2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
11 #include <linux/bio.h>
12 #include <linux/bitops.h>
13 #include <linux/blkdev.h>
14 #include <linux/completion.h>
15 #include <linux/kernel.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
22 #include <linux/blk-mq.h>
23 #include <linux/ratelimit.h>
24 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_driver.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi_dh.h>
35 #include <trace/events/scsi.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
41 struct kmem_cache *scsi_sdb_cache;
44 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
45 * not change behaviour from the previous unplug mechanism, experimentation
46 * may prove this needs changing.
48 #define SCSI_QUEUE_DELAY 3
51 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
53 struct Scsi_Host *host = cmd->device->host;
54 struct scsi_device *device = cmd->device;
55 struct scsi_target *starget = scsi_target(device);
58 * Set the appropriate busy bit for the device/host.
60 * If the host/device isn't busy, assume that something actually
61 * completed, and that we should be able to queue a command now.
63 * Note that the prior mid-layer assumption that any host could
64 * always queue at least one command is now broken. The mid-layer
65 * will implement a user specifiable stall (see
66 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
67 * if a command is requeued with no other commands outstanding
68 * either for the device or for the host.
71 case SCSI_MLQUEUE_HOST_BUSY:
72 atomic_set(&host->host_blocked, host->max_host_blocked);
74 case SCSI_MLQUEUE_DEVICE_BUSY:
75 case SCSI_MLQUEUE_EH_RETRY:
76 atomic_set(&device->device_blocked,
77 device->max_device_blocked);
79 case SCSI_MLQUEUE_TARGET_BUSY:
80 atomic_set(&starget->target_blocked,
81 starget->max_target_blocked);
86 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
88 struct scsi_device *sdev = cmd->device;
89 struct request_queue *q = cmd->request->q;
91 blk_mq_requeue_request(cmd->request);
92 blk_mq_kick_requeue_list(q);
93 put_device(&sdev->sdev_gendev);
97 * __scsi_queue_insert - private queue insertion
98 * @cmd: The SCSI command being requeued
99 * @reason: The reason for the requeue
100 * @unbusy: Whether the queue should be unbusied
102 * This is a private queue insertion. The public interface
103 * scsi_queue_insert() always assumes the queue should be unbusied
104 * because it's always called before the completion. This function is
105 * for a requeue after completion, which should only occur in this
108 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
110 struct scsi_device *device = cmd->device;
111 struct request_queue *q = device->request_queue;
114 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
115 "Inserting command %p into mlqueue\n", cmd));
117 scsi_set_blocked(cmd, reason);
120 * Decrement the counters, since these commands are no longer
121 * active on the host/device.
124 scsi_device_unbusy(device);
127 * Requeue this command. It will go before all other commands
128 * that are already in the queue. Schedule requeue work under
129 * lock such that the kblockd_schedule_work() call happens
130 * before blk_cleanup_queue() finishes.
134 scsi_mq_requeue_cmd(cmd);
137 spin_lock_irqsave(q->queue_lock, flags);
138 blk_requeue_request(q, cmd->request);
139 kblockd_schedule_work(&device->requeue_work);
140 spin_unlock_irqrestore(q->queue_lock, flags);
144 * Function: scsi_queue_insert()
146 * Purpose: Insert a command in the midlevel queue.
148 * Arguments: cmd - command that we are adding to queue.
149 * reason - why we are inserting command to queue.
151 * Lock status: Assumed that lock is not held upon entry.
155 * Notes: We do this for one of two cases. Either the host is busy
156 * and it cannot accept any more commands for the time being,
157 * or the device returned QUEUE_FULL and can accept no more
159 * Notes: This could be called either from an interrupt context or a
160 * normal process context.
162 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
164 __scsi_queue_insert(cmd, reason, 1);
167 * scsi_execute - insert request and wait for the result
170 * @data_direction: data direction
171 * @buffer: data buffer
172 * @bufflen: len of buffer
173 * @sense: optional sense buffer
174 * @timeout: request timeout in seconds
175 * @retries: number of times to retry request
176 * @flags: or into request flags;
177 * @resid: optional residual length
179 * returns the req->errors value which is the scsi_cmnd result
182 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
183 int data_direction, void *buffer, unsigned bufflen,
184 unsigned char *sense, int timeout, int retries, u64 flags,
188 int write = (data_direction == DMA_TO_DEVICE);
189 int ret = DRIVER_ERROR << 24;
191 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
194 blk_rq_set_block_pc(req);
196 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
197 buffer, bufflen, __GFP_RECLAIM))
200 req->cmd_len = COMMAND_SIZE(cmd[0]);
201 memcpy(req->cmd, cmd, req->cmd_len);
204 req->retries = retries;
205 req->timeout = timeout;
206 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
209 * head injection *required* here otherwise quiesce won't work
211 blk_execute_rq(req->q, NULL, req, 1);
214 * Some devices (USB mass-storage in particular) may transfer
215 * garbage data together with a residue indicating that the data
216 * is invalid. Prevent the garbage from being misinterpreted
217 * and prevent security leaks by zeroing out the excess data.
219 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
220 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
223 *resid = req->resid_len;
226 blk_put_request(req);
230 EXPORT_SYMBOL(scsi_execute);
232 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
233 int data_direction, void *buffer, unsigned bufflen,
234 struct scsi_sense_hdr *sshdr, int timeout, int retries,
235 int *resid, u64 flags)
241 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
243 return DRIVER_ERROR << 24;
245 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
246 sense, timeout, retries, flags, resid);
248 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
253 EXPORT_SYMBOL(scsi_execute_req_flags);
256 * Function: scsi_init_cmd_errh()
258 * Purpose: Initialize cmd fields related to error handling.
260 * Arguments: cmd - command that is ready to be queued.
262 * Notes: This function has the job of initializing a number of
263 * fields related to error handling. Typically this will
264 * be called once for each command, as required.
266 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
268 cmd->serial_number = 0;
269 scsi_set_resid(cmd, 0);
270 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
271 if (cmd->cmd_len == 0)
272 cmd->cmd_len = scsi_command_size(cmd->cmnd);
275 void scsi_device_unbusy(struct scsi_device *sdev)
277 struct Scsi_Host *shost = sdev->host;
278 struct scsi_target *starget = scsi_target(sdev);
281 atomic_dec(&shost->host_busy);
282 if (starget->can_queue > 0)
283 atomic_dec(&starget->target_busy);
285 if (unlikely(scsi_host_in_recovery(shost) &&
286 (shost->host_failed || shost->host_eh_scheduled))) {
287 spin_lock_irqsave(shost->host_lock, flags);
288 scsi_eh_wakeup(shost);
289 spin_unlock_irqrestore(shost->host_lock, flags);
292 atomic_dec(&sdev->device_busy);
295 static void scsi_kick_queue(struct request_queue *q)
298 blk_mq_start_hw_queues(q);
304 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
305 * and call blk_run_queue for all the scsi_devices on the target -
306 * including current_sdev first.
308 * Called with *no* scsi locks held.
310 static void scsi_single_lun_run(struct scsi_device *current_sdev)
312 struct Scsi_Host *shost = current_sdev->host;
313 struct scsi_device *sdev, *tmp;
314 struct scsi_target *starget = scsi_target(current_sdev);
317 spin_lock_irqsave(shost->host_lock, flags);
318 starget->starget_sdev_user = NULL;
319 spin_unlock_irqrestore(shost->host_lock, flags);
322 * Call blk_run_queue for all LUNs on the target, starting with
323 * current_sdev. We race with others (to set starget_sdev_user),
324 * but in most cases, we will be first. Ideally, each LU on the
325 * target would get some limited time or requests on the target.
327 scsi_kick_queue(current_sdev->request_queue);
329 spin_lock_irqsave(shost->host_lock, flags);
330 if (starget->starget_sdev_user)
332 list_for_each_entry_safe(sdev, tmp, &starget->devices,
333 same_target_siblings) {
334 if (sdev == current_sdev)
336 if (scsi_device_get(sdev))
339 spin_unlock_irqrestore(shost->host_lock, flags);
340 scsi_kick_queue(sdev->request_queue);
341 spin_lock_irqsave(shost->host_lock, flags);
343 scsi_device_put(sdev);
346 spin_unlock_irqrestore(shost->host_lock, flags);
349 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
351 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
353 if (atomic_read(&sdev->device_blocked) > 0)
358 static inline bool scsi_target_is_busy(struct scsi_target *starget)
360 if (starget->can_queue > 0) {
361 if (atomic_read(&starget->target_busy) >= starget->can_queue)
363 if (atomic_read(&starget->target_blocked) > 0)
369 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
371 if (shost->can_queue > 0 &&
372 atomic_read(&shost->host_busy) >= shost->can_queue)
374 if (atomic_read(&shost->host_blocked) > 0)
376 if (shost->host_self_blocked)
381 static void scsi_starved_list_run(struct Scsi_Host *shost)
383 LIST_HEAD(starved_list);
384 struct scsi_device *sdev;
387 spin_lock_irqsave(shost->host_lock, flags);
388 list_splice_init(&shost->starved_list, &starved_list);
390 while (!list_empty(&starved_list)) {
391 struct request_queue *slq;
394 * As long as shost is accepting commands and we have
395 * starved queues, call blk_run_queue. scsi_request_fn
396 * drops the queue_lock and can add us back to the
399 * host_lock protects the starved_list and starved_entry.
400 * scsi_request_fn must get the host_lock before checking
401 * or modifying starved_list or starved_entry.
403 if (scsi_host_is_busy(shost))
406 sdev = list_entry(starved_list.next,
407 struct scsi_device, starved_entry);
408 list_del_init(&sdev->starved_entry);
409 if (scsi_target_is_busy(scsi_target(sdev))) {
410 list_move_tail(&sdev->starved_entry,
411 &shost->starved_list);
416 * Once we drop the host lock, a racing scsi_remove_device()
417 * call may remove the sdev from the starved list and destroy
418 * it and the queue. Mitigate by taking a reference to the
419 * queue and never touching the sdev again after we drop the
420 * host lock. Note: if __scsi_remove_device() invokes
421 * blk_cleanup_queue() before the queue is run from this
422 * function then blk_run_queue() will return immediately since
423 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
425 slq = sdev->request_queue;
426 if (!blk_get_queue(slq))
428 spin_unlock_irqrestore(shost->host_lock, flags);
430 scsi_kick_queue(slq);
433 spin_lock_irqsave(shost->host_lock, flags);
435 /* put any unprocessed entries back */
436 list_splice(&starved_list, &shost->starved_list);
437 spin_unlock_irqrestore(shost->host_lock, flags);
441 * Function: scsi_run_queue()
443 * Purpose: Select a proper request queue to serve next
445 * Arguments: q - last request's queue
449 * Notes: The previous command was completely finished, start
450 * a new one if possible.
452 static void scsi_run_queue(struct request_queue *q)
454 struct scsi_device *sdev = q->queuedata;
456 if (scsi_target(sdev)->single_lun)
457 scsi_single_lun_run(sdev);
458 if (!list_empty(&sdev->host->starved_list))
459 scsi_starved_list_run(sdev->host);
462 blk_mq_start_stopped_hw_queues(q, false);
467 void scsi_requeue_run_queue(struct work_struct *work)
469 struct scsi_device *sdev;
470 struct request_queue *q;
472 sdev = container_of(work, struct scsi_device, requeue_work);
473 q = sdev->request_queue;
478 * Function: scsi_requeue_command()
480 * Purpose: Handle post-processing of completed commands.
482 * Arguments: q - queue to operate on
483 * cmd - command that may need to be requeued.
487 * Notes: After command completion, there may be blocks left
488 * over which weren't finished by the previous command
489 * this can be for a number of reasons - the main one is
490 * I/O errors in the middle of the request, in which case
491 * we need to request the blocks that come after the bad
493 * Notes: Upon return, cmd is a stale pointer.
495 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
497 struct scsi_device *sdev = cmd->device;
498 struct request *req = cmd->request;
501 spin_lock_irqsave(q->queue_lock, flags);
502 blk_unprep_request(req);
504 scsi_put_command(cmd);
505 blk_requeue_request(q, req);
506 spin_unlock_irqrestore(q->queue_lock, flags);
510 put_device(&sdev->sdev_gendev);
513 void scsi_run_host_queues(struct Scsi_Host *shost)
515 struct scsi_device *sdev;
517 shost_for_each_device(sdev, shost)
518 scsi_run_queue(sdev->request_queue);
521 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
523 if (cmd->request->cmd_type == REQ_TYPE_FS) {
524 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
526 if (drv->uninit_command)
527 drv->uninit_command(cmd);
531 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
533 struct scsi_data_buffer *sdb;
535 if (cmd->sdb.table.nents)
536 sg_free_table_chained(&cmd->sdb.table, true);
537 if (cmd->request->next_rq) {
538 sdb = cmd->request->next_rq->special;
540 sg_free_table_chained(&sdb->table, true);
542 if (scsi_prot_sg_count(cmd))
543 sg_free_table_chained(&cmd->prot_sdb->table, true);
546 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
548 struct scsi_device *sdev = cmd->device;
549 struct Scsi_Host *shost = sdev->host;
552 scsi_mq_free_sgtables(cmd);
553 scsi_uninit_cmd(cmd);
555 if (shost->use_cmd_list) {
556 BUG_ON(list_empty(&cmd->list));
557 spin_lock_irqsave(&sdev->list_lock, flags);
558 list_del_init(&cmd->list);
559 spin_unlock_irqrestore(&sdev->list_lock, flags);
564 * Function: scsi_release_buffers()
566 * Purpose: Free resources allocate for a scsi_command.
568 * Arguments: cmd - command that we are bailing.
570 * Lock status: Assumed that no lock is held upon entry.
574 * Notes: In the event that an upper level driver rejects a
575 * command, we must release resources allocated during
576 * the __init_io() function. Primarily this would involve
577 * the scatter-gather table.
579 static void scsi_release_buffers(struct scsi_cmnd *cmd)
581 if (cmd->sdb.table.nents)
582 sg_free_table_chained(&cmd->sdb.table, false);
584 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
586 if (scsi_prot_sg_count(cmd))
587 sg_free_table_chained(&cmd->prot_sdb->table, false);
590 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
592 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
594 sg_free_table_chained(&bidi_sdb->table, false);
595 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
596 cmd->request->next_rq->special = NULL;
599 static bool scsi_end_request(struct request *req, int error,
600 unsigned int bytes, unsigned int bidi_bytes)
602 struct scsi_cmnd *cmd = req->special;
603 struct scsi_device *sdev = cmd->device;
604 struct request_queue *q = sdev->request_queue;
606 if (blk_update_request(req, error, bytes))
609 /* Bidi request must be completed as a whole */
610 if (unlikely(bidi_bytes) &&
611 blk_update_request(req->next_rq, error, bidi_bytes))
614 if (blk_queue_add_random(q))
615 add_disk_randomness(req->rq_disk);
619 * In the MQ case the command gets freed by __blk_mq_end_request,
620 * so we have to do all cleanup that depends on it earlier.
622 * We also can't kick the queues from irq context, so we
623 * will have to defer it to a workqueue.
625 scsi_mq_uninit_cmd(cmd);
627 __blk_mq_end_request(req, error);
629 if (scsi_target(sdev)->single_lun ||
630 !list_empty(&sdev->host->starved_list))
631 kblockd_schedule_work(&sdev->requeue_work);
633 blk_mq_start_stopped_hw_queues(q, true);
638 scsi_release_bidi_buffers(cmd);
640 spin_lock_irqsave(q->queue_lock, flags);
641 blk_finish_request(req, error);
642 spin_unlock_irqrestore(q->queue_lock, flags);
644 scsi_release_buffers(cmd);
646 scsi_put_command(cmd);
650 put_device(&sdev->sdev_gendev);
655 * __scsi_error_from_host_byte - translate SCSI error code into errno
656 * @cmd: SCSI command (unused)
657 * @result: scsi error code
659 * Translate SCSI error code into standard UNIX errno.
661 * -ENOLINK temporary transport failure
662 * -EREMOTEIO permanent target failure, do not retry
663 * -EBADE permanent nexus failure, retry on other path
664 * -ENOSPC No write space available
665 * -ENODATA Medium error
666 * -EIO unspecified I/O error
668 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
672 switch(host_byte(result)) {
673 case DID_TRANSPORT_FAILFAST:
676 case DID_TARGET_FAILURE:
677 set_host_byte(cmd, DID_OK);
680 case DID_NEXUS_FAILURE:
681 set_host_byte(cmd, DID_OK);
684 case DID_ALLOC_FAILURE:
685 set_host_byte(cmd, DID_OK);
688 case DID_MEDIUM_ERROR:
689 set_host_byte(cmd, DID_OK);
701 * Function: scsi_io_completion()
703 * Purpose: Completion processing for block device I/O requests.
705 * Arguments: cmd - command that is finished.
707 * Lock status: Assumed that no lock is held upon entry.
711 * Notes: We will finish off the specified number of sectors. If we
712 * are done, the command block will be released and the queue
713 * function will be goosed. If we are not done then we have to
714 * figure out what to do next:
716 * a) We can call scsi_requeue_command(). The request
717 * will be unprepared and put back on the queue. Then
718 * a new command will be created for it. This should
719 * be used if we made forward progress, or if we want
720 * to switch from READ(10) to READ(6) for example.
722 * b) We can call __scsi_queue_insert(). The request will
723 * be put back on the queue and retried using the same
724 * command as before, possibly after a delay.
726 * c) We can call scsi_end_request() with -EIO to fail
727 * the remainder of the request.
729 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
731 int result = cmd->result;
732 struct request_queue *q = cmd->device->request_queue;
733 struct request *req = cmd->request;
735 struct scsi_sense_hdr sshdr;
736 bool sense_valid = false;
737 int sense_deferred = 0, level = 0;
738 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
739 ACTION_DELAYED_RETRY} action;
740 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
743 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
745 sense_deferred = scsi_sense_is_deferred(&sshdr);
748 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
750 if (sense_valid && req->sense) {
752 * SG_IO wants current and deferred errors
754 int len = 8 + cmd->sense_buffer[7];
756 if (len > SCSI_SENSE_BUFFERSIZE)
757 len = SCSI_SENSE_BUFFERSIZE;
758 memcpy(req->sense, cmd->sense_buffer, len);
759 req->sense_len = len;
762 error = __scsi_error_from_host_byte(cmd, result);
765 * __scsi_error_from_host_byte may have reset the host_byte
767 req->errors = cmd->result;
769 req->resid_len = scsi_get_resid(cmd);
771 if (scsi_bidi_cmnd(cmd)) {
773 * Bidi commands Must be complete as a whole,
774 * both sides at once.
776 req->next_rq->resid_len = scsi_in(cmd)->resid;
777 if (scsi_end_request(req, 0, blk_rq_bytes(req),
778 blk_rq_bytes(req->next_rq)))
782 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
784 * Certain non BLOCK_PC requests are commands that don't
785 * actually transfer anything (FLUSH), so cannot use
786 * good_bytes != blk_rq_bytes(req) as the signal for an error.
787 * This sets the error explicitly for the problem case.
789 error = __scsi_error_from_host_byte(cmd, result);
792 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
793 BUG_ON(blk_bidi_rq(req));
796 * Next deal with any sectors which we were able to correctly
799 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
800 "%u sectors total, %d bytes done.\n",
801 blk_rq_sectors(req), good_bytes));
804 * Recovered errors need reporting, but they're always treated
805 * as success, so fiddle the result code here. For BLOCK_PC
806 * we already took a copy of the original into rq->errors which
807 * is what gets returned to the user
809 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
810 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
811 * print since caller wants ATA registers. Only occurs on
812 * SCSI ATA PASS_THROUGH commands when CK_COND=1
814 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
816 else if (!(req->cmd_flags & REQ_QUIET))
817 scsi_print_sense(cmd);
819 /* BLOCK_PC may have set error */
824 * special case: failed zero length commands always need to
825 * drop down into the retry code. Otherwise, if we finished
826 * all bytes in the request we are done now.
828 if (!(blk_rq_bytes(req) == 0 && error) &&
829 !scsi_end_request(req, error, good_bytes, 0))
833 * Kill remainder if no retrys.
835 if (error && scsi_noretry_cmd(cmd)) {
836 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
842 * If there had been no error, but we have leftover bytes in the
843 * requeues just queue the command up again.
848 error = __scsi_error_from_host_byte(cmd, result);
850 if (host_byte(result) == DID_RESET) {
851 /* Third party bus reset or reset for error recovery
852 * reasons. Just retry the command and see what
855 action = ACTION_RETRY;
856 } else if (sense_valid && !sense_deferred) {
857 switch (sshdr.sense_key) {
859 if (cmd->device->removable) {
860 /* Detected disc change. Set a bit
861 * and quietly refuse further access.
863 cmd->device->changed = 1;
864 action = ACTION_FAIL;
866 /* Must have been a power glitch, or a
867 * bus reset. Could not have been a
868 * media change, so we just retry the
869 * command and see what happens.
871 action = ACTION_RETRY;
874 case ILLEGAL_REQUEST:
875 /* If we had an ILLEGAL REQUEST returned, then
876 * we may have performed an unsupported
877 * command. The only thing this should be
878 * would be a ten byte read where only a six
879 * byte read was supported. Also, on a system
880 * where READ CAPACITY failed, we may have
881 * read past the end of the disk.
883 if ((cmd->device->use_10_for_rw &&
884 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
885 (cmd->cmnd[0] == READ_10 ||
886 cmd->cmnd[0] == WRITE_10)) {
887 /* This will issue a new 6-byte command. */
888 cmd->device->use_10_for_rw = 0;
889 action = ACTION_REPREP;
890 } else if (sshdr.asc == 0x10) /* DIX */ {
891 action = ACTION_FAIL;
893 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
894 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
895 action = ACTION_FAIL;
898 action = ACTION_FAIL;
900 case ABORTED_COMMAND:
901 action = ACTION_FAIL;
902 if (sshdr.asc == 0x10) /* DIF */
906 /* If the device is in the process of becoming
907 * ready, or has a temporary blockage, retry.
909 if (sshdr.asc == 0x04) {
910 switch (sshdr.ascq) {
911 case 0x01: /* becoming ready */
912 case 0x04: /* format in progress */
913 case 0x05: /* rebuild in progress */
914 case 0x06: /* recalculation in progress */
915 case 0x07: /* operation in progress */
916 case 0x08: /* Long write in progress */
917 case 0x09: /* self test in progress */
918 case 0x11: /* notify (enable spinup) required */
919 case 0x14: /* space allocation in progress */
920 action = ACTION_DELAYED_RETRY;
923 action = ACTION_FAIL;
927 action = ACTION_FAIL;
929 case VOLUME_OVERFLOW:
930 /* See SSC3rXX or current. */
931 action = ACTION_FAIL;
934 action = ACTION_FAIL;
938 action = ACTION_FAIL;
940 if (action != ACTION_FAIL &&
941 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
942 action = ACTION_FAIL;
946 /* Give up and fail the remainder of the request */
947 if (!(req->cmd_flags & REQ_QUIET)) {
948 static DEFINE_RATELIMIT_STATE(_rs,
949 DEFAULT_RATELIMIT_INTERVAL,
950 DEFAULT_RATELIMIT_BURST);
952 if (unlikely(scsi_logging_level))
953 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
954 SCSI_LOG_MLCOMPLETE_BITS);
957 * if logging is enabled the failure will be printed
958 * in scsi_log_completion(), so avoid duplicate messages
960 if (!level && __ratelimit(&_rs)) {
961 scsi_print_result(cmd, NULL, FAILED);
962 if (driver_byte(result) & DRIVER_SENSE)
963 scsi_print_sense(cmd);
964 scsi_print_command(cmd);
967 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
972 /* Unprep the request and put it back at the head of the queue.
973 * A new command will be prepared and issued.
976 cmd->request->cmd_flags &= ~REQ_DONTPREP;
977 scsi_mq_uninit_cmd(cmd);
978 scsi_mq_requeue_cmd(cmd);
980 scsi_release_buffers(cmd);
981 scsi_requeue_command(q, cmd);
985 /* Retry the same command immediately */
986 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
988 case ACTION_DELAYED_RETRY:
989 /* Retry the same command after a delay */
990 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
995 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1000 * If sg table allocation fails, requeue request later.
1002 if (unlikely(sg_alloc_table_chained(&sdb->table, req->nr_phys_segments,
1004 return BLKPREP_DEFER;
1007 * Next, walk the list, and fill in the addresses and sizes of
1010 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1011 BUG_ON(count > sdb->table.nents);
1012 sdb->table.nents = count;
1013 sdb->length = blk_rq_bytes(req);
1018 * Function: scsi_init_io()
1020 * Purpose: SCSI I/O initialize function.
1022 * Arguments: cmd - Command descriptor we wish to initialize
1024 * Returns: 0 on success
1025 * BLKPREP_DEFER if the failure is retryable
1026 * BLKPREP_KILL if the failure is fatal
1028 int scsi_init_io(struct scsi_cmnd *cmd)
1030 struct scsi_device *sdev = cmd->device;
1031 struct request *rq = cmd->request;
1032 bool is_mq = (rq->mq_ctx != NULL);
1033 int error = BLKPREP_KILL;
1035 if (WARN_ON_ONCE(!rq->nr_phys_segments))
1038 error = scsi_init_sgtable(rq, &cmd->sdb);
1042 if (blk_bidi_rq(rq)) {
1043 if (!rq->q->mq_ops) {
1044 struct scsi_data_buffer *bidi_sdb =
1045 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1047 error = BLKPREP_DEFER;
1051 rq->next_rq->special = bidi_sdb;
1054 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1059 if (blk_integrity_rq(rq)) {
1060 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1063 if (prot_sdb == NULL) {
1065 * This can happen if someone (e.g. multipath)
1066 * queues a command to a device on an adapter
1067 * that does not support DIX.
1070 error = BLKPREP_KILL;
1074 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1076 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1077 prot_sdb->table.sgl)) {
1078 error = BLKPREP_DEFER;
1082 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1083 prot_sdb->table.sgl);
1084 BUG_ON(unlikely(count > ivecs));
1085 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1087 cmd->prot_sdb = prot_sdb;
1088 cmd->prot_sdb->table.nents = count;
1094 scsi_mq_free_sgtables(cmd);
1096 scsi_release_buffers(cmd);
1097 cmd->request->special = NULL;
1098 scsi_put_command(cmd);
1099 put_device(&sdev->sdev_gendev);
1103 EXPORT_SYMBOL(scsi_init_io);
1105 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1106 struct request *req)
1108 struct scsi_cmnd *cmd;
1110 if (!req->special) {
1111 /* Bail if we can't get a reference to the device */
1112 if (!get_device(&sdev->sdev_gendev))
1115 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1116 if (unlikely(!cmd)) {
1117 put_device(&sdev->sdev_gendev);
1125 /* pull a tag out of the request if we have one */
1126 cmd->tag = req->tag;
1129 cmd->cmnd = req->cmd;
1130 cmd->prot_op = SCSI_PROT_NORMAL;
1135 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1137 struct scsi_cmnd *cmd = req->special;
1140 * BLOCK_PC requests may transfer data, in which case they must
1141 * a bio attached to them. Or they might contain a SCSI command
1142 * that does not transfer data, in which case they may optionally
1143 * submit a request without an attached bio.
1146 int ret = scsi_init_io(cmd);
1150 BUG_ON(blk_rq_bytes(req));
1152 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1155 cmd->cmd_len = req->cmd_len;
1156 cmd->transfersize = blk_rq_bytes(req);
1157 cmd->allowed = req->retries;
1162 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1163 * that still need to be translated to SCSI CDBs from the ULD.
1165 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1167 struct scsi_cmnd *cmd = req->special;
1169 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1170 int ret = sdev->handler->prep_fn(sdev, req);
1171 if (ret != BLKPREP_OK)
1175 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1176 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1179 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1181 struct scsi_cmnd *cmd = req->special;
1183 if (!blk_rq_bytes(req))
1184 cmd->sc_data_direction = DMA_NONE;
1185 else if (rq_data_dir(req) == WRITE)
1186 cmd->sc_data_direction = DMA_TO_DEVICE;
1188 cmd->sc_data_direction = DMA_FROM_DEVICE;
1190 switch (req->cmd_type) {
1192 return scsi_setup_fs_cmnd(sdev, req);
1193 case REQ_TYPE_BLOCK_PC:
1194 return scsi_setup_blk_pc_cmnd(sdev, req);
1196 return BLKPREP_KILL;
1201 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1203 int ret = BLKPREP_OK;
1206 * If the device is not in running state we will reject some
1209 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1210 switch (sdev->sdev_state) {
1212 case SDEV_TRANSPORT_OFFLINE:
1214 * If the device is offline we refuse to process any
1215 * commands. The device must be brought online
1216 * before trying any recovery commands.
1218 sdev_printk(KERN_ERR, sdev,
1219 "rejecting I/O to offline device\n");
1224 * If the device is fully deleted, we refuse to
1225 * process any commands as well.
1227 sdev_printk(KERN_ERR, sdev,
1228 "rejecting I/O to dead device\n");
1232 case SDEV_CREATED_BLOCK:
1233 ret = BLKPREP_DEFER;
1237 * If the devices is blocked we defer normal commands.
1239 if (!(req->cmd_flags & REQ_PREEMPT))
1240 ret = BLKPREP_DEFER;
1244 * For any other not fully online state we only allow
1245 * special commands. In particular any user initiated
1246 * command is not allowed.
1248 if (!(req->cmd_flags & REQ_PREEMPT))
1257 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1259 struct scsi_device *sdev = q->queuedata;
1263 case BLKPREP_INVALID:
1264 req->errors = DID_NO_CONNECT << 16;
1265 /* release the command and kill it */
1267 struct scsi_cmnd *cmd = req->special;
1268 scsi_release_buffers(cmd);
1269 scsi_put_command(cmd);
1270 put_device(&sdev->sdev_gendev);
1271 req->special = NULL;
1276 * If we defer, the blk_peek_request() returns NULL, but the
1277 * queue must be restarted, so we schedule a callback to happen
1280 if (atomic_read(&sdev->device_busy) == 0)
1281 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1284 req->cmd_flags |= REQ_DONTPREP;
1290 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1292 struct scsi_device *sdev = q->queuedata;
1293 struct scsi_cmnd *cmd;
1296 ret = scsi_prep_state_check(sdev, req);
1297 if (ret != BLKPREP_OK)
1300 cmd = scsi_get_cmd_from_req(sdev, req);
1301 if (unlikely(!cmd)) {
1302 ret = BLKPREP_DEFER;
1306 ret = scsi_setup_cmnd(sdev, req);
1308 return scsi_prep_return(q, req, ret);
1311 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1313 scsi_uninit_cmd(req->special);
1317 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1320 * Called with the queue_lock held.
1322 static inline int scsi_dev_queue_ready(struct request_queue *q,
1323 struct scsi_device *sdev)
1327 busy = atomic_inc_return(&sdev->device_busy) - 1;
1328 if (atomic_read(&sdev->device_blocked)) {
1333 * unblock after device_blocked iterates to zero
1335 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1337 * For the MQ case we take care of this in the caller.
1340 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1343 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1344 "unblocking device at zero depth\n"));
1347 if (busy >= sdev->queue_depth)
1352 atomic_dec(&sdev->device_busy);
1357 * scsi_target_queue_ready: checks if there we can send commands to target
1358 * @sdev: scsi device on starget to check.
1360 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1361 struct scsi_device *sdev)
1363 struct scsi_target *starget = scsi_target(sdev);
1366 if (starget->single_lun) {
1367 spin_lock_irq(shost->host_lock);
1368 if (starget->starget_sdev_user &&
1369 starget->starget_sdev_user != sdev) {
1370 spin_unlock_irq(shost->host_lock);
1373 starget->starget_sdev_user = sdev;
1374 spin_unlock_irq(shost->host_lock);
1377 if (starget->can_queue <= 0)
1380 busy = atomic_inc_return(&starget->target_busy) - 1;
1381 if (atomic_read(&starget->target_blocked) > 0) {
1386 * unblock after target_blocked iterates to zero
1388 if (atomic_dec_return(&starget->target_blocked) > 0)
1391 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1392 "unblocking target at zero depth\n"));
1395 if (busy >= starget->can_queue)
1401 spin_lock_irq(shost->host_lock);
1402 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1403 spin_unlock_irq(shost->host_lock);
1405 if (starget->can_queue > 0)
1406 atomic_dec(&starget->target_busy);
1411 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1412 * return 0. We must end up running the queue again whenever 0 is
1413 * returned, else IO can hang.
1415 static inline int scsi_host_queue_ready(struct request_queue *q,
1416 struct Scsi_Host *shost,
1417 struct scsi_device *sdev)
1421 if (scsi_host_in_recovery(shost))
1424 busy = atomic_inc_return(&shost->host_busy) - 1;
1425 if (atomic_read(&shost->host_blocked) > 0) {
1430 * unblock after host_blocked iterates to zero
1432 if (atomic_dec_return(&shost->host_blocked) > 0)
1436 shost_printk(KERN_INFO, shost,
1437 "unblocking host at zero depth\n"));
1440 if (shost->can_queue > 0 && busy >= shost->can_queue)
1442 if (shost->host_self_blocked)
1445 /* We're OK to process the command, so we can't be starved */
1446 if (!list_empty(&sdev->starved_entry)) {
1447 spin_lock_irq(shost->host_lock);
1448 if (!list_empty(&sdev->starved_entry))
1449 list_del_init(&sdev->starved_entry);
1450 spin_unlock_irq(shost->host_lock);
1456 spin_lock_irq(shost->host_lock);
1457 if (list_empty(&sdev->starved_entry))
1458 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1459 spin_unlock_irq(shost->host_lock);
1461 atomic_dec(&shost->host_busy);
1466 * Busy state exporting function for request stacking drivers.
1468 * For efficiency, no lock is taken to check the busy state of
1469 * shost/starget/sdev, since the returned value is not guaranteed and
1470 * may be changed after request stacking drivers call the function,
1471 * regardless of taking lock or not.
1473 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1474 * needs to return 'not busy'. Otherwise, request stacking drivers
1475 * may hold requests forever.
1477 static int scsi_lld_busy(struct request_queue *q)
1479 struct scsi_device *sdev = q->queuedata;
1480 struct Scsi_Host *shost;
1482 if (blk_queue_dying(q))
1488 * Ignore host/starget busy state.
1489 * Since block layer does not have a concept of fairness across
1490 * multiple queues, congestion of host/starget needs to be handled
1493 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1500 * Kill a request for a dead device
1502 static void scsi_kill_request(struct request *req, struct request_queue *q)
1504 struct scsi_cmnd *cmd = req->special;
1505 struct scsi_device *sdev;
1506 struct scsi_target *starget;
1507 struct Scsi_Host *shost;
1509 blk_start_request(req);
1511 scmd_printk(KERN_INFO, cmd, "killing request\n");
1514 starget = scsi_target(sdev);
1516 scsi_init_cmd_errh(cmd);
1517 cmd->result = DID_NO_CONNECT << 16;
1518 atomic_inc(&cmd->device->iorequest_cnt);
1521 * SCSI request completion path will do scsi_device_unbusy(),
1522 * bump busy counts. To bump the counters, we need to dance
1523 * with the locks as normal issue path does.
1525 atomic_inc(&sdev->device_busy);
1526 atomic_inc(&shost->host_busy);
1527 if (starget->can_queue > 0)
1528 atomic_inc(&starget->target_busy);
1530 blk_complete_request(req);
1533 static void scsi_softirq_done(struct request *rq)
1535 struct scsi_cmnd *cmd = rq->special;
1536 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1539 INIT_LIST_HEAD(&cmd->eh_entry);
1541 atomic_inc(&cmd->device->iodone_cnt);
1543 atomic_inc(&cmd->device->ioerr_cnt);
1545 disposition = scsi_decide_disposition(cmd);
1546 if (disposition != SUCCESS &&
1547 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1548 sdev_printk(KERN_ERR, cmd->device,
1549 "timing out command, waited %lus\n",
1551 disposition = SUCCESS;
1554 scsi_log_completion(cmd, disposition);
1556 switch (disposition) {
1558 scsi_finish_command(cmd);
1561 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1563 case ADD_TO_MLQUEUE:
1564 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1567 if (!scsi_eh_scmd_add(cmd, 0))
1568 scsi_finish_command(cmd);
1573 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1574 * @cmd: command block we are dispatching.
1576 * Return: nonzero return request was rejected and device's queue needs to be
1579 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1581 struct Scsi_Host *host = cmd->device->host;
1584 atomic_inc(&cmd->device->iorequest_cnt);
1586 /* check if the device is still usable */
1587 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1588 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1589 * returns an immediate error upwards, and signals
1590 * that the device is no longer present */
1591 cmd->result = DID_NO_CONNECT << 16;
1595 /* Check to see if the scsi lld made this device blocked. */
1596 if (unlikely(scsi_device_blocked(cmd->device))) {
1598 * in blocked state, the command is just put back on
1599 * the device queue. The suspend state has already
1600 * blocked the queue so future requests should not
1601 * occur until the device transitions out of the
1604 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1605 "queuecommand : device blocked\n"));
1606 return SCSI_MLQUEUE_DEVICE_BUSY;
1609 /* Store the LUN value in cmnd, if needed. */
1610 if (cmd->device->lun_in_cdb)
1611 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1612 (cmd->device->lun << 5 & 0xe0);
1617 * Before we queue this command, check if the command
1618 * length exceeds what the host adapter can handle.
1620 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1621 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1622 "queuecommand : command too long. "
1623 "cdb_size=%d host->max_cmd_len=%d\n",
1624 cmd->cmd_len, cmd->device->host->max_cmd_len));
1625 cmd->result = (DID_ABORT << 16);
1629 if (unlikely(host->shost_state == SHOST_DEL)) {
1630 cmd->result = (DID_NO_CONNECT << 16);
1635 trace_scsi_dispatch_cmd_start(cmd);
1636 rtn = host->hostt->queuecommand(host, cmd);
1638 trace_scsi_dispatch_cmd_error(cmd, rtn);
1639 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1640 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1641 rtn = SCSI_MLQUEUE_HOST_BUSY;
1643 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1644 "queuecommand : request rejected\n"));
1649 cmd->scsi_done(cmd);
1654 * scsi_done - Invoke completion on finished SCSI command.
1655 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1656 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1658 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1659 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1660 * calls blk_complete_request() for further processing.
1662 * This function is interrupt context safe.
1664 static void scsi_done(struct scsi_cmnd *cmd)
1666 trace_scsi_dispatch_cmd_done(cmd);
1667 blk_complete_request(cmd->request);
1671 * Function: scsi_request_fn()
1673 * Purpose: Main strategy routine for SCSI.
1675 * Arguments: q - Pointer to actual queue.
1679 * Lock status: IO request lock assumed to be held when called.
1681 static void scsi_request_fn(struct request_queue *q)
1682 __releases(q->queue_lock)
1683 __acquires(q->queue_lock)
1685 struct scsi_device *sdev = q->queuedata;
1686 struct Scsi_Host *shost;
1687 struct scsi_cmnd *cmd;
1688 struct request *req;
1691 * To start with, we keep looping until the queue is empty, or until
1692 * the host is no longer able to accept any more requests.
1698 * get next queueable request. We do this early to make sure
1699 * that the request is fully prepared even if we cannot
1702 req = blk_peek_request(q);
1706 if (unlikely(!scsi_device_online(sdev))) {
1707 sdev_printk(KERN_ERR, sdev,
1708 "rejecting I/O to offline device\n");
1709 scsi_kill_request(req, q);
1713 if (!scsi_dev_queue_ready(q, sdev))
1717 * Remove the request from the request list.
1719 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1720 blk_start_request(req);
1722 spin_unlock_irq(q->queue_lock);
1724 if (unlikely(cmd == NULL)) {
1725 printk(KERN_CRIT "impossible request in %s.\n"
1726 "please mail a stack trace to "
1727 "linux-scsi@vger.kernel.org\n",
1729 blk_dump_rq_flags(req, "foo");
1734 * We hit this when the driver is using a host wide
1735 * tag map. For device level tag maps the queue_depth check
1736 * in the device ready fn would prevent us from trying
1737 * to allocate a tag. Since the map is a shared host resource
1738 * we add the dev to the starved list so it eventually gets
1739 * a run when a tag is freed.
1741 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1742 spin_lock_irq(shost->host_lock);
1743 if (list_empty(&sdev->starved_entry))
1744 list_add_tail(&sdev->starved_entry,
1745 &shost->starved_list);
1746 spin_unlock_irq(shost->host_lock);
1750 if (!scsi_target_queue_ready(shost, sdev))
1753 if (!scsi_host_queue_ready(q, shost, sdev))
1754 goto host_not_ready;
1756 if (sdev->simple_tags)
1757 cmd->flags |= SCMD_TAGGED;
1759 cmd->flags &= ~SCMD_TAGGED;
1762 * Finally, initialize any error handling parameters, and set up
1763 * the timers for timeouts.
1765 scsi_init_cmd_errh(cmd);
1768 * Dispatch the command to the low-level driver.
1770 cmd->scsi_done = scsi_done;
1771 rtn = scsi_dispatch_cmd(cmd);
1773 scsi_queue_insert(cmd, rtn);
1774 spin_lock_irq(q->queue_lock);
1777 spin_lock_irq(q->queue_lock);
1783 if (scsi_target(sdev)->can_queue > 0)
1784 atomic_dec(&scsi_target(sdev)->target_busy);
1787 * lock q, handle tag, requeue req, and decrement device_busy. We
1788 * must return with queue_lock held.
1790 * Decrementing device_busy without checking it is OK, as all such
1791 * cases (host limits or settings) should run the queue at some
1794 spin_lock_irq(q->queue_lock);
1795 blk_requeue_request(q, req);
1796 atomic_dec(&sdev->device_busy);
1798 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1799 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1802 static inline int prep_to_mq(int ret)
1808 return BLK_MQ_RQ_QUEUE_BUSY;
1810 return BLK_MQ_RQ_QUEUE_ERROR;
1814 static int scsi_mq_prep_fn(struct request *req)
1816 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1817 struct scsi_device *sdev = req->q->queuedata;
1818 struct Scsi_Host *shost = sdev->host;
1819 unsigned char *sense_buf = cmd->sense_buffer;
1820 struct scatterlist *sg;
1822 memset(cmd, 0, sizeof(struct scsi_cmnd));
1828 cmd->sense_buffer = sense_buf;
1830 cmd->tag = req->tag;
1832 cmd->cmnd = req->cmd;
1833 cmd->prot_op = SCSI_PROT_NORMAL;
1835 INIT_LIST_HEAD(&cmd->list);
1836 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1837 cmd->jiffies_at_alloc = jiffies;
1839 if (shost->use_cmd_list) {
1840 spin_lock_irq(&sdev->list_lock);
1841 list_add_tail(&cmd->list, &sdev->cmd_list);
1842 spin_unlock_irq(&sdev->list_lock);
1845 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1846 cmd->sdb.table.sgl = sg;
1848 if (scsi_host_get_prot(shost)) {
1849 cmd->prot_sdb = (void *)sg +
1851 shost->sg_tablesize, SG_CHUNK_SIZE) *
1852 sizeof(struct scatterlist);
1853 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1855 cmd->prot_sdb->table.sgl =
1856 (struct scatterlist *)(cmd->prot_sdb + 1);
1859 if (blk_bidi_rq(req)) {
1860 struct request *next_rq = req->next_rq;
1861 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1863 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1864 bidi_sdb->table.sgl =
1865 (struct scatterlist *)(bidi_sdb + 1);
1867 next_rq->special = bidi_sdb;
1870 blk_mq_start_request(req);
1872 return scsi_setup_cmnd(sdev, req);
1875 static void scsi_mq_done(struct scsi_cmnd *cmd)
1877 trace_scsi_dispatch_cmd_done(cmd);
1878 blk_mq_complete_request(cmd->request, cmd->request->errors);
1881 static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1882 const struct blk_mq_queue_data *bd)
1884 struct request *req = bd->rq;
1885 struct request_queue *q = req->q;
1886 struct scsi_device *sdev = q->queuedata;
1887 struct Scsi_Host *shost = sdev->host;
1888 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1892 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1896 ret = BLK_MQ_RQ_QUEUE_BUSY;
1897 if (!get_device(&sdev->sdev_gendev))
1900 if (!scsi_dev_queue_ready(q, sdev))
1901 goto out_put_device;
1902 if (!scsi_target_queue_ready(shost, sdev))
1903 goto out_dec_device_busy;
1904 if (!scsi_host_queue_ready(q, shost, sdev))
1905 goto out_dec_target_busy;
1908 if (!(req->cmd_flags & REQ_DONTPREP)) {
1909 ret = prep_to_mq(scsi_mq_prep_fn(req));
1911 goto out_dec_host_busy;
1912 req->cmd_flags |= REQ_DONTPREP;
1914 blk_mq_start_request(req);
1917 if (sdev->simple_tags)
1918 cmd->flags |= SCMD_TAGGED;
1920 cmd->flags &= ~SCMD_TAGGED;
1922 scsi_init_cmd_errh(cmd);
1923 cmd->scsi_done = scsi_mq_done;
1925 reason = scsi_dispatch_cmd(cmd);
1927 scsi_set_blocked(cmd, reason);
1928 ret = BLK_MQ_RQ_QUEUE_BUSY;
1929 goto out_dec_host_busy;
1932 return BLK_MQ_RQ_QUEUE_OK;
1935 atomic_dec(&shost->host_busy);
1936 out_dec_target_busy:
1937 if (scsi_target(sdev)->can_queue > 0)
1938 atomic_dec(&scsi_target(sdev)->target_busy);
1939 out_dec_device_busy:
1940 atomic_dec(&sdev->device_busy);
1942 put_device(&sdev->sdev_gendev);
1945 case BLK_MQ_RQ_QUEUE_BUSY:
1946 blk_mq_stop_hw_queue(hctx);
1947 if (atomic_read(&sdev->device_busy) == 0 &&
1948 !scsi_device_blocked(sdev))
1949 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
1951 case BLK_MQ_RQ_QUEUE_ERROR:
1953 * Make sure to release all allocated ressources when
1954 * we hit an error, as we will never see this command
1957 if (req->cmd_flags & REQ_DONTPREP)
1958 scsi_mq_uninit_cmd(cmd);
1966 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1970 return BLK_EH_RESET_TIMER;
1971 return scsi_times_out(req);
1974 static int scsi_init_request(void *data, struct request *rq,
1975 unsigned int hctx_idx, unsigned int request_idx,
1976 unsigned int numa_node)
1978 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1980 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
1982 if (!cmd->sense_buffer)
1987 static void scsi_exit_request(void *data, struct request *rq,
1988 unsigned int hctx_idx, unsigned int request_idx)
1990 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1992 kfree(cmd->sense_buffer);
1995 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1997 struct device *host_dev;
1998 u64 bounce_limit = 0xffffffff;
2000 if (shost->unchecked_isa_dma)
2001 return BLK_BOUNCE_ISA;
2003 * Platforms with virtual-DMA translation
2004 * hardware have no practical limit.
2006 if (!PCI_DMA_BUS_IS_PHYS)
2007 return BLK_BOUNCE_ANY;
2009 host_dev = scsi_get_device(shost);
2010 if (host_dev && host_dev->dma_mask)
2011 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2013 return bounce_limit;
2016 static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2018 struct device *dev = shost->dma_dev;
2021 * this limit is imposed by hardware restrictions
2023 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2026 if (scsi_host_prot_dma(shost)) {
2027 shost->sg_prot_tablesize =
2028 min_not_zero(shost->sg_prot_tablesize,
2029 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2030 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2031 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2034 blk_queue_max_hw_sectors(q, shost->max_sectors);
2035 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2036 blk_queue_segment_boundary(q, shost->dma_boundary);
2037 dma_set_seg_boundary(dev, shost->dma_boundary);
2039 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2041 if (!shost->use_clustering)
2042 q->limits.cluster = 0;
2045 * Set a reasonable default alignment: The larger of 32-byte (dword),
2046 * which is a common minimum for HBAs, and the minimum DMA alignment,
2047 * which is set by the platform.
2049 * Devices that require a bigger alignment can increase it later.
2051 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
2054 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2055 request_fn_proc *request_fn)
2057 struct request_queue *q;
2059 q = blk_init_queue(request_fn, NULL);
2062 __scsi_init_queue(shost, q);
2065 EXPORT_SYMBOL(__scsi_alloc_queue);
2067 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2069 struct request_queue *q;
2071 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2075 blk_queue_prep_rq(q, scsi_prep_fn);
2076 blk_queue_unprep_rq(q, scsi_unprep_fn);
2077 blk_queue_softirq_done(q, scsi_softirq_done);
2078 blk_queue_rq_timed_out(q, scsi_times_out);
2079 blk_queue_lld_busy(q, scsi_lld_busy);
2083 static struct blk_mq_ops scsi_mq_ops = {
2084 .queue_rq = scsi_queue_rq,
2085 .complete = scsi_softirq_done,
2086 .timeout = scsi_timeout,
2087 .init_request = scsi_init_request,
2088 .exit_request = scsi_exit_request,
2091 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2093 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2094 if (IS_ERR(sdev->request_queue))
2097 sdev->request_queue->queuedata = sdev;
2098 __scsi_init_queue(sdev->host, sdev->request_queue);
2099 return sdev->request_queue;
2102 int scsi_mq_setup_tags(struct Scsi_Host *shost)
2104 unsigned int cmd_size, sgl_size, tbl_size;
2106 tbl_size = shost->sg_tablesize;
2107 if (tbl_size > SG_CHUNK_SIZE)
2108 tbl_size = SG_CHUNK_SIZE;
2109 sgl_size = tbl_size * sizeof(struct scatterlist);
2110 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2111 if (scsi_host_get_prot(shost))
2112 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2114 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2115 shost->tag_set.ops = &scsi_mq_ops;
2116 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2117 shost->tag_set.queue_depth = shost->can_queue;
2118 shost->tag_set.cmd_size = cmd_size;
2119 shost->tag_set.numa_node = NUMA_NO_NODE;
2120 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2121 shost->tag_set.flags |=
2122 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2123 shost->tag_set.driver_data = shost;
2125 return blk_mq_alloc_tag_set(&shost->tag_set);
2128 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2130 blk_mq_free_tag_set(&shost->tag_set);
2134 * scsi_device_from_queue - return sdev associated with a request_queue
2135 * @q: The request queue to return the sdev from
2137 * Return the sdev associated with a request queue or NULL if the
2138 * request_queue does not reference a SCSI device.
2140 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2142 struct scsi_device *sdev = NULL;
2145 if (q->mq_ops == &scsi_mq_ops)
2146 sdev = q->queuedata;
2147 } else if (q->request_fn == scsi_request_fn)
2148 sdev = q->queuedata;
2149 if (!sdev || !get_device(&sdev->sdev_gendev))
2154 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2157 * Function: scsi_block_requests()
2159 * Purpose: Utility function used by low-level drivers to prevent further
2160 * commands from being queued to the device.
2162 * Arguments: shost - Host in question
2166 * Lock status: No locks are assumed held.
2168 * Notes: There is no timer nor any other means by which the requests
2169 * get unblocked other than the low-level driver calling
2170 * scsi_unblock_requests().
2172 void scsi_block_requests(struct Scsi_Host *shost)
2174 shost->host_self_blocked = 1;
2176 EXPORT_SYMBOL(scsi_block_requests);
2179 * Function: scsi_unblock_requests()
2181 * Purpose: Utility function used by low-level drivers to allow further
2182 * commands from being queued to the device.
2184 * Arguments: shost - Host in question
2188 * Lock status: No locks are assumed held.
2190 * Notes: There is no timer nor any other means by which the requests
2191 * get unblocked other than the low-level driver calling
2192 * scsi_unblock_requests().
2194 * This is done as an API function so that changes to the
2195 * internals of the scsi mid-layer won't require wholesale
2196 * changes to drivers that use this feature.
2198 void scsi_unblock_requests(struct Scsi_Host *shost)
2200 shost->host_self_blocked = 0;
2201 scsi_run_host_queues(shost);
2203 EXPORT_SYMBOL(scsi_unblock_requests);
2205 int __init scsi_init_queue(void)
2207 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2208 sizeof(struct scsi_data_buffer),
2210 if (!scsi_sdb_cache) {
2211 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2218 void scsi_exit_queue(void)
2220 kmem_cache_destroy(scsi_sdb_cache);
2224 * scsi_mode_select - issue a mode select
2225 * @sdev: SCSI device to be queried
2226 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2227 * @sp: Save page bit (0 == don't save, 1 == save)
2228 * @modepage: mode page being requested
2229 * @buffer: request buffer (may not be smaller than eight bytes)
2230 * @len: length of request buffer.
2231 * @timeout: command timeout
2232 * @retries: number of retries before failing
2233 * @data: returns a structure abstracting the mode header data
2234 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2235 * must be SCSI_SENSE_BUFFERSIZE big.
2237 * Returns zero if successful; negative error number or scsi
2242 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2243 unsigned char *buffer, int len, int timeout, int retries,
2244 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2246 unsigned char cmd[10];
2247 unsigned char *real_buffer;
2250 memset(cmd, 0, sizeof(cmd));
2251 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2253 if (sdev->use_10_for_ms) {
2256 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2259 memcpy(real_buffer + 8, buffer, len);
2263 real_buffer[2] = data->medium_type;
2264 real_buffer[3] = data->device_specific;
2265 real_buffer[4] = data->longlba ? 0x01 : 0;
2267 real_buffer[6] = data->block_descriptor_length >> 8;
2268 real_buffer[7] = data->block_descriptor_length;
2270 cmd[0] = MODE_SELECT_10;
2274 if (len > 255 || data->block_descriptor_length > 255 ||
2278 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2281 memcpy(real_buffer + 4, buffer, len);
2284 real_buffer[1] = data->medium_type;
2285 real_buffer[2] = data->device_specific;
2286 real_buffer[3] = data->block_descriptor_length;
2289 cmd[0] = MODE_SELECT;
2293 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2294 sshdr, timeout, retries, NULL);
2298 EXPORT_SYMBOL_GPL(scsi_mode_select);
2301 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2302 * @sdev: SCSI device to be queried
2303 * @dbd: set if mode sense will allow block descriptors to be returned
2304 * @modepage: mode page being requested
2305 * @buffer: request buffer (may not be smaller than eight bytes)
2306 * @len: length of request buffer.
2307 * @timeout: command timeout
2308 * @retries: number of retries before failing
2309 * @data: returns a structure abstracting the mode header data
2310 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2311 * must be SCSI_SENSE_BUFFERSIZE big.
2313 * Returns zero if unsuccessful, or the header offset (either 4
2314 * or 8 depending on whether a six or ten byte command was
2315 * issued) if successful.
2318 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2319 unsigned char *buffer, int len, int timeout, int retries,
2320 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2322 unsigned char cmd[12];
2325 int result, retry_count = retries;
2326 struct scsi_sense_hdr my_sshdr;
2328 memset(data, 0, sizeof(*data));
2329 memset(&cmd[0], 0, 12);
2330 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2333 /* caller might not be interested in sense, but we need it */
2338 use_10_for_ms = sdev->use_10_for_ms;
2340 if (use_10_for_ms) {
2344 cmd[0] = MODE_SENSE_10;
2351 cmd[0] = MODE_SENSE;
2356 memset(buffer, 0, len);
2358 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2359 sshdr, timeout, retries, NULL);
2361 /* This code looks awful: what it's doing is making sure an
2362 * ILLEGAL REQUEST sense return identifies the actual command
2363 * byte as the problem. MODE_SENSE commands can return
2364 * ILLEGAL REQUEST if the code page isn't supported */
2366 if (use_10_for_ms && !scsi_status_is_good(result) &&
2367 (driver_byte(result) & DRIVER_SENSE)) {
2368 if (scsi_sense_valid(sshdr)) {
2369 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2370 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2372 * Invalid command operation code
2374 sdev->use_10_for_ms = 0;
2380 if(scsi_status_is_good(result)) {
2381 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2382 (modepage == 6 || modepage == 8))) {
2383 /* Initio breakage? */
2386 data->medium_type = 0;
2387 data->device_specific = 0;
2389 data->block_descriptor_length = 0;
2390 } else if(use_10_for_ms) {
2391 data->length = buffer[0]*256 + buffer[1] + 2;
2392 data->medium_type = buffer[2];
2393 data->device_specific = buffer[3];
2394 data->longlba = buffer[4] & 0x01;
2395 data->block_descriptor_length = buffer[6]*256
2398 data->length = buffer[0] + 1;
2399 data->medium_type = buffer[1];
2400 data->device_specific = buffer[2];
2401 data->block_descriptor_length = buffer[3];
2403 data->header_length = header_length;
2404 } else if ((status_byte(result) == CHECK_CONDITION) &&
2405 scsi_sense_valid(sshdr) &&
2406 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2413 EXPORT_SYMBOL(scsi_mode_sense);
2416 * scsi_test_unit_ready - test if unit is ready
2417 * @sdev: scsi device to change the state of.
2418 * @timeout: command timeout
2419 * @retries: number of retries before failing
2420 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2421 * returning sense. Make sure that this is cleared before passing
2424 * Returns zero if unsuccessful or an error if TUR failed. For
2425 * removable media, UNIT_ATTENTION sets ->changed flag.
2428 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2429 struct scsi_sense_hdr *sshdr_external)
2432 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2434 struct scsi_sense_hdr *sshdr;
2437 if (!sshdr_external)
2438 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2440 sshdr = sshdr_external;
2442 /* try to eat the UNIT_ATTENTION if there are enough retries */
2444 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2445 timeout, retries, NULL);
2446 if (sdev->removable && scsi_sense_valid(sshdr) &&
2447 sshdr->sense_key == UNIT_ATTENTION)
2449 } while (scsi_sense_valid(sshdr) &&
2450 sshdr->sense_key == UNIT_ATTENTION && --retries);
2452 if (!sshdr_external)
2456 EXPORT_SYMBOL(scsi_test_unit_ready);
2459 * scsi_device_set_state - Take the given device through the device state model.
2460 * @sdev: scsi device to change the state of.
2461 * @state: state to change to.
2463 * Returns zero if unsuccessful or an error if the requested
2464 * transition is illegal.
2467 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2469 enum scsi_device_state oldstate = sdev->sdev_state;
2471 if (state == oldstate)
2477 case SDEV_CREATED_BLOCK:
2488 case SDEV_TRANSPORT_OFFLINE:
2501 case SDEV_TRANSPORT_OFFLINE:
2509 case SDEV_TRANSPORT_OFFLINE:
2524 case SDEV_CREATED_BLOCK:
2531 case SDEV_CREATED_BLOCK:
2546 case SDEV_TRANSPORT_OFFLINE:
2559 case SDEV_TRANSPORT_OFFLINE:
2561 case SDEV_CREATED_BLOCK:
2569 sdev->sdev_state = state;
2573 SCSI_LOG_ERROR_RECOVERY(1,
2574 sdev_printk(KERN_ERR, sdev,
2575 "Illegal state transition %s->%s",
2576 scsi_device_state_name(oldstate),
2577 scsi_device_state_name(state))
2581 EXPORT_SYMBOL(scsi_device_set_state);
2584 * sdev_evt_emit - emit a single SCSI device uevent
2585 * @sdev: associated SCSI device
2586 * @evt: event to emit
2588 * Send a single uevent (scsi_event) to the associated scsi_device.
2590 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2595 switch (evt->evt_type) {
2596 case SDEV_EVT_MEDIA_CHANGE:
2597 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2599 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2600 scsi_rescan_device(&sdev->sdev_gendev);
2601 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2603 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2604 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2606 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2607 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2609 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2610 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2612 case SDEV_EVT_LUN_CHANGE_REPORTED:
2613 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2615 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2616 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2625 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2629 * sdev_evt_thread - send a uevent for each scsi event
2630 * @work: work struct for scsi_device
2632 * Dispatch queued events to their associated scsi_device kobjects
2635 void scsi_evt_thread(struct work_struct *work)
2637 struct scsi_device *sdev;
2638 enum scsi_device_event evt_type;
2639 LIST_HEAD(event_list);
2641 sdev = container_of(work, struct scsi_device, event_work);
2643 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2644 if (test_and_clear_bit(evt_type, sdev->pending_events))
2645 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2648 struct scsi_event *evt;
2649 struct list_head *this, *tmp;
2650 unsigned long flags;
2652 spin_lock_irqsave(&sdev->list_lock, flags);
2653 list_splice_init(&sdev->event_list, &event_list);
2654 spin_unlock_irqrestore(&sdev->list_lock, flags);
2656 if (list_empty(&event_list))
2659 list_for_each_safe(this, tmp, &event_list) {
2660 evt = list_entry(this, struct scsi_event, node);
2661 list_del(&evt->node);
2662 scsi_evt_emit(sdev, evt);
2669 * sdev_evt_send - send asserted event to uevent thread
2670 * @sdev: scsi_device event occurred on
2671 * @evt: event to send
2673 * Assert scsi device event asynchronously.
2675 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2677 unsigned long flags;
2680 /* FIXME: currently this check eliminates all media change events
2681 * for polled devices. Need to update to discriminate between AN
2682 * and polled events */
2683 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2689 spin_lock_irqsave(&sdev->list_lock, flags);
2690 list_add_tail(&evt->node, &sdev->event_list);
2691 schedule_work(&sdev->event_work);
2692 spin_unlock_irqrestore(&sdev->list_lock, flags);
2694 EXPORT_SYMBOL_GPL(sdev_evt_send);
2697 * sdev_evt_alloc - allocate a new scsi event
2698 * @evt_type: type of event to allocate
2699 * @gfpflags: GFP flags for allocation
2701 * Allocates and returns a new scsi_event.
2703 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2706 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2710 evt->evt_type = evt_type;
2711 INIT_LIST_HEAD(&evt->node);
2713 /* evt_type-specific initialization, if any */
2715 case SDEV_EVT_MEDIA_CHANGE:
2716 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2717 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2718 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2719 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2720 case SDEV_EVT_LUN_CHANGE_REPORTED:
2721 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2729 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2732 * sdev_evt_send_simple - send asserted event to uevent thread
2733 * @sdev: scsi_device event occurred on
2734 * @evt_type: type of event to send
2735 * @gfpflags: GFP flags for allocation
2737 * Assert scsi device event asynchronously, given an event type.
2739 void sdev_evt_send_simple(struct scsi_device *sdev,
2740 enum scsi_device_event evt_type, gfp_t gfpflags)
2742 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2744 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2749 sdev_evt_send(sdev, evt);
2751 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2754 * scsi_device_quiesce - Block user issued commands.
2755 * @sdev: scsi device to quiesce.
2757 * This works by trying to transition to the SDEV_QUIESCE state
2758 * (which must be a legal transition). When the device is in this
2759 * state, only special requests will be accepted, all others will
2760 * be deferred. Since special requests may also be requeued requests,
2761 * a successful return doesn't guarantee the device will be
2762 * totally quiescent.
2764 * Must be called with user context, may sleep.
2766 * Returns zero if unsuccessful or an error if not.
2769 scsi_device_quiesce(struct scsi_device *sdev)
2771 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2775 scsi_run_queue(sdev->request_queue);
2776 while (atomic_read(&sdev->device_busy)) {
2777 msleep_interruptible(200);
2778 scsi_run_queue(sdev->request_queue);
2782 EXPORT_SYMBOL(scsi_device_quiesce);
2785 * scsi_device_resume - Restart user issued commands to a quiesced device.
2786 * @sdev: scsi device to resume.
2788 * Moves the device from quiesced back to running and restarts the
2791 * Must be called with user context, may sleep.
2793 void scsi_device_resume(struct scsi_device *sdev)
2795 /* check if the device state was mutated prior to resume, and if
2796 * so assume the state is being managed elsewhere (for example
2797 * device deleted during suspend)
2799 if (sdev->sdev_state != SDEV_QUIESCE ||
2800 scsi_device_set_state(sdev, SDEV_RUNNING))
2802 scsi_run_queue(sdev->request_queue);
2804 EXPORT_SYMBOL(scsi_device_resume);
2807 device_quiesce_fn(struct scsi_device *sdev, void *data)
2809 scsi_device_quiesce(sdev);
2813 scsi_target_quiesce(struct scsi_target *starget)
2815 starget_for_each_device(starget, NULL, device_quiesce_fn);
2817 EXPORT_SYMBOL(scsi_target_quiesce);
2820 device_resume_fn(struct scsi_device *sdev, void *data)
2822 scsi_device_resume(sdev);
2826 scsi_target_resume(struct scsi_target *starget)
2828 starget_for_each_device(starget, NULL, device_resume_fn);
2830 EXPORT_SYMBOL(scsi_target_resume);
2833 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2834 * @sdev: device to block
2836 * Block request made by scsi lld's to temporarily stop all
2837 * scsi commands on the specified device. Called from interrupt
2838 * or normal process context.
2840 * Returns zero if successful or error if not
2843 * This routine transitions the device to the SDEV_BLOCK state
2844 * (which must be a legal transition). When the device is in this
2845 * state, all commands are deferred until the scsi lld reenables
2846 * the device with scsi_device_unblock or device_block_tmo fires.
2849 scsi_internal_device_block(struct scsi_device *sdev)
2851 struct request_queue *q = sdev->request_queue;
2852 unsigned long flags;
2855 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2857 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2864 * The device has transitioned to SDEV_BLOCK. Stop the
2865 * block layer from calling the midlayer with this device's
2869 blk_mq_stop_hw_queues(q);
2871 spin_lock_irqsave(q->queue_lock, flags);
2873 spin_unlock_irqrestore(q->queue_lock, flags);
2878 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2881 * scsi_internal_device_unblock - resume a device after a block request
2882 * @sdev: device to resume
2883 * @new_state: state to set devices to after unblocking
2885 * Called by scsi lld's or the midlayer to restart the device queue
2886 * for the previously suspended scsi device. Called from interrupt or
2887 * normal process context.
2889 * Returns zero if successful or error if not.
2892 * This routine transitions the device to the SDEV_RUNNING state
2893 * or to one of the offline states (which must be a legal transition)
2894 * allowing the midlayer to goose the queue for this device.
2897 scsi_internal_device_unblock(struct scsi_device *sdev,
2898 enum scsi_device_state new_state)
2900 struct request_queue *q = sdev->request_queue;
2901 unsigned long flags;
2904 * Try to transition the scsi device to SDEV_RUNNING or one of the
2905 * offlined states and goose the device queue if successful.
2907 if ((sdev->sdev_state == SDEV_BLOCK) ||
2908 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2909 sdev->sdev_state = new_state;
2910 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2911 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2912 new_state == SDEV_OFFLINE)
2913 sdev->sdev_state = new_state;
2915 sdev->sdev_state = SDEV_CREATED;
2916 } else if (sdev->sdev_state != SDEV_CANCEL &&
2917 sdev->sdev_state != SDEV_OFFLINE)
2921 blk_mq_start_stopped_hw_queues(q, false);
2923 spin_lock_irqsave(q->queue_lock, flags);
2925 spin_unlock_irqrestore(q->queue_lock, flags);
2930 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2933 device_block(struct scsi_device *sdev, void *data)
2935 scsi_internal_device_block(sdev);
2939 target_block(struct device *dev, void *data)
2941 if (scsi_is_target_device(dev))
2942 starget_for_each_device(to_scsi_target(dev), NULL,
2948 scsi_target_block(struct device *dev)
2950 if (scsi_is_target_device(dev))
2951 starget_for_each_device(to_scsi_target(dev), NULL,
2954 device_for_each_child(dev, NULL, target_block);
2956 EXPORT_SYMBOL_GPL(scsi_target_block);
2959 device_unblock(struct scsi_device *sdev, void *data)
2961 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2965 target_unblock(struct device *dev, void *data)
2967 if (scsi_is_target_device(dev))
2968 starget_for_each_device(to_scsi_target(dev), data,
2974 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2976 if (scsi_is_target_device(dev))
2977 starget_for_each_device(to_scsi_target(dev), &new_state,
2980 device_for_each_child(dev, &new_state, target_unblock);
2982 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2985 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2986 * @sgl: scatter-gather list
2987 * @sg_count: number of segments in sg
2988 * @offset: offset in bytes into sg, on return offset into the mapped area
2989 * @len: bytes to map, on return number of bytes mapped
2991 * Returns virtual address of the start of the mapped page
2993 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2994 size_t *offset, size_t *len)
2997 size_t sg_len = 0, len_complete = 0;
2998 struct scatterlist *sg;
3001 WARN_ON(!irqs_disabled());
3003 for_each_sg(sgl, sg, sg_count, i) {
3004 len_complete = sg_len; /* Complete sg-entries */
3005 sg_len += sg->length;
3006 if (sg_len > *offset)
3010 if (unlikely(i == sg_count)) {
3011 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3013 __func__, sg_len, *offset, sg_count);
3018 /* Offset starting from the beginning of first page in this sg-entry */
3019 *offset = *offset - len_complete + sg->offset;
3021 /* Assumption: contiguous pages can be accessed as "page + i" */
3022 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3023 *offset &= ~PAGE_MASK;
3025 /* Bytes in this sg-entry from *offset to the end of the page */
3026 sg_len = PAGE_SIZE - *offset;
3030 return kmap_atomic(page);
3032 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3035 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3036 * @virt: virtual address to be unmapped
3038 void scsi_kunmap_atomic_sg(void *virt)
3040 kunmap_atomic(virt);
3042 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3044 void sdev_disable_disk_events(struct scsi_device *sdev)
3046 atomic_inc(&sdev->disk_events_disable_depth);
3048 EXPORT_SYMBOL(sdev_disable_disk_events);
3050 void sdev_enable_disk_events(struct scsi_device *sdev)
3052 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3054 atomic_dec(&sdev->disk_events_disable_depth);
3056 EXPORT_SYMBOL(sdev_enable_disk_events);
3059 * scsi_vpd_lun_id - return a unique device identification
3060 * @sdev: SCSI device
3061 * @id: buffer for the identification
3062 * @id_len: length of the buffer
3064 * Copies a unique device identification into @id based
3065 * on the information in the VPD page 0x83 of the device.
3066 * The string will be formatted as a SCSI name string.
3068 * Returns the length of the identification or error on failure.
3069 * If the identifier is longer than the supplied buffer the actual
3070 * identifier length is returned and the buffer is not zero-padded.
3072 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3074 u8 cur_id_type = 0xff;
3076 unsigned char *d, *cur_id_str;
3077 unsigned char __rcu *vpd_pg83;
3078 int id_size = -EINVAL;
3081 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3088 * Look for the correct descriptor.
3089 * Order of preference for lun descriptor:
3090 * - SCSI name string
3091 * - NAA IEEE Registered Extended
3092 * - EUI-64 based 16-byte
3093 * - EUI-64 based 12-byte
3094 * - NAA IEEE Registered
3095 * - NAA IEEE Extended
3097 * as longer descriptors reduce the likelyhood
3098 * of identification clashes.
3101 /* The id string must be at least 20 bytes + terminating NULL byte */
3107 memset(id, 0, id_len);
3109 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3110 /* Skip designators not referring to the LUN */
3111 if ((d[1] & 0x30) != 0x00)
3114 switch (d[1] & 0xf) {
3117 if (cur_id_size > d[3])
3119 /* Prefer anything */
3120 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3123 if (cur_id_size + 4 > id_len)
3124 cur_id_size = id_len - 4;
3126 cur_id_type = d[1] & 0xf;
3127 id_size = snprintf(id, id_len, "t10.%*pE",
3128 cur_id_size, cur_id_str);
3132 if (cur_id_size > d[3])
3134 /* Prefer NAA IEEE Registered Extended */
3135 if (cur_id_type == 0x3 &&
3136 cur_id_size == d[3])
3140 cur_id_type = d[1] & 0xf;
3141 switch (cur_id_size) {
3143 id_size = snprintf(id, id_len,
3148 id_size = snprintf(id, id_len,
3153 id_size = snprintf(id, id_len,
3164 if (cur_id_size > d[3])
3168 cur_id_type = d[1] & 0xf;
3169 switch (cur_id_size) {
3171 id_size = snprintf(id, id_len,
3176 id_size = snprintf(id, id_len,
3186 /* SCSI name string */
3187 if (cur_id_size + 4 > d[3])
3189 /* Prefer others for truncated descriptor */
3190 if (cur_id_size && d[3] > id_len)
3192 cur_id_size = id_size = d[3];
3194 cur_id_type = d[1] & 0xf;
3195 if (cur_id_size >= id_len)
3196 cur_id_size = id_len - 1;
3197 memcpy(id, cur_id_str, cur_id_size);
3198 /* Decrease priority for truncated descriptor */
3199 if (cur_id_size != id_size)
3212 EXPORT_SYMBOL(scsi_vpd_lun_id);
3215 * scsi_vpd_tpg_id - return a target port group identifier
3216 * @sdev: SCSI device
3218 * Returns the Target Port Group identifier from the information
3219 * froom VPD page 0x83 of the device.
3221 * Returns the identifier or error on failure.
3223 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3226 unsigned char __rcu *vpd_pg83;
3227 int group_id = -EAGAIN, rel_port = -1;
3230 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3236 d = sdev->vpd_pg83 + 4;
3237 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3238 switch (d[1] & 0xf) {
3240 /* Relative target port */
3241 rel_port = get_unaligned_be16(&d[6]);
3244 /* Target port group */
3245 group_id = get_unaligned_be16(&d[6]);
3254 if (group_id >= 0 && rel_id && rel_port != -1)
3259 EXPORT_SYMBOL(scsi_vpd_tpg_id);