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/mempool.h>
16 #include <linux/slab.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>
23 #include <scsi/scsi.h>
24 #include <scsi/scsi_cmnd.h>
25 #include <scsi/scsi_dbg.h>
26 #include <scsi/scsi_device.h>
27 #include <scsi/scsi_driver.h>
28 #include <scsi/scsi_eh.h>
29 #include <scsi/scsi_host.h>
31 #include "scsi_priv.h"
32 #include "scsi_logging.h"
35 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
36 #define SG_MEMPOOL_SIZE 2
38 struct scsi_host_sg_pool
{
41 struct kmem_cache
*slab
;
45 #define SP(x) { x, "sgpool-" __stringify(x) }
46 #if (SCSI_MAX_SG_SEGMENTS < 32)
47 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
49 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
52 #if (SCSI_MAX_SG_SEGMENTS > 32)
54 #if (SCSI_MAX_SG_SEGMENTS > 64)
56 #if (SCSI_MAX_SG_SEGMENTS > 128)
58 #if (SCSI_MAX_SG_SEGMENTS > 256)
59 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
64 SP(SCSI_MAX_SG_SEGMENTS
)
68 struct kmem_cache
*scsi_sdb_cache
;
70 static void scsi_run_queue(struct request_queue
*q
);
73 * Function: scsi_unprep_request()
75 * Purpose: Remove all preparation done for a request, including its
76 * associated scsi_cmnd, so that it can be requeued.
78 * Arguments: req - request to unprepare
80 * Lock status: Assumed that no locks are held upon entry.
84 static void scsi_unprep_request(struct request
*req
)
86 struct scsi_cmnd
*cmd
= req
->special
;
88 blk_unprep_request(req
);
91 scsi_put_command(cmd
);
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
106 static int __scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
, int unbusy
)
108 struct Scsi_Host
*host
= cmd
->device
->host
;
109 struct scsi_device
*device
= cmd
->device
;
110 struct scsi_target
*starget
= scsi_target(device
);
111 struct request_queue
*q
= device
->request_queue
;
115 printk("Inserting command %p into mlqueue\n", cmd
));
118 * Set the appropriate busy bit for the device/host.
120 * If the host/device isn't busy, assume that something actually
121 * completed, and that we should be able to queue a command now.
123 * Note that the prior mid-layer assumption that any host could
124 * always queue at least one command is now broken. The mid-layer
125 * will implement a user specifiable stall (see
126 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
127 * if a command is requeued with no other commands outstanding
128 * either for the device or for the host.
131 case SCSI_MLQUEUE_HOST_BUSY
:
132 host
->host_blocked
= host
->max_host_blocked
;
134 case SCSI_MLQUEUE_DEVICE_BUSY
:
135 device
->device_blocked
= device
->max_device_blocked
;
137 case SCSI_MLQUEUE_TARGET_BUSY
:
138 starget
->target_blocked
= starget
->max_target_blocked
;
143 * Decrement the counters, since these commands are no longer
144 * active on the host/device.
147 scsi_device_unbusy(device
);
150 * Requeue this command. It will go before all other commands
151 * that are already in the queue.
153 * NOTE: there is magic here about the way the queue is plugged if
154 * we have no outstanding commands.
156 * Although we *don't* plug the queue, we call the request
157 * function. The SCSI request function detects the blocked condition
158 * and plugs the queue appropriately.
160 spin_lock_irqsave(q
->queue_lock
, flags
);
161 blk_requeue_request(q
, cmd
->request
);
162 spin_unlock_irqrestore(q
->queue_lock
, flags
);
170 * Function: scsi_queue_insert()
172 * Purpose: Insert a command in the midlevel queue.
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
177 * Lock status: Assumed that lock is not held upon entry.
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
188 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
190 return __scsi_queue_insert(cmd
, reason
, 1);
193 * scsi_execute - insert request and wait for the result
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
205 * returns the req->errors value which is the scsi_cmnd result
208 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
209 int data_direction
, void *buffer
, unsigned bufflen
,
210 unsigned char *sense
, int timeout
, int retries
, int flags
,
214 int write
= (data_direction
== DMA_TO_DEVICE
);
215 int ret
= DRIVER_ERROR
<< 24;
217 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
219 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
220 buffer
, bufflen
, __GFP_WAIT
))
223 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
224 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
227 req
->retries
= retries
;
228 req
->timeout
= timeout
;
229 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
230 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
233 * head injection *required* here otherwise quiesce won't work
235 blk_execute_rq(req
->q
, NULL
, req
, 1);
238 * Some devices (USB mass-storage in particular) may transfer
239 * garbage data together with a residue indicating that the data
240 * is invalid. Prevent the garbage from being misinterpreted
241 * and prevent security leaks by zeroing out the excess data.
243 if (unlikely(req
->resid_len
> 0 && req
->resid_len
<= bufflen
))
244 memset(buffer
+ (bufflen
- req
->resid_len
), 0, req
->resid_len
);
247 *resid
= req
->resid_len
;
250 blk_put_request(req
);
254 EXPORT_SYMBOL(scsi_execute
);
257 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
258 int data_direction
, void *buffer
, unsigned bufflen
,
259 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
,
266 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
268 return DRIVER_ERROR
<< 24;
270 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
271 sense
, timeout
, retries
, 0, resid
);
273 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
278 EXPORT_SYMBOL(scsi_execute_req
);
281 * Function: scsi_init_cmd_errh()
283 * Purpose: Initialize cmd fields related to error handling.
285 * Arguments: cmd - command that is ready to be queued.
287 * Notes: This function has the job of initializing a number of
288 * fields related to error handling. Typically this will
289 * be called once for each command, as required.
291 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
293 cmd
->serial_number
= 0;
294 scsi_set_resid(cmd
, 0);
295 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
296 if (cmd
->cmd_len
== 0)
297 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
300 void scsi_device_unbusy(struct scsi_device
*sdev
)
302 struct Scsi_Host
*shost
= sdev
->host
;
303 struct scsi_target
*starget
= scsi_target(sdev
);
306 spin_lock_irqsave(shost
->host_lock
, flags
);
308 starget
->target_busy
--;
309 if (unlikely(scsi_host_in_recovery(shost
) &&
310 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
311 scsi_eh_wakeup(shost
);
312 spin_unlock(shost
->host_lock
);
313 spin_lock(sdev
->request_queue
->queue_lock
);
315 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
319 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
320 * and call blk_run_queue for all the scsi_devices on the target -
321 * including current_sdev first.
323 * Called with *no* scsi locks held.
325 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
327 struct Scsi_Host
*shost
= current_sdev
->host
;
328 struct scsi_device
*sdev
, *tmp
;
329 struct scsi_target
*starget
= scsi_target(current_sdev
);
332 spin_lock_irqsave(shost
->host_lock
, flags
);
333 starget
->starget_sdev_user
= NULL
;
334 spin_unlock_irqrestore(shost
->host_lock
, flags
);
337 * Call blk_run_queue for all LUNs on the target, starting with
338 * current_sdev. We race with others (to set starget_sdev_user),
339 * but in most cases, we will be first. Ideally, each LU on the
340 * target would get some limited time or requests on the target.
342 blk_run_queue(current_sdev
->request_queue
);
344 spin_lock_irqsave(shost
->host_lock
, flags
);
345 if (starget
->starget_sdev_user
)
347 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
348 same_target_siblings
) {
349 if (sdev
== current_sdev
)
351 if (scsi_device_get(sdev
))
354 spin_unlock_irqrestore(shost
->host_lock
, flags
);
355 blk_run_queue(sdev
->request_queue
);
356 spin_lock_irqsave(shost
->host_lock
, flags
);
358 scsi_device_put(sdev
);
361 spin_unlock_irqrestore(shost
->host_lock
, flags
);
364 static inline int scsi_device_is_busy(struct scsi_device
*sdev
)
366 if (sdev
->device_busy
>= sdev
->queue_depth
|| sdev
->device_blocked
)
372 static inline int scsi_target_is_busy(struct scsi_target
*starget
)
374 return ((starget
->can_queue
> 0 &&
375 starget
->target_busy
>= starget
->can_queue
) ||
376 starget
->target_blocked
);
379 static inline int scsi_host_is_busy(struct Scsi_Host
*shost
)
381 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
382 shost
->host_blocked
|| shost
->host_self_blocked
)
389 * Function: scsi_run_queue()
391 * Purpose: Select a proper request queue to serve next
393 * Arguments: q - last request's queue
397 * Notes: The previous command was completely finished, start
398 * a new one if possible.
400 static void scsi_run_queue(struct request_queue
*q
)
402 struct scsi_device
*sdev
= q
->queuedata
;
403 struct Scsi_Host
*shost
= sdev
->host
;
404 LIST_HEAD(starved_list
);
407 if (scsi_target(sdev
)->single_lun
)
408 scsi_single_lun_run(sdev
);
410 spin_lock_irqsave(shost
->host_lock
, flags
);
411 list_splice_init(&shost
->starved_list
, &starved_list
);
413 while (!list_empty(&starved_list
)) {
417 * As long as shost is accepting commands and we have
418 * starved queues, call blk_run_queue. scsi_request_fn
419 * drops the queue_lock and can add us back to the
422 * host_lock protects the starved_list and starved_entry.
423 * scsi_request_fn must get the host_lock before checking
424 * or modifying starved_list or starved_entry.
426 if (scsi_host_is_busy(shost
))
429 sdev
= list_entry(starved_list
.next
,
430 struct scsi_device
, starved_entry
);
431 list_del_init(&sdev
->starved_entry
);
432 if (scsi_target_is_busy(scsi_target(sdev
))) {
433 list_move_tail(&sdev
->starved_entry
,
434 &shost
->starved_list
);
438 spin_unlock(shost
->host_lock
);
440 spin_lock(sdev
->request_queue
->queue_lock
);
441 flagset
= test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
) &&
442 !test_bit(QUEUE_FLAG_REENTER
,
443 &sdev
->request_queue
->queue_flags
);
445 queue_flag_set(QUEUE_FLAG_REENTER
, sdev
->request_queue
);
446 __blk_run_queue(sdev
->request_queue
);
448 queue_flag_clear(QUEUE_FLAG_REENTER
, sdev
->request_queue
);
449 spin_unlock(sdev
->request_queue
->queue_lock
);
451 spin_lock(shost
->host_lock
);
453 /* put any unprocessed entries back */
454 list_splice(&starved_list
, &shost
->starved_list
);
455 spin_unlock_irqrestore(shost
->host_lock
, flags
);
461 * Function: scsi_requeue_command()
463 * Purpose: Handle post-processing of completed commands.
465 * Arguments: q - queue to operate on
466 * cmd - command that may need to be requeued.
470 * Notes: After command completion, there may be blocks left
471 * over which weren't finished by the previous command
472 * this can be for a number of reasons - the main one is
473 * I/O errors in the middle of the request, in which case
474 * we need to request the blocks that come after the bad
476 * Notes: Upon return, cmd is a stale pointer.
478 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
480 struct request
*req
= cmd
->request
;
483 spin_lock_irqsave(q
->queue_lock
, flags
);
484 scsi_unprep_request(req
);
485 blk_requeue_request(q
, req
);
486 spin_unlock_irqrestore(q
->queue_lock
, flags
);
491 void scsi_next_command(struct scsi_cmnd
*cmd
)
493 struct scsi_device
*sdev
= cmd
->device
;
494 struct request_queue
*q
= sdev
->request_queue
;
496 /* need to hold a reference on the device before we let go of the cmd */
497 get_device(&sdev
->sdev_gendev
);
499 scsi_put_command(cmd
);
502 /* ok to remove device now */
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 * Function: scsi_io_completion()
673 * Purpose: Completion processing for block device I/O requests.
675 * Arguments: cmd - command that is finished.
677 * Lock status: Assumed that no lock is held upon entry.
681 * Notes: This function is matched in terms of capabilities to
682 * the function that created the scatter-gather list.
683 * In other words, if there are no bounce buffers
684 * (the normal case for most drivers), we don't need
685 * the logic to deal with cleaning up afterwards.
687 * We must call scsi_end_request(). This will finish off
688 * the specified number of sectors. If we are done, the
689 * command block will be released and the queue function
690 * will be goosed. If we are not done then we have to
691 * figure out what to do next:
693 * a) We can call scsi_requeue_command(). The request
694 * will be unprepared and put back on the queue. Then
695 * a new command will be created for it. This should
696 * be used if we made forward progress, or if we want
697 * to switch from READ(10) to READ(6) for example.
699 * b) We can call scsi_queue_insert(). The request will
700 * be put back on the queue and retried using the same
701 * command as before, possibly after a delay.
703 * c) We can call blk_end_request() with -EIO to fail
704 * the remainder of the request.
706 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
708 int result
= cmd
->result
;
709 struct request_queue
*q
= cmd
->device
->request_queue
;
710 struct request
*req
= cmd
->request
;
712 struct scsi_sense_hdr sshdr
;
714 int sense_deferred
= 0;
715 enum {ACTION_FAIL
, ACTION_REPREP
, ACTION_RETRY
,
716 ACTION_DELAYED_RETRY
} action
;
717 char *description
= NULL
;
720 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
722 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
725 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
) { /* SG_IO ioctl from block level */
726 req
->errors
= result
;
728 if (sense_valid
&& req
->sense
) {
730 * SG_IO wants current and deferred errors
732 int len
= 8 + cmd
->sense_buffer
[7];
734 if (len
> SCSI_SENSE_BUFFERSIZE
)
735 len
= SCSI_SENSE_BUFFERSIZE
;
736 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
737 req
->sense_len
= len
;
743 req
->resid_len
= scsi_get_resid(cmd
);
745 if (scsi_bidi_cmnd(cmd
)) {
747 * Bidi commands Must be complete as a whole,
748 * both sides at once.
750 req
->next_rq
->resid_len
= scsi_in(cmd
)->resid
;
752 scsi_release_buffers(cmd
);
753 blk_end_request_all(req
, 0);
755 scsi_next_command(cmd
);
760 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
761 BUG_ON(blk_bidi_rq(req
));
764 * Next deal with any sectors which we were able to correctly
767 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
769 blk_rq_sectors(req
), good_bytes
));
772 * Recovered errors need reporting, but they're always treated
773 * as success, so fiddle the result code here. For BLOCK_PC
774 * we already took a copy of the original into rq->errors which
775 * is what gets returned to the user
777 if (sense_valid
&& (sshdr
.sense_key
== RECOVERED_ERROR
)) {
778 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
779 * print since caller wants ATA registers. Only occurs on
780 * SCSI ATA PASS_THROUGH commands when CK_COND=1
782 if ((sshdr
.asc
== 0x0) && (sshdr
.ascq
== 0x1d))
784 else if (!(req
->cmd_flags
& REQ_QUIET
))
785 scsi_print_sense("", cmd
);
787 /* BLOCK_PC may have set error */
792 * A number of bytes were successfully read. If there
793 * are leftovers and there is some kind of error
794 * (result != 0), retry the rest.
796 if (scsi_end_request(cmd
, error
, good_bytes
, result
== 0) == NULL
)
801 if (host_byte(result
) == DID_RESET
) {
802 /* Third party bus reset or reset for error recovery
803 * reasons. Just retry the command and see what
806 action
= ACTION_RETRY
;
807 } else if (sense_valid
&& !sense_deferred
) {
808 switch (sshdr
.sense_key
) {
810 if (cmd
->device
->removable
) {
811 /* Detected disc change. Set a bit
812 * and quietly refuse further access.
814 cmd
->device
->changed
= 1;
815 description
= "Media Changed";
816 action
= ACTION_FAIL
;
818 /* Must have been a power glitch, or a
819 * bus reset. Could not have been a
820 * media change, so we just retry the
821 * command and see what happens.
823 action
= ACTION_RETRY
;
826 case ILLEGAL_REQUEST
:
827 /* If we had an ILLEGAL REQUEST returned, then
828 * we may have performed an unsupported
829 * command. The only thing this should be
830 * would be a ten byte read where only a six
831 * byte read was supported. Also, on a system
832 * where READ CAPACITY failed, we may have
833 * read past the end of the disk.
835 if ((cmd
->device
->use_10_for_rw
&&
836 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
837 (cmd
->cmnd
[0] == READ_10
||
838 cmd
->cmnd
[0] == WRITE_10
)) {
839 /* This will issue a new 6-byte command. */
840 cmd
->device
->use_10_for_rw
= 0;
841 action
= ACTION_REPREP
;
842 } else if (sshdr
.asc
== 0x10) /* DIX */ {
843 description
= "Host Data Integrity Failure";
844 action
= ACTION_FAIL
;
847 action
= ACTION_FAIL
;
849 case ABORTED_COMMAND
:
850 action
= ACTION_FAIL
;
851 if (sshdr
.asc
== 0x10) { /* DIF */
852 description
= "Target Data Integrity Failure";
857 /* If the device is in the process of becoming
858 * ready, or has a temporary blockage, retry.
860 if (sshdr
.asc
== 0x04) {
861 switch (sshdr
.ascq
) {
862 case 0x01: /* becoming ready */
863 case 0x04: /* format in progress */
864 case 0x05: /* rebuild in progress */
865 case 0x06: /* recalculation in progress */
866 case 0x07: /* operation in progress */
867 case 0x08: /* Long write in progress */
868 case 0x09: /* self test in progress */
869 case 0x14: /* space allocation in progress */
870 action
= ACTION_DELAYED_RETRY
;
873 description
= "Device not ready";
874 action
= ACTION_FAIL
;
878 description
= "Device not ready";
879 action
= ACTION_FAIL
;
882 case VOLUME_OVERFLOW
:
883 /* See SSC3rXX or current. */
884 action
= ACTION_FAIL
;
887 description
= "Unhandled sense code";
888 action
= ACTION_FAIL
;
892 description
= "Unhandled error code";
893 action
= ACTION_FAIL
;
898 /* Give up and fail the remainder of the request */
899 scsi_release_buffers(cmd
);
900 if (!(req
->cmd_flags
& REQ_QUIET
)) {
902 scmd_printk(KERN_INFO
, cmd
, "%s\n",
904 scsi_print_result(cmd
);
905 if (driver_byte(result
) & DRIVER_SENSE
)
906 scsi_print_sense("", cmd
);
907 scsi_print_command(cmd
);
909 if (blk_end_request_err(req
, error
))
910 scsi_requeue_command(q
, cmd
);
912 scsi_next_command(cmd
);
915 /* Unprep the request and put it back at the head of the queue.
916 * A new command will be prepared and issued.
918 scsi_release_buffers(cmd
);
919 scsi_requeue_command(q
, cmd
);
922 /* Retry the same command immediately */
923 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
925 case ACTION_DELAYED_RETRY
:
926 /* Retry the same command after a delay */
927 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
932 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
938 * If sg table allocation fails, requeue request later.
940 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
942 return BLKPREP_DEFER
;
948 * Next, walk the list, and fill in the addresses and sizes of
951 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
952 BUG_ON(count
> sdb
->table
.nents
);
953 sdb
->table
.nents
= count
;
954 sdb
->length
= blk_rq_bytes(req
);
959 * Function: scsi_init_io()
961 * Purpose: SCSI I/O initialize function.
963 * Arguments: cmd - Command descriptor we wish to initialize
965 * Returns: 0 on success
966 * BLKPREP_DEFER if the failure is retryable
967 * BLKPREP_KILL if the failure is fatal
969 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
971 struct request
*rq
= cmd
->request
;
973 int error
= scsi_init_sgtable(rq
, &cmd
->sdb
, gfp_mask
);
977 if (blk_bidi_rq(rq
)) {
978 struct scsi_data_buffer
*bidi_sdb
= kmem_cache_zalloc(
979 scsi_sdb_cache
, GFP_ATOMIC
);
981 error
= BLKPREP_DEFER
;
985 rq
->next_rq
->special
= bidi_sdb
;
986 error
= scsi_init_sgtable(rq
->next_rq
, bidi_sdb
, GFP_ATOMIC
);
991 if (blk_integrity_rq(rq
)) {
992 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
995 BUG_ON(prot_sdb
== NULL
);
996 ivecs
= blk_rq_count_integrity_sg(rq
->q
, rq
->bio
);
998 if (scsi_alloc_sgtable(prot_sdb
, ivecs
, gfp_mask
)) {
999 error
= BLKPREP_DEFER
;
1003 count
= blk_rq_map_integrity_sg(rq
->q
, rq
->bio
,
1004 prot_sdb
->table
.sgl
);
1005 BUG_ON(unlikely(count
> ivecs
));
1006 BUG_ON(unlikely(count
> queue_max_integrity_segments(rq
->q
)));
1008 cmd
->prot_sdb
= prot_sdb
;
1009 cmd
->prot_sdb
->table
.nents
= count
;
1015 scsi_release_buffers(cmd
);
1016 cmd
->request
->special
= NULL
;
1017 scsi_put_command(cmd
);
1020 EXPORT_SYMBOL(scsi_init_io
);
1022 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1023 struct request
*req
)
1025 struct scsi_cmnd
*cmd
;
1027 if (!req
->special
) {
1028 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1036 /* pull a tag out of the request if we have one */
1037 cmd
->tag
= req
->tag
;
1040 cmd
->cmnd
= req
->cmd
;
1045 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1047 struct scsi_cmnd
*cmd
;
1048 int ret
= scsi_prep_state_check(sdev
, req
);
1050 if (ret
!= BLKPREP_OK
)
1053 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1055 return BLKPREP_DEFER
;
1058 * BLOCK_PC requests may transfer data, in which case they must
1059 * a bio attached to them. Or they might contain a SCSI command
1060 * that does not transfer data, in which case they may optionally
1061 * submit a request without an attached bio.
1066 BUG_ON(!req
->nr_phys_segments
);
1068 ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1072 BUG_ON(blk_rq_bytes(req
));
1074 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1078 cmd
->cmd_len
= req
->cmd_len
;
1079 if (!blk_rq_bytes(req
))
1080 cmd
->sc_data_direction
= DMA_NONE
;
1081 else if (rq_data_dir(req
) == WRITE
)
1082 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1084 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1086 cmd
->transfersize
= blk_rq_bytes(req
);
1087 cmd
->allowed
= req
->retries
;
1090 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1093 * Setup a REQ_TYPE_FS command. These are simple read/write request
1094 * from filesystems that still need to be translated to SCSI CDBs from
1097 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1099 struct scsi_cmnd
*cmd
;
1100 int ret
= scsi_prep_state_check(sdev
, req
);
1102 if (ret
!= BLKPREP_OK
)
1105 if (unlikely(sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
1106 && sdev
->scsi_dh_data
->scsi_dh
->prep_fn
)) {
1107 ret
= sdev
->scsi_dh_data
->scsi_dh
->prep_fn(sdev
, req
);
1108 if (ret
!= BLKPREP_OK
)
1113 * Filesystem requests must transfer data.
1115 BUG_ON(!req
->nr_phys_segments
);
1117 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1119 return BLKPREP_DEFER
;
1121 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1122 return scsi_init_io(cmd
, GFP_ATOMIC
);
1124 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1126 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1128 int ret
= BLKPREP_OK
;
1131 * If the device is not in running state we will reject some
1134 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1135 switch (sdev
->sdev_state
) {
1138 * If the device is offline we refuse to process any
1139 * commands. The device must be brought online
1140 * before trying any recovery commands.
1142 sdev_printk(KERN_ERR
, sdev
,
1143 "rejecting I/O to offline device\n");
1148 * If the device is fully deleted, we refuse to
1149 * process any commands as well.
1151 sdev_printk(KERN_ERR
, sdev
,
1152 "rejecting I/O to dead device\n");
1157 case SDEV_CREATED_BLOCK
:
1159 * If the devices is blocked we defer normal commands.
1161 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1162 ret
= BLKPREP_DEFER
;
1166 * For any other not fully online state we only allow
1167 * special commands. In particular any user initiated
1168 * command is not allowed.
1170 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1177 EXPORT_SYMBOL(scsi_prep_state_check
);
1179 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1181 struct scsi_device
*sdev
= q
->queuedata
;
1185 req
->errors
= DID_NO_CONNECT
<< 16;
1186 /* release the command and kill it */
1188 struct scsi_cmnd
*cmd
= req
->special
;
1189 scsi_release_buffers(cmd
);
1190 scsi_put_command(cmd
);
1191 req
->special
= NULL
;
1196 * If we defer, the blk_peek_request() returns NULL, but the
1197 * queue must be restarted, so we plug here if no returning
1198 * command will automatically do that.
1200 if (sdev
->device_busy
== 0)
1204 req
->cmd_flags
|= REQ_DONTPREP
;
1209 EXPORT_SYMBOL(scsi_prep_return
);
1211 int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1213 struct scsi_device
*sdev
= q
->queuedata
;
1214 int ret
= BLKPREP_KILL
;
1216 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1217 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1218 return scsi_prep_return(q
, req
, ret
);
1220 EXPORT_SYMBOL(scsi_prep_fn
);
1223 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1226 * Called with the queue_lock held.
1228 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1229 struct scsi_device
*sdev
)
1231 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1233 * unblock after device_blocked iterates to zero
1235 if (--sdev
->device_blocked
== 0) {
1237 sdev_printk(KERN_INFO
, sdev
,
1238 "unblocking device at zero depth\n"));
1244 if (scsi_device_is_busy(sdev
))
1252 * scsi_target_queue_ready: checks if there we can send commands to target
1253 * @sdev: scsi device on starget to check.
1255 * Called with the host lock held.
1257 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1258 struct scsi_device
*sdev
)
1260 struct scsi_target
*starget
= scsi_target(sdev
);
1262 if (starget
->single_lun
) {
1263 if (starget
->starget_sdev_user
&&
1264 starget
->starget_sdev_user
!= sdev
)
1266 starget
->starget_sdev_user
= sdev
;
1269 if (starget
->target_busy
== 0 && starget
->target_blocked
) {
1271 * unblock after target_blocked iterates to zero
1273 if (--starget
->target_blocked
== 0) {
1274 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1275 "unblocking target at zero depth\n"));
1280 if (scsi_target_is_busy(starget
)) {
1281 if (list_empty(&sdev
->starved_entry
)) {
1282 list_add_tail(&sdev
->starved_entry
,
1283 &shost
->starved_list
);
1288 /* We're OK to process the command, so we can't be starved */
1289 if (!list_empty(&sdev
->starved_entry
))
1290 list_del_init(&sdev
->starved_entry
);
1295 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1296 * return 0. We must end up running the queue again whenever 0 is
1297 * returned, else IO can hang.
1299 * Called with host_lock held.
1301 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1302 struct Scsi_Host
*shost
,
1303 struct scsi_device
*sdev
)
1305 if (scsi_host_in_recovery(shost
))
1307 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1309 * unblock after host_blocked iterates to zero
1311 if (--shost
->host_blocked
== 0) {
1313 printk("scsi%d unblocking host at zero depth\n",
1319 if (scsi_host_is_busy(shost
)) {
1320 if (list_empty(&sdev
->starved_entry
))
1321 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1325 /* We're OK to process the command, so we can't be starved */
1326 if (!list_empty(&sdev
->starved_entry
))
1327 list_del_init(&sdev
->starved_entry
);
1333 * Busy state exporting function for request stacking drivers.
1335 * For efficiency, no lock is taken to check the busy state of
1336 * shost/starget/sdev, since the returned value is not guaranteed and
1337 * may be changed after request stacking drivers call the function,
1338 * regardless of taking lock or not.
1340 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1341 * (e.g. !sdev), scsi needs to return 'not busy'.
1342 * Otherwise, request stacking drivers may hold requests forever.
1344 static int scsi_lld_busy(struct request_queue
*q
)
1346 struct scsi_device
*sdev
= q
->queuedata
;
1347 struct Scsi_Host
*shost
;
1348 struct scsi_target
*starget
;
1354 starget
= scsi_target(sdev
);
1356 if (scsi_host_in_recovery(shost
) || scsi_host_is_busy(shost
) ||
1357 scsi_target_is_busy(starget
) || scsi_device_is_busy(sdev
))
1364 * Kill a request for a dead device
1366 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1368 struct scsi_cmnd
*cmd
= req
->special
;
1369 struct scsi_device
*sdev
;
1370 struct scsi_target
*starget
;
1371 struct Scsi_Host
*shost
;
1373 blk_start_request(req
);
1376 starget
= scsi_target(sdev
);
1378 scsi_init_cmd_errh(cmd
);
1379 cmd
->result
= DID_NO_CONNECT
<< 16;
1380 atomic_inc(&cmd
->device
->iorequest_cnt
);
1383 * SCSI request completion path will do scsi_device_unbusy(),
1384 * bump busy counts. To bump the counters, we need to dance
1385 * with the locks as normal issue path does.
1387 sdev
->device_busy
++;
1388 spin_unlock(sdev
->request_queue
->queue_lock
);
1389 spin_lock(shost
->host_lock
);
1391 starget
->target_busy
++;
1392 spin_unlock(shost
->host_lock
);
1393 spin_lock(sdev
->request_queue
->queue_lock
);
1395 blk_complete_request(req
);
1398 static void scsi_softirq_done(struct request
*rq
)
1400 struct scsi_cmnd
*cmd
= rq
->special
;
1401 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1404 INIT_LIST_HEAD(&cmd
->eh_entry
);
1406 atomic_inc(&cmd
->device
->iodone_cnt
);
1408 atomic_inc(&cmd
->device
->ioerr_cnt
);
1410 disposition
= scsi_decide_disposition(cmd
);
1411 if (disposition
!= SUCCESS
&&
1412 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1413 sdev_printk(KERN_ERR
, cmd
->device
,
1414 "timing out command, waited %lus\n",
1416 disposition
= SUCCESS
;
1419 scsi_log_completion(cmd
, disposition
);
1421 switch (disposition
) {
1423 scsi_finish_command(cmd
);
1426 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1428 case ADD_TO_MLQUEUE
:
1429 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1432 if (!scsi_eh_scmd_add(cmd
, 0))
1433 scsi_finish_command(cmd
);
1438 * Function: scsi_request_fn()
1440 * Purpose: Main strategy routine for SCSI.
1442 * Arguments: q - Pointer to actual queue.
1446 * Lock status: IO request lock assumed to be held when called.
1448 static void scsi_request_fn(struct request_queue
*q
)
1450 struct scsi_device
*sdev
= q
->queuedata
;
1451 struct Scsi_Host
*shost
;
1452 struct scsi_cmnd
*cmd
;
1453 struct request
*req
;
1456 printk("scsi: killing requests for dead queue\n");
1457 while ((req
= blk_peek_request(q
)) != NULL
)
1458 scsi_kill_request(req
, q
);
1462 if(!get_device(&sdev
->sdev_gendev
))
1463 /* We must be tearing the block queue down already */
1467 * To start with, we keep looping until the queue is empty, or until
1468 * the host is no longer able to accept any more requests.
1471 while (!blk_queue_plugged(q
)) {
1474 * get next queueable request. We do this early to make sure
1475 * that the request is fully prepared even if we cannot
1478 req
= blk_peek_request(q
);
1479 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1482 if (unlikely(!scsi_device_online(sdev
))) {
1483 sdev_printk(KERN_ERR
, sdev
,
1484 "rejecting I/O to offline device\n");
1485 scsi_kill_request(req
, q
);
1491 * Remove the request from the request list.
1493 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1494 blk_start_request(req
);
1495 sdev
->device_busy
++;
1497 spin_unlock(q
->queue_lock
);
1499 if (unlikely(cmd
== NULL
)) {
1500 printk(KERN_CRIT
"impossible request in %s.\n"
1501 "please mail a stack trace to "
1502 "linux-scsi@vger.kernel.org\n",
1504 blk_dump_rq_flags(req
, "foo");
1507 spin_lock(shost
->host_lock
);
1510 * We hit this when the driver is using a host wide
1511 * tag map. For device level tag maps the queue_depth check
1512 * in the device ready fn would prevent us from trying
1513 * to allocate a tag. Since the map is a shared host resource
1514 * we add the dev to the starved list so it eventually gets
1515 * a run when a tag is freed.
1517 if (blk_queue_tagged(q
) && !blk_rq_tagged(req
)) {
1518 if (list_empty(&sdev
->starved_entry
))
1519 list_add_tail(&sdev
->starved_entry
,
1520 &shost
->starved_list
);
1524 if (!scsi_target_queue_ready(shost
, sdev
))
1527 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1530 scsi_target(sdev
)->target_busy
++;
1534 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1535 * take the lock again.
1537 spin_unlock_irq(shost
->host_lock
);
1540 * Finally, initialize any error handling parameters, and set up
1541 * the timers for timeouts.
1543 scsi_init_cmd_errh(cmd
);
1546 * Dispatch the command to the low-level driver.
1548 rtn
= scsi_dispatch_cmd(cmd
);
1549 spin_lock_irq(q
->queue_lock
);
1551 /* we're refusing the command; because of
1552 * the way locks get dropped, we need to
1553 * check here if plugging is required */
1554 if(sdev
->device_busy
== 0)
1564 spin_unlock_irq(shost
->host_lock
);
1567 * lock q, handle tag, requeue req, and decrement device_busy. We
1568 * must return with queue_lock held.
1570 * Decrementing device_busy without checking it is OK, as all such
1571 * cases (host limits or settings) should run the queue at some
1574 spin_lock_irq(q
->queue_lock
);
1575 blk_requeue_request(q
, req
);
1576 sdev
->device_busy
--;
1577 if(sdev
->device_busy
== 0)
1580 /* must be careful here...if we trigger the ->remove() function
1581 * we cannot be holding the q lock */
1582 spin_unlock_irq(q
->queue_lock
);
1583 put_device(&sdev
->sdev_gendev
);
1584 spin_lock_irq(q
->queue_lock
);
1587 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1589 struct device
*host_dev
;
1590 u64 bounce_limit
= 0xffffffff;
1592 if (shost
->unchecked_isa_dma
)
1593 return BLK_BOUNCE_ISA
;
1595 * Platforms with virtual-DMA translation
1596 * hardware have no practical limit.
1598 if (!PCI_DMA_BUS_IS_PHYS
)
1599 return BLK_BOUNCE_ANY
;
1601 host_dev
= scsi_get_device(shost
);
1602 if (host_dev
&& host_dev
->dma_mask
)
1603 bounce_limit
= *host_dev
->dma_mask
;
1605 return bounce_limit
;
1607 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1609 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1610 request_fn_proc
*request_fn
)
1612 struct request_queue
*q
;
1613 struct device
*dev
= shost
->shost_gendev
.parent
;
1615 q
= blk_init_queue(request_fn
, NULL
);
1620 * this limit is imposed by hardware restrictions
1622 blk_queue_max_segments(q
, min_t(unsigned short, shost
->sg_tablesize
,
1623 SCSI_MAX_SG_CHAIN_SEGMENTS
));
1625 if (scsi_host_prot_dma(shost
)) {
1626 shost
->sg_prot_tablesize
=
1627 min_not_zero(shost
->sg_prot_tablesize
,
1628 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS
);
1629 BUG_ON(shost
->sg_prot_tablesize
< shost
->sg_tablesize
);
1630 blk_queue_max_integrity_segments(q
, shost
->sg_prot_tablesize
);
1633 blk_queue_max_hw_sectors(q
, shost
->max_sectors
);
1634 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1635 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1636 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1638 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
1640 /* New queue, no concurrency on queue_flags */
1641 if (!shost
->use_clustering
)
1642 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER
, q
);
1645 * set a reasonable default alignment on word boundaries: the
1646 * host and device may alter it using
1647 * blk_queue_update_dma_alignment() later.
1649 blk_queue_dma_alignment(q
, 0x03);
1653 EXPORT_SYMBOL(__scsi_alloc_queue
);
1655 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1657 struct request_queue
*q
;
1659 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1663 blk_queue_prep_rq(q
, scsi_prep_fn
);
1664 blk_queue_softirq_done(q
, scsi_softirq_done
);
1665 blk_queue_rq_timed_out(q
, scsi_times_out
);
1666 blk_queue_lld_busy(q
, scsi_lld_busy
);
1670 void scsi_free_queue(struct request_queue
*q
)
1672 blk_cleanup_queue(q
);
1676 * Function: scsi_block_requests()
1678 * Purpose: Utility function used by low-level drivers to prevent further
1679 * commands from being queued to the device.
1681 * Arguments: shost - Host in question
1685 * Lock status: No locks are assumed held.
1687 * Notes: There is no timer nor any other means by which the requests
1688 * get unblocked other than the low-level driver calling
1689 * scsi_unblock_requests().
1691 void scsi_block_requests(struct Scsi_Host
*shost
)
1693 shost
->host_self_blocked
= 1;
1695 EXPORT_SYMBOL(scsi_block_requests
);
1698 * Function: scsi_unblock_requests()
1700 * Purpose: Utility function used by low-level drivers to allow further
1701 * commands from being queued to the device.
1703 * Arguments: shost - Host in question
1707 * Lock status: No locks are assumed held.
1709 * Notes: There is no timer nor any other means by which the requests
1710 * get unblocked other than the low-level driver calling
1711 * scsi_unblock_requests().
1713 * This is done as an API function so that changes to the
1714 * internals of the scsi mid-layer won't require wholesale
1715 * changes to drivers that use this feature.
1717 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1719 shost
->host_self_blocked
= 0;
1720 scsi_run_host_queues(shost
);
1722 EXPORT_SYMBOL(scsi_unblock_requests
);
1724 int __init
scsi_init_queue(void)
1728 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
1729 sizeof(struct scsi_data_buffer
),
1731 if (!scsi_sdb_cache
) {
1732 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
1736 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1737 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1738 int size
= sgp
->size
* sizeof(struct scatterlist
);
1740 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1741 SLAB_HWCACHE_ALIGN
, NULL
);
1743 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1748 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1751 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1760 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1761 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1763 mempool_destroy(sgp
->pool
);
1765 kmem_cache_destroy(sgp
->slab
);
1767 kmem_cache_destroy(scsi_sdb_cache
);
1772 void scsi_exit_queue(void)
1776 kmem_cache_destroy(scsi_sdb_cache
);
1778 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1779 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1780 mempool_destroy(sgp
->pool
);
1781 kmem_cache_destroy(sgp
->slab
);
1786 * scsi_mode_select - issue a mode select
1787 * @sdev: SCSI device to be queried
1788 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1789 * @sp: Save page bit (0 == don't save, 1 == save)
1790 * @modepage: mode page being requested
1791 * @buffer: request buffer (may not be smaller than eight bytes)
1792 * @len: length of request buffer.
1793 * @timeout: command timeout
1794 * @retries: number of retries before failing
1795 * @data: returns a structure abstracting the mode header data
1796 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1797 * must be SCSI_SENSE_BUFFERSIZE big.
1799 * Returns zero if successful; negative error number or scsi
1804 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1805 unsigned char *buffer
, int len
, int timeout
, int retries
,
1806 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1808 unsigned char cmd
[10];
1809 unsigned char *real_buffer
;
1812 memset(cmd
, 0, sizeof(cmd
));
1813 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1815 if (sdev
->use_10_for_ms
) {
1818 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1821 memcpy(real_buffer
+ 8, buffer
, len
);
1825 real_buffer
[2] = data
->medium_type
;
1826 real_buffer
[3] = data
->device_specific
;
1827 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1829 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1830 real_buffer
[7] = data
->block_descriptor_length
;
1832 cmd
[0] = MODE_SELECT_10
;
1836 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1840 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1843 memcpy(real_buffer
+ 4, buffer
, len
);
1846 real_buffer
[1] = data
->medium_type
;
1847 real_buffer
[2] = data
->device_specific
;
1848 real_buffer
[3] = data
->block_descriptor_length
;
1851 cmd
[0] = MODE_SELECT
;
1855 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1856 sshdr
, timeout
, retries
, NULL
);
1860 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1863 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1864 * @sdev: SCSI device to be queried
1865 * @dbd: set if mode sense will allow block descriptors to be returned
1866 * @modepage: mode page being requested
1867 * @buffer: request buffer (may not be smaller than eight bytes)
1868 * @len: length of request buffer.
1869 * @timeout: command timeout
1870 * @retries: number of retries before failing
1871 * @data: returns a structure abstracting the mode header data
1872 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1873 * must be SCSI_SENSE_BUFFERSIZE big.
1875 * Returns zero if unsuccessful, or the header offset (either 4
1876 * or 8 depending on whether a six or ten byte command was
1877 * issued) if successful.
1880 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1881 unsigned char *buffer
, int len
, int timeout
, int retries
,
1882 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1884 unsigned char cmd
[12];
1888 struct scsi_sense_hdr my_sshdr
;
1890 memset(data
, 0, sizeof(*data
));
1891 memset(&cmd
[0], 0, 12);
1892 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1895 /* caller might not be interested in sense, but we need it */
1900 use_10_for_ms
= sdev
->use_10_for_ms
;
1902 if (use_10_for_ms
) {
1906 cmd
[0] = MODE_SENSE_10
;
1913 cmd
[0] = MODE_SENSE
;
1918 memset(buffer
, 0, len
);
1920 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1921 sshdr
, timeout
, retries
, NULL
);
1923 /* This code looks awful: what it's doing is making sure an
1924 * ILLEGAL REQUEST sense return identifies the actual command
1925 * byte as the problem. MODE_SENSE commands can return
1926 * ILLEGAL REQUEST if the code page isn't supported */
1928 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1929 (driver_byte(result
) & DRIVER_SENSE
)) {
1930 if (scsi_sense_valid(sshdr
)) {
1931 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1932 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1934 * Invalid command operation code
1936 sdev
->use_10_for_ms
= 0;
1942 if(scsi_status_is_good(result
)) {
1943 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1944 (modepage
== 6 || modepage
== 8))) {
1945 /* Initio breakage? */
1948 data
->medium_type
= 0;
1949 data
->device_specific
= 0;
1951 data
->block_descriptor_length
= 0;
1952 } else if(use_10_for_ms
) {
1953 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1954 data
->medium_type
= buffer
[2];
1955 data
->device_specific
= buffer
[3];
1956 data
->longlba
= buffer
[4] & 0x01;
1957 data
->block_descriptor_length
= buffer
[6]*256
1960 data
->length
= buffer
[0] + 1;
1961 data
->medium_type
= buffer
[1];
1962 data
->device_specific
= buffer
[2];
1963 data
->block_descriptor_length
= buffer
[3];
1965 data
->header_length
= header_length
;
1970 EXPORT_SYMBOL(scsi_mode_sense
);
1973 * scsi_test_unit_ready - test if unit is ready
1974 * @sdev: scsi device to change the state of.
1975 * @timeout: command timeout
1976 * @retries: number of retries before failing
1977 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1978 * returning sense. Make sure that this is cleared before passing
1981 * Returns zero if unsuccessful or an error if TUR failed. For
1982 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1983 * translated to success, with the ->changed flag updated.
1986 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
1987 struct scsi_sense_hdr
*sshdr_external
)
1990 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1992 struct scsi_sense_hdr
*sshdr
;
1995 if (!sshdr_external
)
1996 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1998 sshdr
= sshdr_external
;
2000 /* try to eat the UNIT_ATTENTION if there are enough retries */
2002 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
2003 timeout
, retries
, NULL
);
2004 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2005 sshdr
->sense_key
== UNIT_ATTENTION
)
2007 } while (scsi_sense_valid(sshdr
) &&
2008 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2011 /* could not allocate sense buffer, so can't process it */
2014 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2015 (sshdr
->sense_key
== UNIT_ATTENTION
||
2016 sshdr
->sense_key
== NOT_READY
)) {
2020 if (!sshdr_external
)
2024 EXPORT_SYMBOL(scsi_test_unit_ready
);
2027 * scsi_device_set_state - Take the given device through the device state model.
2028 * @sdev: scsi device to change the state of.
2029 * @state: state to change to.
2031 * Returns zero if unsuccessful or an error if the requested
2032 * transition is illegal.
2035 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2037 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2039 if (state
== oldstate
)
2045 case SDEV_CREATED_BLOCK
:
2089 case SDEV_CREATED_BLOCK
:
2096 case SDEV_CREATED_BLOCK
:
2131 sdev
->sdev_state
= state
;
2135 SCSI_LOG_ERROR_RECOVERY(1,
2136 sdev_printk(KERN_ERR
, sdev
,
2137 "Illegal state transition %s->%s\n",
2138 scsi_device_state_name(oldstate
),
2139 scsi_device_state_name(state
))
2143 EXPORT_SYMBOL(scsi_device_set_state
);
2146 * sdev_evt_emit - emit a single SCSI device uevent
2147 * @sdev: associated SCSI device
2148 * @evt: event to emit
2150 * Send a single uevent (scsi_event) to the associated scsi_device.
2152 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2157 switch (evt
->evt_type
) {
2158 case SDEV_EVT_MEDIA_CHANGE
:
2159 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2169 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2173 * sdev_evt_thread - send a uevent for each scsi event
2174 * @work: work struct for scsi_device
2176 * Dispatch queued events to their associated scsi_device kobjects
2179 void scsi_evt_thread(struct work_struct
*work
)
2181 struct scsi_device
*sdev
;
2182 LIST_HEAD(event_list
);
2184 sdev
= container_of(work
, struct scsi_device
, event_work
);
2187 struct scsi_event
*evt
;
2188 struct list_head
*this, *tmp
;
2189 unsigned long flags
;
2191 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2192 list_splice_init(&sdev
->event_list
, &event_list
);
2193 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2195 if (list_empty(&event_list
))
2198 list_for_each_safe(this, tmp
, &event_list
) {
2199 evt
= list_entry(this, struct scsi_event
, node
);
2200 list_del(&evt
->node
);
2201 scsi_evt_emit(sdev
, evt
);
2208 * sdev_evt_send - send asserted event to uevent thread
2209 * @sdev: scsi_device event occurred on
2210 * @evt: event to send
2212 * Assert scsi device event asynchronously.
2214 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2216 unsigned long flags
;
2219 /* FIXME: currently this check eliminates all media change events
2220 * for polled devices. Need to update to discriminate between AN
2221 * and polled events */
2222 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2228 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2229 list_add_tail(&evt
->node
, &sdev
->event_list
);
2230 schedule_work(&sdev
->event_work
);
2231 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2233 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2236 * sdev_evt_alloc - allocate a new scsi event
2237 * @evt_type: type of event to allocate
2238 * @gfpflags: GFP flags for allocation
2240 * Allocates and returns a new scsi_event.
2242 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2245 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2249 evt
->evt_type
= evt_type
;
2250 INIT_LIST_HEAD(&evt
->node
);
2252 /* evt_type-specific initialization, if any */
2254 case SDEV_EVT_MEDIA_CHANGE
:
2262 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2265 * sdev_evt_send_simple - send asserted event to uevent thread
2266 * @sdev: scsi_device event occurred on
2267 * @evt_type: type of event to send
2268 * @gfpflags: GFP flags for allocation
2270 * Assert scsi device event asynchronously, given an event type.
2272 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2273 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2275 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2277 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2282 sdev_evt_send(sdev
, evt
);
2284 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2287 * scsi_device_quiesce - Block user issued commands.
2288 * @sdev: scsi device to quiesce.
2290 * This works by trying to transition to the SDEV_QUIESCE state
2291 * (which must be a legal transition). When the device is in this
2292 * state, only special requests will be accepted, all others will
2293 * be deferred. Since special requests may also be requeued requests,
2294 * a successful return doesn't guarantee the device will be
2295 * totally quiescent.
2297 * Must be called with user context, may sleep.
2299 * Returns zero if unsuccessful or an error if not.
2302 scsi_device_quiesce(struct scsi_device
*sdev
)
2304 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2308 scsi_run_queue(sdev
->request_queue
);
2309 while (sdev
->device_busy
) {
2310 msleep_interruptible(200);
2311 scsi_run_queue(sdev
->request_queue
);
2315 EXPORT_SYMBOL(scsi_device_quiesce
);
2318 * scsi_device_resume - Restart user issued commands to a quiesced device.
2319 * @sdev: scsi device to resume.
2321 * Moves the device from quiesced back to running and restarts the
2324 * Must be called with user context, may sleep.
2327 scsi_device_resume(struct scsi_device
*sdev
)
2329 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2331 scsi_run_queue(sdev
->request_queue
);
2333 EXPORT_SYMBOL(scsi_device_resume
);
2336 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2338 scsi_device_quiesce(sdev
);
2342 scsi_target_quiesce(struct scsi_target
*starget
)
2344 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2346 EXPORT_SYMBOL(scsi_target_quiesce
);
2349 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2351 scsi_device_resume(sdev
);
2355 scsi_target_resume(struct scsi_target
*starget
)
2357 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2359 EXPORT_SYMBOL(scsi_target_resume
);
2362 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2363 * @sdev: device to block
2365 * Block request made by scsi lld's to temporarily stop all
2366 * scsi commands on the specified device. Called from interrupt
2367 * or normal process context.
2369 * Returns zero if successful or error if not
2372 * This routine transitions the device to the SDEV_BLOCK state
2373 * (which must be a legal transition). When the device is in this
2374 * state, all commands are deferred until the scsi lld reenables
2375 * the device with scsi_device_unblock or device_block_tmo fires.
2376 * This routine assumes the host_lock is held on entry.
2379 scsi_internal_device_block(struct scsi_device
*sdev
)
2381 struct request_queue
*q
= sdev
->request_queue
;
2382 unsigned long flags
;
2385 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2387 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2394 * The device has transitioned to SDEV_BLOCK. Stop the
2395 * block layer from calling the midlayer with this device's
2398 spin_lock_irqsave(q
->queue_lock
, flags
);
2400 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2404 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2407 * scsi_internal_device_unblock - resume a device after a block request
2408 * @sdev: device to resume
2410 * Called by scsi lld's or the midlayer to restart the device queue
2411 * for the previously suspended scsi device. Called from interrupt or
2412 * normal process context.
2414 * Returns zero if successful or error if not.
2417 * This routine transitions the device to the SDEV_RUNNING state
2418 * (which must be a legal transition) allowing the midlayer to
2419 * goose the queue for this device. This routine assumes the
2420 * host_lock is held upon entry.
2423 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2425 struct request_queue
*q
= sdev
->request_queue
;
2426 unsigned long flags
;
2429 * Try to transition the scsi device to SDEV_RUNNING
2430 * and goose the device queue if successful.
2432 if (sdev
->sdev_state
== SDEV_BLOCK
)
2433 sdev
->sdev_state
= SDEV_RUNNING
;
2434 else if (sdev
->sdev_state
== SDEV_CREATED_BLOCK
)
2435 sdev
->sdev_state
= SDEV_CREATED
;
2436 else if (sdev
->sdev_state
!= SDEV_CANCEL
&&
2437 sdev
->sdev_state
!= SDEV_OFFLINE
)
2440 spin_lock_irqsave(q
->queue_lock
, flags
);
2442 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2446 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2449 device_block(struct scsi_device
*sdev
, void *data
)
2451 scsi_internal_device_block(sdev
);
2455 target_block(struct device
*dev
, void *data
)
2457 if (scsi_is_target_device(dev
))
2458 starget_for_each_device(to_scsi_target(dev
), NULL
,
2464 scsi_target_block(struct device
*dev
)
2466 if (scsi_is_target_device(dev
))
2467 starget_for_each_device(to_scsi_target(dev
), NULL
,
2470 device_for_each_child(dev
, NULL
, target_block
);
2472 EXPORT_SYMBOL_GPL(scsi_target_block
);
2475 device_unblock(struct scsi_device
*sdev
, void *data
)
2477 scsi_internal_device_unblock(sdev
);
2481 target_unblock(struct device
*dev
, void *data
)
2483 if (scsi_is_target_device(dev
))
2484 starget_for_each_device(to_scsi_target(dev
), NULL
,
2490 scsi_target_unblock(struct device
*dev
)
2492 if (scsi_is_target_device(dev
))
2493 starget_for_each_device(to_scsi_target(dev
), NULL
,
2496 device_for_each_child(dev
, NULL
, target_unblock
);
2498 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2501 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2502 * @sgl: scatter-gather list
2503 * @sg_count: number of segments in sg
2504 * @offset: offset in bytes into sg, on return offset into the mapped area
2505 * @len: bytes to map, on return number of bytes mapped
2507 * Returns virtual address of the start of the mapped page
2509 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2510 size_t *offset
, size_t *len
)
2513 size_t sg_len
= 0, len_complete
= 0;
2514 struct scatterlist
*sg
;
2517 WARN_ON(!irqs_disabled());
2519 for_each_sg(sgl
, sg
, sg_count
, i
) {
2520 len_complete
= sg_len
; /* Complete sg-entries */
2521 sg_len
+= sg
->length
;
2522 if (sg_len
> *offset
)
2526 if (unlikely(i
== sg_count
)) {
2527 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2529 __func__
, sg_len
, *offset
, sg_count
);
2534 /* Offset starting from the beginning of first page in this sg-entry */
2535 *offset
= *offset
- len_complete
+ sg
->offset
;
2537 /* Assumption: contiguous pages can be accessed as "page + i" */
2538 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2539 *offset
&= ~PAGE_MASK
;
2541 /* Bytes in this sg-entry from *offset to the end of the page */
2542 sg_len
= PAGE_SIZE
- *offset
;
2546 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2548 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2551 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2552 * @virt: virtual address to be unmapped
2554 void scsi_kunmap_atomic_sg(void *virt
)
2556 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2558 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);