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 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
78 mutex_lock(&scsi_sense_cache_mutex);
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_flags(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_RECLAIM))
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_start_hw_queues(q);
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 static bool scsi_end_request(struct request *req, blk_status_t error,
666 unsigned int bytes, unsigned int bidi_bytes)
668 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
669 struct scsi_device *sdev = cmd->device;
670 struct request_queue *q = sdev->request_queue;
672 if (blk_update_request(req, error, bytes))
675 /* Bidi request must be completed as a whole */
676 if (unlikely(bidi_bytes) &&
677 blk_update_request(req->next_rq, error, bidi_bytes))
680 if (blk_queue_add_random(q))
681 add_disk_randomness(req->rq_disk);
683 if (!blk_rq_is_scsi(req)) {
684 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
685 cmd->flags &= ~SCMD_INITIALIZED;
686 destroy_rcu_head(&cmd->rcu);
691 * In the MQ case the command gets freed by __blk_mq_end_request,
692 * so we have to do all cleanup that depends on it earlier.
694 * We also can't kick the queues from irq context, so we
695 * will have to defer it to a workqueue.
697 scsi_mq_uninit_cmd(cmd);
699 __blk_mq_end_request(req, error);
701 if (scsi_target(sdev)->single_lun ||
702 !list_empty(&sdev->host->starved_list))
703 kblockd_schedule_work(&sdev->requeue_work);
705 blk_mq_run_hw_queues(q, true);
710 scsi_release_bidi_buffers(cmd);
711 scsi_release_buffers(cmd);
712 scsi_put_command(cmd);
714 spin_lock_irqsave(q->queue_lock, flags);
715 blk_finish_request(req, error);
716 spin_unlock_irqrestore(q->queue_lock, flags);
721 put_device(&sdev->sdev_gendev);
726 * __scsi_error_from_host_byte - translate SCSI error code into errno
727 * @cmd: SCSI command (unused)
728 * @result: scsi error code
730 * Translate SCSI error code into block errors.
732 static blk_status_t __scsi_error_from_host_byte(struct scsi_cmnd *cmd,
735 switch (host_byte(result)) {
738 case DID_TRANSPORT_FAILFAST:
739 return BLK_STS_TRANSPORT;
740 case DID_TARGET_FAILURE:
741 set_host_byte(cmd, DID_OK);
742 return BLK_STS_TARGET;
743 case DID_NEXUS_FAILURE:
744 return BLK_STS_NEXUS;
745 case DID_ALLOC_FAILURE:
746 set_host_byte(cmd, DID_OK);
747 return BLK_STS_NOSPC;
748 case DID_MEDIUM_ERROR:
749 set_host_byte(cmd, DID_OK);
750 return BLK_STS_MEDIUM;
752 return BLK_STS_IOERR;
757 * Function: scsi_io_completion()
759 * Purpose: Completion processing for block device I/O requests.
761 * Arguments: cmd - command that is finished.
763 * Lock status: Assumed that no lock is held upon entry.
767 * Notes: We will finish off the specified number of sectors. If we
768 * are done, the command block will be released and the queue
769 * function will be goosed. If we are not done then we have to
770 * figure out what to do next:
772 * a) We can call scsi_requeue_command(). The request
773 * will be unprepared and put back on the queue. Then
774 * a new command will be created for it. This should
775 * be used if we made forward progress, or if we want
776 * to switch from READ(10) to READ(6) for example.
778 * b) We can call __scsi_queue_insert(). The request will
779 * be put back on the queue and retried using the same
780 * command as before, possibly after a delay.
782 * c) We can call scsi_end_request() with -EIO to fail
783 * the remainder of the request.
785 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
787 int result = cmd->result;
788 struct request_queue *q = cmd->device->request_queue;
789 struct request *req = cmd->request;
790 blk_status_t error = BLK_STS_OK;
791 struct scsi_sense_hdr sshdr;
792 bool sense_valid = false;
793 int sense_deferred = 0, level = 0;
794 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
795 ACTION_DELAYED_RETRY} action;
796 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
799 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
801 sense_deferred = scsi_sense_is_deferred(&sshdr);
804 if (blk_rq_is_passthrough(req)) {
808 * SG_IO wants current and deferred errors
810 scsi_req(req)->sense_len =
811 min(8 + cmd->sense_buffer[7],
812 SCSI_SENSE_BUFFERSIZE);
815 error = __scsi_error_from_host_byte(cmd, result);
818 * __scsi_error_from_host_byte may have reset the host_byte
820 scsi_req(req)->result = cmd->result;
821 scsi_req(req)->resid_len = scsi_get_resid(cmd);
823 if (scsi_bidi_cmnd(cmd)) {
825 * Bidi commands Must be complete as a whole,
826 * both sides at once.
828 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
829 if (scsi_end_request(req, BLK_STS_OK, blk_rq_bytes(req),
830 blk_rq_bytes(req->next_rq)))
834 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
836 * Flush commands do not transfers any data, and thus cannot use
837 * good_bytes != blk_rq_bytes(req) as the signal for an error.
838 * This sets the error explicitly for the problem case.
840 error = __scsi_error_from_host_byte(cmd, result);
843 /* no bidi support for !blk_rq_is_passthrough yet */
844 BUG_ON(blk_bidi_rq(req));
847 * Next deal with any sectors which we were able to correctly
850 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
851 "%u sectors total, %d bytes done.\n",
852 blk_rq_sectors(req), good_bytes));
855 * Recovered errors need reporting, but they're always treated as
856 * success, so fiddle the result code here. For passthrough requests
857 * we already took a copy of the original into sreq->result which
858 * is what gets returned to the user
860 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
861 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
862 * print since caller wants ATA registers. Only occurs on
863 * SCSI ATA PASS_THROUGH commands when CK_COND=1
865 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
867 else if (!(req->rq_flags & RQF_QUIET))
868 scsi_print_sense(cmd);
870 /* for passthrough error may be set */
874 * Another corner case: the SCSI status byte is non-zero but 'good'.
875 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
876 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
877 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
878 * intermediate statuses (both obsolete in SAM-4) as good.
880 if (status_byte(result) && scsi_status_is_good(result)) {
886 * special case: failed zero length commands always need to
887 * drop down into the retry code. Otherwise, if we finished
888 * all bytes in the request we are done now.
890 if (!(blk_rq_bytes(req) == 0 && error) &&
891 !scsi_end_request(req, error, good_bytes, 0))
895 * Kill remainder if no retrys.
897 if (error && scsi_noretry_cmd(cmd)) {
898 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
904 * If there had been no error, but we have leftover bytes in the
905 * requeues just queue the command up again.
910 error = __scsi_error_from_host_byte(cmd, result);
912 if (host_byte(result) == DID_RESET) {
913 /* Third party bus reset or reset for error recovery
914 * reasons. Just retry the command and see what
917 action = ACTION_RETRY;
918 } else if (sense_valid && !sense_deferred) {
919 switch (sshdr.sense_key) {
921 if (cmd->device->removable) {
922 /* Detected disc change. Set a bit
923 * and quietly refuse further access.
925 cmd->device->changed = 1;
926 action = ACTION_FAIL;
928 /* Must have been a power glitch, or a
929 * bus reset. Could not have been a
930 * media change, so we just retry the
931 * command and see what happens.
933 action = ACTION_RETRY;
936 case ILLEGAL_REQUEST:
937 /* If we had an ILLEGAL REQUEST returned, then
938 * we may have performed an unsupported
939 * command. The only thing this should be
940 * would be a ten byte read where only a six
941 * byte read was supported. Also, on a system
942 * where READ CAPACITY failed, we may have
943 * read past the end of the disk.
945 if ((cmd->device->use_10_for_rw &&
946 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
947 (cmd->cmnd[0] == READ_10 ||
948 cmd->cmnd[0] == WRITE_10)) {
949 /* This will issue a new 6-byte command. */
950 cmd->device->use_10_for_rw = 0;
951 action = ACTION_REPREP;
952 } else if (sshdr.asc == 0x10) /* DIX */ {
953 action = ACTION_FAIL;
954 error = BLK_STS_PROTECTION;
955 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
956 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
957 action = ACTION_FAIL;
958 error = BLK_STS_TARGET;
960 action = ACTION_FAIL;
962 case ABORTED_COMMAND:
963 action = ACTION_FAIL;
964 if (sshdr.asc == 0x10) /* DIF */
965 error = BLK_STS_PROTECTION;
968 /* If the device is in the process of becoming
969 * ready, or has a temporary blockage, retry.
971 if (sshdr.asc == 0x04) {
972 switch (sshdr.ascq) {
973 case 0x01: /* becoming ready */
974 case 0x04: /* format in progress */
975 case 0x05: /* rebuild in progress */
976 case 0x06: /* recalculation in progress */
977 case 0x07: /* operation in progress */
978 case 0x08: /* Long write in progress */
979 case 0x09: /* self test in progress */
980 case 0x14: /* space allocation in progress */
981 action = ACTION_DELAYED_RETRY;
984 action = ACTION_FAIL;
988 action = ACTION_FAIL;
990 case VOLUME_OVERFLOW:
991 /* See SSC3rXX or current. */
992 action = ACTION_FAIL;
995 action = ACTION_FAIL;
999 action = ACTION_FAIL;
1001 if (action != ACTION_FAIL &&
1002 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1003 action = ACTION_FAIL;
1007 /* Give up and fail the remainder of the request */
1008 if (!(req->rq_flags & RQF_QUIET)) {
1009 static DEFINE_RATELIMIT_STATE(_rs,
1010 DEFAULT_RATELIMIT_INTERVAL,
1011 DEFAULT_RATELIMIT_BURST);
1013 if (unlikely(scsi_logging_level))
1014 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1015 SCSI_LOG_MLCOMPLETE_BITS);
1018 * if logging is enabled the failure will be printed
1019 * in scsi_log_completion(), so avoid duplicate messages
1021 if (!level && __ratelimit(&_rs)) {
1022 scsi_print_result(cmd, NULL, FAILED);
1023 if (driver_byte(result) & DRIVER_SENSE)
1024 scsi_print_sense(cmd);
1025 scsi_print_command(cmd);
1028 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1033 /* Unprep the request and put it back at the head of the queue.
1034 * A new command will be prepared and issued.
1037 scsi_mq_requeue_cmd(cmd);
1039 scsi_release_buffers(cmd);
1040 scsi_requeue_command(q, cmd);
1044 /* Retry the same command immediately */
1045 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
1047 case ACTION_DELAYED_RETRY:
1048 /* Retry the same command after a delay */
1049 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
1054 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1059 * If sg table allocation fails, requeue request later.
1061 if (unlikely(sg_alloc_table_chained(&sdb->table,
1062 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1063 return BLKPREP_DEFER;
1066 * Next, walk the list, and fill in the addresses and sizes of
1069 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1070 BUG_ON(count > sdb->table.nents);
1071 sdb->table.nents = count;
1072 sdb->length = blk_rq_payload_bytes(req);
1077 * Function: scsi_init_io()
1079 * Purpose: SCSI I/O initialize function.
1081 * Arguments: cmd - Command descriptor we wish to initialize
1083 * Returns: 0 on success
1084 * BLKPREP_DEFER if the failure is retryable
1085 * BLKPREP_KILL if the failure is fatal
1087 int scsi_init_io(struct scsi_cmnd *cmd)
1089 struct scsi_device *sdev = cmd->device;
1090 struct request *rq = cmd->request;
1091 bool is_mq = (rq->mq_ctx != NULL);
1092 int error = BLKPREP_KILL;
1094 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1097 error = scsi_init_sgtable(rq, &cmd->sdb);
1101 if (blk_bidi_rq(rq)) {
1102 if (!rq->q->mq_ops) {
1103 struct scsi_data_buffer *bidi_sdb =
1104 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1106 error = BLKPREP_DEFER;
1110 rq->next_rq->special = bidi_sdb;
1113 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1118 if (blk_integrity_rq(rq)) {
1119 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1122 if (prot_sdb == NULL) {
1124 * This can happen if someone (e.g. multipath)
1125 * queues a command to a device on an adapter
1126 * that does not support DIX.
1129 error = BLKPREP_KILL;
1133 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1135 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1136 prot_sdb->table.sgl)) {
1137 error = BLKPREP_DEFER;
1141 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1142 prot_sdb->table.sgl);
1143 BUG_ON(unlikely(count > ivecs));
1144 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1146 cmd->prot_sdb = prot_sdb;
1147 cmd->prot_sdb->table.nents = count;
1153 scsi_mq_free_sgtables(cmd);
1155 scsi_release_buffers(cmd);
1156 cmd->request->special = NULL;
1157 scsi_put_command(cmd);
1158 put_device(&sdev->sdev_gendev);
1162 EXPORT_SYMBOL(scsi_init_io);
1165 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1166 * @rq: Request associated with the SCSI command to be initialized.
1168 * This function initializes the members of struct scsi_cmnd that must be
1169 * initialized before request processing starts and that won't be
1170 * reinitialized if a SCSI command is requeued.
1172 * Called from inside blk_get_request() for pass-through requests and from
1173 * inside scsi_init_command() for filesystem requests.
1175 static void scsi_initialize_rq(struct request *rq)
1177 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1179 scsi_req_init(&cmd->req);
1180 init_rcu_head(&cmd->rcu);
1181 cmd->jiffies_at_alloc = jiffies;
1185 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1186 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1188 struct scsi_device *sdev = cmd->device;
1189 struct Scsi_Host *shost = sdev->host;
1190 unsigned long flags;
1192 if (shost->use_cmd_list) {
1193 spin_lock_irqsave(&sdev->list_lock, flags);
1194 list_add_tail(&cmd->list, &sdev->cmd_list);
1195 spin_unlock_irqrestore(&sdev->list_lock, flags);
1199 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1200 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1202 struct scsi_device *sdev = cmd->device;
1203 struct Scsi_Host *shost = sdev->host;
1204 unsigned long flags;
1206 if (shost->use_cmd_list) {
1207 spin_lock_irqsave(&sdev->list_lock, flags);
1208 BUG_ON(list_empty(&cmd->list));
1209 list_del_init(&cmd->list);
1210 spin_unlock_irqrestore(&sdev->list_lock, flags);
1214 /* Called after a request has been started. */
1215 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1217 void *buf = cmd->sense_buffer;
1218 void *prot = cmd->prot_sdb;
1219 struct request *rq = blk_mq_rq_from_pdu(cmd);
1220 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1221 unsigned long jiffies_at_alloc;
1224 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1225 flags |= SCMD_INITIALIZED;
1226 scsi_initialize_rq(rq);
1229 jiffies_at_alloc = cmd->jiffies_at_alloc;
1230 retries = cmd->retries;
1231 /* zero out the cmd, except for the embedded scsi_request */
1232 memset((char *)cmd + sizeof(cmd->req), 0,
1233 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1236 cmd->sense_buffer = buf;
1237 cmd->prot_sdb = prot;
1239 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1240 cmd->jiffies_at_alloc = jiffies_at_alloc;
1241 cmd->retries = retries;
1243 scsi_add_cmd_to_list(cmd);
1246 static int scsi_setup_scsi_cmnd(struct scsi_device *sdev, struct request *req)
1248 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1251 * Passthrough requests may transfer data, in which case they must
1252 * a bio attached to them. Or they might contain a SCSI command
1253 * that does not transfer data, in which case they may optionally
1254 * submit a request without an attached bio.
1257 int ret = scsi_init_io(cmd);
1261 BUG_ON(blk_rq_bytes(req));
1263 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1266 cmd->cmd_len = scsi_req(req)->cmd_len;
1267 cmd->cmnd = scsi_req(req)->cmd;
1268 cmd->transfersize = blk_rq_bytes(req);
1269 cmd->allowed = scsi_req(req)->retries;
1274 * Setup a normal block command. These are simple request from filesystems
1275 * that still need to be translated to SCSI CDBs from the ULD.
1277 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1279 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1281 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1282 int ret = sdev->handler->prep_fn(sdev, req);
1283 if (ret != BLKPREP_OK)
1287 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1288 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1289 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1292 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1294 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1296 if (!blk_rq_bytes(req))
1297 cmd->sc_data_direction = DMA_NONE;
1298 else if (rq_data_dir(req) == WRITE)
1299 cmd->sc_data_direction = DMA_TO_DEVICE;
1301 cmd->sc_data_direction = DMA_FROM_DEVICE;
1303 if (blk_rq_is_scsi(req))
1304 return scsi_setup_scsi_cmnd(sdev, req);
1306 return scsi_setup_fs_cmnd(sdev, req);
1310 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1312 int ret = BLKPREP_OK;
1315 * If the device is not in running state we will reject some
1318 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1319 switch (sdev->sdev_state) {
1321 case SDEV_TRANSPORT_OFFLINE:
1323 * If the device is offline we refuse to process any
1324 * commands. The device must be brought online
1325 * before trying any recovery commands.
1327 sdev_printk(KERN_ERR, sdev,
1328 "rejecting I/O to offline device\n");
1333 * If the device is fully deleted, we refuse to
1334 * process any commands as well.
1336 sdev_printk(KERN_ERR, sdev,
1337 "rejecting I/O to dead device\n");
1341 case SDEV_CREATED_BLOCK:
1342 ret = BLKPREP_DEFER;
1346 * If the devices is blocked we defer normal commands.
1348 if (req && !(req->rq_flags & RQF_PREEMPT))
1349 ret = BLKPREP_DEFER;
1353 * For any other not fully online state we only allow
1354 * special commands. In particular any user initiated
1355 * command is not allowed.
1357 if (req && !(req->rq_flags & RQF_PREEMPT))
1366 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1368 struct scsi_device *sdev = q->queuedata;
1372 case BLKPREP_INVALID:
1373 scsi_req(req)->result = DID_NO_CONNECT << 16;
1374 /* release the command and kill it */
1376 struct scsi_cmnd *cmd = req->special;
1377 scsi_release_buffers(cmd);
1378 scsi_put_command(cmd);
1379 put_device(&sdev->sdev_gendev);
1380 req->special = NULL;
1385 * If we defer, the blk_peek_request() returns NULL, but the
1386 * queue must be restarted, so we schedule a callback to happen
1389 if (atomic_read(&sdev->device_busy) == 0)
1390 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1393 req->rq_flags |= RQF_DONTPREP;
1399 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1401 struct scsi_device *sdev = q->queuedata;
1402 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1405 ret = scsi_prep_state_check(sdev, req);
1406 if (ret != BLKPREP_OK)
1409 if (!req->special) {
1410 /* Bail if we can't get a reference to the device */
1411 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1412 ret = BLKPREP_DEFER;
1416 scsi_init_command(sdev, cmd);
1420 cmd->tag = req->tag;
1422 cmd->prot_op = SCSI_PROT_NORMAL;
1424 ret = scsi_setup_cmnd(sdev, req);
1426 return scsi_prep_return(q, req, ret);
1429 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1431 scsi_uninit_cmd(blk_mq_rq_to_pdu(req));
1435 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1438 * Called with the queue_lock held.
1440 static inline int scsi_dev_queue_ready(struct request_queue *q,
1441 struct scsi_device *sdev)
1445 busy = atomic_inc_return(&sdev->device_busy) - 1;
1446 if (atomic_read(&sdev->device_blocked)) {
1451 * unblock after device_blocked iterates to zero
1453 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1455 * For the MQ case we take care of this in the caller.
1458 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1461 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1462 "unblocking device at zero depth\n"));
1465 if (busy >= sdev->queue_depth)
1470 atomic_dec(&sdev->device_busy);
1475 * scsi_target_queue_ready: checks if there we can send commands to target
1476 * @sdev: scsi device on starget to check.
1478 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1479 struct scsi_device *sdev)
1481 struct scsi_target *starget = scsi_target(sdev);
1484 if (starget->single_lun) {
1485 spin_lock_irq(shost->host_lock);
1486 if (starget->starget_sdev_user &&
1487 starget->starget_sdev_user != sdev) {
1488 spin_unlock_irq(shost->host_lock);
1491 starget->starget_sdev_user = sdev;
1492 spin_unlock_irq(shost->host_lock);
1495 if (starget->can_queue <= 0)
1498 busy = atomic_inc_return(&starget->target_busy) - 1;
1499 if (atomic_read(&starget->target_blocked) > 0) {
1504 * unblock after target_blocked iterates to zero
1506 if (atomic_dec_return(&starget->target_blocked) > 0)
1509 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1510 "unblocking target at zero depth\n"));
1513 if (busy >= starget->can_queue)
1519 spin_lock_irq(shost->host_lock);
1520 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1521 spin_unlock_irq(shost->host_lock);
1523 if (starget->can_queue > 0)
1524 atomic_dec(&starget->target_busy);
1529 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1530 * return 0. We must end up running the queue again whenever 0 is
1531 * returned, else IO can hang.
1533 static inline int scsi_host_queue_ready(struct request_queue *q,
1534 struct Scsi_Host *shost,
1535 struct scsi_device *sdev)
1539 if (scsi_host_in_recovery(shost))
1542 busy = atomic_inc_return(&shost->host_busy) - 1;
1543 if (atomic_read(&shost->host_blocked) > 0) {
1548 * unblock after host_blocked iterates to zero
1550 if (atomic_dec_return(&shost->host_blocked) > 0)
1554 shost_printk(KERN_INFO, shost,
1555 "unblocking host at zero depth\n"));
1558 if (shost->can_queue > 0 && busy >= shost->can_queue)
1560 if (shost->host_self_blocked)
1563 /* We're OK to process the command, so we can't be starved */
1564 if (!list_empty(&sdev->starved_entry)) {
1565 spin_lock_irq(shost->host_lock);
1566 if (!list_empty(&sdev->starved_entry))
1567 list_del_init(&sdev->starved_entry);
1568 spin_unlock_irq(shost->host_lock);
1574 spin_lock_irq(shost->host_lock);
1575 if (list_empty(&sdev->starved_entry))
1576 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1577 spin_unlock_irq(shost->host_lock);
1579 scsi_dec_host_busy(shost);
1584 * Busy state exporting function for request stacking drivers.
1586 * For efficiency, no lock is taken to check the busy state of
1587 * shost/starget/sdev, since the returned value is not guaranteed and
1588 * may be changed after request stacking drivers call the function,
1589 * regardless of taking lock or not.
1591 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1592 * needs to return 'not busy'. Otherwise, request stacking drivers
1593 * may hold requests forever.
1595 static int scsi_lld_busy(struct request_queue *q)
1597 struct scsi_device *sdev = q->queuedata;
1598 struct Scsi_Host *shost;
1600 if (blk_queue_dying(q))
1606 * Ignore host/starget busy state.
1607 * Since block layer does not have a concept of fairness across
1608 * multiple queues, congestion of host/starget needs to be handled
1611 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1618 * Kill a request for a dead device
1620 static void scsi_kill_request(struct request *req, struct request_queue *q)
1622 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1623 struct scsi_device *sdev;
1624 struct scsi_target *starget;
1625 struct Scsi_Host *shost;
1627 blk_start_request(req);
1629 scmd_printk(KERN_INFO, cmd, "killing request\n");
1632 starget = scsi_target(sdev);
1634 scsi_init_cmd_errh(cmd);
1635 cmd->result = DID_NO_CONNECT << 16;
1636 atomic_inc(&cmd->device->iorequest_cnt);
1639 * SCSI request completion path will do scsi_device_unbusy(),
1640 * bump busy counts. To bump the counters, we need to dance
1641 * with the locks as normal issue path does.
1643 atomic_inc(&sdev->device_busy);
1644 atomic_inc(&shost->host_busy);
1645 if (starget->can_queue > 0)
1646 atomic_inc(&starget->target_busy);
1648 blk_complete_request(req);
1651 static void scsi_softirq_done(struct request *rq)
1653 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1654 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1657 INIT_LIST_HEAD(&cmd->eh_entry);
1659 atomic_inc(&cmd->device->iodone_cnt);
1661 atomic_inc(&cmd->device->ioerr_cnt);
1663 disposition = scsi_decide_disposition(cmd);
1664 if (disposition != SUCCESS &&
1665 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1666 sdev_printk(KERN_ERR, cmd->device,
1667 "timing out command, waited %lus\n",
1669 disposition = SUCCESS;
1672 scsi_log_completion(cmd, disposition);
1674 switch (disposition) {
1676 scsi_finish_command(cmd);
1679 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1681 case ADD_TO_MLQUEUE:
1682 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1685 scsi_eh_scmd_add(cmd);
1691 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1692 * @cmd: command block we are dispatching.
1694 * Return: nonzero return request was rejected and device's queue needs to be
1697 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1699 struct Scsi_Host *host = cmd->device->host;
1702 atomic_inc(&cmd->device->iorequest_cnt);
1704 /* check if the device is still usable */
1705 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1706 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1707 * returns an immediate error upwards, and signals
1708 * that the device is no longer present */
1709 cmd->result = DID_NO_CONNECT << 16;
1713 /* Check to see if the scsi lld made this device blocked. */
1714 if (unlikely(scsi_device_blocked(cmd->device))) {
1716 * in blocked state, the command is just put back on
1717 * the device queue. The suspend state has already
1718 * blocked the queue so future requests should not
1719 * occur until the device transitions out of the
1722 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1723 "queuecommand : device blocked\n"));
1724 return SCSI_MLQUEUE_DEVICE_BUSY;
1727 /* Store the LUN value in cmnd, if needed. */
1728 if (cmd->device->lun_in_cdb)
1729 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1730 (cmd->device->lun << 5 & 0xe0);
1735 * Before we queue this command, check if the command
1736 * length exceeds what the host adapter can handle.
1738 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1739 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1740 "queuecommand : command too long. "
1741 "cdb_size=%d host->max_cmd_len=%d\n",
1742 cmd->cmd_len, cmd->device->host->max_cmd_len));
1743 cmd->result = (DID_ABORT << 16);
1747 if (unlikely(host->shost_state == SHOST_DEL)) {
1748 cmd->result = (DID_NO_CONNECT << 16);
1753 trace_scsi_dispatch_cmd_start(cmd);
1754 rtn = host->hostt->queuecommand(host, cmd);
1756 trace_scsi_dispatch_cmd_error(cmd, rtn);
1757 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1758 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1759 rtn = SCSI_MLQUEUE_HOST_BUSY;
1761 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1762 "queuecommand : request rejected\n"));
1767 cmd->scsi_done(cmd);
1772 * scsi_done - Invoke completion on finished SCSI command.
1773 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1774 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1776 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1777 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1778 * calls blk_complete_request() for further processing.
1780 * This function is interrupt context safe.
1782 static void scsi_done(struct scsi_cmnd *cmd)
1784 trace_scsi_dispatch_cmd_done(cmd);
1785 blk_complete_request(cmd->request);
1789 * Function: scsi_request_fn()
1791 * Purpose: Main strategy routine for SCSI.
1793 * Arguments: q - Pointer to actual queue.
1797 * Lock status: request queue lock assumed to be held when called.
1799 * Note: See sd_zbc.c sd_zbc_write_lock_zone() for write order
1800 * protection for ZBC disks.
1802 static void scsi_request_fn(struct request_queue *q)
1803 __releases(q->queue_lock)
1804 __acquires(q->queue_lock)
1806 struct scsi_device *sdev = q->queuedata;
1807 struct Scsi_Host *shost;
1808 struct scsi_cmnd *cmd;
1809 struct request *req;
1812 * To start with, we keep looping until the queue is empty, or until
1813 * the host is no longer able to accept any more requests.
1819 * get next queueable request. We do this early to make sure
1820 * that the request is fully prepared even if we cannot
1823 req = blk_peek_request(q);
1827 if (unlikely(!scsi_device_online(sdev))) {
1828 sdev_printk(KERN_ERR, sdev,
1829 "rejecting I/O to offline device\n");
1830 scsi_kill_request(req, q);
1834 if (!scsi_dev_queue_ready(q, sdev))
1838 * Remove the request from the request list.
1840 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1841 blk_start_request(req);
1843 spin_unlock_irq(q->queue_lock);
1844 cmd = blk_mq_rq_to_pdu(req);
1845 if (cmd != req->special) {
1846 printk(KERN_CRIT "impossible request in %s.\n"
1847 "please mail a stack trace to "
1848 "linux-scsi@vger.kernel.org\n",
1850 blk_dump_rq_flags(req, "foo");
1855 * We hit this when the driver is using a host wide
1856 * tag map. For device level tag maps the queue_depth check
1857 * in the device ready fn would prevent us from trying
1858 * to allocate a tag. Since the map is a shared host resource
1859 * we add the dev to the starved list so it eventually gets
1860 * a run when a tag is freed.
1862 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1863 spin_lock_irq(shost->host_lock);
1864 if (list_empty(&sdev->starved_entry))
1865 list_add_tail(&sdev->starved_entry,
1866 &shost->starved_list);
1867 spin_unlock_irq(shost->host_lock);
1871 if (!scsi_target_queue_ready(shost, sdev))
1874 if (!scsi_host_queue_ready(q, shost, sdev))
1875 goto host_not_ready;
1877 if (sdev->simple_tags)
1878 cmd->flags |= SCMD_TAGGED;
1880 cmd->flags &= ~SCMD_TAGGED;
1883 * Finally, initialize any error handling parameters, and set up
1884 * the timers for timeouts.
1886 scsi_init_cmd_errh(cmd);
1889 * Dispatch the command to the low-level driver.
1891 cmd->scsi_done = scsi_done;
1892 rtn = scsi_dispatch_cmd(cmd);
1894 scsi_queue_insert(cmd, rtn);
1895 spin_lock_irq(q->queue_lock);
1898 spin_lock_irq(q->queue_lock);
1904 if (scsi_target(sdev)->can_queue > 0)
1905 atomic_dec(&scsi_target(sdev)->target_busy);
1908 * lock q, handle tag, requeue req, and decrement device_busy. We
1909 * must return with queue_lock held.
1911 * Decrementing device_busy without checking it is OK, as all such
1912 * cases (host limits or settings) should run the queue at some
1915 spin_lock_irq(q->queue_lock);
1916 blk_requeue_request(q, req);
1917 atomic_dec(&sdev->device_busy);
1919 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1920 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1923 static inline blk_status_t prep_to_mq(int ret)
1929 return BLK_STS_RESOURCE;
1931 return BLK_STS_IOERR;
1935 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1936 static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1938 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1939 sizeof(struct scatterlist);
1942 static int scsi_mq_prep_fn(struct request *req)
1944 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1945 struct scsi_device *sdev = req->q->queuedata;
1946 struct Scsi_Host *shost = sdev->host;
1947 struct scatterlist *sg;
1949 scsi_init_command(sdev, cmd);
1955 cmd->tag = req->tag;
1956 cmd->prot_op = SCSI_PROT_NORMAL;
1958 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1959 cmd->sdb.table.sgl = sg;
1961 if (scsi_host_get_prot(shost)) {
1962 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1964 cmd->prot_sdb->table.sgl =
1965 (struct scatterlist *)(cmd->prot_sdb + 1);
1968 if (blk_bidi_rq(req)) {
1969 struct request *next_rq = req->next_rq;
1970 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1972 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1973 bidi_sdb->table.sgl =
1974 (struct scatterlist *)(bidi_sdb + 1);
1976 next_rq->special = bidi_sdb;
1979 blk_mq_start_request(req);
1981 return scsi_setup_cmnd(sdev, req);
1984 static void scsi_mq_done(struct scsi_cmnd *cmd)
1986 trace_scsi_dispatch_cmd_done(cmd);
1987 blk_mq_complete_request(cmd->request);
1990 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1992 struct request_queue *q = hctx->queue;
1993 struct scsi_device *sdev = q->queuedata;
1995 atomic_dec(&sdev->device_busy);
1996 put_device(&sdev->sdev_gendev);
1999 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
2001 struct request_queue *q = hctx->queue;
2002 struct scsi_device *sdev = q->queuedata;
2004 if (!get_device(&sdev->sdev_gendev))
2006 if (!scsi_dev_queue_ready(q, sdev))
2007 goto out_put_device;
2012 put_device(&sdev->sdev_gendev);
2014 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
2015 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
2019 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
2020 const struct blk_mq_queue_data *bd)
2022 struct request *req = bd->rq;
2023 struct request_queue *q = req->q;
2024 struct scsi_device *sdev = q->queuedata;
2025 struct Scsi_Host *shost = sdev->host;
2026 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
2030 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
2031 if (ret != BLK_STS_OK)
2032 goto out_put_budget;
2034 ret = BLK_STS_RESOURCE;
2035 if (!scsi_target_queue_ready(shost, sdev))
2036 goto out_put_budget;
2037 if (!scsi_host_queue_ready(q, shost, sdev))
2038 goto out_dec_target_busy;
2040 if (!(req->rq_flags & RQF_DONTPREP)) {
2041 ret = prep_to_mq(scsi_mq_prep_fn(req));
2042 if (ret != BLK_STS_OK)
2043 goto out_dec_host_busy;
2044 req->rq_flags |= RQF_DONTPREP;
2046 blk_mq_start_request(req);
2049 if (sdev->simple_tags)
2050 cmd->flags |= SCMD_TAGGED;
2052 cmd->flags &= ~SCMD_TAGGED;
2054 scsi_init_cmd_errh(cmd);
2055 cmd->scsi_done = scsi_mq_done;
2057 reason = scsi_dispatch_cmd(cmd);
2059 scsi_set_blocked(cmd, reason);
2060 ret = BLK_STS_RESOURCE;
2061 goto out_dec_host_busy;
2067 scsi_dec_host_busy(shost);
2068 out_dec_target_busy:
2069 if (scsi_target(sdev)->can_queue > 0)
2070 atomic_dec(&scsi_target(sdev)->target_busy);
2072 scsi_mq_put_budget(hctx);
2076 case BLK_STS_RESOURCE:
2077 if (atomic_read(&sdev->device_busy) ||
2078 scsi_device_blocked(sdev))
2079 ret = BLK_STS_DEV_RESOURCE;
2083 * Make sure to release all allocated ressources when
2084 * we hit an error, as we will never see this command
2087 if (req->rq_flags & RQF_DONTPREP)
2088 scsi_mq_uninit_cmd(cmd);
2094 static enum blk_eh_timer_return scsi_timeout(struct request *req,
2098 return BLK_EH_RESET_TIMER;
2099 return scsi_times_out(req);
2102 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
2103 unsigned int hctx_idx, unsigned int numa_node)
2105 struct Scsi_Host *shost = set->driver_data;
2106 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2107 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2108 struct scatterlist *sg;
2110 if (unchecked_isa_dma)
2111 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2112 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
2113 GFP_KERNEL, numa_node);
2114 if (!cmd->sense_buffer)
2116 cmd->req.sense = cmd->sense_buffer;
2118 if (scsi_host_get_prot(shost)) {
2119 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
2120 shost->hostt->cmd_size;
2121 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
2127 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
2128 unsigned int hctx_idx)
2130 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2132 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2136 static int scsi_map_queues(struct blk_mq_tag_set *set)
2138 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2140 if (shost->hostt->map_queues)
2141 return shost->hostt->map_queues(shost);
2142 return blk_mq_map_queues(set);
2145 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2147 struct device *host_dev;
2148 u64 bounce_limit = 0xffffffff;
2150 if (shost->unchecked_isa_dma)
2151 return BLK_BOUNCE_ISA;
2153 * Platforms with virtual-DMA translation
2154 * hardware have no practical limit.
2156 if (!PCI_DMA_BUS_IS_PHYS)
2157 return BLK_BOUNCE_ANY;
2159 host_dev = scsi_get_device(shost);
2160 if (host_dev && host_dev->dma_mask)
2161 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2163 return bounce_limit;
2166 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2168 struct device *dev = shost->dma_dev;
2170 queue_flag_set_unlocked(QUEUE_FLAG_SCSI_PASSTHROUGH, q);
2173 * this limit is imposed by hardware restrictions
2175 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2178 if (scsi_host_prot_dma(shost)) {
2179 shost->sg_prot_tablesize =
2180 min_not_zero(shost->sg_prot_tablesize,
2181 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2182 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2183 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2186 blk_queue_max_hw_sectors(q, shost->max_sectors);
2187 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2188 blk_queue_segment_boundary(q, shost->dma_boundary);
2189 dma_set_seg_boundary(dev, shost->dma_boundary);
2191 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2193 if (!shost->use_clustering)
2194 q->limits.cluster = 0;
2197 * Set a reasonable default alignment: The larger of 32-byte (dword),
2198 * which is a common minimum for HBAs, and the minimum DMA alignment,
2199 * which is set by the platform.
2201 * Devices that require a bigger alignment can increase it later.
2203 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
2205 EXPORT_SYMBOL_GPL(__scsi_init_queue);
2207 static int scsi_old_init_rq(struct request_queue *q, struct request *rq,
2210 struct Scsi_Host *shost = q->rq_alloc_data;
2211 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
2212 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2214 memset(cmd, 0, sizeof(*cmd));
2216 if (unchecked_isa_dma)
2217 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
2218 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma, gfp,
2220 if (!cmd->sense_buffer)
2222 cmd->req.sense = cmd->sense_buffer;
2224 if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2225 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2227 goto fail_free_sense;
2233 scsi_free_sense_buffer(unchecked_isa_dma, cmd->sense_buffer);
2238 static void scsi_old_exit_rq(struct request_queue *q, struct request *rq)
2240 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2243 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2244 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
2248 struct request_queue *scsi_old_alloc_queue(struct scsi_device *sdev)
2250 struct Scsi_Host *shost = sdev->host;
2251 struct request_queue *q;
2253 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE);
2256 q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2257 q->rq_alloc_data = shost;
2258 q->request_fn = scsi_request_fn;
2259 q->init_rq_fn = scsi_old_init_rq;
2260 q->exit_rq_fn = scsi_old_exit_rq;
2261 q->initialize_rq_fn = scsi_initialize_rq;
2263 if (blk_init_allocated_queue(q) < 0) {
2264 blk_cleanup_queue(q);
2268 __scsi_init_queue(shost, q);
2269 blk_queue_prep_rq(q, scsi_prep_fn);
2270 blk_queue_unprep_rq(q, scsi_unprep_fn);
2271 blk_queue_softirq_done(q, scsi_softirq_done);
2272 blk_queue_rq_timed_out(q, scsi_times_out);
2273 blk_queue_lld_busy(q, scsi_lld_busy);
2277 static const struct blk_mq_ops scsi_mq_ops = {
2278 .get_budget = scsi_mq_get_budget,
2279 .put_budget = scsi_mq_put_budget,
2280 .queue_rq = scsi_queue_rq,
2281 .complete = scsi_softirq_done,
2282 .timeout = scsi_timeout,
2283 #ifdef CONFIG_BLK_DEBUG_FS
2284 .show_rq = scsi_show_rq,
2286 .init_request = scsi_mq_init_request,
2287 .exit_request = scsi_mq_exit_request,
2288 .initialize_rq_fn = scsi_initialize_rq,
2289 .map_queues = scsi_map_queues,
2292 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2294 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2295 if (IS_ERR(sdev->request_queue))
2298 sdev->request_queue->queuedata = sdev;
2299 __scsi_init_queue(sdev->host, sdev->request_queue);
2300 return sdev->request_queue;
2303 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2305 unsigned int cmd_size, sgl_size;
2307 sgl_size = scsi_mq_sgl_size(shost);
2308 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2309 if (scsi_host_get_prot(shost))
2310 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2312 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2313 shost->tag_set.ops = &scsi_mq_ops;
2314 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2315 shost->tag_set.queue_depth = shost->can_queue;
2316 shost->tag_set.cmd_size = cmd_size;
2317 shost->tag_set.numa_node = NUMA_NO_NODE;
2318 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2319 shost->tag_set.flags |=
2320 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2321 shost->tag_set.driver_data = shost;
2323 return blk_mq_alloc_tag_set(&shost->tag_set);
2326 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2328 blk_mq_free_tag_set(&shost->tag_set);
2332 * scsi_device_from_queue - return sdev associated with a request_queue
2333 * @q: The request queue to return the sdev from
2335 * Return the sdev associated with a request queue or NULL if the
2336 * request_queue does not reference a SCSI device.
2338 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2340 struct scsi_device *sdev = NULL;
2343 if (q->mq_ops == &scsi_mq_ops)
2344 sdev = q->queuedata;
2345 } else if (q->request_fn == scsi_request_fn)
2346 sdev = q->queuedata;
2347 if (!sdev || !get_device(&sdev->sdev_gendev))
2352 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2355 * Function: scsi_block_requests()
2357 * Purpose: Utility function used by low-level drivers to prevent further
2358 * commands from being queued to the device.
2360 * Arguments: shost - Host in question
2364 * Lock status: No locks are assumed held.
2366 * Notes: There is no timer nor any other means by which the requests
2367 * get unblocked other than the low-level driver calling
2368 * scsi_unblock_requests().
2370 void scsi_block_requests(struct Scsi_Host *shost)
2372 shost->host_self_blocked = 1;
2374 EXPORT_SYMBOL(scsi_block_requests);
2377 * Function: scsi_unblock_requests()
2379 * Purpose: Utility function used by low-level drivers to allow further
2380 * commands from being queued to the device.
2382 * Arguments: shost - Host in question
2386 * Lock status: No locks are assumed held.
2388 * Notes: There is no timer nor any other means by which the requests
2389 * get unblocked other than the low-level driver calling
2390 * scsi_unblock_requests().
2392 * This is done as an API function so that changes to the
2393 * internals of the scsi mid-layer won't require wholesale
2394 * changes to drivers that use this feature.
2396 void scsi_unblock_requests(struct Scsi_Host *shost)
2398 shost->host_self_blocked = 0;
2399 scsi_run_host_queues(shost);
2401 EXPORT_SYMBOL(scsi_unblock_requests);
2403 int __init scsi_init_queue(void)
2405 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2406 sizeof(struct scsi_data_buffer),
2408 if (!scsi_sdb_cache) {
2409 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2416 void scsi_exit_queue(void)
2418 kmem_cache_destroy(scsi_sense_cache);
2419 kmem_cache_destroy(scsi_sense_isadma_cache);
2420 kmem_cache_destroy(scsi_sdb_cache);
2424 * scsi_mode_select - issue a mode select
2425 * @sdev: SCSI device to be queried
2426 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2427 * @sp: Save page bit (0 == don't save, 1 == save)
2428 * @modepage: mode page being requested
2429 * @buffer: request buffer (may not be smaller than eight bytes)
2430 * @len: length of request buffer.
2431 * @timeout: command timeout
2432 * @retries: number of retries before failing
2433 * @data: returns a structure abstracting the mode header data
2434 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2435 * must be SCSI_SENSE_BUFFERSIZE big.
2437 * Returns zero if successful; negative error number or scsi
2442 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2443 unsigned char *buffer, int len, int timeout, int retries,
2444 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2446 unsigned char cmd[10];
2447 unsigned char *real_buffer;
2450 memset(cmd, 0, sizeof(cmd));
2451 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2453 if (sdev->use_10_for_ms) {
2456 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2459 memcpy(real_buffer + 8, buffer, len);
2463 real_buffer[2] = data->medium_type;
2464 real_buffer[3] = data->device_specific;
2465 real_buffer[4] = data->longlba ? 0x01 : 0;
2467 real_buffer[6] = data->block_descriptor_length >> 8;
2468 real_buffer[7] = data->block_descriptor_length;
2470 cmd[0] = MODE_SELECT_10;
2474 if (len > 255 || data->block_descriptor_length > 255 ||
2478 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2481 memcpy(real_buffer + 4, buffer, len);
2484 real_buffer[1] = data->medium_type;
2485 real_buffer[2] = data->device_specific;
2486 real_buffer[3] = data->block_descriptor_length;
2489 cmd[0] = MODE_SELECT;
2493 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2494 sshdr, timeout, retries, NULL);
2498 EXPORT_SYMBOL_GPL(scsi_mode_select);
2501 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2502 * @sdev: SCSI device to be queried
2503 * @dbd: set if mode sense will allow block descriptors to be returned
2504 * @modepage: mode page being requested
2505 * @buffer: request buffer (may not be smaller than eight bytes)
2506 * @len: length of request buffer.
2507 * @timeout: command timeout
2508 * @retries: number of retries before failing
2509 * @data: returns a structure abstracting the mode header data
2510 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2511 * must be SCSI_SENSE_BUFFERSIZE big.
2513 * Returns zero if unsuccessful, or the header offset (either 4
2514 * or 8 depending on whether a six or ten byte command was
2515 * issued) if successful.
2518 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2519 unsigned char *buffer, int len, int timeout, int retries,
2520 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2522 unsigned char cmd[12];
2525 int result, retry_count = retries;
2526 struct scsi_sense_hdr my_sshdr;
2528 memset(data, 0, sizeof(*data));
2529 memset(&cmd[0], 0, 12);
2530 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2533 /* caller might not be interested in sense, but we need it */
2538 use_10_for_ms = sdev->use_10_for_ms;
2540 if (use_10_for_ms) {
2544 cmd[0] = MODE_SENSE_10;
2551 cmd[0] = MODE_SENSE;
2556 memset(buffer, 0, len);
2558 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2559 sshdr, timeout, retries, NULL);
2561 /* This code looks awful: what it's doing is making sure an
2562 * ILLEGAL REQUEST sense return identifies the actual command
2563 * byte as the problem. MODE_SENSE commands can return
2564 * ILLEGAL REQUEST if the code page isn't supported */
2566 if (use_10_for_ms && !scsi_status_is_good(result) &&
2567 (driver_byte(result) & DRIVER_SENSE)) {
2568 if (scsi_sense_valid(sshdr)) {
2569 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2570 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2572 * Invalid command operation code
2574 sdev->use_10_for_ms = 0;
2580 if(scsi_status_is_good(result)) {
2581 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2582 (modepage == 6 || modepage == 8))) {
2583 /* Initio breakage? */
2586 data->medium_type = 0;
2587 data->device_specific = 0;
2589 data->block_descriptor_length = 0;
2590 } else if(use_10_for_ms) {
2591 data->length = buffer[0]*256 + buffer[1] + 2;
2592 data->medium_type = buffer[2];
2593 data->device_specific = buffer[3];
2594 data->longlba = buffer[4] & 0x01;
2595 data->block_descriptor_length = buffer[6]*256
2598 data->length = buffer[0] + 1;
2599 data->medium_type = buffer[1];
2600 data->device_specific = buffer[2];
2601 data->block_descriptor_length = buffer[3];
2603 data->header_length = header_length;
2604 } else if ((status_byte(result) == CHECK_CONDITION) &&
2605 scsi_sense_valid(sshdr) &&
2606 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2613 EXPORT_SYMBOL(scsi_mode_sense);
2616 * scsi_test_unit_ready - test if unit is ready
2617 * @sdev: scsi device to change the state of.
2618 * @timeout: command timeout
2619 * @retries: number of retries before failing
2620 * @sshdr: outpout pointer for decoded sense information.
2622 * Returns zero if unsuccessful or an error if TUR failed. For
2623 * removable media, UNIT_ATTENTION sets ->changed flag.
2626 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2627 struct scsi_sense_hdr *sshdr)
2630 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2634 /* try to eat the UNIT_ATTENTION if there are enough retries */
2636 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2638 if (sdev->removable && scsi_sense_valid(sshdr) &&
2639 sshdr->sense_key == UNIT_ATTENTION)
2641 } while (scsi_sense_valid(sshdr) &&
2642 sshdr->sense_key == UNIT_ATTENTION && --retries);
2646 EXPORT_SYMBOL(scsi_test_unit_ready);
2649 * scsi_device_set_state - Take the given device through the device state model.
2650 * @sdev: scsi device to change the state of.
2651 * @state: state to change to.
2653 * Returns zero if successful or an error if the requested
2654 * transition is illegal.
2657 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2659 enum scsi_device_state oldstate = sdev->sdev_state;
2661 if (state == oldstate)
2667 case SDEV_CREATED_BLOCK:
2678 case SDEV_TRANSPORT_OFFLINE:
2691 case SDEV_TRANSPORT_OFFLINE:
2699 case SDEV_TRANSPORT_OFFLINE:
2714 case SDEV_CREATED_BLOCK:
2721 case SDEV_CREATED_BLOCK:
2736 case SDEV_TRANSPORT_OFFLINE:
2748 case SDEV_TRANSPORT_OFFLINE:
2751 case SDEV_CREATED_BLOCK:
2759 sdev->sdev_state = state;
2763 SCSI_LOG_ERROR_RECOVERY(1,
2764 sdev_printk(KERN_ERR, sdev,
2765 "Illegal state transition %s->%s",
2766 scsi_device_state_name(oldstate),
2767 scsi_device_state_name(state))
2771 EXPORT_SYMBOL(scsi_device_set_state);
2774 * sdev_evt_emit - emit a single SCSI device uevent
2775 * @sdev: associated SCSI device
2776 * @evt: event to emit
2778 * Send a single uevent (scsi_event) to the associated scsi_device.
2780 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2785 switch (evt->evt_type) {
2786 case SDEV_EVT_MEDIA_CHANGE:
2787 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2789 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2790 scsi_rescan_device(&sdev->sdev_gendev);
2791 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2793 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2794 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2796 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2797 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2799 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2800 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2802 case SDEV_EVT_LUN_CHANGE_REPORTED:
2803 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2805 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2806 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2808 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2809 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2818 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2822 * sdev_evt_thread - send a uevent for each scsi event
2823 * @work: work struct for scsi_device
2825 * Dispatch queued events to their associated scsi_device kobjects
2828 void scsi_evt_thread(struct work_struct *work)
2830 struct scsi_device *sdev;
2831 enum scsi_device_event evt_type;
2832 LIST_HEAD(event_list);
2834 sdev = container_of(work, struct scsi_device, event_work);
2836 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2837 if (test_and_clear_bit(evt_type, sdev->pending_events))
2838 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2841 struct scsi_event *evt;
2842 struct list_head *this, *tmp;
2843 unsigned long flags;
2845 spin_lock_irqsave(&sdev->list_lock, flags);
2846 list_splice_init(&sdev->event_list, &event_list);
2847 spin_unlock_irqrestore(&sdev->list_lock, flags);
2849 if (list_empty(&event_list))
2852 list_for_each_safe(this, tmp, &event_list) {
2853 evt = list_entry(this, struct scsi_event, node);
2854 list_del(&evt->node);
2855 scsi_evt_emit(sdev, evt);
2862 * sdev_evt_send - send asserted event to uevent thread
2863 * @sdev: scsi_device event occurred on
2864 * @evt: event to send
2866 * Assert scsi device event asynchronously.
2868 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2870 unsigned long flags;
2873 /* FIXME: currently this check eliminates all media change events
2874 * for polled devices. Need to update to discriminate between AN
2875 * and polled events */
2876 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2882 spin_lock_irqsave(&sdev->list_lock, flags);
2883 list_add_tail(&evt->node, &sdev->event_list);
2884 schedule_work(&sdev->event_work);
2885 spin_unlock_irqrestore(&sdev->list_lock, flags);
2887 EXPORT_SYMBOL_GPL(sdev_evt_send);
2890 * sdev_evt_alloc - allocate a new scsi event
2891 * @evt_type: type of event to allocate
2892 * @gfpflags: GFP flags for allocation
2894 * Allocates and returns a new scsi_event.
2896 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2899 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2903 evt->evt_type = evt_type;
2904 INIT_LIST_HEAD(&evt->node);
2906 /* evt_type-specific initialization, if any */
2908 case SDEV_EVT_MEDIA_CHANGE:
2909 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2910 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2911 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2912 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2913 case SDEV_EVT_LUN_CHANGE_REPORTED:
2914 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2915 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2923 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2926 * sdev_evt_send_simple - send asserted event to uevent thread
2927 * @sdev: scsi_device event occurred on
2928 * @evt_type: type of event to send
2929 * @gfpflags: GFP flags for allocation
2931 * Assert scsi device event asynchronously, given an event type.
2933 void sdev_evt_send_simple(struct scsi_device *sdev,
2934 enum scsi_device_event evt_type, gfp_t gfpflags)
2936 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2938 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2943 sdev_evt_send(sdev, evt);
2945 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2948 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2949 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2951 static int scsi_request_fn_active(struct scsi_device *sdev)
2953 struct request_queue *q = sdev->request_queue;
2954 int request_fn_active;
2956 WARN_ON_ONCE(sdev->host->use_blk_mq);
2958 spin_lock_irq(q->queue_lock);
2959 request_fn_active = q->request_fn_active;
2960 spin_unlock_irq(q->queue_lock);
2962 return request_fn_active;
2966 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2967 * @sdev: SCSI device pointer.
2969 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2970 * invoked from scsi_request_fn() have finished.
2972 static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2974 WARN_ON_ONCE(sdev->host->use_blk_mq);
2976 while (scsi_request_fn_active(sdev))
2981 * scsi_device_quiesce - Block user issued commands.
2982 * @sdev: scsi device to quiesce.
2984 * This works by trying to transition to the SDEV_QUIESCE state
2985 * (which must be a legal transition). When the device is in this
2986 * state, only special requests will be accepted, all others will
2987 * be deferred. Since special requests may also be requeued requests,
2988 * a successful return doesn't guarantee the device will be
2989 * totally quiescent.
2991 * Must be called with user context, may sleep.
2993 * Returns zero if unsuccessful or an error if not.
2996 scsi_device_quiesce(struct scsi_device *sdev)
2998 struct request_queue *q = sdev->request_queue;
3002 * It is allowed to call scsi_device_quiesce() multiple times from
3003 * the same context but concurrent scsi_device_quiesce() calls are
3006 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
3008 blk_set_preempt_only(q);
3010 blk_mq_freeze_queue(q);
3012 * Ensure that the effect of blk_set_preempt_only() will be visible
3013 * for percpu_ref_tryget() callers that occur after the queue
3014 * unfreeze even if the queue was already frozen before this function
3015 * was called. See also https://lwn.net/Articles/573497/.
3018 blk_mq_unfreeze_queue(q);
3020 mutex_lock(&sdev->state_mutex);
3021 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
3023 sdev->quiesced_by = current;
3025 blk_clear_preempt_only(q);
3026 mutex_unlock(&sdev->state_mutex);
3030 EXPORT_SYMBOL(scsi_device_quiesce);
3033 * scsi_device_resume - Restart user issued commands to a quiesced device.
3034 * @sdev: scsi device to resume.
3036 * Moves the device from quiesced back to running and restarts the
3039 * Must be called with user context, may sleep.
3041 void scsi_device_resume(struct scsi_device *sdev)
3043 /* check if the device state was mutated prior to resume, and if
3044 * so assume the state is being managed elsewhere (for example
3045 * device deleted during suspend)
3047 mutex_lock(&sdev->state_mutex);
3048 WARN_ON_ONCE(!sdev->quiesced_by);
3049 sdev->quiesced_by = NULL;
3050 blk_clear_preempt_only(sdev->request_queue);
3051 if (sdev->sdev_state == SDEV_QUIESCE)
3052 scsi_device_set_state(sdev, SDEV_RUNNING);
3053 mutex_unlock(&sdev->state_mutex);
3055 EXPORT_SYMBOL(scsi_device_resume);
3058 device_quiesce_fn(struct scsi_device *sdev, void *data)
3060 scsi_device_quiesce(sdev);
3064 scsi_target_quiesce(struct scsi_target *starget)
3066 starget_for_each_device(starget, NULL, device_quiesce_fn);
3068 EXPORT_SYMBOL(scsi_target_quiesce);
3071 device_resume_fn(struct scsi_device *sdev, void *data)
3073 scsi_device_resume(sdev);
3077 scsi_target_resume(struct scsi_target *starget)
3079 starget_for_each_device(starget, NULL, device_resume_fn);
3081 EXPORT_SYMBOL(scsi_target_resume);
3084 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
3085 * @sdev: device to block
3087 * Pause SCSI command processing on the specified device. Does not sleep.
3089 * Returns zero if successful or a negative error code upon failure.
3092 * This routine transitions the device to the SDEV_BLOCK state (which must be
3093 * a legal transition). When the device is in this state, command processing
3094 * is paused until the device leaves the SDEV_BLOCK state. See also
3095 * scsi_internal_device_unblock_nowait().
3097 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
3099 struct request_queue *q = sdev->request_queue;
3100 unsigned long flags;
3103 err = scsi_device_set_state(sdev, SDEV_BLOCK);
3105 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
3112 * The device has transitioned to SDEV_BLOCK. Stop the
3113 * block layer from calling the midlayer with this device's
3117 blk_mq_quiesce_queue_nowait(q);
3119 spin_lock_irqsave(q->queue_lock, flags);
3121 spin_unlock_irqrestore(q->queue_lock, flags);
3126 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
3129 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
3130 * @sdev: device to block
3132 * Pause SCSI command processing on the specified device and wait until all
3133 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
3135 * Returns zero if successful or a negative error code upon failure.
3138 * This routine transitions the device to the SDEV_BLOCK state (which must be
3139 * a legal transition). When the device is in this state, command processing
3140 * is paused until the device leaves the SDEV_BLOCK state. See also
3141 * scsi_internal_device_unblock().
3143 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
3144 * scsi_internal_device_block() has blocked a SCSI device and also
3145 * remove the rport mutex lock and unlock calls from srp_queuecommand().
3147 static int scsi_internal_device_block(struct scsi_device *sdev)
3149 struct request_queue *q = sdev->request_queue;
3152 mutex_lock(&sdev->state_mutex);
3153 err = scsi_internal_device_block_nowait(sdev);
3156 blk_mq_quiesce_queue(q);
3158 scsi_wait_for_queuecommand(sdev);
3160 mutex_unlock(&sdev->state_mutex);
3165 void scsi_start_queue(struct scsi_device *sdev)
3167 struct request_queue *q = sdev->request_queue;
3168 unsigned long flags;
3171 blk_mq_unquiesce_queue(q);
3173 spin_lock_irqsave(q->queue_lock, flags);
3175 spin_unlock_irqrestore(q->queue_lock, flags);
3180 * scsi_internal_device_unblock_nowait - resume a device after a block request
3181 * @sdev: device to resume
3182 * @new_state: state to set the device to after unblocking
3184 * Restart the device queue for a previously suspended SCSI device. Does not
3187 * Returns zero if successful or a negative error code upon failure.
3190 * This routine transitions the device to the SDEV_RUNNING state or to one of
3191 * the offline states (which must be a legal transition) allowing the midlayer
3192 * to goose the queue for this device.
3194 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
3195 enum scsi_device_state new_state)
3198 * Try to transition the scsi device to SDEV_RUNNING or one of the
3199 * offlined states and goose the device queue if successful.
3201 switch (sdev->sdev_state) {
3203 case SDEV_TRANSPORT_OFFLINE:
3204 sdev->sdev_state = new_state;
3206 case SDEV_CREATED_BLOCK:
3207 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3208 new_state == SDEV_OFFLINE)
3209 sdev->sdev_state = new_state;
3211 sdev->sdev_state = SDEV_CREATED;
3219 scsi_start_queue(sdev);
3223 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
3226 * scsi_internal_device_unblock - resume a device after a block request
3227 * @sdev: device to resume
3228 * @new_state: state to set the device to after unblocking
3230 * Restart the device queue for a previously suspended SCSI device. May sleep.
3232 * Returns zero if successful or a negative error code upon failure.
3235 * This routine transitions the device to the SDEV_RUNNING state or to one of
3236 * the offline states (which must be a legal transition) allowing the midlayer
3237 * to goose the queue for this device.
3239 static int scsi_internal_device_unblock(struct scsi_device *sdev,
3240 enum scsi_device_state new_state)
3244 mutex_lock(&sdev->state_mutex);
3245 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
3246 mutex_unlock(&sdev->state_mutex);
3252 device_block(struct scsi_device *sdev, void *data)
3254 scsi_internal_device_block(sdev);
3258 target_block(struct device *dev, void *data)
3260 if (scsi_is_target_device(dev))
3261 starget_for_each_device(to_scsi_target(dev), NULL,
3267 scsi_target_block(struct device *dev)
3269 if (scsi_is_target_device(dev))
3270 starget_for_each_device(to_scsi_target(dev), NULL,
3273 device_for_each_child(dev, NULL, target_block);
3275 EXPORT_SYMBOL_GPL(scsi_target_block);
3278 device_unblock(struct scsi_device *sdev, void *data)
3280 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3284 target_unblock(struct device *dev, void *data)
3286 if (scsi_is_target_device(dev))
3287 starget_for_each_device(to_scsi_target(dev), data,
3293 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3295 if (scsi_is_target_device(dev))
3296 starget_for_each_device(to_scsi_target(dev), &new_state,
3299 device_for_each_child(dev, &new_state, target_unblock);
3301 EXPORT_SYMBOL_GPL(scsi_target_unblock);
3304 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3305 * @sgl: scatter-gather list
3306 * @sg_count: number of segments in sg
3307 * @offset: offset in bytes into sg, on return offset into the mapped area
3308 * @len: bytes to map, on return number of bytes mapped
3310 * Returns virtual address of the start of the mapped page
3312 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3313 size_t *offset, size_t *len)
3316 size_t sg_len = 0, len_complete = 0;
3317 struct scatterlist *sg;
3320 WARN_ON(!irqs_disabled());
3322 for_each_sg(sgl, sg, sg_count, i) {
3323 len_complete = sg_len; /* Complete sg-entries */
3324 sg_len += sg->length;
3325 if (sg_len > *offset)
3329 if (unlikely(i == sg_count)) {
3330 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3332 __func__, sg_len, *offset, sg_count);
3337 /* Offset starting from the beginning of first page in this sg-entry */
3338 *offset = *offset - len_complete + sg->offset;
3340 /* Assumption: contiguous pages can be accessed as "page + i" */
3341 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3342 *offset &= ~PAGE_MASK;
3344 /* Bytes in this sg-entry from *offset to the end of the page */
3345 sg_len = PAGE_SIZE - *offset;
3349 return kmap_atomic(page);
3351 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3354 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3355 * @virt: virtual address to be unmapped
3357 void scsi_kunmap_atomic_sg(void *virt)
3359 kunmap_atomic(virt);
3361 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3363 void sdev_disable_disk_events(struct scsi_device *sdev)
3365 atomic_inc(&sdev->disk_events_disable_depth);
3367 EXPORT_SYMBOL(sdev_disable_disk_events);
3369 void sdev_enable_disk_events(struct scsi_device *sdev)
3371 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3373 atomic_dec(&sdev->disk_events_disable_depth);
3375 EXPORT_SYMBOL(sdev_enable_disk_events);
3378 * scsi_vpd_lun_id - return a unique device identification
3379 * @sdev: SCSI device
3380 * @id: buffer for the identification
3381 * @id_len: length of the buffer
3383 * Copies a unique device identification into @id based
3384 * on the information in the VPD page 0x83 of the device.
3385 * The string will be formatted as a SCSI name string.
3387 * Returns the length of the identification or error on failure.
3388 * If the identifier is longer than the supplied buffer the actual
3389 * identifier length is returned and the buffer is not zero-padded.
3391 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3393 u8 cur_id_type = 0xff;
3395 const unsigned char *d, *cur_id_str;
3396 const struct scsi_vpd *vpd_pg83;
3397 int id_size = -EINVAL;
3400 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3407 * Look for the correct descriptor.
3408 * Order of preference for lun descriptor:
3409 * - SCSI name string
3410 * - NAA IEEE Registered Extended
3411 * - EUI-64 based 16-byte
3412 * - EUI-64 based 12-byte
3413 * - NAA IEEE Registered
3414 * - NAA IEEE Extended
3416 * as longer descriptors reduce the likelyhood
3417 * of identification clashes.
3420 /* The id string must be at least 20 bytes + terminating NULL byte */
3426 memset(id, 0, id_len);
3427 d = vpd_pg83->data + 4;
3428 while (d < vpd_pg83->data + vpd_pg83->len) {
3429 /* Skip designators not referring to the LUN */
3430 if ((d[1] & 0x30) != 0x00)
3433 switch (d[1] & 0xf) {
3436 if (cur_id_size > d[3])
3438 /* Prefer anything */
3439 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3442 if (cur_id_size + 4 > id_len)
3443 cur_id_size = id_len - 4;
3445 cur_id_type = d[1] & 0xf;
3446 id_size = snprintf(id, id_len, "t10.%*pE",
3447 cur_id_size, cur_id_str);
3451 if (cur_id_size > d[3])
3453 /* Prefer NAA IEEE Registered Extended */
3454 if (cur_id_type == 0x3 &&
3455 cur_id_size == d[3])
3459 cur_id_type = d[1] & 0xf;
3460 switch (cur_id_size) {
3462 id_size = snprintf(id, id_len,
3467 id_size = snprintf(id, id_len,
3472 id_size = snprintf(id, id_len,
3483 if (cur_id_size > d[3])
3487 cur_id_type = d[1] & 0xf;
3488 switch (cur_id_size) {
3490 id_size = snprintf(id, id_len,
3495 id_size = snprintf(id, id_len,
3505 /* SCSI name string */
3506 if (cur_id_size + 4 > d[3])
3508 /* Prefer others for truncated descriptor */
3509 if (cur_id_size && d[3] > id_len)
3511 cur_id_size = id_size = d[3];
3513 cur_id_type = d[1] & 0xf;
3514 if (cur_id_size >= id_len)
3515 cur_id_size = id_len - 1;
3516 memcpy(id, cur_id_str, cur_id_size);
3517 /* Decrease priority for truncated descriptor */
3518 if (cur_id_size != id_size)
3531 EXPORT_SYMBOL(scsi_vpd_lun_id);
3534 * scsi_vpd_tpg_id - return a target port group identifier
3535 * @sdev: SCSI device
3537 * Returns the Target Port Group identifier from the information
3538 * froom VPD page 0x83 of the device.
3540 * Returns the identifier or error on failure.
3542 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3544 const unsigned char *d;
3545 const struct scsi_vpd *vpd_pg83;
3546 int group_id = -EAGAIN, rel_port = -1;
3549 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3555 d = vpd_pg83->data + 4;
3556 while (d < vpd_pg83->data + vpd_pg83->len) {
3557 switch (d[1] & 0xf) {
3559 /* Relative target port */
3560 rel_port = get_unaligned_be16(&d[6]);
3563 /* Target port group */
3564 group_id = get_unaligned_be16(&d[6]);
3573 if (group_id >= 0 && rel_id && rel_port != -1)
3578 EXPORT_SYMBOL(scsi_vpd_tpg_id);