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 req
->cmd_flags
&= ~REQ_DONTPREP
;
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 (blk_pc_request(req
)) { /* 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 blk_end_request_all(req
, 0);
754 scsi_release_buffers(cmd
);
755 scsi_next_command(cmd
);
760 BUG_ON(blk_bidi_rq(req
)); /* bidi not support for !blk_pc_request yet */
763 * Next deal with any sectors which we were able to correctly
766 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
768 blk_rq_sectors(req
), good_bytes
));
771 * Recovered errors need reporting, but they're always treated
772 * as success, so fiddle the result code here. For BLOCK_PC
773 * we already took a copy of the original into rq->errors which
774 * is what gets returned to the user
776 if (sense_valid
&& sshdr
.sense_key
== RECOVERED_ERROR
) {
777 if (!(req
->cmd_flags
& REQ_QUIET
))
778 scsi_print_sense("", cmd
);
780 /* BLOCK_PC may have set error */
785 * A number of bytes were successfully read. If there
786 * are leftovers and there is some kind of error
787 * (result != 0), retry the rest.
789 if (scsi_end_request(cmd
, error
, good_bytes
, result
== 0) == NULL
)
794 if (host_byte(result
) == DID_RESET
) {
795 /* Third party bus reset or reset for error recovery
796 * reasons. Just retry the command and see what
799 action
= ACTION_RETRY
;
800 } else if (sense_valid
&& !sense_deferred
) {
801 switch (sshdr
.sense_key
) {
803 if (cmd
->device
->removable
) {
804 /* Detected disc change. Set a bit
805 * and quietly refuse further access.
807 cmd
->device
->changed
= 1;
808 description
= "Media Changed";
809 action
= ACTION_FAIL
;
811 /* Must have been a power glitch, or a
812 * bus reset. Could not have been a
813 * media change, so we just retry the
814 * command and see what happens.
816 action
= ACTION_RETRY
;
819 case ILLEGAL_REQUEST
:
820 /* If we had an ILLEGAL REQUEST returned, then
821 * we may have performed an unsupported
822 * command. The only thing this should be
823 * would be a ten byte read where only a six
824 * byte read was supported. Also, on a system
825 * where READ CAPACITY failed, we may have
826 * read past the end of the disk.
828 if ((cmd
->device
->use_10_for_rw
&&
829 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
830 (cmd
->cmnd
[0] == READ_10
||
831 cmd
->cmnd
[0] == WRITE_10
)) {
832 /* This will issue a new 6-byte command. */
833 cmd
->device
->use_10_for_rw
= 0;
834 action
= ACTION_REPREP
;
835 } else if (sshdr
.asc
== 0x10) /* DIX */ {
836 description
= "Host Data Integrity Failure";
837 action
= ACTION_FAIL
;
840 action
= ACTION_FAIL
;
842 case ABORTED_COMMAND
:
843 action
= ACTION_FAIL
;
844 if (sshdr
.asc
== 0x10) { /* DIF */
845 description
= "Target Data Integrity Failure";
850 /* If the device is in the process of becoming
851 * ready, or has a temporary blockage, retry.
853 if (sshdr
.asc
== 0x04) {
854 switch (sshdr
.ascq
) {
855 case 0x01: /* becoming ready */
856 case 0x04: /* format in progress */
857 case 0x05: /* rebuild in progress */
858 case 0x06: /* recalculation in progress */
859 case 0x07: /* operation in progress */
860 case 0x08: /* Long write in progress */
861 case 0x09: /* self test in progress */
862 action
= ACTION_DELAYED_RETRY
;
865 description
= "Device not ready";
866 action
= ACTION_FAIL
;
870 description
= "Device not ready";
871 action
= ACTION_FAIL
;
874 case VOLUME_OVERFLOW
:
875 /* See SSC3rXX or current. */
876 action
= ACTION_FAIL
;
879 description
= "Unhandled sense code";
880 action
= ACTION_FAIL
;
884 description
= "Unhandled error code";
885 action
= ACTION_FAIL
;
890 /* Give up and fail the remainder of the request */
891 scsi_release_buffers(cmd
);
892 if (!(req
->cmd_flags
& REQ_QUIET
)) {
894 scmd_printk(KERN_INFO
, cmd
, "%s\n",
896 scsi_print_result(cmd
);
897 if (driver_byte(result
) & DRIVER_SENSE
)
898 scsi_print_sense("", cmd
);
900 blk_end_request_all(req
, -EIO
);
901 scsi_next_command(cmd
);
904 /* Unprep the request and put it back at the head of the queue.
905 * A new command will be prepared and issued.
907 scsi_release_buffers(cmd
);
908 scsi_requeue_command(q
, cmd
);
911 /* Retry the same command immediately */
912 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
, 0);
914 case ACTION_DELAYED_RETRY
:
915 /* Retry the same command after a delay */
916 __scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
, 0);
921 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
927 * If sg table allocation fails, requeue request later.
929 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
931 return BLKPREP_DEFER
;
937 * Next, walk the list, and fill in the addresses and sizes of
940 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
941 BUG_ON(count
> sdb
->table
.nents
);
942 sdb
->table
.nents
= count
;
943 sdb
->length
= blk_rq_bytes(req
);
948 * Function: scsi_init_io()
950 * Purpose: SCSI I/O initialize function.
952 * Arguments: cmd - Command descriptor we wish to initialize
954 * Returns: 0 on success
955 * BLKPREP_DEFER if the failure is retryable
956 * BLKPREP_KILL if the failure is fatal
958 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
960 int error
= scsi_init_sgtable(cmd
->request
, &cmd
->sdb
, gfp_mask
);
964 if (blk_bidi_rq(cmd
->request
)) {
965 struct scsi_data_buffer
*bidi_sdb
= kmem_cache_zalloc(
966 scsi_sdb_cache
, GFP_ATOMIC
);
968 error
= BLKPREP_DEFER
;
972 cmd
->request
->next_rq
->special
= bidi_sdb
;
973 error
= scsi_init_sgtable(cmd
->request
->next_rq
, bidi_sdb
,
979 if (blk_integrity_rq(cmd
->request
)) {
980 struct scsi_data_buffer
*prot_sdb
= cmd
->prot_sdb
;
983 BUG_ON(prot_sdb
== NULL
);
984 ivecs
= blk_rq_count_integrity_sg(cmd
->request
);
986 if (scsi_alloc_sgtable(prot_sdb
, ivecs
, gfp_mask
)) {
987 error
= BLKPREP_DEFER
;
991 count
= blk_rq_map_integrity_sg(cmd
->request
,
992 prot_sdb
->table
.sgl
);
993 BUG_ON(unlikely(count
> ivecs
));
995 cmd
->prot_sdb
= prot_sdb
;
996 cmd
->prot_sdb
->table
.nents
= count
;
1002 scsi_release_buffers(cmd
);
1003 if (error
== BLKPREP_KILL
)
1004 scsi_put_command(cmd
);
1005 else /* BLKPREP_DEFER */
1006 scsi_unprep_request(cmd
->request
);
1010 EXPORT_SYMBOL(scsi_init_io
);
1012 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1013 struct request
*req
)
1015 struct scsi_cmnd
*cmd
;
1017 if (!req
->special
) {
1018 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1026 /* pull a tag out of the request if we have one */
1027 cmd
->tag
= req
->tag
;
1030 cmd
->cmnd
= req
->cmd
;
1035 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1037 struct scsi_cmnd
*cmd
;
1038 int ret
= scsi_prep_state_check(sdev
, req
);
1040 if (ret
!= BLKPREP_OK
)
1043 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1045 return BLKPREP_DEFER
;
1048 * BLOCK_PC requests may transfer data, in which case they must
1049 * a bio attached to them. Or they might contain a SCSI command
1050 * that does not transfer data, in which case they may optionally
1051 * submit a request without an attached bio.
1056 BUG_ON(!req
->nr_phys_segments
);
1058 ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1062 BUG_ON(blk_rq_bytes(req
));
1064 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1068 cmd
->cmd_len
= req
->cmd_len
;
1069 if (!blk_rq_bytes(req
))
1070 cmd
->sc_data_direction
= DMA_NONE
;
1071 else if (rq_data_dir(req
) == WRITE
)
1072 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1074 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1076 cmd
->transfersize
= blk_rq_bytes(req
);
1077 cmd
->allowed
= req
->retries
;
1080 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1083 * Setup a REQ_TYPE_FS command. These are simple read/write request
1084 * from filesystems that still need to be translated to SCSI CDBs from
1087 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1089 struct scsi_cmnd
*cmd
;
1090 int ret
= scsi_prep_state_check(sdev
, req
);
1092 if (ret
!= BLKPREP_OK
)
1095 if (unlikely(sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
1096 && sdev
->scsi_dh_data
->scsi_dh
->prep_fn
)) {
1097 ret
= sdev
->scsi_dh_data
->scsi_dh
->prep_fn(sdev
, req
);
1098 if (ret
!= BLKPREP_OK
)
1103 * Filesystem requests must transfer data.
1105 BUG_ON(!req
->nr_phys_segments
);
1107 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1109 return BLKPREP_DEFER
;
1111 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1112 return scsi_init_io(cmd
, GFP_ATOMIC
);
1114 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1116 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1118 int ret
= BLKPREP_OK
;
1121 * If the device is not in running state we will reject some
1124 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1125 switch (sdev
->sdev_state
) {
1128 * If the device is offline we refuse to process any
1129 * commands. The device must be brought online
1130 * before trying any recovery commands.
1132 sdev_printk(KERN_ERR
, sdev
,
1133 "rejecting I/O to offline device\n");
1138 * If the device is fully deleted, we refuse to
1139 * process any commands as well.
1141 sdev_printk(KERN_ERR
, sdev
,
1142 "rejecting I/O to dead device\n");
1147 case SDEV_CREATED_BLOCK
:
1149 * If the devices is blocked we defer normal commands.
1151 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1152 ret
= BLKPREP_DEFER
;
1156 * For any other not fully online state we only allow
1157 * special commands. In particular any user initiated
1158 * command is not allowed.
1160 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1167 EXPORT_SYMBOL(scsi_prep_state_check
);
1169 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1171 struct scsi_device
*sdev
= q
->queuedata
;
1175 req
->errors
= DID_NO_CONNECT
<< 16;
1176 /* release the command and kill it */
1178 struct scsi_cmnd
*cmd
= req
->special
;
1179 scsi_release_buffers(cmd
);
1180 scsi_put_command(cmd
);
1181 req
->special
= NULL
;
1186 * If we defer, the blk_peek_request() returns NULL, but the
1187 * queue must be restarted, so we plug here if no returning
1188 * command will automatically do that.
1190 if (sdev
->device_busy
== 0)
1194 req
->cmd_flags
|= REQ_DONTPREP
;
1199 EXPORT_SYMBOL(scsi_prep_return
);
1201 int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1203 struct scsi_device
*sdev
= q
->queuedata
;
1204 int ret
= BLKPREP_KILL
;
1206 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1207 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1208 return scsi_prep_return(q
, req
, ret
);
1212 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1215 * Called with the queue_lock held.
1217 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1218 struct scsi_device
*sdev
)
1220 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1222 * unblock after device_blocked iterates to zero
1224 if (--sdev
->device_blocked
== 0) {
1226 sdev_printk(KERN_INFO
, sdev
,
1227 "unblocking device at zero depth\n"));
1233 if (scsi_device_is_busy(sdev
))
1241 * scsi_target_queue_ready: checks if there we can send commands to target
1242 * @sdev: scsi device on starget to check.
1244 * Called with the host lock held.
1246 static inline int scsi_target_queue_ready(struct Scsi_Host
*shost
,
1247 struct scsi_device
*sdev
)
1249 struct scsi_target
*starget
= scsi_target(sdev
);
1251 if (starget
->single_lun
) {
1252 if (starget
->starget_sdev_user
&&
1253 starget
->starget_sdev_user
!= sdev
)
1255 starget
->starget_sdev_user
= sdev
;
1258 if (starget
->target_busy
== 0 && starget
->target_blocked
) {
1260 * unblock after target_blocked iterates to zero
1262 if (--starget
->target_blocked
== 0) {
1263 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO
, starget
,
1264 "unblocking target at zero depth\n"));
1266 blk_plug_device(sdev
->request_queue
);
1271 if (scsi_target_is_busy(starget
)) {
1272 if (list_empty(&sdev
->starved_entry
)) {
1273 list_add_tail(&sdev
->starved_entry
,
1274 &shost
->starved_list
);
1279 /* We're OK to process the command, so we can't be starved */
1280 if (!list_empty(&sdev
->starved_entry
))
1281 list_del_init(&sdev
->starved_entry
);
1286 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1287 * return 0. We must end up running the queue again whenever 0 is
1288 * returned, else IO can hang.
1290 * Called with host_lock held.
1292 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1293 struct Scsi_Host
*shost
,
1294 struct scsi_device
*sdev
)
1296 if (scsi_host_in_recovery(shost
))
1298 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1300 * unblock after host_blocked iterates to zero
1302 if (--shost
->host_blocked
== 0) {
1304 printk("scsi%d unblocking host at zero depth\n",
1310 if (scsi_host_is_busy(shost
)) {
1311 if (list_empty(&sdev
->starved_entry
))
1312 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1316 /* We're OK to process the command, so we can't be starved */
1317 if (!list_empty(&sdev
->starved_entry
))
1318 list_del_init(&sdev
->starved_entry
);
1324 * Busy state exporting function for request stacking drivers.
1326 * For efficiency, no lock is taken to check the busy state of
1327 * shost/starget/sdev, since the returned value is not guaranteed and
1328 * may be changed after request stacking drivers call the function,
1329 * regardless of taking lock or not.
1331 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1332 * (e.g. !sdev), scsi needs to return 'not busy'.
1333 * Otherwise, request stacking drivers may hold requests forever.
1335 static int scsi_lld_busy(struct request_queue
*q
)
1337 struct scsi_device
*sdev
= q
->queuedata
;
1338 struct Scsi_Host
*shost
;
1339 struct scsi_target
*starget
;
1345 starget
= scsi_target(sdev
);
1347 if (scsi_host_in_recovery(shost
) || scsi_host_is_busy(shost
) ||
1348 scsi_target_is_busy(starget
) || scsi_device_is_busy(sdev
))
1355 * Kill a request for a dead device
1357 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1359 struct scsi_cmnd
*cmd
= req
->special
;
1360 struct scsi_device
*sdev
= cmd
->device
;
1361 struct scsi_target
*starget
= scsi_target(sdev
);
1362 struct Scsi_Host
*shost
= sdev
->host
;
1364 blk_start_request(req
);
1366 if (unlikely(cmd
== NULL
)) {
1367 printk(KERN_CRIT
"impossible request in %s.\n",
1372 scsi_init_cmd_errh(cmd
);
1373 cmd
->result
= DID_NO_CONNECT
<< 16;
1374 atomic_inc(&cmd
->device
->iorequest_cnt
);
1377 * SCSI request completion path will do scsi_device_unbusy(),
1378 * bump busy counts. To bump the counters, we need to dance
1379 * with the locks as normal issue path does.
1381 sdev
->device_busy
++;
1382 spin_unlock(sdev
->request_queue
->queue_lock
);
1383 spin_lock(shost
->host_lock
);
1385 starget
->target_busy
++;
1386 spin_unlock(shost
->host_lock
);
1387 spin_lock(sdev
->request_queue
->queue_lock
);
1389 blk_complete_request(req
);
1392 static void scsi_softirq_done(struct request
*rq
)
1394 struct scsi_cmnd
*cmd
= rq
->special
;
1395 unsigned long wait_for
= (cmd
->allowed
+ 1) * rq
->timeout
;
1398 INIT_LIST_HEAD(&cmd
->eh_entry
);
1401 * Set the serial numbers back to zero
1403 cmd
->serial_number
= 0;
1405 atomic_inc(&cmd
->device
->iodone_cnt
);
1407 atomic_inc(&cmd
->device
->ioerr_cnt
);
1409 disposition
= scsi_decide_disposition(cmd
);
1410 if (disposition
!= SUCCESS
&&
1411 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1412 sdev_printk(KERN_ERR
, cmd
->device
,
1413 "timing out command, waited %lus\n",
1415 disposition
= SUCCESS
;
1418 scsi_log_completion(cmd
, disposition
);
1420 switch (disposition
) {
1422 scsi_finish_command(cmd
);
1425 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1427 case ADD_TO_MLQUEUE
:
1428 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1431 if (!scsi_eh_scmd_add(cmd
, 0))
1432 scsi_finish_command(cmd
);
1437 * Function: scsi_request_fn()
1439 * Purpose: Main strategy routine for SCSI.
1441 * Arguments: q - Pointer to actual queue.
1445 * Lock status: IO request lock assumed to be held when called.
1447 static void scsi_request_fn(struct request_queue
*q
)
1449 struct scsi_device
*sdev
= q
->queuedata
;
1450 struct Scsi_Host
*shost
;
1451 struct scsi_cmnd
*cmd
;
1452 struct request
*req
;
1455 printk("scsi: killing requests for dead queue\n");
1456 while ((req
= blk_peek_request(q
)) != NULL
)
1457 scsi_kill_request(req
, q
);
1461 if(!get_device(&sdev
->sdev_gendev
))
1462 /* We must be tearing the block queue down already */
1466 * To start with, we keep looping until the queue is empty, or until
1467 * the host is no longer able to accept any more requests.
1470 while (!blk_queue_plugged(q
)) {
1473 * get next queueable request. We do this early to make sure
1474 * that the request is fully prepared even if we cannot
1477 req
= blk_peek_request(q
);
1478 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1481 if (unlikely(!scsi_device_online(sdev
))) {
1482 sdev_printk(KERN_ERR
, sdev
,
1483 "rejecting I/O to offline device\n");
1484 scsi_kill_request(req
, q
);
1490 * Remove the request from the request list.
1492 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1493 blk_start_request(req
);
1494 sdev
->device_busy
++;
1496 spin_unlock(q
->queue_lock
);
1498 if (unlikely(cmd
== NULL
)) {
1499 printk(KERN_CRIT
"impossible request in %s.\n"
1500 "please mail a stack trace to "
1501 "linux-scsi@vger.kernel.org\n",
1503 blk_dump_rq_flags(req
, "foo");
1506 spin_lock(shost
->host_lock
);
1509 * We hit this when the driver is using a host wide
1510 * tag map. For device level tag maps the queue_depth check
1511 * in the device ready fn would prevent us from trying
1512 * to allocate a tag. Since the map is a shared host resource
1513 * we add the dev to the starved list so it eventually gets
1514 * a run when a tag is freed.
1516 if (blk_queue_tagged(q
) && !blk_rq_tagged(req
)) {
1517 if (list_empty(&sdev
->starved_entry
))
1518 list_add_tail(&sdev
->starved_entry
,
1519 &shost
->starved_list
);
1523 if (!scsi_target_queue_ready(shost
, sdev
))
1526 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1529 scsi_target(sdev
)->target_busy
++;
1533 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1534 * take the lock again.
1536 spin_unlock_irq(shost
->host_lock
);
1539 * Finally, initialize any error handling parameters, and set up
1540 * the timers for timeouts.
1542 scsi_init_cmd_errh(cmd
);
1545 * Dispatch the command to the low-level driver.
1547 rtn
= scsi_dispatch_cmd(cmd
);
1548 spin_lock_irq(q
->queue_lock
);
1550 /* we're refusing the command; because of
1551 * the way locks get dropped, we need to
1552 * check here if plugging is required */
1553 if(sdev
->device_busy
== 0)
1563 spin_unlock_irq(shost
->host_lock
);
1566 * lock q, handle tag, requeue req, and decrement device_busy. We
1567 * must return with queue_lock held.
1569 * Decrementing device_busy without checking it is OK, as all such
1570 * cases (host limits or settings) should run the queue at some
1573 spin_lock_irq(q
->queue_lock
);
1574 blk_requeue_request(q
, req
);
1575 sdev
->device_busy
--;
1576 if(sdev
->device_busy
== 0)
1579 /* must be careful here...if we trigger the ->remove() function
1580 * we cannot be holding the q lock */
1581 spin_unlock_irq(q
->queue_lock
);
1582 put_device(&sdev
->sdev_gendev
);
1583 spin_lock_irq(q
->queue_lock
);
1586 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1588 struct device
*host_dev
;
1589 u64 bounce_limit
= 0xffffffff;
1591 if (shost
->unchecked_isa_dma
)
1592 return BLK_BOUNCE_ISA
;
1594 * Platforms with virtual-DMA translation
1595 * hardware have no practical limit.
1597 if (!PCI_DMA_BUS_IS_PHYS
)
1598 return BLK_BOUNCE_ANY
;
1600 host_dev
= scsi_get_device(shost
);
1601 if (host_dev
&& host_dev
->dma_mask
)
1602 bounce_limit
= *host_dev
->dma_mask
;
1604 return bounce_limit
;
1606 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1608 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1609 request_fn_proc
*request_fn
)
1611 struct request_queue
*q
;
1612 struct device
*dev
= shost
->shost_gendev
.parent
;
1614 q
= blk_init_queue(request_fn
, NULL
);
1619 * this limit is imposed by hardware restrictions
1621 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1622 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_CHAIN_SEGMENTS
);
1624 blk_queue_max_sectors(q
, shost
->max_sectors
);
1625 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1626 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1627 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1629 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
1631 /* New queue, no concurrency on queue_flags */
1632 if (!shost
->use_clustering
)
1633 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER
, q
);
1636 * set a reasonable default alignment on word boundaries: the
1637 * host and device may alter it using
1638 * blk_queue_update_dma_alignment() later.
1640 blk_queue_dma_alignment(q
, 0x03);
1644 EXPORT_SYMBOL(__scsi_alloc_queue
);
1646 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1648 struct request_queue
*q
;
1650 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1654 blk_queue_prep_rq(q
, scsi_prep_fn
);
1655 blk_queue_softirq_done(q
, scsi_softirq_done
);
1656 blk_queue_rq_timed_out(q
, scsi_times_out
);
1657 blk_queue_lld_busy(q
, scsi_lld_busy
);
1661 void scsi_free_queue(struct request_queue
*q
)
1663 blk_cleanup_queue(q
);
1667 * Function: scsi_block_requests()
1669 * Purpose: Utility function used by low-level drivers to prevent further
1670 * commands from being queued to the device.
1672 * Arguments: shost - Host in question
1676 * Lock status: No locks are assumed held.
1678 * Notes: There is no timer nor any other means by which the requests
1679 * get unblocked other than the low-level driver calling
1680 * scsi_unblock_requests().
1682 void scsi_block_requests(struct Scsi_Host
*shost
)
1684 shost
->host_self_blocked
= 1;
1686 EXPORT_SYMBOL(scsi_block_requests
);
1689 * Function: scsi_unblock_requests()
1691 * Purpose: Utility function used by low-level drivers to allow further
1692 * commands from being queued to the device.
1694 * Arguments: shost - Host in question
1698 * Lock status: No locks are assumed held.
1700 * Notes: There is no timer nor any other means by which the requests
1701 * get unblocked other than the low-level driver calling
1702 * scsi_unblock_requests().
1704 * This is done as an API function so that changes to the
1705 * internals of the scsi mid-layer won't require wholesale
1706 * changes to drivers that use this feature.
1708 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1710 shost
->host_self_blocked
= 0;
1711 scsi_run_host_queues(shost
);
1713 EXPORT_SYMBOL(scsi_unblock_requests
);
1715 int __init
scsi_init_queue(void)
1719 scsi_sdb_cache
= kmem_cache_create("scsi_data_buffer",
1720 sizeof(struct scsi_data_buffer
),
1722 if (!scsi_sdb_cache
) {
1723 printk(KERN_ERR
"SCSI: can't init scsi sdb cache\n");
1727 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1728 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1729 int size
= sgp
->size
* sizeof(struct scatterlist
);
1731 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1732 SLAB_HWCACHE_ALIGN
, NULL
);
1734 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1739 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1742 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1751 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1752 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1754 mempool_destroy(sgp
->pool
);
1756 kmem_cache_destroy(sgp
->slab
);
1758 kmem_cache_destroy(scsi_sdb_cache
);
1763 void scsi_exit_queue(void)
1767 kmem_cache_destroy(scsi_sdb_cache
);
1769 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1770 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1771 mempool_destroy(sgp
->pool
);
1772 kmem_cache_destroy(sgp
->slab
);
1777 * scsi_mode_select - issue a mode select
1778 * @sdev: SCSI device to be queried
1779 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1780 * @sp: Save page bit (0 == don't save, 1 == save)
1781 * @modepage: mode page being requested
1782 * @buffer: request buffer (may not be smaller than eight bytes)
1783 * @len: length of request buffer.
1784 * @timeout: command timeout
1785 * @retries: number of retries before failing
1786 * @data: returns a structure abstracting the mode header data
1787 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1788 * must be SCSI_SENSE_BUFFERSIZE big.
1790 * Returns zero if successful; negative error number or scsi
1795 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1796 unsigned char *buffer
, int len
, int timeout
, int retries
,
1797 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1799 unsigned char cmd
[10];
1800 unsigned char *real_buffer
;
1803 memset(cmd
, 0, sizeof(cmd
));
1804 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1806 if (sdev
->use_10_for_ms
) {
1809 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1812 memcpy(real_buffer
+ 8, buffer
, len
);
1816 real_buffer
[2] = data
->medium_type
;
1817 real_buffer
[3] = data
->device_specific
;
1818 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1820 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1821 real_buffer
[7] = data
->block_descriptor_length
;
1823 cmd
[0] = MODE_SELECT_10
;
1827 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1831 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1834 memcpy(real_buffer
+ 4, buffer
, len
);
1837 real_buffer
[1] = data
->medium_type
;
1838 real_buffer
[2] = data
->device_specific
;
1839 real_buffer
[3] = data
->block_descriptor_length
;
1842 cmd
[0] = MODE_SELECT
;
1846 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1847 sshdr
, timeout
, retries
, NULL
);
1851 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1854 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1855 * @sdev: SCSI device to be queried
1856 * @dbd: set if mode sense will allow block descriptors to be returned
1857 * @modepage: mode page being requested
1858 * @buffer: request buffer (may not be smaller than eight bytes)
1859 * @len: length of request buffer.
1860 * @timeout: command timeout
1861 * @retries: number of retries before failing
1862 * @data: returns a structure abstracting the mode header data
1863 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1864 * must be SCSI_SENSE_BUFFERSIZE big.
1866 * Returns zero if unsuccessful, or the header offset (either 4
1867 * or 8 depending on whether a six or ten byte command was
1868 * issued) if successful.
1871 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1872 unsigned char *buffer
, int len
, int timeout
, int retries
,
1873 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1875 unsigned char cmd
[12];
1879 struct scsi_sense_hdr my_sshdr
;
1881 memset(data
, 0, sizeof(*data
));
1882 memset(&cmd
[0], 0, 12);
1883 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1886 /* caller might not be interested in sense, but we need it */
1891 use_10_for_ms
= sdev
->use_10_for_ms
;
1893 if (use_10_for_ms
) {
1897 cmd
[0] = MODE_SENSE_10
;
1904 cmd
[0] = MODE_SENSE
;
1909 memset(buffer
, 0, len
);
1911 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1912 sshdr
, timeout
, retries
, NULL
);
1914 /* This code looks awful: what it's doing is making sure an
1915 * ILLEGAL REQUEST sense return identifies the actual command
1916 * byte as the problem. MODE_SENSE commands can return
1917 * ILLEGAL REQUEST if the code page isn't supported */
1919 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1920 (driver_byte(result
) & DRIVER_SENSE
)) {
1921 if (scsi_sense_valid(sshdr
)) {
1922 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1923 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1925 * Invalid command operation code
1927 sdev
->use_10_for_ms
= 0;
1933 if(scsi_status_is_good(result
)) {
1934 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1935 (modepage
== 6 || modepage
== 8))) {
1936 /* Initio breakage? */
1939 data
->medium_type
= 0;
1940 data
->device_specific
= 0;
1942 data
->block_descriptor_length
= 0;
1943 } else if(use_10_for_ms
) {
1944 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1945 data
->medium_type
= buffer
[2];
1946 data
->device_specific
= buffer
[3];
1947 data
->longlba
= buffer
[4] & 0x01;
1948 data
->block_descriptor_length
= buffer
[6]*256
1951 data
->length
= buffer
[0] + 1;
1952 data
->medium_type
= buffer
[1];
1953 data
->device_specific
= buffer
[2];
1954 data
->block_descriptor_length
= buffer
[3];
1956 data
->header_length
= header_length
;
1961 EXPORT_SYMBOL(scsi_mode_sense
);
1964 * scsi_test_unit_ready - test if unit is ready
1965 * @sdev: scsi device to change the state of.
1966 * @timeout: command timeout
1967 * @retries: number of retries before failing
1968 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1969 * returning sense. Make sure that this is cleared before passing
1972 * Returns zero if unsuccessful or an error if TUR failed. For
1973 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1974 * translated to success, with the ->changed flag updated.
1977 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
1978 struct scsi_sense_hdr
*sshdr_external
)
1981 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1983 struct scsi_sense_hdr
*sshdr
;
1986 if (!sshdr_external
)
1987 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1989 sshdr
= sshdr_external
;
1991 /* try to eat the UNIT_ATTENTION if there are enough retries */
1993 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
1994 timeout
, retries
, NULL
);
1995 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
1996 sshdr
->sense_key
== UNIT_ATTENTION
)
1998 } while (scsi_sense_valid(sshdr
) &&
1999 sshdr
->sense_key
== UNIT_ATTENTION
&& --retries
);
2002 /* could not allocate sense buffer, so can't process it */
2005 if (sdev
->removable
&& scsi_sense_valid(sshdr
) &&
2006 (sshdr
->sense_key
== UNIT_ATTENTION
||
2007 sshdr
->sense_key
== NOT_READY
)) {
2011 if (!sshdr_external
)
2015 EXPORT_SYMBOL(scsi_test_unit_ready
);
2018 * scsi_device_set_state - Take the given device through the device state model.
2019 * @sdev: scsi device to change the state of.
2020 * @state: state to change to.
2022 * Returns zero if unsuccessful or an error if the requested
2023 * transition is illegal.
2026 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2028 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2030 if (state
== oldstate
)
2036 case SDEV_CREATED_BLOCK
:
2080 case SDEV_CREATED_BLOCK
:
2087 case SDEV_CREATED_BLOCK
:
2122 sdev
->sdev_state
= state
;
2126 SCSI_LOG_ERROR_RECOVERY(1,
2127 sdev_printk(KERN_ERR
, sdev
,
2128 "Illegal state transition %s->%s\n",
2129 scsi_device_state_name(oldstate
),
2130 scsi_device_state_name(state
))
2134 EXPORT_SYMBOL(scsi_device_set_state
);
2137 * sdev_evt_emit - emit a single SCSI device uevent
2138 * @sdev: associated SCSI device
2139 * @evt: event to emit
2141 * Send a single uevent (scsi_event) to the associated scsi_device.
2143 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2148 switch (evt
->evt_type
) {
2149 case SDEV_EVT_MEDIA_CHANGE
:
2150 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2160 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2164 * sdev_evt_thread - send a uevent for each scsi event
2165 * @work: work struct for scsi_device
2167 * Dispatch queued events to their associated scsi_device kobjects
2170 void scsi_evt_thread(struct work_struct
*work
)
2172 struct scsi_device
*sdev
;
2173 LIST_HEAD(event_list
);
2175 sdev
= container_of(work
, struct scsi_device
, event_work
);
2178 struct scsi_event
*evt
;
2179 struct list_head
*this, *tmp
;
2180 unsigned long flags
;
2182 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2183 list_splice_init(&sdev
->event_list
, &event_list
);
2184 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2186 if (list_empty(&event_list
))
2189 list_for_each_safe(this, tmp
, &event_list
) {
2190 evt
= list_entry(this, struct scsi_event
, node
);
2191 list_del(&evt
->node
);
2192 scsi_evt_emit(sdev
, evt
);
2199 * sdev_evt_send - send asserted event to uevent thread
2200 * @sdev: scsi_device event occurred on
2201 * @evt: event to send
2203 * Assert scsi device event asynchronously.
2205 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2207 unsigned long flags
;
2210 /* FIXME: currently this check eliminates all media change events
2211 * for polled devices. Need to update to discriminate between AN
2212 * and polled events */
2213 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2219 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2220 list_add_tail(&evt
->node
, &sdev
->event_list
);
2221 schedule_work(&sdev
->event_work
);
2222 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2224 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2227 * sdev_evt_alloc - allocate a new scsi event
2228 * @evt_type: type of event to allocate
2229 * @gfpflags: GFP flags for allocation
2231 * Allocates and returns a new scsi_event.
2233 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2236 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2240 evt
->evt_type
= evt_type
;
2241 INIT_LIST_HEAD(&evt
->node
);
2243 /* evt_type-specific initialization, if any */
2245 case SDEV_EVT_MEDIA_CHANGE
:
2253 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2256 * sdev_evt_send_simple - send asserted event to uevent thread
2257 * @sdev: scsi_device event occurred on
2258 * @evt_type: type of event to send
2259 * @gfpflags: GFP flags for allocation
2261 * Assert scsi device event asynchronously, given an event type.
2263 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2264 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2266 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2268 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2273 sdev_evt_send(sdev
, evt
);
2275 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2278 * scsi_device_quiesce - Block user issued commands.
2279 * @sdev: scsi device to quiesce.
2281 * This works by trying to transition to the SDEV_QUIESCE state
2282 * (which must be a legal transition). When the device is in this
2283 * state, only special requests will be accepted, all others will
2284 * be deferred. Since special requests may also be requeued requests,
2285 * a successful return doesn't guarantee the device will be
2286 * totally quiescent.
2288 * Must be called with user context, may sleep.
2290 * Returns zero if unsuccessful or an error if not.
2293 scsi_device_quiesce(struct scsi_device
*sdev
)
2295 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2299 scsi_run_queue(sdev
->request_queue
);
2300 while (sdev
->device_busy
) {
2301 msleep_interruptible(200);
2302 scsi_run_queue(sdev
->request_queue
);
2306 EXPORT_SYMBOL(scsi_device_quiesce
);
2309 * scsi_device_resume - Restart user issued commands to a quiesced device.
2310 * @sdev: scsi device to resume.
2312 * Moves the device from quiesced back to running and restarts the
2315 * Must be called with user context, may sleep.
2318 scsi_device_resume(struct scsi_device
*sdev
)
2320 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2322 scsi_run_queue(sdev
->request_queue
);
2324 EXPORT_SYMBOL(scsi_device_resume
);
2327 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2329 scsi_device_quiesce(sdev
);
2333 scsi_target_quiesce(struct scsi_target
*starget
)
2335 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2337 EXPORT_SYMBOL(scsi_target_quiesce
);
2340 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2342 scsi_device_resume(sdev
);
2346 scsi_target_resume(struct scsi_target
*starget
)
2348 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2350 EXPORT_SYMBOL(scsi_target_resume
);
2353 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2354 * @sdev: device to block
2356 * Block request made by scsi lld's to temporarily stop all
2357 * scsi commands on the specified device. Called from interrupt
2358 * or normal process context.
2360 * Returns zero if successful or error if not
2363 * This routine transitions the device to the SDEV_BLOCK state
2364 * (which must be a legal transition). When the device is in this
2365 * state, all commands are deferred until the scsi lld reenables
2366 * the device with scsi_device_unblock or device_block_tmo fires.
2367 * This routine assumes the host_lock is held on entry.
2370 scsi_internal_device_block(struct scsi_device
*sdev
)
2372 struct request_queue
*q
= sdev
->request_queue
;
2373 unsigned long flags
;
2376 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2378 err
= scsi_device_set_state(sdev
, SDEV_CREATED_BLOCK
);
2385 * The device has transitioned to SDEV_BLOCK. Stop the
2386 * block layer from calling the midlayer with this device's
2389 spin_lock_irqsave(q
->queue_lock
, flags
);
2391 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2395 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2398 * scsi_internal_device_unblock - resume a device after a block request
2399 * @sdev: device to resume
2401 * Called by scsi lld's or the midlayer to restart the device queue
2402 * for the previously suspended scsi device. Called from interrupt or
2403 * normal process context.
2405 * Returns zero if successful or error if not.
2408 * This routine transitions the device to the SDEV_RUNNING state
2409 * (which must be a legal transition) allowing the midlayer to
2410 * goose the queue for this device. This routine assumes the
2411 * host_lock is held upon entry.
2414 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2416 struct request_queue
*q
= sdev
->request_queue
;
2418 unsigned long flags
;
2421 * Try to transition the scsi device to SDEV_RUNNING
2422 * and goose the device queue if successful.
2424 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2426 err
= scsi_device_set_state(sdev
, SDEV_CREATED
);
2432 spin_lock_irqsave(q
->queue_lock
, flags
);
2434 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2438 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2441 device_block(struct scsi_device
*sdev
, void *data
)
2443 scsi_internal_device_block(sdev
);
2447 target_block(struct device
*dev
, void *data
)
2449 if (scsi_is_target_device(dev
))
2450 starget_for_each_device(to_scsi_target(dev
), NULL
,
2456 scsi_target_block(struct device
*dev
)
2458 if (scsi_is_target_device(dev
))
2459 starget_for_each_device(to_scsi_target(dev
), NULL
,
2462 device_for_each_child(dev
, NULL
, target_block
);
2464 EXPORT_SYMBOL_GPL(scsi_target_block
);
2467 device_unblock(struct scsi_device
*sdev
, void *data
)
2469 scsi_internal_device_unblock(sdev
);
2473 target_unblock(struct device
*dev
, void *data
)
2475 if (scsi_is_target_device(dev
))
2476 starget_for_each_device(to_scsi_target(dev
), NULL
,
2482 scsi_target_unblock(struct device
*dev
)
2484 if (scsi_is_target_device(dev
))
2485 starget_for_each_device(to_scsi_target(dev
), NULL
,
2488 device_for_each_child(dev
, NULL
, target_unblock
);
2490 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2493 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2494 * @sgl: scatter-gather list
2495 * @sg_count: number of segments in sg
2496 * @offset: offset in bytes into sg, on return offset into the mapped area
2497 * @len: bytes to map, on return number of bytes mapped
2499 * Returns virtual address of the start of the mapped page
2501 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2502 size_t *offset
, size_t *len
)
2505 size_t sg_len
= 0, len_complete
= 0;
2506 struct scatterlist
*sg
;
2509 WARN_ON(!irqs_disabled());
2511 for_each_sg(sgl
, sg
, sg_count
, i
) {
2512 len_complete
= sg_len
; /* Complete sg-entries */
2513 sg_len
+= sg
->length
;
2514 if (sg_len
> *offset
)
2518 if (unlikely(i
== sg_count
)) {
2519 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2521 __func__
, sg_len
, *offset
, sg_count
);
2526 /* Offset starting from the beginning of first page in this sg-entry */
2527 *offset
= *offset
- len_complete
+ sg
->offset
;
2529 /* Assumption: contiguous pages can be accessed as "page + i" */
2530 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2531 *offset
&= ~PAGE_MASK
;
2533 /* Bytes in this sg-entry from *offset to the end of the page */
2534 sg_len
= PAGE_SIZE
- *offset
;
2538 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2540 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2543 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2544 * @virt: virtual address to be unmapped
2546 void scsi_kunmap_atomic_sg(void *virt
)
2548 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2550 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);