q->queue_lock)
1887 __acquires(q->queue_lock)
1889 struct scsi_device *sdev = q->queuedata;
1890 struct Scsi_Host *shost;
1891 struct scsi_cmnd *cmd;
1892 struct request *req;
1895 * To start with, we keep looping until the queue is empty, or until
1896 * the host is no longer able to accept any more requests.
1902 * get next queueable request. We do this early to make sure
1903 * that the request is fully prepared even if we cannot
1906 req = blk_peek_request(q);
1910 if (unlikely(!scsi_device_online(sdev))) {
1911 sdev_printk(KERN_ERR, sdev,
1912 "rejecting I/O to offline device\n");
1913 scsi_kill_request(req, q);
1917 if (!scsi_dev_queue_ready(q, sdev))
1921 * Remove the request from the request list.
1923 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1924 blk_start_request(req);
1926 spin_unlock_irq(q->queue_lock);
1927 cmd = blk_mq_rq_to_pdu(req);
1928 if (cmd != req->special) {
1929 printk(KERN_CRIT "impossible request in %s.\n"
1930 "please mail a stack trace to "
1931 "linux-scsi@vger.kernel.org\n",
1933 blk_dump_rq_flags(req, "foo");
1938 * We hit this when the driver is using a host wide
1939 * tag map. For device level tag maps the queue_depth check
1940 * in the device ready fn would prevent us from trying
1941 * to allocate a tag. Since the map is a shared host resource
1942 * we add the dev to the starved list so it eventually gets
1943 * a run when a tag is freed.
1945 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1946 spin_lock_irq(shost->host_lock);
1947 if (list_empty(&sdev->starved_entry))
1948 list_add_tail(&sdev->starved_entry,
1949 &shost->starved_list);
1950 spin_unlock_irq(shost->host_lock);
1954 if (!scsi_target_queue_ready(shost, sdev))
1957 if (!scsi_host_queue_ready(q, shost, sdev))
1958 goto host_not_ready;
1960 if (sdev->simple_tags)
1961 cmd->flags |= SCMD_TAGGED;
1963 cmd->flags &= ~SCMD_TAGGED;
1966 * Finally, initialize any error handling parameters, and set up
1967 * the timers for timeouts.
1969 scsi_init_cmd_errh(cmd);
1972 * Dispatch the command to the low-level driver.
1974 cmd->scsi_done = scsi_done;
1975 rtn = scsi_dispatch_cmd(cmd);
1977 scsi_queue_insert(cmd, rtn);
1978 spin_lock_irq(q->queue_lock);
1981 spin_lock_irq(q->queue_lock);
1987 if (scsi_target(sdev)->can_queue > 0)
1988 atomic_dec(&scsi_target(sdev)->target_busy);
1991 * lock q, handle tag, requeue req, and decrement device_busy. We
1992 * must return with queue_lock held.
1994 * Decrementing device_busy without checking it is OK, as all such
1995 * cases (host limits or settings) should run the queue at some
1998 spin_lock_irq(q->queue_lock);
1999 blk_requeue_request(q, req);
2000 atomic_dec(&sdev->device_busy);
2002 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
2003 blk_delay_queue(q, SCSI_QUEUE_DELAY);
2006 static inline blk_status_t prep_to_mq(int ret)
2012 return BLK_STS_RESOURCE;
2014 return BLK_STS_IOERR;
2018 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
2019 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
2021 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
2022 sizeof(struct scatterlist);
2025 static int scsi_mq_prep_fn(struct request *req)
2027 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
2028 struct scsi_device *sdev = req->q->queuedata;
2029 struct Scsi_Host *shost = sdev->host;
2030 struct scatterlist *sg;
2032 scsi_init_command(sdev, cmd);
2038 cmd->tag = req->tag;
2039 cmd->prot_op = SCSI_PROT_NORMAL;
2041 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2042 cmd->sdb.table.sgl = sg;
2044 if (scsi_host_get_prot(shost)) {
2045 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
2047 cmd->prot_sdb->table.sgl =
2048 (struct scatterlist *)(cmd->prot_sdb + 1);
2051 if (blk_bidi_rq(req)) {
2052 struct request *next_rq = req->next_rq;
2053 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
2055 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
2056 bidi_sdb->table.sgl =
2057 (struct scatterlist *)(bidi_sdb + 1);
2059 next_rq->special = bidi_sdb;
2062 blk_mq_start_request(req);
2064 return scsi_setup_cmnd(sdev, req);
2067 static void scsi_mq_done(struct scsi_cmnd *cmd)
2069 trace_scsi_dispatch_cmd_done(cmd);
2070 blk_mq_complete_request(cmd->request);
2073 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
2075 struct request_queue *q = hctx->queue;
2076 struct scsi_device *sdev = q->queuedata;
2078 atomic_dec(&sdev->device_busy);
2079 put_device(&sdev->sdev_gendev);
2082 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
2084 struct request_queue *q = hctx->queue;
2085 struct scsi_device *sdev = q->queuedata;
2087 if (!get_device(&sdev->sdev_gendev))
2089 if (!scsi_dev_queue_ready(q, sdev))
2090 goto out_put_device;
2095 put_device(&sdev->sdev_gendev);
2097 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
2098 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
2102 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
2103 const struct blk_mq_queue_data *bd)
2105 struct request *req = bd->rq;
2106 struct request_queue *q = req->q;
2107 struct scsi_device *sdev = q->queuedata;
2108 struct Scsi_Host *shost = sdev->host;
2109 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
2113 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
2114 if (ret != BLK_STS_OK)
2115 goto out_put_budget;
2117 ret = BLK_STS_RESOURCE;
2118 if (!scsi_target_queue_ready(shost, sdev))
2119 goto out_put_budget;
2120 if (!scsi_host_queue_ready(q, shost, sdev))
2121 goto out_dec_target_busy;
2123 if (!(req->rq_flags & RQF_DONTPREP)) {
2124 ret = prep_to_mq(scsi_mq_prep_fn(req));
2125 if (ret != BLK_STS_OK)
2126 goto out_dec_host_busy;
2127 req->rq_flags |= RQF_DONTPREP;
2129 blk_mq_start_request(req);
2132 if (sdev->simple_tags)
2133 cmd->flags |= SCMD_TAGGED;
2135 cmd->flags &= ~SCMD_TAGGED;
2137 scsi_init_cmd_errh(cmd);
2138 cmd->scsi_done = scsi_mq_done;
2140 reason = scsi_dispatch_cmd(cmd);
2142 scsi_set_blocked(cmd, reason);
2143 ret = BLK_STS_RESOURCE;
2144 goto out_dec_host_busy;
2150 scsi_dec_host_busy(shost);
2151 out_dec_target_busy:
2152 if (scsi_target(sdev)->can_queue > 0)
2153 atomic_dec(&scsi_target(sdev)->target_busy);
2155 scsi_mq_put_budget(hctx);
2159 case BLK_STS_RESOURCE:
2160 if (atomic_read(&sdev->device_busy) ||
2161 scsi_device_blocked(sdev))
2162 ret = BLK_STS_DEV_RESOURCE;
2165 if (unlikely(!scsi_device_online(sdev)))
2166 scsi_req(req)->result = DID_NO_CONNECT << 16;
2168 scsi_req(req)->result = DID_ERROR << 16;
2170 * Make sure to release all allocated resources when
2171 * we hit an error, as we will never see this command
2174 if (req->rq_flags & RQF_DONTPREP)
2175 scsi_mq_uninit_cmd(cmd);
2181 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2185 return BLK_EH_RESET_TIMER;
2186 return scsi_times_out(req);
2189 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
2190 unsigned int hctx_idx, unsigned int numa_node)
2192 struct Scsi_Host *shost = set->driver_data;
2193 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2194 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2195 struct scatterlist *sg;
2197 if (unchecked_isa_dma)
2198 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2199 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
2200 GFP_KERNEL, numa_node);
2201 if (!cmd->sense_buffer)
2203 cmd->req.sense = cmd->sense_buffer;
2205 if (scsi_host_get_prot(shost)) {
2206 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
2207 shost->hostt->cmd_size;
2208 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
2214 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
2215 unsigned int hctx_idx)
2217 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2219 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2223 static int scsi_map_queues(struct blk_mq_tag_set *set)
2225 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2227 if (shost->hostt->map_queues)
2228 return shost->hostt->map_queues(shost);
2229 return blk_mq_map_queues(set);
2232 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2234 struct device *dev = shost->dma_dev;
2237 * this limit is imposed by hardware restrictions
2239 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2242 if (scsi_host_prot_dma(shost)) {
2243 shost->sg_prot_tablesize =
2244 min_not_zero(shost->sg_prot_tablesize,
2245 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2246 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2247 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2250 blk_queue_max_hw_sectors(q, shost->max_sectors);
2251 if (shost->unchecked_isa_dma)
2252 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
2253 blk_queue_segment_boundary(q, shost->dma_boundary);
2254 dma_set_seg_boundary(dev, shost->dma_boundary);
2256 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2258 if (!shost->use_clustering)
2259 q->limits.cluster = 0;
2262 * Set a reasonable default alignment: The larger of 32-byte (dword),
2263 * which is a common minimum for HBAs, and the minimum DMA alignment,
2264 * which is set by the platform.
2266 * Devices that require a bigger alignment can increase it later.
2268 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
2270 EXPORT_SYMBOL_GPL(__scsi_init_queue);
2272 static int scsi_old_init_rq(struct request_queue *q, struct request *rq,
2275 struct Scsi_Host *shost = q->rq_alloc_data;
2276 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2277 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2279 memset(cmd, 0, sizeof(*cmd));
2281 if (unchecked_isa_dma)
2282 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2283 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma, gfp,
2285 if (!cmd->sense_buffer)
2287 cmd->req.sense = cmd->sense_buffer;
2289 if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2290 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2292 goto fail_free_sense;
2298 scsi_free_sense_buffer(unchecked_isa_dma, cmd->sense_buffer);
2303 static void scsi_old_exit_rq(struct request_queue *q, struct request *rq)
2305 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2308 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2309 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2313 struct request_queue *scsi_old_alloc_queue(struct scsi_device *sdev)
2315 struct Scsi_Host *shost = sdev->host;
2316 struct request_queue *q;
2318 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE, NULL);
2321 q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2322 q->rq_alloc_data = shost;
2323 q->request_fn = scsi_request_fn;
2324 q->init_rq_fn = scsi_old_init_rq;
2325 q->exit_rq_fn = scsi_old_exit_rq;
2326 q->initialize_rq_fn = scsi_initialize_rq;
2328 if (blk_init_allocated_queue(q) < 0) {
2329 blk_cleanup_queue(q);
2333 __scsi_init_queue(shost, q);
2334 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, q);
2335 blk_queue_prep_rq(q, scsi_prep_fn);
2336 blk_queue_unprep_rq(q, scsi_unprep_fn);
2337 blk_queue_softirq_done(q, scsi_softirq_done);
2338 blk_queue_rq_timed_out(q, scsi_times_out);
2339 blk_queue_lld_busy(q, scsi_lld_busy);
2343 static const struct blk_mq_ops scsi_mq_ops = {
2344 .get_budget = scsi_mq_get_budget,
2345 .put_budget = scsi_mq_put_budget,
2346 .queue_rq = scsi_queue_rq,
2347 .complete = scsi_softirq_done,
2348 .timeout = scsi_timeout,
2349 #ifdef CONFIG_BLK_DEBUG_FS
2350 .show_rq = scsi_show_rq,
2352 .init_request = scsi_mq_init_request,
2353 .exit_request = scsi_mq_exit_request,
2354 .initialize_rq_fn = scsi_initialize_rq,
2355 .cleanup_rq = scsi_cleanup_rq,
2356 .map_queues = scsi_map_queues,
2359 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2361 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2362 if (IS_ERR(sdev->request_queue))
2365 sdev->request_queue->queuedata = sdev;
2366 __scsi_init_queue(sdev->host, sdev->request_queue);
2367 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
2368 return sdev->request_queue;
2371 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2373 unsigned int cmd_size, sgl_size;
2375 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2376 scsi_mq_sgl_size(shost));
2377 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2378 if (scsi_host_get_prot(shost))
2379 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2381 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2382 shost->tag_set.ops = &scsi_mq_ops;
2383 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2384 shost->tag_set.queue_depth = shost->can_queue;
2385 shost->tag_set.cmd_size = cmd_size;
2386 shost->tag_set.numa_node = NUMA_NO_NODE;
2387 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2388 shost->tag_set.flags |=
2389 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2390 shost->tag_set.driver_data = shost;
2392 return blk_mq_alloc_tag_set(&shost->tag_set);
2395 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2397 blk_mq_free_tag_set(&shost->tag_set);
2401 * scsi_device_from_queue - return sdev associated with a request_queue
2402 * @q: The request queue to return the sdev from
2404 * Return the sdev associated with a request queue or NULL if the
2405 * request_queue does not reference a SCSI device.
2407 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2409 struct scsi_device *sdev = NULL;
2412 if (q->mq_ops == &scsi_mq_ops)
2413 sdev = q->queuedata;
2414 } else if (q->request_fn == scsi_request_fn)
2415 sdev = q->queuedata;
2416 if (!sdev || !get_device(&sdev->sdev_gendev))
2421 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2424 * Function: scsi_block_requests()
2426 * Purpose: Utility function used by low-level drivers to prevent further
2427 * commands from being queued to the device.
2429 * Arguments: shost - Host in question
2433 * Lock status: No locks are assumed held.
2435 * Notes: There is no timer nor any other means by which the requests
2436 * get unblocked other than the low-level driver calling
2437 * scsi_unblock_requests().
2439 void scsi_block_requests(struct Scsi_Host *shost)
2441 shost->host_self_blocked = 1;
2443 EXPORT_SYMBOL(scsi_block_requests);
2446 * Function: scsi_unblock_requests()
2448 * Purpose: Utility function used by low-level drivers to allow further
2449 * commands from being queued to the device.
2451 * Arguments: shost - Host in question
2455 * Lock status: No locks are assumed held.
2457 * Notes: There is no timer nor any other means by which the requests
2458 * get unblocked other than the low-level driver calling
2459 * scsi_unblock_requests().
2461 * This is done as an API function so that changes to the
2462 * internals of the scsi mid-layer won't require wholesale
2463 * changes to drivers that use this feature.
2465 void scsi_unblock_requests(struct Scsi_Host *shost)
2467 shost->host_self_blocked = 0;
2468 scsi_run_host_queues(shost);
2470 EXPORT_SYMBOL(scsi_unblock_requests);
2472 int __init scsi_init_queue(void)
2474 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2475 sizeof(struct scsi_data_buffer),
2477 if (!scsi_sdb_cache) {
2478 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2485 void scsi_exit_queue(void)
2487 kmem_cache_destroy(scsi_sense_cache);
2488 kmem_cache_destroy(scsi_sense_isadma_cache);
2489 kmem_cache_destroy(scsi_sdb_cache);
2493 * scsi_mode_select - issue a mode select
2494 * @sdev: SCSI device to be queried
2495 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2496 * @sp: Save page bit (0 == don't save, 1 == save)
2497 * @modepage: mode page being requested
2498 * @buffer: request buffer (may not be smaller than eight bytes)
2499 * @len: length of request buffer.
2500 * @timeout: command timeout
2501 * @retries: number of retries before failing
2502 * @data: returns a structure abstracting the mode header data
2503 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2504 * must be SCSI_SENSE_BUFFERSIZE big.
2506 * Returns zero if successful; negative error number or scsi
2511 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2512 unsigned char *buffer, int len, int timeout, int retries,
2513 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2515 unsigned char cmd[10];
2516 unsigned char *real_buffer;
2519 memset(cmd, 0, sizeof(cmd));
2520 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2522 if (sdev->use_10_for_ms) {
2525 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2528 memcpy(real_buffer + 8, buffer, len);
2532 real_buffer[2] = data->medium_type;
2533 real_buffer[3] = data->device_specific;
2534 real_buffer[4] = data->longlba ? 0x01 : 0;
2536 real_buffer[6] = data->block_descriptor_length >> 8;
2537 real_buffer[7] = data->block_descriptor_length;
2539 cmd[0] = MODE_SELECT_10;
2543 if (len > 255 || data->block_descriptor_length > 255 ||
2547 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2550 memcpy(real_buffer + 4, buffer, len);
2553 real_buffer[1] = data->medium_type;
2554 real_buffer[2] = data->device_specific;
2555 real_buffer[3] = data->block_descriptor_length;
2558 cmd[0] = MODE_SELECT;
2562 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2563 sshdr, timeout, retries, NULL);
2567 EXPORT_SYMBOL_GPL(scsi_mode_select);
2570 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2571 * @sdev: SCSI device to be queried
2572 * @dbd: set if mode sense will allow block descriptors to be returned
2573 * @modepage: mode page being requested
2574 * @buffer: request buffer (may not be smaller than eight bytes)
2575 * @len: length of request buffer.
2576 * @timeout: command timeout
2577 * @retries: number of retries before failing
2578 * @data: returns a structure abstracting the mode header data
2579 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2580 * must be SCSI_SENSE_BUFFERSIZE big.
2582 * Returns zero if unsuccessful, or the header offset (either 4
2583 * or 8 depending on whether a six or ten byte command was
2584 * issued) if successful.
2587 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2588 unsigned char *buffer, int len, int timeout, int retries,
2589 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2591 unsigned char cmd[12];
2594 int result, retry_count = retries;
2595 struct scsi_sense_hdr my_sshdr;
2597 memset(data, 0, sizeof(*data));
2598 memset(&cmd[0], 0, 12);
2599 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2602 /* caller might not be interested in sense, but we need it */
2607 use_10_for_ms = sdev->use_10_for_ms;
2609 if (use_10_for_ms) {
2613 cmd[0] = MODE_SENSE_10;
2620 cmd[0] = MODE_SENSE;
2625 memset(buffer, 0, len);
2627 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2628 sshdr, timeout, retries, NULL);
2630 /* This code looks awful: what it's doing is making sure an
2631 * ILLEGAL REQUEST sense return identifies the actual command
2632 * byte as the problem. MODE_SENSE commands can return
2633 * ILLEGAL REQUEST if the code page isn't supported */
2635 if (use_10_for_ms && !scsi_status_is_good(result) &&
2636 driver_byte(result) == DRIVER_SENSE) {
2637 if (scsi_sense_valid(sshdr)) {
2638 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2639 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2641 * Invalid command operation code
2643 sdev->use_10_for_ms = 0;
2649 if(scsi_status_is_good(result)) {
2650 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2651 (modepage == 6 || modepage == 8))) {
2652 /* Initio breakage? */
2655 data->medium_type = 0;
2656 data->device_specific = 0;
2658 data->block_descriptor_length = 0;
2659 } else if(use_10_for_ms) {
2660 data->length = buffer[0]*256 + buffer[1] + 2;
2661 data->medium_type = buffer[2];
2662 data->device_specific = buffer[3];
2663 data->longlba = buffer[4] & 0x01;
2664 data->block_descriptor_length = buffer[6]*256
2667 data->length = buffer[0] + 1;
2668 data->medium_type = buffer[1];
2669 data->device_specific = buffer[2];
2670 data->block_descriptor_length = buffer[3];
2672 data->header_length = header_length;
2673 } else if ((status_byte(result) == CHECK_CONDITION) &&
2674 scsi_sense_valid(sshdr) &&
2675 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2682 EXPORT_SYMBOL(scsi_mode_sense);
2685 * scsi_test_unit_ready - test if unit is ready
2686 * @sdev: scsi device to change the state of.
2687 * @timeout: command timeout
2688 * @retries: number of retries before failing
2689 * @sshdr: outpout pointer for decoded sense information.
2691 * Returns zero if unsuccessful or an error if TUR failed. For
2692 * removable media, UNIT_ATTENTION sets ->changed flag.
2695 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2696 struct scsi_sense_hdr *sshdr)
2699 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2703 /* try to eat the UNIT_ATTENTION if there are enough retries */
2705 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2707 if (sdev->removable && scsi_sense_valid(sshdr) &&
2708 sshdr->sense_key == UNIT_ATTENTION)
2710 } while (scsi_sense_valid(sshdr) &&
2711 sshdr->sense_key == UNIT_ATTENTION && --retries);
2715 EXPORT_SYMBOL(scsi_test_unit_ready);
2718 * scsi_device_set_state - Take the given device through the device state model.
2719 * @sdev: scsi device to change the state of.
2720 * @state: state to change to.
2722 * Returns zero if successful or an error if the requested
2723 * transition is illegal.
2726 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2728 enum scsi_device_state oldstate = sdev->sdev_state;
2730 if (state == oldstate)
2736 case SDEV_CREATED_BLOCK:
2747 case SDEV_TRANSPORT_OFFLINE:
2760 case SDEV_TRANSPORT_OFFLINE:
2768 case SDEV_TRANSPORT_OFFLINE:
2783 case SDEV_CREATED_BLOCK:
2790 case SDEV_CREATED_BLOCK:
2805 case SDEV_TRANSPORT_OFFLINE:
2817 case SDEV_TRANSPORT_OFFLINE:
2820 case SDEV_CREATED_BLOCK:
2828 sdev->sdev_state = state;
2832 SCSI_LOG_ERROR_RECOVERY(1,
2833 sdev_printk(KERN_ERR, sdev,
2834 "Illegal state transition %s->%s",
2835 scsi_device_state_name(oldstate),
2836 scsi_device_state_name(state))
2840 EXPORT_SYMBOL(scsi_device_set_state);
2843 * sdev_evt_emit - emit a single SCSI device uevent
2844 * @sdev: associated SCSI device
2845 * @evt: event to emit
2847 * Send a single uevent (scsi_event) to the associated scsi_device.
2849 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2854 switch (evt->evt_type) {
2855 case SDEV_EVT_MEDIA_CHANGE:
2856 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2858 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2859 scsi_rescan_device(&sdev->sdev_gendev);
2860 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2862 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2863 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2865 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2866 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2868 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2869 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2871 case SDEV_EVT_LUN_CHANGE_REPORTED:
2872 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2874 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2875 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2877 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2878 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2887 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2891 * sdev_evt_thread - send a uevent for each scsi event
2892 * @work: work struct for scsi_device
2894 * Dispatch queued events to their associated scsi_device kobjects
2897 void scsi_evt_thread(struct work_struct *work)
2899 struct scsi_device *sdev;
2900 enum scsi_device_event evt_type;
2901 LIST_HEAD(event_list);
2903 sdev = container_of(work, struct scsi_device, event_work);
2905 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2906 if (test_and_clear_bit(evt_type, sdev->pending_events))
2907 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2910 struct scsi_event *evt;
2911 struct list_head *this, *tmp;
2912 unsigned long flags;
2914 spin_lock_irqsave(&sdev->list_lock, flags);
2915 list_splice_init(&sdev->event_list, &event_list);
2916 spin_unlock_irqrestore(&sdev->list_lock, flags);
2918 if (list_empty(&event_list))
2921 list_for_each_safe(this, tmp, &event_list) {
2922 evt = list_entry(this, struct scsi_event, node);
2923 list_del(&evt->node);
2924 scsi_evt_emit(sdev, evt);
2931 * sdev_evt_send - send asserted event to uevent thread
2932 * @sdev: scsi_device event occurred on
2933 * @evt: event to send
2935 * Assert scsi device event asynchronously.
2937 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2939 unsigned long flags;
2942 /* FIXME: currently this check eliminates all media change events
2943 * for polled devices. Need to update to discriminate between AN
2944 * and polled events */
2945 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2951 spin_lock_irqsave(&sdev->list_lock, flags);
2952 list_add_tail(&evt->node, &sdev->event_list);
2953 schedule_work(&sdev->event_work);
2954 spin_unlock_irqrestore(&sdev->list_lock, flags);
2956 EXPORT_SYMBOL_GPL(sdev_evt_send);
2959 * sdev_evt_alloc - allocate a new scsi event
2960 * @evt_type: type of event to allocate
2961 * @gfpflags: GFP flags for allocation
2963 * Allocates and returns a new scsi_event.
2965 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2968 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2972 evt->evt_type = evt_type;
2973 INIT_LIST_HEAD(&evt->node);
2975 /* evt_type-specific initialization, if any */
2977 case SDEV_EVT_MEDIA_CHANGE:
2978 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2979 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2980 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2981 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2982 case SDEV_EVT_LUN_CHANGE_REPORTED:
2983 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2984 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2992 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2995 * sdev_evt_send_simple - send asserted event to uevent thread
2996 * @sdev: scsi_device event occurred on
2997 * @evt_type: type of event to send
2998 * @gfpflags: GFP flags for allocation
3000 * Assert scsi device event asynchronously, given an event type.
3002 void sdev_evt_send_simple(struct scsi_device *sdev,
3003 enum scsi_device_event evt_type, gfp_t gfpflags)
3005 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
3007 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
3012 sdev_evt_send(sdev, evt);
3014 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
3017 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
3018 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
3020 static int scsi_request_fn_active(struct scsi_device *sdev)
3022 struct request_queue *q = sdev->request_queue;
3023 int request_fn_active;
3025 WARN_ON_ONCE(sdev->host->use_blk_mq);
3027 spin_lock_irq(q->queue_lock);
3028 request_fn_active = q->request_fn_active;
3029 spin_unlock_irq(q->queue_lock);
3031 return request_fn_active;
3035 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
3036 * @sdev: SCSI device pointer.
3038 * Wait until the ongoing shost->hostt->queuecommand() calls that are
3039 * invoked from scsi_request_fn() have finished.
3041 static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
3043 WARN_ON_ONCE(sdev->host->use_blk_mq);
3045 while (scsi_request_fn_active(sdev))
3050 * scsi_device_quiesce - Block user issued commands.
3051 * @sdev: scsi device to quiesce.
3053 * This works by trying to transition to the SDEV_QUIESCE state
3054 * (which must be a legal transition). When the device is in this
3055 * state, only special requests will be accepted, all others will
3056 * be deferred. Since special requests may also be requeued requests,
3057 * a successful return doesn't guarantee the device will be
3058 * totally quiescent.
3060 * Must be called with user context, may sleep.
3062 * Returns zero if unsuccessful or an error if not.
3065 scsi_device_quiesce(struct scsi_device *sdev)
3067 struct request_queue *q = sdev->request_queue;
3071 * It is allowed to call scsi_device_quiesce() multiple times from
3072 * the same context but concurrent scsi_device_quiesce() calls are
3075 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
3077 if (sdev->quiesced_by == current)
3082 blk_mq_freeze_queue(q);
3084 * Ensure that the effect of blk_set_pm_only() will be visible
3085 * for percpu_ref_tryget() callers that occur after the queue
3086 * unfreeze even if the queue was already frozen before this function
3087 * was called. See also https://lwn.net/Articles/573497/.
3090 blk_mq_unfreeze_queue(q);
3092 mutex_lock(&sdev->state_mutex);
3093 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
3095 sdev->quiesced_by = current;
3097 blk_clear_pm_only(q);
3098 mutex_unlock(&sdev->state_mutex);
3102 EXPORT_SYMBOL(scsi_device_quiesce);
3105 * scsi_device_resume - Restart user issued commands to a quiesced device.
3106 * @sdev: scsi device to resume.
3108 * Moves the device from quiesced back to running and restarts the
3111 * Must be called with user context, may sleep.
3113 void scsi_device_resume(struct scsi_device *sdev)
3115 /* check if the device state was mutated prior to resume, and if
3116 * so assume the state is being managed elsewhere (for example
3117 * device deleted during suspend)
3119 mutex_lock(&sdev->state_mutex);
3120 if (sdev->quiesced_by) {
3121 sdev->quiesced_by = NULL;
3122 blk_clear_pm_only(sdev->request_queue);
3124 if (sdev->sdev_state == SDEV_QUIESCE)
3125 scsi_device_set_state(sdev, SDEV_RUNNING);
3126 mutex_unlock(&sdev->state_mutex);
3128 EXPORT_SYMBOL(scsi_device_resume);
3131 device_quiesce_fn(struct scsi_device *sdev, void *data)
3133 scsi_device_quiesce(sdev);
3137 scsi_target_quiesce(struct scsi_target *starget)
3139 starget_for_each_device(starget, NULL, device_quiesce_fn);
3141 EXPORT_SYMBOL(scsi_target_quiesce);
3144 device_resume_fn(struct scsi_device *sdev, void *data)
3146 scsi_device_resume(sdev);
3150 scsi_target_resume(struct scsi_target *starget)
3152 starget_for_each_device(starget, NULL, device_resume_fn);
3154 EXPORT_SYMBOL(scsi_target_resume);
3157 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
3158 * @sdev: device to block
3160 * Pause SCSI command processing on the specified device. Does not sleep.
3162 * Returns zero if successful or a negative error code upon failure.
3165 * This routine transitions the device to the SDEV_BLOCK state (which must be
3166 * a legal transition). When the device is in this state, command processing
3167 * is paused until the device leaves the SDEV_BLOCK state. See also
3168 * scsi_internal_device_unblock_nowait().
3170 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
3172 struct request_queue *q = sdev->request_queue;
3173 unsigned long flags;
3176 err = scsi_device_set_state(sdev, SDEV_BLOCK);
3178 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
3185 * The device has transitioned to SDEV_BLOCK. Stop the
3186 * block layer from calling the midlayer with this device's
3190 blk_mq_quiesce_queue_nowait(q);
3192 spin_lock_irqsave(q->queue_lock, flags);
3194 spin_unlock_irqrestore(q->queue_lock, flags);
3199 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
3202 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
3203 * @sdev: device to block
3205 * Pause SCSI command processing on the specified device and wait until all
3206 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
3208 * Returns zero if successful or a negative error code upon failure.
3211 * This routine transitions the device to the SDEV_BLOCK state (which must be
3212 * a legal transition). When the device is in this state, command processing
3213 * is paused until the device leaves the SDEV_BLOCK state. See also
3214 * scsi_internal_device_unblock().
3216 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
3217 * scsi_internal_device_block() has blocked a SCSI device and also
3218 * remove the rport mutex lock and unlock calls from srp_queuecommand().
3220 static int scsi_internal_device_block(struct scsi_device *sdev)
3222 struct request_queue *q = sdev->request_queue;
3225 mutex_lock(&sdev->state_mutex);
3226 err = scsi_internal_device_block_nowait(sdev);
3229 blk_mq_quiesce_queue(q);
3231 scsi_wait_for_queuecommand(sdev);
3233 mutex_unlock(&sdev->state_mutex);
3238 void scsi_start_queue(struct scsi_device *sdev)
3240 struct request_queue *q = sdev->request_queue;
3241 unsigned long flags;
3244 blk_mq_unquiesce_queue(q);
3246 spin_lock_irqsave(q->queue_lock, flags);
3248 spin_unlock_irqrestore(q->queue_lock, flags);
3253 * scsi_internal_device_unblock_nowait - resume a device after a block request
3254 * @sdev: device to resume
3255 * @new_state: state to set the device to after unblocking
3257 * Restart the device queue for a previously suspended SCSI device. Does not
3260 * Returns zero if successful or a negative error code upon failure.
3263 * This routine transitions the device to the SDEV_RUNNING state or to one of
3264 * the offline states (which must be a legal transition) allowing the midlayer
3265 * to goose the queue for this device.
3267 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
3268 enum scsi_device_state new_state)
3271 * Try to transition the scsi device to SDEV_RUNNING or one of the
3272 * offlined states and goose the device queue if successful.
3274 switch (sdev->sdev_state) {
3276 case SDEV_TRANSPORT_OFFLINE:
3277 sdev->sdev_state = new_state;
3279 case SDEV_CREATED_BLOCK:
3280 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3281 new_state == SDEV_OFFLINE)
3282 sdev->sdev_state = new_state;
3284 sdev->sdev_state = SDEV_CREATED;
3292 scsi_start_queue(sdev);
3296 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
3299 * scsi_internal_device_unblock - resume a device after a block request
3300 * @sdev: device to resume
3301 * @new_state: state to set the device to after unblocking
3303 * Restart the device queue for a previously suspended SCSI device. May sleep.
3305 * Returns zero if successful or a negative error code upon failure.
3308 * This routine transitions the device to the SDEV_RUNNING state or to one of
3309 * the offline states (which must be a legal transition) allowing the midlayer
3310 * to goose the queue for this device.
3312 static int scsi_internal_device_unblock(struct scsi_device *sdev,
3313 enum scsi_device_state new_state)
3317 mutex_lock(&sdev->state_mutex);
3318 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
3319 mutex_unlock(&sdev->state_mutex);
3325 device_block(struct scsi_device *sdev, void *data)
3327 scsi_internal_device_block(sdev);
3331 target_block(struct device *dev, void *data)
3333 if (scsi_is_target_device(dev))
3334 starget_for_each_device(to_scsi_target(dev), NULL,
3340 scsi_target_block(struct device *dev)
3342 if (scsi_is_target_device(dev))
3343 starget_for_each_device(to_scsi_target(dev), NULL,
3346 device_for_each_child(dev, NULL, target_block);
3348 EXPORT_SYMBOL_GPL(scsi_target_block);
3351 device_unblock(struct scsi_device *sdev, void *data)
3353 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3357 target_unblock(struct device *dev, void *data)
3359 if (scsi_is_target_device(dev))
3360 starget_for_each_device(to_scsi_target(dev), data,
3366 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3368 if (scsi_is_target_device(dev))
3369 starget_for_each_device(to_scsi_target(dev), &new_state,
3372 device_for_each_child(dev, &new_state, target_unblock);
3374 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3377 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3378 * @sgl: scatter-gather list
3379 * @sg_count: number of segments in sg
3380 * @offset: offset in bytes into sg, on return offset into the mapped area
3381 * @len: bytes to map, on return number of bytes mapped
3383 * Returns virtual address of the start of the mapped page
3385 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3386 size_t *offset, size_t *len)
3389 size_t sg_len = 0, len_complete = 0;
3390 struct scatterlist *sg;
3393 WARN_ON(!irqs_disabled());
3395 for_each_sg(sgl, sg, sg_count, i) {
3396 len_complete = sg_len; /* Complete sg-entries */
3397 sg_len += sg->length;
3398 if (sg_len > *offset)
3402 if (unlikely(i == sg_count)) {
3403 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3405 __func__, sg_len, *offset, sg_count);
3410 /* Offset starting from the beginning of first page in this sg-entry */
3411 *offset = *offset - len_complete + sg->offset;
3413 /* Assumption: contiguous pages can be accessed as "page + i" */
3414 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3415 *offset &= ~PAGE_MASK;
3417 /* Bytes in this sg-entry from *offset to the end of the page */
3418 sg_len = PAGE_SIZE - *offset;
3422 return kmap_atomic(page);
3424 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3427 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3428 * @virt: virtual address to be unmapped
3430 void scsi_kunmap_atomic_sg(void *virt)
3432 kunmap_atomic(virt);
3434 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3436 void sdev_disable_disk_events(struct scsi_device *sdev)
3438 atomic_inc(&sdev->disk_events_disable_depth);
3440 EXPORT_SYMBOL(sdev_disable_disk_events);
3442 void sdev_enable_disk_events(struct scsi_device *sdev)
3444 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3446 atomic_dec(&sdev->disk_events_disable_depth);
3448 EXPORT_SYMBOL(sdev_enable_disk_events);
3450 static unsigned char designator_prio(const unsigned char *d)
3453 /* not associated with LUN */
3457 /* invalid length */
3461 * Order of preference for lun descriptor:
3462 * - SCSI name string
3463 * - NAA IEEE Registered Extended
3464 * - EUI-64 based 16-byte
3465 * - EUI-64 based 12-byte
3466 * - NAA IEEE Registered
3467 * - NAA IEEE Extended
3468 * - EUI-64 based 8-byte
3469 * - SCSI name string (truncated)
3471 * as longer descriptors reduce the likelyhood
3472 * of identification clashes.
3475 switch (d[1] & 0xf) {
3477 /* SCSI name string, variable-length UTF-8 */
3480 switch (d[4] >> 4) {
3482 /* NAA registered extended */
3485 /* NAA registered */
3491 /* NAA locally assigned */
3500 /* EUI64-based, 16 byte */
3503 /* EUI64-based, 12 byte */
3506 /* EUI64-based, 8 byte */
3523 * scsi_vpd_lun_id - return a unique device identification
3524 * @sdev: SCSI device
3525 * @id: buffer for the identification
3526 * @id_len: length of the buffer
3528 * Copies a unique device identification into @id based
3529 * on the information in the VPD page 0x83 of the device.
3530 * The string will be formatted as a SCSI name string.
3532 * Returns the length of the identification or error on failure.
3533 * If the identifier is longer than the supplied buffer the actual
3534 * identifier length is returned and the buffer is not zero-padded.
3536 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3540 const unsigned char *d, *cur_id_str;
3541 const struct scsi_vpd *vpd_pg83;
3542 int id_size = -EINVAL;
3545 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3551 /* The id string must be at least 20 bytes + terminating NULL byte */
3557 memset(id, 0, id_len);
3558 d = vpd_pg83->data + 4;
3559 while (d < vpd_pg83->data + vpd_pg83->len) {
3560 u8 prio = designator_prio(d);
3562 if (prio == 0 || cur_id_prio > prio)
3565 switch (d[1] & 0xf) {
3568 if (cur_id_size > d[3])
3572 if (cur_id_size + 4 > id_len)
3573 cur_id_size = id_len - 4;
3575 id_size = snprintf(id, id_len, "t10.%*pE",
3576 cur_id_size, cur_id_str);
3583 switch (cur_id_size) {
3585 id_size = snprintf(id, id_len,
3590 id_size = snprintf(id, id_len,
3595 id_size = snprintf(id, id_len,
3608 switch (cur_id_size) {
3610 id_size = snprintf(id, id_len,
3615 id_size = snprintf(id, id_len,
3624 /* SCSI name string */
3625 if (cur_id_size > d[3])
3627 /* Prefer others for truncated descriptor */
3628 if (d[3] > id_len) {
3630 if (cur_id_prio > prio)
3634 cur_id_size = id_size = d[3];
3636 if (cur_id_size >= id_len)
3637 cur_id_size = id_len - 1;
3638 memcpy(id, cur_id_str, cur_id_size);
3650 EXPORT_SYMBOL(scsi_vpd_lun_id);
3653 * scsi_vpd_tpg_id - return a target port group identifier
3654 * @sdev: SCSI device
3656 * Returns the Target Port Group identifier from the information
3657 * froom VPD page 0x83 of the device.
3659 * Returns the identifier or error on failure.
3661 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3663 const unsigned char *d;
3664 const struct scsi_vpd *vpd_pg83;
3665 int group_id = -EAGAIN, rel_port = -1;
3668 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3674 d = vpd_pg83->data + 4;
3675 while (d < vpd_pg83->data + vpd_pg83->len) {
3676 switch (d[1] & 0xf) {
3678 /* Relative target port */
3679 rel_port = get_unaligned_be16(&d[6]);
3682 /* Target port group */
3683 group_id = get_unaligned_be16(&d[6]);
3692 if (group_id >= 0 && rel_id && rel_port != -1)
3697 EXPORT_SYMBOL(scsi_vpd_tpg_id);