2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/bitops.h>
12 #include <linux/blkdev.h>
13 #include <linux/completion.h>
14 #include <linux/kernel.h>
15 #include <linux/export.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_dbg.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_driver.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_host.h>
32 #include "scsi_priv.h"
33 #include "scsi_logging.h"
36 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
37 #define SG_MEMPOOL_SIZE 2
39 struct scsi_host_sg_pool {
42 struct kmem_cache *slab;
46 #define SP(x) { x, "sgpool-" __stringify(x) }
47 #if (SCSI_MAX_SG_SEGMENTS < 32)
48 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
50 static struct scsi_host_sg_pool scsi_sg_pools[] = {
53 #if (SCSI_MAX_SG_SEGMENTS > 32)
55 #if (SCSI_MAX_SG_SEGMENTS > 64)
57 #if (SCSI_MAX_SG_SEGMENTS > 128)
59 #if (SCSI_MAX_SG_SEGMENTS > 256)
60 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
65 SP(SCSI_MAX_SG_SEGMENTS)
69 struct kmem_cache *scsi_sdb_cache;
72 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
73 * not change behaviour from the previous unplug mechanism, experimentation
74 * may prove this needs changing.
76 #define SCSI_QUEUE_DELAY 3
79 * __scsi_queue_insert - private queue insertion
80 * @cmd: The SCSI command being requeued
81 * @reason: The reason for the requeue
82 * @unbusy: Whether the queue should be unbusied
84 * This is a private queue insertion. The public interface
85 * scsi_queue_insert() always assumes the queue should be unbusied
86 * because it's always called before the completion. This function is
87 * for a requeue after completion, which should only occur in this
90 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
92 struct Scsi_Host *host = cmd->device->host;
93 struct scsi_device *device = cmd->device;
94 struct scsi_target *starget = scsi_target(device);
95 struct request_queue *q = device->request_queue;
99 printk("Inserting command %p into mlqueue\n", cmd));
102 * Set the appropriate busy bit for the device/host.
104 * If the host/device isn't busy, assume that something actually
105 * completed, and that we should be able to queue a command now.
107 * Note that the prior mid-layer assumption that any host could
108 * always queue at least one command is now broken. The mid-layer
109 * will implement a user specifiable stall (see
110 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
111 * if a command is requeued with no other commands outstanding
112 * either for the device or for the host.
115 case SCSI_MLQUEUE_HOST_BUSY:
116 host->host_blocked = host->max_host_blocked;
118 case SCSI_MLQUEUE_DEVICE_BUSY:
119 case SCSI_MLQUEUE_EH_RETRY:
120 device->device_blocked = device->max_device_blocked;
122 case SCSI_MLQUEUE_TARGET_BUSY:
123 starget->target_blocked = starget->max_target_blocked;
128 * Decrement the counters, since these commands are no longer
129 * active on the host/device.
132 scsi_device_unbusy(device);
135 * Requeue this command. It will go before all other commands
136 * that are already in the queue. Schedule requeue work under
137 * lock such that the kblockd_schedule_work() call happens
138 * before blk_cleanup_queue() finishes.
140 spin_lock_irqsave(q->queue_lock, flags);
141 blk_requeue_request(q, cmd->request);
142 kblockd_schedule_work(q, &device->requeue_work);
143 spin_unlock_irqrestore(q->queue_lock, flags);
147 * Function: scsi_queue_insert()
149 * Purpose: Insert a command in the midlevel queue.
151 * Arguments: cmd - command that we are adding to queue.
152 * reason - why we are inserting command to queue.
154 * Lock status: Assumed that lock is not held upon entry.
158 * Notes: We do this for one of two cases. Either the host is busy
159 * and it cannot accept any more commands for the time being,
160 * or the device returned QUEUE_FULL and can accept no more
162 * Notes: This could be called either from an interrupt context or a
163 * normal process context.
165 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
167 __scsi_queue_insert(cmd, reason, 1);
170 * scsi_execute - insert request and wait for the result
173 * @data_direction: data direction
174 * @buffer: data buffer
175 * @bufflen: len of buffer
176 * @sense: optional sense buffer
177 * @timeout: request timeout in seconds
178 * @retries: number of times to retry request
179 * @flags: or into request flags;
180 * @resid: optional residual length
182 * returns the req->errors value which is the scsi_cmnd result
185 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
186 int data_direction, void *buffer, unsigned bufflen,
187 unsigned char *sense, int timeout, int retries, u64 flags,
191 int write = (data_direction == DMA_TO_DEVICE);
192 int ret = DRIVER_ERROR << 24;
194 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
198 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
199 buffer, bufflen, __GFP_WAIT))
202 req->cmd_len = COMMAND_SIZE(cmd[0]);
203 memcpy(req->cmd, cmd, req->cmd_len);
206 req->retries = retries;
207 req->timeout = timeout;
208 req->cmd_type = REQ_TYPE_BLOCK_PC;
209 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
212 * head injection *required* here otherwise quiesce won't work
214 blk_execute_rq(req->q, NULL, req, 1);
217 * Some devices (USB mass-storage in particular) may transfer
218 * garbage data together with a residue indicating that the data
219 * is invalid. Prevent the garbage from being misinterpreted
220 * and prevent security leaks by zeroing out the excess data.
222 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
223 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
226 *resid = req->resid_len;
229 blk_put_request(req);
233 EXPORT_SYMBOL(scsi_execute);
235 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
236 int data_direction, void *buffer, unsigned bufflen,
237 struct scsi_sense_hdr *sshdr, int timeout, int retries,
238 int *resid, u64 flags)
244 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
246 return DRIVER_ERROR << 24;
248 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
249 sense, timeout, retries, flags, resid);
251 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
256 EXPORT_SYMBOL(scsi_execute_req_flags);
259 * Function: scsi_init_cmd_errh()
261 * Purpose: Initialize cmd fields related to error handling.
263 * Arguments: cmd - command that is ready to be queued.
265 * Notes: This function has the job of initializing a number of
266 * fields related to error handling. Typically this will
267 * be called once for each command, as required.
269 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
271 cmd->serial_number = 0;
272 scsi_set_resid(cmd, 0);
273 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
274 if (cmd->cmd_len == 0)
275 cmd->cmd_len = scsi_command_size(cmd->cmnd);
278 void scsi_device_unbusy(struct scsi_device *sdev)
280 struct Scsi_Host *shost = sdev->host;
281 struct scsi_target *starget = scsi_target(sdev);
284 spin_lock_irqsave(shost->host_lock, flags);
286 starget->target_busy--;
287 if (unlikely(scsi_host_in_recovery(shost) &&
288 (shost->host_failed || shost->host_eh_scheduled)))
289 scsi_eh_wakeup(shost);
290 spin_unlock(shost->host_lock);
291 spin_lock(sdev->request_queue->queue_lock);
293 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
297 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
298 * and call blk_run_queue for all the scsi_devices on the target -
299 * including current_sdev first.
301 * Called with *no* scsi locks held.
303 static void scsi_single_lun_run(struct scsi_device *current_sdev)
305 struct Scsi_Host *shost = current_sdev->host;
306 struct scsi_device *sdev, *tmp;
307 struct scsi_target *starget = scsi_target(current_sdev);
310 spin_lock_irqsave(shost->host_lock, flags);
311 starget->starget_sdev_user = NULL;
312 spin_unlock_irqrestore(shost->host_lock, flags);
315 * Call blk_run_queue for all LUNs on the target, starting with
316 * current_sdev. We race with others (to set starget_sdev_user),
317 * but in most cases, we will be first. Ideally, each LU on the
318 * target would get some limited time or requests on the target.
320 blk_run_queue(current_sdev->request_queue);
322 spin_lock_irqsave(shost->host_lock, flags);
323 if (starget->starget_sdev_user)
325 list_for_each_entry_safe(sdev, tmp, &starget->devices,
326 same_target_siblings) {
327 if (sdev == current_sdev)
329 if (scsi_device_get(sdev))
332 spin_unlock_irqrestore(shost->host_lock, flags);
333 blk_run_queue(sdev->request_queue);
334 spin_lock_irqsave(shost->host_lock, flags);
336 scsi_device_put(sdev);
339 spin_unlock_irqrestore(shost->host_lock, flags);
342 static inline int scsi_device_is_busy(struct scsi_device *sdev)
344 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
350 static inline int scsi_target_is_busy(struct scsi_target *starget)
352 return ((starget->can_queue > 0 &&
353 starget->target_busy >= starget->can_queue) ||
354 starget->target_blocked);
357 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
359 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
360 shost->host_blocked || shost->host_self_blocked)
366 static void scsi_starved_list_run(struct Scsi_Host *shost)
368 LIST_HEAD(starved_list);
369 struct scsi_device *sdev;
372 spin_lock_irqsave(shost->host_lock, flags);
373 list_splice_init(&shost->starved_list, &starved_list);
375 while (!list_empty(&starved_list)) {
376 struct request_queue *slq;
379 * As long as shost is accepting commands and we have
380 * starved queues, call blk_run_queue. scsi_request_fn
381 * drops the queue_lock and can add us back to the
384 * host_lock protects the starved_list and starved_entry.
385 * scsi_request_fn must get the host_lock before checking
386 * or modifying starved_list or starved_entry.
388 if (scsi_host_is_busy(shost))
391 sdev = list_entry(starved_list.next,
392 struct scsi_device, starved_entry);
393 list_del_init(&sdev->starved_entry);
394 if (scsi_target_is_busy(scsi_target(sdev))) {
395 list_move_tail(&sdev->starved_entry,
396 &shost->starved_list);
401 * Once we drop the host lock, a racing scsi_remove_device()
402 * call may remove the sdev from the starved list and destroy
403 * it and the queue. Mitigate by taking a reference to the
404 * queue and never touching the sdev again after we drop the
405 * host lock. Note: if __scsi_remove_device() invokes
406 * blk_cleanup_queue() before the queue is run from this
407 * function then blk_run_queue() will return immediately since
408 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
410 slq = sdev->request_queue;
411 if (!blk_get_queue(slq))
413 spin_unlock_irqrestore(shost->host_lock, flags);
418 spin_lock_irqsave(shost->host_lock, flags);
420 /* put any unprocessed entries back */
421 list_splice(&starved_list, &shost->starved_list);
422 spin_unlock_irqrestore(shost->host_lock, flags);
426 * Function: scsi_run_queue()
428 * Purpose: Select a proper request queue to serve next
430 * Arguments: q - last request's queue
434 * Notes: The previous command was completely finished, start
435 * a new one if possible.
437 static void scsi_run_queue(struct request_queue *q)
439 struct scsi_device *sdev = q->queuedata;
441 if (scsi_target(sdev)->single_lun)
442 scsi_single_lun_run(sdev);
443 if (!list_empty(&sdev->host->starved_list))
444 scsi_starved_list_run(sdev->host);
449 void scsi_requeue_run_queue(struct work_struct *work)
451 struct scsi_device *sdev;
452 struct request_queue *q;
454 sdev = container_of(work, struct scsi_device, requeue_work);
455 q = sdev->request_queue;
460 * Function: scsi_requeue_command()
462 * Purpose: Handle post-processing of completed commands.
464 * Arguments: q - queue to operate on
465 * cmd - command that may need to be requeued.
469 * Notes: After command completion, there may be blocks left
470 * over which weren't finished by the previous command
471 * this can be for a number of reasons - the main one is
472 * I/O errors in the middle of the request, in which case
473 * we need to request the blocks that come after the bad
475 * Notes: Upon return, cmd is a stale pointer.
477 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
479 struct scsi_device *sdev = cmd->device;
480 struct request *req = cmd->request;
483 spin_lock_irqsave(q->queue_lock, flags);
484 blk_unprep_request(req);
486 scsi_put_command(cmd);
487 blk_requeue_request(q, req);
488 spin_unlock_irqrestore(q->queue_lock, flags);
492 put_device(&sdev->sdev_gendev);
495 void scsi_next_command(struct scsi_cmnd *cmd)
497 struct scsi_device *sdev = cmd->device;
498 struct request_queue *q = sdev->request_queue;
500 scsi_put_command(cmd);
503 put_device(&sdev->sdev_gendev);
506 void scsi_run_host_queues(struct Scsi_Host *shost)
508 struct scsi_device *sdev;
510 shost_for_each_device(sdev, shost)
511 scsi_run_queue(sdev->request_queue);
514 static void __scsi_release_buffers(struct scsi_cmnd *, int);
517 * Function: scsi_end_request()
519 * Purpose: Post-processing of completed commands (usually invoked at end
520 * of upper level post-processing and scsi_io_completion).
522 * Arguments: cmd - command that is complete.
523 * error - 0 if I/O indicates success, < 0 for I/O error.
524 * bytes - number of bytes of completed I/O
525 * requeue - indicates whether we should requeue leftovers.
527 * Lock status: Assumed that lock is not held upon entry.
529 * Returns: cmd if requeue required, NULL otherwise.
531 * Notes: This is called for block device requests in order to
532 * mark some number of sectors as complete.
534 * We are guaranteeing that the request queue will be goosed
535 * at some point during this call.
536 * Notes: If cmd was requeued, upon return it will be a stale pointer.
538 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
539 int bytes, int requeue)
541 struct request_queue *q = cmd->device->request_queue;
542 struct request *req = cmd->request;
545 * If there are blocks left over at the end, set up the command
546 * to queue the remainder of them.
548 if (blk_end_request(req, error, bytes)) {
549 /* kill remainder if no retrys */
550 if (error && scsi_noretry_cmd(cmd))
551 blk_end_request_all(req, error);
555 * Bleah. Leftovers again. Stick the
556 * leftovers in the front of the
557 * queue, and goose the queue again.
559 scsi_release_buffers(cmd);
560 scsi_requeue_command(q, cmd);
568 * This will goose the queue request function at the end, so we don't
569 * need to worry about launching another command.
571 __scsi_release_buffers(cmd, 0);
572 scsi_next_command(cmd);
576 static inline unsigned int scsi_sgtable_index(unsigned short nents)
580 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
585 index = get_count_order(nents) - 3;
590 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
592 struct scsi_host_sg_pool *sgp;
594 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
595 mempool_free(sgl, sgp->pool);
598 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
600 struct scsi_host_sg_pool *sgp;
602 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
603 return mempool_alloc(sgp->pool, gfp_mask);
606 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
613 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
614 gfp_mask, scsi_sg_alloc);
616 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
622 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
624 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
627 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
630 if (cmd->sdb.table.nents)
631 scsi_free_sgtable(&cmd->sdb);
633 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
635 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
636 struct scsi_data_buffer *bidi_sdb =
637 cmd->request->next_rq->special;
638 scsi_free_sgtable(bidi_sdb);
639 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
640 cmd->request->next_rq->special = NULL;
643 if (scsi_prot_sg_count(cmd))
644 scsi_free_sgtable(cmd->prot_sdb);
648 * Function: scsi_release_buffers()
650 * Purpose: Completion processing for block device I/O requests.
652 * Arguments: cmd - command that we are bailing.
654 * Lock status: Assumed that no lock is held upon entry.
658 * Notes: In the event that an upper level driver rejects a
659 * command, we must release resources allocated during
660 * the __init_io() function. Primarily this would involve
661 * the scatter-gather table, and potentially any bounce
664 void scsi_release_buffers(struct scsi_cmnd *cmd)
666 __scsi_release_buffers(cmd, 1);
668 EXPORT_SYMBOL(scsi_release_buffers);
671 * __scsi_error_from_host_byte - translate SCSI error code into errno
672 * @cmd: SCSI command (unused)
673 * @result: scsi error code
675 * Translate SCSI error code into standard UNIX errno.
677 * -ENOLINK temporary transport failure
678 * -EREMOTEIO permanent target failure, do not retry
679 * -EBADE permanent nexus failure, retry on other path
680 * -ENOSPC No write space available
681 * -ENODATA Medium error
682 * -EIO unspecified I/O error
684 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
688 switch(host_byte(result)) {
689 case DID_TRANSPORT_FAILFAST:
692 case DID_TARGET_FAILURE:
693 set_host_byte(cmd, DID_OK);
696 case DID_NEXUS_FAILURE:
697 set_host_byte(cmd, DID_OK);
700 case DID_ALLOC_FAILURE:
701 set_host_byte(cmd, DID_OK);
704 case DID_MEDIUM_ERROR:
705 set_host_byte(cmd, DID_OK);
717 * Function: scsi_io_completion()
719 * Purpose: Completion processing for block device I/O requests.
721 * Arguments: cmd - command that is finished.
723 * Lock status: Assumed that no lock is held upon entry.
727 * Notes: This function is matched in terms of capabilities to
728 * the function that created the scatter-gather list.
729 * In other words, if there are no bounce buffers
730 * (the normal case for most drivers), we don't need
731 * the logic to deal with cleaning up afterwards.
733 * We must call scsi_end_request(). This will finish off
734 * the specified number of sectors. If we are done, the
735 * command block will be released and the queue function
736 * will be goosed. If we are not done then we have to
737 * figure out what to do next:
739 * a) We can call scsi_requeue_command(). The request
740 * will be unprepared and put back on the queue. Then
741 * a new command will be created for it. This should
742 * be used if we made forward progress, or if we want
743 * to switch from READ(10) to READ(6) for example.
745 * b) We can call scsi_queue_insert(). The request will
746 * be put back on the queue and retried using the same
747 * command as before, possibly after a delay.
749 * c) We can call blk_end_request() with -EIO to fail
750 * the remainder of the request.
752 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
754 int result = cmd->result;
755 struct request_queue *q = cmd->device->request_queue;
756 struct request *req = cmd->request;
758 struct scsi_sense_hdr sshdr;
760 int sense_deferred = 0;
761 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
762 ACTION_DELAYED_RETRY} action;
763 char *description = NULL;
764 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
767 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
769 sense_deferred = scsi_sense_is_deferred(&sshdr);
772 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
774 if (sense_valid && req->sense) {
776 * SG_IO wants current and deferred errors
778 int len = 8 + cmd->sense_buffer[7];
780 if (len > SCSI_SENSE_BUFFERSIZE)
781 len = SCSI_SENSE_BUFFERSIZE;
782 memcpy(req->sense, cmd->sense_buffer, len);
783 req->sense_len = len;
786 error = __scsi_error_from_host_byte(cmd, result);
789 * __scsi_error_from_host_byte may have reset the host_byte
791 req->errors = cmd->result;
793 req->resid_len = scsi_get_resid(cmd);
795 if (scsi_bidi_cmnd(cmd)) {
797 * Bidi commands Must be complete as a whole,
798 * both sides at once.
800 req->next_rq->resid_len = scsi_in(cmd)->resid;
802 scsi_release_buffers(cmd);
803 blk_end_request_all(req, 0);
805 scsi_next_command(cmd);
810 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
811 BUG_ON(blk_bidi_rq(req));
814 * Next deal with any sectors which we were able to correctly
817 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
819 blk_rq_sectors(req), good_bytes));
822 * Recovered errors need reporting, but they're always treated
823 * as success, so fiddle the result code here. For BLOCK_PC
824 * we already took a copy of the original into rq->errors which
825 * is what gets returned to the user
827 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
828 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
829 * print since caller wants ATA registers. Only occurs on
830 * SCSI ATA PASS_THROUGH commands when CK_COND=1
832 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
834 else if (!(req->cmd_flags & REQ_QUIET))
835 scsi_print_sense("", cmd);
837 /* BLOCK_PC may have set error */
842 * A number of bytes were successfully read. If there
843 * are leftovers and there is some kind of error
844 * (result != 0), retry the rest.
846 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
849 error = __scsi_error_from_host_byte(cmd, result);
851 if (host_byte(result) == DID_RESET) {
852 /* Third party bus reset or reset for error recovery
853 * reasons. Just retry the command and see what
856 action = ACTION_RETRY;
857 } else if (sense_valid && !sense_deferred) {
858 switch (sshdr.sense_key) {
860 if (cmd->device->removable) {
861 /* Detected disc change. Set a bit
862 * and quietly refuse further access.
864 cmd->device->changed = 1;
865 description = "Media Changed";
866 action = ACTION_FAIL;
868 /* Must have been a power glitch, or a
869 * bus reset. Could not have been a
870 * media change, so we just retry the
871 * command and see what happens.
873 action = ACTION_RETRY;
876 case ILLEGAL_REQUEST:
877 /* If we had an ILLEGAL REQUEST returned, then
878 * we may have performed an unsupported
879 * command. The only thing this should be
880 * would be a ten byte read where only a six
881 * byte read was supported. Also, on a system
882 * where READ CAPACITY failed, we may have
883 * read past the end of the disk.
885 if ((cmd->device->use_10_for_rw &&
886 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
887 (cmd->cmnd[0] == READ_10 ||
888 cmd->cmnd[0] == WRITE_10)) {
889 /* This will issue a new 6-byte command. */
890 cmd->device->use_10_for_rw = 0;
891 action = ACTION_REPREP;
892 } else if (sshdr.asc == 0x10) /* DIX */ {
893 description = "Host Data Integrity Failure";
894 action = ACTION_FAIL;
896 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
897 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
898 switch (cmd->cmnd[0]) {
900 description = "Discard failure";
904 if (cmd->cmnd[1] & 0x8)
905 description = "Discard failure";
908 "Write same failure";
911 description = "Invalid command failure";
914 action = ACTION_FAIL;
917 action = ACTION_FAIL;
919 case ABORTED_COMMAND:
920 action = ACTION_FAIL;
921 if (sshdr.asc == 0x10) { /* DIF */
922 description = "Target Data Integrity Failure";
927 /* If the device is in the process of becoming
928 * ready, or has a temporary blockage, retry.
930 if (sshdr.asc == 0x04) {
931 switch (sshdr.ascq) {
932 case 0x01: /* becoming ready */
933 case 0x04: /* format in progress */
934 case 0x05: /* rebuild in progress */
935 case 0x06: /* recalculation in progress */
936 case 0x07: /* operation in progress */
937 case 0x08: /* Long write in progress */
938 case 0x09: /* self test in progress */
939 case 0x14: /* space allocation in progress */
940 action = ACTION_DELAYED_RETRY;
943 description = "Device not ready";
944 action = ACTION_FAIL;
948 description = "Device not ready";
949 action = ACTION_FAIL;
952 case VOLUME_OVERFLOW:
953 /* See SSC3rXX or current. */
954 action = ACTION_FAIL;
957 description = "Unhandled sense code";
958 action = ACTION_FAIL;
962 description = "Unhandled error code";
963 action = ACTION_FAIL;
966 if (action != ACTION_FAIL &&
967 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
968 action = ACTION_FAIL;
969 description = "Command timed out";
974 /* Give up and fail the remainder of the request */
975 scsi_release_buffers(cmd);
976 if (!(req->cmd_flags & REQ_QUIET)) {
978 scmd_printk(KERN_INFO, cmd, "%s\n",
980 scsi_print_result(cmd);
981 if (driver_byte(result) & DRIVER_SENSE)
982 scsi_print_sense("", cmd);
983 scsi_print_command(cmd);
985 if (blk_end_request_err(req, error))
986 scsi_requeue_command(q, cmd);
988 scsi_next_command(cmd);
991 /* Unprep the request and put it back at the head of the queue.
992 * A new command will be prepared and issued.
994 scsi_release_buffers(cmd);
995 scsi_requeue_command(q, cmd);
998 /* Retry the same command immediately */
999 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1001 case ACTION_DELAYED_RETRY:
1002 /* Retry the same command after a delay */
1003 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1008 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1014 * If sg table allocation fails, requeue request later.
1016 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1018 return BLKPREP_DEFER;
1024 * Next, walk the list, and fill in the addresses and sizes of
1027 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1028 BUG_ON(count > sdb->table.nents);
1029 sdb->table.nents = count;
1030 sdb->length = blk_rq_bytes(req);
1035 * Function: scsi_init_io()
1037 * Purpose: SCSI I/O initialize function.
1039 * Arguments: cmd - Command descriptor we wish to initialize
1041 * Returns: 0 on success
1042 * BLKPREP_DEFER if the failure is retryable
1043 * BLKPREP_KILL if the failure is fatal
1045 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1047 struct request *rq = cmd->request;
1049 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1053 if (blk_bidi_rq(rq)) {
1054 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1055 scsi_sdb_cache, GFP_ATOMIC);
1057 error = BLKPREP_DEFER;
1061 rq->next_rq->special = bidi_sdb;
1062 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1067 if (blk_integrity_rq(rq)) {
1068 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1071 BUG_ON(prot_sdb == NULL);
1072 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1074 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1075 error = BLKPREP_DEFER;
1079 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1080 prot_sdb->table.sgl);
1081 BUG_ON(unlikely(count > ivecs));
1082 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1084 cmd->prot_sdb = prot_sdb;
1085 cmd->prot_sdb->table.nents = count;
1091 scsi_release_buffers(cmd);
1092 cmd->request->special = NULL;
1093 scsi_put_command(cmd);
1094 put_device(&cmd->device->sdev_gendev);
1097 EXPORT_SYMBOL(scsi_init_io);
1099 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1100 struct request *req)
1102 struct scsi_cmnd *cmd;
1104 if (!req->special) {
1105 /* Bail if we can't get a reference to the device */
1106 if (!get_device(&sdev->sdev_gendev))
1109 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1110 if (unlikely(!cmd)) {
1111 put_device(&sdev->sdev_gendev);
1119 /* pull a tag out of the request if we have one */
1120 cmd->tag = req->tag;
1123 cmd->cmnd = req->cmd;
1124 cmd->prot_op = SCSI_PROT_NORMAL;
1129 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1131 struct scsi_cmnd *cmd;
1132 int ret = scsi_prep_state_check(sdev, req);
1134 if (ret != BLKPREP_OK)
1137 cmd = scsi_get_cmd_from_req(sdev, req);
1139 return BLKPREP_DEFER;
1142 * BLOCK_PC requests may transfer data, in which case they must
1143 * a bio attached to them. Or they might contain a SCSI command
1144 * that does not transfer data, in which case they may optionally
1145 * submit a request without an attached bio.
1150 BUG_ON(!req->nr_phys_segments);
1152 ret = scsi_init_io(cmd, GFP_ATOMIC);
1156 BUG_ON(blk_rq_bytes(req));
1158 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1162 cmd->cmd_len = req->cmd_len;
1163 if (!blk_rq_bytes(req))
1164 cmd->sc_data_direction = DMA_NONE;
1165 else if (rq_data_dir(req) == WRITE)
1166 cmd->sc_data_direction = DMA_TO_DEVICE;
1168 cmd->sc_data_direction = DMA_FROM_DEVICE;
1170 cmd->transfersize = blk_rq_bytes(req);
1171 cmd->allowed = req->retries;
1174 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1177 * Setup a REQ_TYPE_FS command. These are simple read/write request
1178 * from filesystems that still need to be translated to SCSI CDBs from
1181 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1183 struct scsi_cmnd *cmd;
1184 int ret = scsi_prep_state_check(sdev, req);
1186 if (ret != BLKPREP_OK)
1189 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1190 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1191 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1192 if (ret != BLKPREP_OK)
1197 * Filesystem requests must transfer data.
1199 BUG_ON(!req->nr_phys_segments);
1201 cmd = scsi_get_cmd_from_req(sdev, req);
1203 return BLKPREP_DEFER;
1205 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1206 return scsi_init_io(cmd, GFP_ATOMIC);
1208 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1210 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1212 int ret = BLKPREP_OK;
1215 * If the device is not in running state we will reject some
1218 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1219 switch (sdev->sdev_state) {
1221 case SDEV_TRANSPORT_OFFLINE:
1223 * If the device is offline we refuse to process any
1224 * commands. The device must be brought online
1225 * before trying any recovery commands.
1227 sdev_printk(KERN_ERR, sdev,
1228 "rejecting I/O to offline device\n");
1233 * If the device is fully deleted, we refuse to
1234 * process any commands as well.
1236 sdev_printk(KERN_ERR, sdev,
1237 "rejecting I/O to dead device\n");
1242 case SDEV_CREATED_BLOCK:
1244 * If the devices is blocked we defer normal commands.
1246 if (!(req->cmd_flags & REQ_PREEMPT))
1247 ret = BLKPREP_DEFER;
1251 * For any other not fully online state we only allow
1252 * special commands. In particular any user initiated
1253 * command is not allowed.
1255 if (!(req->cmd_flags & REQ_PREEMPT))
1262 EXPORT_SYMBOL(scsi_prep_state_check);
1264 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1266 struct scsi_device *sdev = q->queuedata;
1270 req->errors = DID_NO_CONNECT << 16;
1271 /* release the command and kill it */
1273 struct scsi_cmnd *cmd = req->special;
1274 scsi_release_buffers(cmd);
1275 scsi_put_command(cmd);
1276 put_device(&cmd->device->sdev_gendev);
1277 req->special = NULL;
1282 * If we defer, the blk_peek_request() returns NULL, but the
1283 * queue must be restarted, so we schedule a callback to happen
1286 if (sdev->device_busy == 0)
1287 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 req->cmd_flags |= REQ_DONTPREP;
1295 EXPORT_SYMBOL(scsi_prep_return);
1297 int scsi_prep_fn(struct request_queue *q, struct request *req)
1299 struct scsi_device *sdev = q->queuedata;
1300 int ret = BLKPREP_KILL;
1302 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1303 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1304 return scsi_prep_return(q, req, ret);
1306 EXPORT_SYMBOL(scsi_prep_fn);
1309 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1312 * Called with the queue_lock held.
1314 static inline int scsi_dev_queue_ready(struct request_queue *q,
1315 struct scsi_device *sdev)
1317 if (sdev->device_busy == 0 && sdev->device_blocked) {
1319 * unblock after device_blocked iterates to zero
1321 if (--sdev->device_blocked == 0) {
1323 sdev_printk(KERN_INFO, sdev,
1324 "unblocking device at zero depth\n"));
1326 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1330 if (scsi_device_is_busy(sdev))
1338 * scsi_target_queue_ready: checks if there we can send commands to target
1339 * @sdev: scsi device on starget to check.
1341 * Called with the host lock held.
1343 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1344 struct scsi_device *sdev)
1346 struct scsi_target *starget = scsi_target(sdev);
1348 if (starget->single_lun) {
1349 if (starget->starget_sdev_user &&
1350 starget->starget_sdev_user != sdev)
1352 starget->starget_sdev_user = sdev;
1355 if (starget->target_busy == 0 && starget->target_blocked) {
1357 * unblock after target_blocked iterates to zero
1359 if (--starget->target_blocked == 0) {
1360 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1361 "unblocking target at zero depth\n"));
1366 if (scsi_target_is_busy(starget)) {
1367 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1375 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1376 * return 0. We must end up running the queue again whenever 0 is
1377 * returned, else IO can hang.
1379 * Called with host_lock held.
1381 static inline int scsi_host_queue_ready(struct request_queue *q,
1382 struct Scsi_Host *shost,
1383 struct scsi_device *sdev)
1385 if (scsi_host_in_recovery(shost))
1387 if (shost->host_busy == 0 && shost->host_blocked) {
1389 * unblock after host_blocked iterates to zero
1391 if (--shost->host_blocked == 0) {
1393 printk("scsi%d unblocking host at zero depth\n",
1399 if (scsi_host_is_busy(shost)) {
1400 if (list_empty(&sdev->starved_entry))
1401 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1405 /* We're OK to process the command, so we can't be starved */
1406 if (!list_empty(&sdev->starved_entry))
1407 list_del_init(&sdev->starved_entry);
1413 * Busy state exporting function for request stacking drivers.
1415 * For efficiency, no lock is taken to check the busy state of
1416 * shost/starget/sdev, since the returned value is not guaranteed and
1417 * may be changed after request stacking drivers call the function,
1418 * regardless of taking lock or not.
1420 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1421 * needs to return 'not busy'. Otherwise, request stacking drivers
1422 * may hold requests forever.
1424 static int scsi_lld_busy(struct request_queue *q)
1426 struct scsi_device *sdev = q->queuedata;
1427 struct Scsi_Host *shost;
1429 if (blk_queue_dying(q))
1435 * Ignore host/starget busy state.
1436 * Since block layer does not have a concept of fairness across
1437 * multiple queues, congestion of host/starget needs to be handled
1440 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1447 * Kill a request for a dead device
1449 static void scsi_kill_request(struct request *req, struct request_queue *q)
1451 struct scsi_cmnd *cmd = req->special;
1452 struct scsi_device *sdev;
1453 struct scsi_target *starget;
1454 struct Scsi_Host *shost;
1456 blk_start_request(req);
1458 scmd_printk(KERN_INFO, cmd, "killing request\n");
1461 starget = scsi_target(sdev);
1463 scsi_init_cmd_errh(cmd);
1464 cmd->result = DID_NO_CONNECT << 16;
1465 atomic_inc(&cmd->device->iorequest_cnt);
1468 * SCSI request completion path will do scsi_device_unbusy(),
1469 * bump busy counts. To bump the counters, we need to dance
1470 * with the locks as normal issue path does.
1472 sdev->device_busy++;
1473 spin_unlock(sdev->request_queue->queue_lock);
1474 spin_lock(shost->host_lock);
1476 starget->target_busy++;
1477 spin_unlock(shost->host_lock);
1478 spin_lock(sdev->request_queue->queue_lock);
1480 blk_complete_request(req);
1483 static void scsi_softirq_done(struct request *rq)
1485 struct scsi_cmnd *cmd = rq->special;
1486 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1489 INIT_LIST_HEAD(&cmd->eh_entry);
1491 atomic_inc(&cmd->device->iodone_cnt);
1493 atomic_inc(&cmd->device->ioerr_cnt);
1495 disposition = scsi_decide_disposition(cmd);
1496 if (disposition != SUCCESS &&
1497 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1498 sdev_printk(KERN_ERR, cmd->device,
1499 "timing out command, waited %lus\n",
1501 disposition = SUCCESS;
1504 scsi_log_completion(cmd, disposition);
1506 switch (disposition) {
1508 scsi_finish_command(cmd);
1511 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1513 case ADD_TO_MLQUEUE:
1514 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1517 if (!scsi_eh_scmd_add(cmd, 0))
1518 scsi_finish_command(cmd);
1523 * Function: scsi_request_fn()
1525 * Purpose: Main strategy routine for SCSI.
1527 * Arguments: q - Pointer to actual queue.
1531 * Lock status: IO request lock assumed to be held when called.
1533 static void scsi_request_fn(struct request_queue *q)
1534 __releases(q->queue_lock)
1535 __acquires(q->queue_lock)
1537 struct scsi_device *sdev = q->queuedata;
1538 struct Scsi_Host *shost;
1539 struct scsi_cmnd *cmd;
1540 struct request *req;
1543 * To start with, we keep looping until the queue is empty, or until
1544 * the host is no longer able to accept any more requests.
1550 * get next queueable request. We do this early to make sure
1551 * that the request is fully prepared even if we cannot
1554 req = blk_peek_request(q);
1555 if (!req || !scsi_dev_queue_ready(q, sdev))
1558 if (unlikely(!scsi_device_online(sdev))) {
1559 sdev_printk(KERN_ERR, sdev,
1560 "rejecting I/O to offline device\n");
1561 scsi_kill_request(req, q);
1567 * Remove the request from the request list.
1569 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1570 blk_start_request(req);
1571 sdev->device_busy++;
1573 spin_unlock(q->queue_lock);
1575 if (unlikely(cmd == NULL)) {
1576 printk(KERN_CRIT "impossible request in %s.\n"
1577 "please mail a stack trace to "
1578 "linux-scsi@vger.kernel.org\n",
1580 blk_dump_rq_flags(req, "foo");
1583 spin_lock(shost->host_lock);
1586 * We hit this when the driver is using a host wide
1587 * tag map. For device level tag maps the queue_depth check
1588 * in the device ready fn would prevent us from trying
1589 * to allocate a tag. Since the map is a shared host resource
1590 * we add the dev to the starved list so it eventually gets
1591 * a run when a tag is freed.
1593 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1594 if (list_empty(&sdev->starved_entry))
1595 list_add_tail(&sdev->starved_entry,
1596 &shost->starved_list);
1600 if (!scsi_target_queue_ready(shost, sdev))
1603 if (!scsi_host_queue_ready(q, shost, sdev))
1606 scsi_target(sdev)->target_busy++;
1610 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1611 * take the lock again.
1613 spin_unlock_irq(shost->host_lock);
1616 * Finally, initialize any error handling parameters, and set up
1617 * the timers for timeouts.
1619 scsi_init_cmd_errh(cmd);
1622 * Dispatch the command to the low-level driver.
1624 rtn = scsi_dispatch_cmd(cmd);
1625 spin_lock_irq(q->queue_lock);
1633 spin_unlock_irq(shost->host_lock);
1636 * lock q, handle tag, requeue req, and decrement device_busy. We
1637 * must return with queue_lock held.
1639 * Decrementing device_busy without checking it is OK, as all such
1640 * cases (host limits or settings) should run the queue at some
1643 spin_lock_irq(q->queue_lock);
1644 blk_requeue_request(q, req);
1645 sdev->device_busy--;
1647 if (sdev->device_busy == 0)
1648 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1651 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1653 struct device *host_dev;
1654 u64 bounce_limit = 0xffffffff;
1656 if (shost->unchecked_isa_dma)
1657 return BLK_BOUNCE_ISA;
1659 * Platforms with virtual-DMA translation
1660 * hardware have no practical limit.
1662 if (!PCI_DMA_BUS_IS_PHYS)
1663 return BLK_BOUNCE_ANY;
1665 host_dev = scsi_get_device(shost);
1666 if (host_dev && host_dev->dma_mask)
1667 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1669 return bounce_limit;
1671 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1673 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1674 request_fn_proc *request_fn)
1676 struct request_queue *q;
1677 struct device *dev = shost->dma_dev;
1679 q = blk_init_queue(request_fn, NULL);
1684 * this limit is imposed by hardware restrictions
1686 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1687 SCSI_MAX_SG_CHAIN_SEGMENTS));
1689 if (scsi_host_prot_dma(shost)) {
1690 shost->sg_prot_tablesize =
1691 min_not_zero(shost->sg_prot_tablesize,
1692 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1693 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1694 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1697 blk_queue_max_hw_sectors(q, shost->max_sectors);
1698 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1699 blk_queue_segment_boundary(q, shost->dma_boundary);
1700 dma_set_seg_boundary(dev, shost->dma_boundary);
1702 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1704 if (!shost->use_clustering)
1705 q->limits.cluster = 0;
1708 * set a reasonable default alignment on word boundaries: the
1709 * host and device may alter it using
1710 * blk_queue_update_dma_alignment() later.
1712 blk_queue_dma_alignment(q, 0x03);
1716 EXPORT_SYMBOL(__scsi_alloc_queue);
1718 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1720 struct request_queue *q;
1722 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1726 blk_queue_prep_rq(q, scsi_prep_fn);
1727 blk_queue_softirq_done(q, scsi_softirq_done);
1728 blk_queue_rq_timed_out(q, scsi_times_out);
1729 blk_queue_lld_busy(q, scsi_lld_busy);
1734 * Function: scsi_block_requests()
1736 * Purpose: Utility function used by low-level drivers to prevent further
1737 * commands from being queued to the device.
1739 * Arguments: shost - Host in question
1743 * Lock status: No locks are assumed held.
1745 * Notes: There is no timer nor any other means by which the requests
1746 * get unblocked other than the low-level driver calling
1747 * scsi_unblock_requests().
1749 void scsi_block_requests(struct Scsi_Host *shost)
1751 shost->host_self_blocked = 1;
1753 EXPORT_SYMBOL(scsi_block_requests);
1756 * Function: scsi_unblock_requests()
1758 * Purpose: Utility function used by low-level drivers to allow further
1759 * commands from being queued to the device.
1761 * Arguments: shost - Host in question
1765 * Lock status: No locks are assumed held.
1767 * Notes: There is no timer nor any other means by which the requests
1768 * get unblocked other than the low-level driver calling
1769 * scsi_unblock_requests().
1771 * This is done as an API function so that changes to the
1772 * internals of the scsi mid-layer won't require wholesale
1773 * changes to drivers that use this feature.
1775 void scsi_unblock_requests(struct Scsi_Host *shost)
1777 shost->host_self_blocked = 0;
1778 scsi_run_host_queues(shost);
1780 EXPORT_SYMBOL(scsi_unblock_requests);
1782 int __init scsi_init_queue(void)
1786 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1787 sizeof(struct scsi_data_buffer),
1789 if (!scsi_sdb_cache) {
1790 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1794 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1795 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1796 int size = sgp->size * sizeof(struct scatterlist);
1798 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1799 SLAB_HWCACHE_ALIGN, NULL);
1801 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1806 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1809 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1818 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1819 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1821 mempool_destroy(sgp->pool);
1823 kmem_cache_destroy(sgp->slab);
1825 kmem_cache_destroy(scsi_sdb_cache);
1830 void scsi_exit_queue(void)
1834 kmem_cache_destroy(scsi_sdb_cache);
1836 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1837 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1838 mempool_destroy(sgp->pool);
1839 kmem_cache_destroy(sgp->slab);
1844 * scsi_mode_select - issue a mode select
1845 * @sdev: SCSI device to be queried
1846 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1847 * @sp: Save page bit (0 == don't save, 1 == save)
1848 * @modepage: mode page being requested
1849 * @buffer: request buffer (may not be smaller than eight bytes)
1850 * @len: length of request buffer.
1851 * @timeout: command timeout
1852 * @retries: number of retries before failing
1853 * @data: returns a structure abstracting the mode header data
1854 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1855 * must be SCSI_SENSE_BUFFERSIZE big.
1857 * Returns zero if successful; negative error number or scsi
1862 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1863 unsigned char *buffer, int len, int timeout, int retries,
1864 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1866 unsigned char cmd[10];
1867 unsigned char *real_buffer;
1870 memset(cmd, 0, sizeof(cmd));
1871 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1873 if (sdev->use_10_for_ms) {
1876 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1879 memcpy(real_buffer + 8, buffer, len);
1883 real_buffer[2] = data->medium_type;
1884 real_buffer[3] = data->device_specific;
1885 real_buffer[4] = data->longlba ? 0x01 : 0;
1887 real_buffer[6] = data->block_descriptor_length >> 8;
1888 real_buffer[7] = data->block_descriptor_length;
1890 cmd[0] = MODE_SELECT_10;
1894 if (len > 255 || data->block_descriptor_length > 255 ||
1898 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1901 memcpy(real_buffer + 4, buffer, len);
1904 real_buffer[1] = data->medium_type;
1905 real_buffer[2] = data->device_specific;
1906 real_buffer[3] = data->block_descriptor_length;
1909 cmd[0] = MODE_SELECT;
1913 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1914 sshdr, timeout, retries, NULL);
1918 EXPORT_SYMBOL_GPL(scsi_mode_select);
1921 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1922 * @sdev: SCSI device to be queried
1923 * @dbd: set if mode sense will allow block descriptors to be returned
1924 * @modepage: mode page being requested
1925 * @buffer: request buffer (may not be smaller than eight bytes)
1926 * @len: length of request buffer.
1927 * @timeout: command timeout
1928 * @retries: number of retries before failing
1929 * @data: returns a structure abstracting the mode header data
1930 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1931 * must be SCSI_SENSE_BUFFERSIZE big.
1933 * Returns zero if unsuccessful, or the header offset (either 4
1934 * or 8 depending on whether a six or ten byte command was
1935 * issued) if successful.
1938 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1939 unsigned char *buffer, int len, int timeout, int retries,
1940 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1942 unsigned char cmd[12];
1946 struct scsi_sense_hdr my_sshdr;
1948 memset(data, 0, sizeof(*data));
1949 memset(&cmd[0], 0, 12);
1950 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1953 /* caller might not be interested in sense, but we need it */
1958 use_10_for_ms = sdev->use_10_for_ms;
1960 if (use_10_for_ms) {
1964 cmd[0] = MODE_SENSE_10;
1971 cmd[0] = MODE_SENSE;
1976 memset(buffer, 0, len);
1978 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1979 sshdr, timeout, retries, NULL);
1981 /* This code looks awful: what it's doing is making sure an
1982 * ILLEGAL REQUEST sense return identifies the actual command
1983 * byte as the problem. MODE_SENSE commands can return
1984 * ILLEGAL REQUEST if the code page isn't supported */
1986 if (use_10_for_ms && !scsi_status_is_good(result) &&
1987 (driver_byte(result) & DRIVER_SENSE)) {
1988 if (scsi_sense_valid(sshdr)) {
1989 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1990 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1992 * Invalid command operation code
1994 sdev->use_10_for_ms = 0;
2000 if(scsi_status_is_good(result)) {
2001 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2002 (modepage == 6 || modepage == 8))) {
2003 /* Initio breakage? */
2006 data->medium_type = 0;
2007 data->device_specific = 0;
2009 data->block_descriptor_length = 0;
2010 } else if(use_10_for_ms) {
2011 data->length = buffer[0]*256 + buffer[1] + 2;
2012 data->medium_type = buffer[2];
2013 data->device_specific = buffer[3];
2014 data->longlba = buffer[4] & 0x01;
2015 data->block_descriptor_length = buffer[6]*256
2018 data->length = buffer[0] + 1;
2019 data->medium_type = buffer[1];
2020 data->device_specific = buffer[2];
2021 data->block_descriptor_length = buffer[3];
2023 data->header_length = header_length;
2028 EXPORT_SYMBOL(scsi_mode_sense);
2031 * scsi_test_unit_ready - test if unit is ready
2032 * @sdev: scsi device to change the state of.
2033 * @timeout: command timeout
2034 * @retries: number of retries before failing
2035 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2036 * returning sense. Make sure that this is cleared before passing
2039 * Returns zero if unsuccessful or an error if TUR failed. For
2040 * removable media, UNIT_ATTENTION sets ->changed flag.
2043 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2044 struct scsi_sense_hdr *sshdr_external)
2047 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2049 struct scsi_sense_hdr *sshdr;
2052 if (!sshdr_external)
2053 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2055 sshdr = sshdr_external;
2057 /* try to eat the UNIT_ATTENTION if there are enough retries */
2059 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2060 timeout, retries, NULL);
2061 if (sdev->removable && scsi_sense_valid(sshdr) &&
2062 sshdr->sense_key == UNIT_ATTENTION)
2064 } while (scsi_sense_valid(sshdr) &&
2065 sshdr->sense_key == UNIT_ATTENTION && --retries);
2067 if (!sshdr_external)
2071 EXPORT_SYMBOL(scsi_test_unit_ready);
2074 * scsi_device_set_state - Take the given device through the device state model.
2075 * @sdev: scsi device to change the state of.
2076 * @state: state to change to.
2078 * Returns zero if unsuccessful or an error if the requested
2079 * transition is illegal.
2082 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2084 enum scsi_device_state oldstate = sdev->sdev_state;
2086 if (state == oldstate)
2092 case SDEV_CREATED_BLOCK:
2103 case SDEV_TRANSPORT_OFFLINE:
2116 case SDEV_TRANSPORT_OFFLINE:
2124 case SDEV_TRANSPORT_OFFLINE:
2139 case SDEV_CREATED_BLOCK:
2146 case SDEV_CREATED_BLOCK:
2161 case SDEV_TRANSPORT_OFFLINE:
2174 case SDEV_TRANSPORT_OFFLINE:
2176 case SDEV_CREATED_BLOCK:
2184 sdev->sdev_state = state;
2188 SCSI_LOG_ERROR_RECOVERY(1,
2189 sdev_printk(KERN_ERR, sdev,
2190 "Illegal state transition %s->%s\n",
2191 scsi_device_state_name(oldstate),
2192 scsi_device_state_name(state))
2196 EXPORT_SYMBOL(scsi_device_set_state);
2199 * sdev_evt_emit - emit a single SCSI device uevent
2200 * @sdev: associated SCSI device
2201 * @evt: event to emit
2203 * Send a single uevent (scsi_event) to the associated scsi_device.
2205 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2210 switch (evt->evt_type) {
2211 case SDEV_EVT_MEDIA_CHANGE:
2212 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2214 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2215 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2217 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2218 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2220 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2221 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2223 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2224 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2226 case SDEV_EVT_LUN_CHANGE_REPORTED:
2227 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2236 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2240 * sdev_evt_thread - send a uevent for each scsi event
2241 * @work: work struct for scsi_device
2243 * Dispatch queued events to their associated scsi_device kobjects
2246 void scsi_evt_thread(struct work_struct *work)
2248 struct scsi_device *sdev;
2249 enum scsi_device_event evt_type;
2250 LIST_HEAD(event_list);
2252 sdev = container_of(work, struct scsi_device, event_work);
2254 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2255 if (test_and_clear_bit(evt_type, sdev->pending_events))
2256 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2259 struct scsi_event *evt;
2260 struct list_head *this, *tmp;
2261 unsigned long flags;
2263 spin_lock_irqsave(&sdev->list_lock, flags);
2264 list_splice_init(&sdev->event_list, &event_list);
2265 spin_unlock_irqrestore(&sdev->list_lock, flags);
2267 if (list_empty(&event_list))
2270 list_for_each_safe(this, tmp, &event_list) {
2271 evt = list_entry(this, struct scsi_event, node);
2272 list_del(&evt->node);
2273 scsi_evt_emit(sdev, evt);
2280 * sdev_evt_send - send asserted event to uevent thread
2281 * @sdev: scsi_device event occurred on
2282 * @evt: event to send
2284 * Assert scsi device event asynchronously.
2286 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2288 unsigned long flags;
2291 /* FIXME: currently this check eliminates all media change events
2292 * for polled devices. Need to update to discriminate between AN
2293 * and polled events */
2294 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2300 spin_lock_irqsave(&sdev->list_lock, flags);
2301 list_add_tail(&evt->node, &sdev->event_list);
2302 schedule_work(&sdev->event_work);
2303 spin_unlock_irqrestore(&sdev->list_lock, flags);
2305 EXPORT_SYMBOL_GPL(sdev_evt_send);
2308 * sdev_evt_alloc - allocate a new scsi event
2309 * @evt_type: type of event to allocate
2310 * @gfpflags: GFP flags for allocation
2312 * Allocates and returns a new scsi_event.
2314 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2317 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2321 evt->evt_type = evt_type;
2322 INIT_LIST_HEAD(&evt->node);
2324 /* evt_type-specific initialization, if any */
2326 case SDEV_EVT_MEDIA_CHANGE:
2327 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2328 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2329 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2330 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2331 case SDEV_EVT_LUN_CHANGE_REPORTED:
2339 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2342 * sdev_evt_send_simple - send asserted event to uevent thread
2343 * @sdev: scsi_device event occurred on
2344 * @evt_type: type of event to send
2345 * @gfpflags: GFP flags for allocation
2347 * Assert scsi device event asynchronously, given an event type.
2349 void sdev_evt_send_simple(struct scsi_device *sdev,
2350 enum scsi_device_event evt_type, gfp_t gfpflags)
2352 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2354 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2359 sdev_evt_send(sdev, evt);
2361 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2364 * scsi_device_quiesce - Block user issued commands.
2365 * @sdev: scsi device to quiesce.
2367 * This works by trying to transition to the SDEV_QUIESCE state
2368 * (which must be a legal transition). When the device is in this
2369 * state, only special requests will be accepted, all others will
2370 * be deferred. Since special requests may also be requeued requests,
2371 * a successful return doesn't guarantee the device will be
2372 * totally quiescent.
2374 * Must be called with user context, may sleep.
2376 * Returns zero if unsuccessful or an error if not.
2379 scsi_device_quiesce(struct scsi_device *sdev)
2381 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2385 scsi_run_queue(sdev->request_queue);
2386 while (sdev->device_busy) {
2387 msleep_interruptible(200);
2388 scsi_run_queue(sdev->request_queue);
2392 EXPORT_SYMBOL(scsi_device_quiesce);
2395 * scsi_device_resume - Restart user issued commands to a quiesced device.
2396 * @sdev: scsi device to resume.
2398 * Moves the device from quiesced back to running and restarts the
2401 * Must be called with user context, may sleep.
2403 void scsi_device_resume(struct scsi_device *sdev)
2405 /* check if the device state was mutated prior to resume, and if
2406 * so assume the state is being managed elsewhere (for example
2407 * device deleted during suspend)
2409 if (sdev->sdev_state != SDEV_QUIESCE ||
2410 scsi_device_set_state(sdev, SDEV_RUNNING))
2412 scsi_run_queue(sdev->request_queue);
2414 EXPORT_SYMBOL(scsi_device_resume);
2417 device_quiesce_fn(struct scsi_device *sdev, void *data)
2419 scsi_device_quiesce(sdev);
2423 scsi_target_quiesce(struct scsi_target *starget)
2425 starget_for_each_device(starget, NULL, device_quiesce_fn);
2427 EXPORT_SYMBOL(scsi_target_quiesce);
2430 device_resume_fn(struct scsi_device *sdev, void *data)
2432 scsi_device_resume(sdev);
2436 scsi_target_resume(struct scsi_target *starget)
2438 starget_for_each_device(starget, NULL, device_resume_fn);
2440 EXPORT_SYMBOL(scsi_target_resume);
2443 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2444 * @sdev: device to block
2446 * Block request made by scsi lld's to temporarily stop all
2447 * scsi commands on the specified device. Called from interrupt
2448 * or normal process context.
2450 * Returns zero if successful or error if not
2453 * This routine transitions the device to the SDEV_BLOCK state
2454 * (which must be a legal transition). When the device is in this
2455 * state, all commands are deferred until the scsi lld reenables
2456 * the device with scsi_device_unblock or device_block_tmo fires.
2459 scsi_internal_device_block(struct scsi_device *sdev)
2461 struct request_queue *q = sdev->request_queue;
2462 unsigned long flags;
2465 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2467 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2474 * The device has transitioned to SDEV_BLOCK. Stop the
2475 * block layer from calling the midlayer with this device's
2478 spin_lock_irqsave(q->queue_lock, flags);
2480 spin_unlock_irqrestore(q->queue_lock, flags);
2484 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2487 * scsi_internal_device_unblock - resume a device after a block request
2488 * @sdev: device to resume
2489 * @new_state: state to set devices to after unblocking
2491 * Called by scsi lld's or the midlayer to restart the device queue
2492 * for the previously suspended scsi device. Called from interrupt or
2493 * normal process context.
2495 * Returns zero if successful or error if not.
2498 * This routine transitions the device to the SDEV_RUNNING state
2499 * or to one of the offline states (which must be a legal transition)
2500 * allowing the midlayer to goose the queue for this device.
2503 scsi_internal_device_unblock(struct scsi_device *sdev,
2504 enum scsi_device_state new_state)
2506 struct request_queue *q = sdev->request_queue;
2507 unsigned long flags;
2510 * Try to transition the scsi device to SDEV_RUNNING or one of the
2511 * offlined states and goose the device queue if successful.
2513 if ((sdev->sdev_state == SDEV_BLOCK) ||
2514 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2515 sdev->sdev_state = new_state;
2516 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2517 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2518 new_state == SDEV_OFFLINE)
2519 sdev->sdev_state = new_state;
2521 sdev->sdev_state = SDEV_CREATED;
2522 } else if (sdev->sdev_state != SDEV_CANCEL &&
2523 sdev->sdev_state != SDEV_OFFLINE)
2526 spin_lock_irqsave(q->queue_lock, flags);
2528 spin_unlock_irqrestore(q->queue_lock, flags);
2532 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2535 device_block(struct scsi_device *sdev, void *data)
2537 scsi_internal_device_block(sdev);
2541 target_block(struct device *dev, void *data)
2543 if (scsi_is_target_device(dev))
2544 starget_for_each_device(to_scsi_target(dev), NULL,
2550 scsi_target_block(struct device *dev)
2552 if (scsi_is_target_device(dev))
2553 starget_for_each_device(to_scsi_target(dev), NULL,
2556 device_for_each_child(dev, NULL, target_block);
2558 EXPORT_SYMBOL_GPL(scsi_target_block);
2561 device_unblock(struct scsi_device *sdev, void *data)
2563 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2567 target_unblock(struct device *dev, void *data)
2569 if (scsi_is_target_device(dev))
2570 starget_for_each_device(to_scsi_target(dev), data,
2576 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2578 if (scsi_is_target_device(dev))
2579 starget_for_each_device(to_scsi_target(dev), &new_state,
2582 device_for_each_child(dev, &new_state, target_unblock);
2584 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2587 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2588 * @sgl: scatter-gather list
2589 * @sg_count: number of segments in sg
2590 * @offset: offset in bytes into sg, on return offset into the mapped area
2591 * @len: bytes to map, on return number of bytes mapped
2593 * Returns virtual address of the start of the mapped page
2595 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2596 size_t *offset, size_t *len)
2599 size_t sg_len = 0, len_complete = 0;
2600 struct scatterlist *sg;
2603 WARN_ON(!irqs_disabled());
2605 for_each_sg(sgl, sg, sg_count, i) {
2606 len_complete = sg_len; /* Complete sg-entries */
2607 sg_len += sg->length;
2608 if (sg_len > *offset)
2612 if (unlikely(i == sg_count)) {
2613 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2615 __func__, sg_len, *offset, sg_count);
2620 /* Offset starting from the beginning of first page in this sg-entry */
2621 *offset = *offset - len_complete + sg->offset;
2623 /* Assumption: contiguous pages can be accessed as "page + i" */
2624 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2625 *offset &= ~PAGE_MASK;
2627 /* Bytes in this sg-entry from *offset to the end of the page */
2628 sg_len = PAGE_SIZE - *offset;
2632 return kmap_atomic(page);
2634 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2637 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2638 * @virt: virtual address to be unmapped
2640 void scsi_kunmap_atomic_sg(void *virt)
2642 kunmap_atomic(virt);
2644 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2646 void sdev_disable_disk_events(struct scsi_device *sdev)
2648 atomic_inc(&sdev->disk_events_disable_depth);
2650 EXPORT_SYMBOL(sdev_disable_disk_events);
2652 void sdev_enable_disk_events(struct scsi_device *sdev)
2654 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2656 atomic_dec(&sdev->disk_events_disable_depth);
2658 EXPORT_SYMBOL(sdev_enable_disk_events);