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/blkdev.h>
12 #include <linux/completion.h>
13 #include <linux/kernel.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/pci.h>
18 #include <linux/delay.h>
19 #include <linux/hardirq.h>
20 #include <linux/scatterlist.h>
22 #include <scsi/scsi.h>
23 #include <scsi/scsi_cmnd.h>
24 #include <scsi/scsi_dbg.h>
25 #include <scsi/scsi_device.h>
26 #include <scsi/scsi_driver.h>
27 #include <scsi/scsi_eh.h>
28 #include <scsi/scsi_host.h>
30 #include "scsi_priv.h"
31 #include "scsi_logging.h"
34 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
35 #define SG_MEMPOOL_SIZE 2
38 * The maximum number of SG segments that we will put inside a scatterlist
39 * (unless chaining is used). Should ideally fit inside a single page, to
40 * avoid a higher order allocation.
42 #define SCSI_MAX_SG_SEGMENTS 128
44 struct scsi_host_sg_pool
{
47 struct kmem_cache
*slab
;
51 #define SP(x) { x, "sgpool-" #x }
52 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
55 #if (SCSI_MAX_SG_SEGMENTS > 16)
57 #if (SCSI_MAX_SG_SEGMENTS > 32)
59 #if (SCSI_MAX_SG_SEGMENTS > 64)
67 static void scsi_run_queue(struct request_queue
*q
);
70 * Function: scsi_unprep_request()
72 * Purpose: Remove all preparation done for a request, including its
73 * associated scsi_cmnd, so that it can be requeued.
75 * Arguments: req - request to unprepare
77 * Lock status: Assumed that no locks are held upon entry.
81 static void scsi_unprep_request(struct request
*req
)
83 struct scsi_cmnd
*cmd
= req
->special
;
85 req
->cmd_flags
&= ~REQ_DONTPREP
;
88 scsi_put_command(cmd
);
92 * Function: scsi_queue_insert()
94 * Purpose: Insert a command in the midlevel queue.
96 * Arguments: cmd - command that we are adding to queue.
97 * reason - why we are inserting command to queue.
99 * Lock status: Assumed that lock is not held upon entry.
103 * Notes: We do this for one of two cases. Either the host is busy
104 * and it cannot accept any more commands for the time being,
105 * or the device returned QUEUE_FULL and can accept no more
107 * Notes: This could be called either from an interrupt context or a
108 * normal process context.
110 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
112 struct Scsi_Host
*host
= cmd
->device
->host
;
113 struct scsi_device
*device
= cmd
->device
;
114 struct request_queue
*q
= device
->request_queue
;
118 printk("Inserting command %p into mlqueue\n", cmd
));
121 * Set the appropriate busy bit for the device/host.
123 * If the host/device isn't busy, assume that something actually
124 * completed, and that we should be able to queue a command now.
126 * Note that the prior mid-layer assumption that any host could
127 * always queue at least one command is now broken. The mid-layer
128 * will implement a user specifiable stall (see
129 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
130 * if a command is requeued with no other commands outstanding
131 * either for the device or for the host.
133 if (reason
== SCSI_MLQUEUE_HOST_BUSY
)
134 host
->host_blocked
= host
->max_host_blocked
;
135 else if (reason
== SCSI_MLQUEUE_DEVICE_BUSY
)
136 device
->device_blocked
= device
->max_device_blocked
;
139 * Decrement the counters, since these commands are no longer
140 * active on the host/device.
142 scsi_device_unbusy(device
);
145 * Requeue this command. It will go before all other commands
146 * that are already in the queue.
148 * NOTE: there is magic here about the way the queue is plugged if
149 * we have no outstanding commands.
151 * Although we *don't* plug the queue, we call the request
152 * function. The SCSI request function detects the blocked condition
153 * and plugs the queue appropriately.
155 spin_lock_irqsave(q
->queue_lock
, flags
);
156 blk_requeue_request(q
, cmd
->request
);
157 spin_unlock_irqrestore(q
->queue_lock
, flags
);
165 * scsi_execute - insert request and wait for the result
168 * @data_direction: data direction
169 * @buffer: data buffer
170 * @bufflen: len of buffer
171 * @sense: optional sense buffer
172 * @timeout: request timeout in seconds
173 * @retries: number of times to retry request
174 * @flags: or into request flags;
176 * returns the req->errors value which is the scsi_cmnd result
179 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
180 int data_direction
, void *buffer
, unsigned bufflen
,
181 unsigned char *sense
, int timeout
, int retries
, int flags
)
184 int write
= (data_direction
== DMA_TO_DEVICE
);
185 int ret
= DRIVER_ERROR
<< 24;
187 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
189 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
190 buffer
, bufflen
, __GFP_WAIT
))
193 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
194 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
197 req
->retries
= retries
;
198 req
->timeout
= timeout
;
199 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
200 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
203 * head injection *required* here otherwise quiesce won't work
205 blk_execute_rq(req
->q
, NULL
, req
, 1);
209 blk_put_request(req
);
213 EXPORT_SYMBOL(scsi_execute
);
216 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
217 int data_direction
, void *buffer
, unsigned bufflen
,
218 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
224 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
226 return DRIVER_ERROR
<< 24;
228 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
229 sense
, timeout
, retries
, 0);
231 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
236 EXPORT_SYMBOL(scsi_execute_req
);
238 struct scsi_io_context
{
240 void (*done
)(void *data
, char *sense
, int result
, int resid
);
241 char sense
[SCSI_SENSE_BUFFERSIZE
];
244 static struct kmem_cache
*scsi_io_context_cache
;
246 static void scsi_end_async(struct request
*req
, int uptodate
)
248 struct scsi_io_context
*sioc
= req
->end_io_data
;
251 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
253 kmem_cache_free(scsi_io_context_cache
, sioc
);
254 __blk_put_request(req
->q
, req
);
257 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
259 struct request_queue
*q
= rq
->q
;
261 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
262 if (rq_data_dir(rq
) == WRITE
)
263 bio
->bi_rw
|= (1 << BIO_RW
);
264 blk_queue_bounce(q
, &bio
);
266 return blk_rq_append_bio(q
, rq
, bio
);
269 static void scsi_bi_endio(struct bio
*bio
, int error
)
275 * scsi_req_map_sg - map a scatterlist into a request
276 * @rq: request to fill
278 * @nsegs: number of elements
279 * @bufflen: len of buffer
280 * @gfp: memory allocation flags
282 * scsi_req_map_sg maps a scatterlist into a request so that the
283 * request can be sent to the block layer. We do not trust the scatterlist
284 * sent to use, as some ULDs use that struct to only organize the pages.
286 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
287 int nsegs
, unsigned bufflen
, gfp_t gfp
)
289 struct request_queue
*q
= rq
->q
;
290 int nr_pages
= (bufflen
+ sgl
[0].offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
291 unsigned int data_len
= bufflen
, len
, bytes
, off
;
292 struct scatterlist
*sg
;
294 struct bio
*bio
= NULL
;
295 int i
, err
, nr_vecs
= 0;
297 for_each_sg(sgl
, sg
, nsegs
, i
) {
303 while (len
> 0 && data_len
> 0) {
305 * sg sends a scatterlist that is larger than
306 * the data_len it wants transferred for certain
309 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
310 bytes
= min(bytes
, data_len
);
313 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
316 bio
= bio_alloc(gfp
, nr_vecs
);
321 bio
->bi_end_io
= scsi_bi_endio
;
324 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
331 if (bio
->bi_vcnt
>= nr_vecs
) {
332 err
= scsi_merge_bio(rq
, bio
);
347 rq
->buffer
= rq
->data
= NULL
;
348 rq
->data_len
= bufflen
;
352 while ((bio
= rq
->bio
) != NULL
) {
353 rq
->bio
= bio
->bi_next
;
355 * call endio instead of bio_put incase it was bounced
364 * scsi_execute_async - insert request
367 * @cmd_len: length of scsi cdb
368 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
369 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
370 * @bufflen: len of buffer
371 * @use_sg: if buffer is a scatterlist this is the number of elements
372 * @timeout: request timeout in seconds
373 * @retries: number of times to retry request
374 * @privdata: data passed to done()
375 * @done: callback function when done
376 * @gfp: memory allocation flags
378 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
379 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
380 int use_sg
, int timeout
, int retries
, void *privdata
,
381 void (*done
)(void *, char *, int, int), gfp_t gfp
)
384 struct scsi_io_context
*sioc
;
386 int write
= (data_direction
== DMA_TO_DEVICE
);
388 sioc
= kmem_cache_zalloc(scsi_io_context_cache
, gfp
);
390 return DRIVER_ERROR
<< 24;
392 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
395 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
396 req
->cmd_flags
|= REQ_QUIET
;
399 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
401 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
406 req
->cmd_len
= cmd_len
;
407 memset(req
->cmd
, 0, BLK_MAX_CDB
); /* ATAPI hates garbage after CDB */
408 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
409 req
->sense
= sioc
->sense
;
411 req
->timeout
= timeout
;
412 req
->retries
= retries
;
413 req
->end_io_data
= sioc
;
415 sioc
->data
= privdata
;
418 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
422 blk_put_request(req
);
424 kmem_cache_free(scsi_io_context_cache
, sioc
);
425 return DRIVER_ERROR
<< 24;
427 EXPORT_SYMBOL_GPL(scsi_execute_async
);
430 * Function: scsi_init_cmd_errh()
432 * Purpose: Initialize cmd fields related to error handling.
434 * Arguments: cmd - command that is ready to be queued.
436 * Notes: This function has the job of initializing a number of
437 * fields related to error handling. Typically this will
438 * be called once for each command, as required.
440 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
442 cmd
->serial_number
= 0;
444 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
445 if (cmd
->cmd_len
== 0)
446 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
449 void scsi_device_unbusy(struct scsi_device
*sdev
)
451 struct Scsi_Host
*shost
= sdev
->host
;
454 spin_lock_irqsave(shost
->host_lock
, flags
);
456 if (unlikely(scsi_host_in_recovery(shost
) &&
457 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
458 scsi_eh_wakeup(shost
);
459 spin_unlock(shost
->host_lock
);
460 spin_lock(sdev
->request_queue
->queue_lock
);
462 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
466 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
467 * and call blk_run_queue for all the scsi_devices on the target -
468 * including current_sdev first.
470 * Called with *no* scsi locks held.
472 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
474 struct Scsi_Host
*shost
= current_sdev
->host
;
475 struct scsi_device
*sdev
, *tmp
;
476 struct scsi_target
*starget
= scsi_target(current_sdev
);
479 spin_lock_irqsave(shost
->host_lock
, flags
);
480 starget
->starget_sdev_user
= NULL
;
481 spin_unlock_irqrestore(shost
->host_lock
, flags
);
484 * Call blk_run_queue for all LUNs on the target, starting with
485 * current_sdev. We race with others (to set starget_sdev_user),
486 * but in most cases, we will be first. Ideally, each LU on the
487 * target would get some limited time or requests on the target.
489 blk_run_queue(current_sdev
->request_queue
);
491 spin_lock_irqsave(shost
->host_lock
, flags
);
492 if (starget
->starget_sdev_user
)
494 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
495 same_target_siblings
) {
496 if (sdev
== current_sdev
)
498 if (scsi_device_get(sdev
))
501 spin_unlock_irqrestore(shost
->host_lock
, flags
);
502 blk_run_queue(sdev
->request_queue
);
503 spin_lock_irqsave(shost
->host_lock
, flags
);
505 scsi_device_put(sdev
);
508 spin_unlock_irqrestore(shost
->host_lock
, flags
);
512 * Function: scsi_run_queue()
514 * Purpose: Select a proper request queue to serve next
516 * Arguments: q - last request's queue
520 * Notes: The previous command was completely finished, start
521 * a new one if possible.
523 static void scsi_run_queue(struct request_queue
*q
)
525 struct scsi_device
*sdev
= q
->queuedata
;
526 struct Scsi_Host
*shost
= sdev
->host
;
529 if (scsi_target(sdev
)->single_lun
)
530 scsi_single_lun_run(sdev
);
532 spin_lock_irqsave(shost
->host_lock
, flags
);
533 while (!list_empty(&shost
->starved_list
) &&
534 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
535 !((shost
->can_queue
> 0) &&
536 (shost
->host_busy
>= shost
->can_queue
))) {
538 * As long as shost is accepting commands and we have
539 * starved queues, call blk_run_queue. scsi_request_fn
540 * drops the queue_lock and can add us back to the
543 * host_lock protects the starved_list and starved_entry.
544 * scsi_request_fn must get the host_lock before checking
545 * or modifying starved_list or starved_entry.
547 sdev
= list_entry(shost
->starved_list
.next
,
548 struct scsi_device
, starved_entry
);
549 list_del_init(&sdev
->starved_entry
);
550 spin_unlock_irqrestore(shost
->host_lock
, flags
);
553 if (test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
) &&
554 !test_and_set_bit(QUEUE_FLAG_REENTER
,
555 &sdev
->request_queue
->queue_flags
)) {
556 blk_run_queue(sdev
->request_queue
);
557 clear_bit(QUEUE_FLAG_REENTER
,
558 &sdev
->request_queue
->queue_flags
);
560 blk_run_queue(sdev
->request_queue
);
562 spin_lock_irqsave(shost
->host_lock
, flags
);
563 if (unlikely(!list_empty(&sdev
->starved_entry
)))
565 * sdev lost a race, and was put back on the
566 * starved list. This is unlikely but without this
567 * in theory we could loop forever.
571 spin_unlock_irqrestore(shost
->host_lock
, flags
);
577 * Function: scsi_requeue_command()
579 * Purpose: Handle post-processing of completed commands.
581 * Arguments: q - queue to operate on
582 * cmd - command that may need to be requeued.
586 * Notes: After command completion, there may be blocks left
587 * over which weren't finished by the previous command
588 * this can be for a number of reasons - the main one is
589 * I/O errors in the middle of the request, in which case
590 * we need to request the blocks that come after the bad
592 * Notes: Upon return, cmd is a stale pointer.
594 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
596 struct request
*req
= cmd
->request
;
599 scsi_unprep_request(req
);
600 spin_lock_irqsave(q
->queue_lock
, flags
);
601 blk_requeue_request(q
, req
);
602 spin_unlock_irqrestore(q
->queue_lock
, flags
);
607 void scsi_next_command(struct scsi_cmnd
*cmd
)
609 struct scsi_device
*sdev
= cmd
->device
;
610 struct request_queue
*q
= sdev
->request_queue
;
612 /* need to hold a reference on the device before we let go of the cmd */
613 get_device(&sdev
->sdev_gendev
);
615 scsi_put_command(cmd
);
618 /* ok to remove device now */
619 put_device(&sdev
->sdev_gendev
);
622 void scsi_run_host_queues(struct Scsi_Host
*shost
)
624 struct scsi_device
*sdev
;
626 shost_for_each_device(sdev
, shost
)
627 scsi_run_queue(sdev
->request_queue
);
631 * Function: scsi_end_request()
633 * Purpose: Post-processing of completed commands (usually invoked at end
634 * of upper level post-processing and scsi_io_completion).
636 * Arguments: cmd - command that is complete.
637 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
638 * bytes - number of bytes of completed I/O
639 * requeue - indicates whether we should requeue leftovers.
641 * Lock status: Assumed that lock is not held upon entry.
643 * Returns: cmd if requeue required, NULL otherwise.
645 * Notes: This is called for block device requests in order to
646 * mark some number of sectors as complete.
648 * We are guaranteeing that the request queue will be goosed
649 * at some point during this call.
650 * Notes: If cmd was requeued, upon return it will be a stale pointer.
652 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int uptodate
,
653 int bytes
, int requeue
)
655 struct request_queue
*q
= cmd
->device
->request_queue
;
656 struct request
*req
= cmd
->request
;
660 * If there are blocks left over at the end, set up the command
661 * to queue the remainder of them.
663 if (end_that_request_chunk(req
, uptodate
, bytes
)) {
664 int leftover
= (req
->hard_nr_sectors
<< 9);
666 if (blk_pc_request(req
))
667 leftover
= req
->data_len
;
669 /* kill remainder if no retrys */
670 if (!uptodate
&& blk_noretry_request(req
))
671 end_that_request_chunk(req
, 0, leftover
);
675 * Bleah. Leftovers again. Stick the
676 * leftovers in the front of the
677 * queue, and goose the queue again.
679 scsi_requeue_command(q
, cmd
);
686 add_disk_randomness(req
->rq_disk
);
688 spin_lock_irqsave(q
->queue_lock
, flags
);
689 if (blk_rq_tagged(req
))
690 blk_queue_end_tag(q
, req
);
691 end_that_request_last(req
, uptodate
);
692 spin_unlock_irqrestore(q
->queue_lock
, flags
);
695 * This will goose the queue request function at the end, so we don't
696 * need to worry about launching another command.
698 scsi_next_command(cmd
);
703 * Like SCSI_MAX_SG_SEGMENTS, but for archs that have sg chaining. This limit
704 * is totally arbitrary, a setting of 2048 will get you at least 8mb ios.
706 #define SCSI_MAX_SG_CHAIN_SEGMENTS 2048
708 static inline unsigned int scsi_sgtable_index(unsigned short nents
)
719 #if (SCSI_MAX_SG_SEGMENTS > 16)
723 #if (SCSI_MAX_SG_SEGMENTS > 32)
727 #if (SCSI_MAX_SG_SEGMENTS > 64)
735 printk(KERN_ERR
"scsi: bad segment count=%d\n", nents
);
742 static void scsi_sg_free(struct scatterlist
*sgl
, unsigned int nents
)
744 struct scsi_host_sg_pool
*sgp
;
746 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
747 mempool_free(sgl
, sgp
->pool
);
750 static struct scatterlist
*scsi_sg_alloc(unsigned int nents
, gfp_t gfp_mask
)
752 struct scsi_host_sg_pool
*sgp
;
754 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
755 return mempool_alloc(sgp
->pool
, gfp_mask
);
758 int scsi_alloc_sgtable(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
762 BUG_ON(!cmd
->use_sg
);
764 ret
= __sg_alloc_table(&cmd
->sg_table
, cmd
->use_sg
, gfp_mask
, scsi_sg_alloc
);
766 __sg_free_table(&cmd
->sg_table
, scsi_sg_free
);
768 cmd
->request_buffer
= cmd
->sg_table
.sgl
;
772 EXPORT_SYMBOL(scsi_alloc_sgtable
);
774 void scsi_free_sgtable(struct scsi_cmnd
*cmd
)
776 __sg_free_table(&cmd
->sg_table
, scsi_sg_free
);
779 EXPORT_SYMBOL(scsi_free_sgtable
);
782 * Function: scsi_release_buffers()
784 * Purpose: Completion processing for block device I/O requests.
786 * Arguments: cmd - command that we are bailing.
788 * Lock status: Assumed that no lock is held upon entry.
792 * Notes: In the event that an upper level driver rejects a
793 * command, we must release resources allocated during
794 * the __init_io() function. Primarily this would involve
795 * the scatter-gather table, and potentially any bounce
798 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
801 scsi_free_sgtable(cmd
);
804 * Zero these out. They now point to freed memory, and it is
805 * dangerous to hang onto the pointers.
807 cmd
->request_buffer
= NULL
;
808 cmd
->request_bufflen
= 0;
812 * Function: scsi_io_completion()
814 * Purpose: Completion processing for block device I/O requests.
816 * Arguments: cmd - command that is finished.
818 * Lock status: Assumed that no lock is held upon entry.
822 * Notes: This function is matched in terms of capabilities to
823 * the function that created the scatter-gather list.
824 * In other words, if there are no bounce buffers
825 * (the normal case for most drivers), we don't need
826 * the logic to deal with cleaning up afterwards.
828 * We must do one of several things here:
830 * a) Call scsi_end_request. This will finish off the
831 * specified number of sectors. If we are done, the
832 * command block will be released, and the queue
833 * function will be goosed. If we are not done, then
834 * scsi_end_request will directly goose the queue.
836 * b) We can just use scsi_requeue_command() here. This would
837 * be used if we just wanted to retry, for example.
839 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
841 int result
= cmd
->result
;
842 int this_count
= cmd
->request_bufflen
;
843 struct request_queue
*q
= cmd
->device
->request_queue
;
844 struct request
*req
= cmd
->request
;
845 int clear_errors
= 1;
846 struct scsi_sense_hdr sshdr
;
848 int sense_deferred
= 0;
850 scsi_release_buffers(cmd
);
853 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
855 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
858 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
859 req
->errors
= result
;
862 if (sense_valid
&& req
->sense
) {
864 * SG_IO wants current and deferred errors
866 int len
= 8 + cmd
->sense_buffer
[7];
868 if (len
> SCSI_SENSE_BUFFERSIZE
)
869 len
= SCSI_SENSE_BUFFERSIZE
;
870 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
871 req
->sense_len
= len
;
874 req
->data_len
= cmd
->resid
;
878 * Next deal with any sectors which we were able to correctly
881 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
883 req
->nr_sectors
, good_bytes
));
884 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd
->use_sg
));
889 /* A number of bytes were successfully read. If there
890 * are leftovers and there is some kind of error
891 * (result != 0), retry the rest.
893 if (scsi_end_request(cmd
, 1, good_bytes
, result
== 0) == NULL
)
896 /* good_bytes = 0, or (inclusive) there were leftovers and
897 * result = 0, so scsi_end_request couldn't retry.
899 if (sense_valid
&& !sense_deferred
) {
900 switch (sshdr
.sense_key
) {
902 if (cmd
->device
->removable
) {
903 /* Detected disc change. Set a bit
904 * and quietly refuse further access.
906 cmd
->device
->changed
= 1;
907 scsi_end_request(cmd
, 0, this_count
, 1);
910 /* Must have been a power glitch, or a
911 * bus reset. Could not have been a
912 * media change, so we just retry the
913 * request and see what happens.
915 scsi_requeue_command(q
, cmd
);
919 case ILLEGAL_REQUEST
:
920 /* If we had an ILLEGAL REQUEST returned, then
921 * we may have performed an unsupported
922 * command. The only thing this should be
923 * would be a ten byte read where only a six
924 * byte read was supported. Also, on a system
925 * where READ CAPACITY failed, we may have
926 * read past the end of the disk.
928 if ((cmd
->device
->use_10_for_rw
&&
929 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
930 (cmd
->cmnd
[0] == READ_10
||
931 cmd
->cmnd
[0] == WRITE_10
)) {
932 cmd
->device
->use_10_for_rw
= 0;
933 /* This will cause a retry with a
936 scsi_requeue_command(q
, cmd
);
939 scsi_end_request(cmd
, 0, this_count
, 1);
944 /* If the device is in the process of becoming
945 * ready, or has a temporary blockage, retry.
947 if (sshdr
.asc
== 0x04) {
948 switch (sshdr
.ascq
) {
949 case 0x01: /* becoming ready */
950 case 0x04: /* format in progress */
951 case 0x05: /* rebuild in progress */
952 case 0x06: /* recalculation in progress */
953 case 0x07: /* operation in progress */
954 case 0x08: /* Long write in progress */
955 case 0x09: /* self test in progress */
956 scsi_requeue_command(q
, cmd
);
962 if (!(req
->cmd_flags
& REQ_QUIET
))
963 scsi_cmd_print_sense_hdr(cmd
,
967 scsi_end_request(cmd
, 0, this_count
, 1);
969 case VOLUME_OVERFLOW
:
970 if (!(req
->cmd_flags
& REQ_QUIET
)) {
971 scmd_printk(KERN_INFO
, cmd
,
972 "Volume overflow, CDB: ");
973 __scsi_print_command(cmd
->cmnd
);
974 scsi_print_sense("", cmd
);
976 /* See SSC3rXX or current. */
977 scsi_end_request(cmd
, 0, this_count
, 1);
983 if (host_byte(result
) == DID_RESET
) {
984 /* Third party bus reset or reset for error recovery
985 * reasons. Just retry the request and see what
988 scsi_requeue_command(q
, cmd
);
992 if (!(req
->cmd_flags
& REQ_QUIET
)) {
993 scsi_print_result(cmd
);
994 if (driver_byte(result
) & DRIVER_SENSE
)
995 scsi_print_sense("", cmd
);
998 scsi_end_request(cmd
, 0, this_count
, !result
);
1002 * Function: scsi_init_io()
1004 * Purpose: SCSI I/O initialize function.
1006 * Arguments: cmd - Command descriptor we wish to initialize
1008 * Returns: 0 on success
1009 * BLKPREP_DEFER if the failure is retryable
1011 static int scsi_init_io(struct scsi_cmnd
*cmd
)
1013 struct request
*req
= cmd
->request
;
1017 * We used to not use scatter-gather for single segment request,
1018 * but now we do (it makes highmem I/O easier to support without
1021 cmd
->use_sg
= req
->nr_phys_segments
;
1024 * If sg table allocation fails, requeue request later.
1026 if (unlikely(scsi_alloc_sgtable(cmd
, GFP_ATOMIC
))) {
1027 scsi_unprep_request(req
);
1028 return BLKPREP_DEFER
;
1032 if (blk_pc_request(req
))
1033 cmd
->request_bufflen
= req
->data_len
;
1035 cmd
->request_bufflen
= req
->nr_sectors
<< 9;
1038 * Next, walk the list, and fill in the addresses and sizes of
1041 count
= blk_rq_map_sg(req
->q
, req
, cmd
->request_buffer
);
1042 BUG_ON(count
> cmd
->use_sg
);
1043 cmd
->use_sg
= count
;
1047 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1048 struct request
*req
)
1050 struct scsi_cmnd
*cmd
;
1052 if (!req
->special
) {
1053 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1061 /* pull a tag out of the request if we have one */
1062 cmd
->tag
= req
->tag
;
1068 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1070 struct scsi_cmnd
*cmd
;
1071 int ret
= scsi_prep_state_check(sdev
, req
);
1073 if (ret
!= BLKPREP_OK
)
1076 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1078 return BLKPREP_DEFER
;
1081 * BLOCK_PC requests may transfer data, in which case they must
1082 * a bio attached to them. Or they might contain a SCSI command
1083 * that does not transfer data, in which case they may optionally
1084 * submit a request without an attached bio.
1089 BUG_ON(!req
->nr_phys_segments
);
1091 ret
= scsi_init_io(cmd
);
1095 BUG_ON(req
->data_len
);
1098 cmd
->request_bufflen
= 0;
1099 cmd
->request_buffer
= NULL
;
1104 BUILD_BUG_ON(sizeof(req
->cmd
) > sizeof(cmd
->cmnd
));
1105 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1106 cmd
->cmd_len
= req
->cmd_len
;
1108 cmd
->sc_data_direction
= DMA_NONE
;
1109 else if (rq_data_dir(req
) == WRITE
)
1110 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1112 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1114 cmd
->transfersize
= req
->data_len
;
1115 cmd
->allowed
= req
->retries
;
1116 cmd
->timeout_per_command
= req
->timeout
;
1119 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1122 * Setup a REQ_TYPE_FS command. These are simple read/write request
1123 * from filesystems that still need to be translated to SCSI CDBs from
1126 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1128 struct scsi_cmnd
*cmd
;
1129 int ret
= scsi_prep_state_check(sdev
, req
);
1131 if (ret
!= BLKPREP_OK
)
1134 * Filesystem requests must transfer data.
1136 BUG_ON(!req
->nr_phys_segments
);
1138 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1140 return BLKPREP_DEFER
;
1142 return scsi_init_io(cmd
);
1144 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1146 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1148 int ret
= BLKPREP_OK
;
1151 * If the device is not in running state we will reject some
1154 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1155 switch (sdev
->sdev_state
) {
1158 * If the device is offline we refuse to process any
1159 * commands. The device must be brought online
1160 * before trying any recovery commands.
1162 sdev_printk(KERN_ERR
, sdev
,
1163 "rejecting I/O to offline device\n");
1168 * If the device is fully deleted, we refuse to
1169 * process any commands as well.
1171 sdev_printk(KERN_ERR
, sdev
,
1172 "rejecting I/O to dead device\n");
1178 * If the devices is blocked we defer normal commands.
1180 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1181 ret
= BLKPREP_DEFER
;
1185 * For any other not fully online state we only allow
1186 * special commands. In particular any user initiated
1187 * command is not allowed.
1189 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1196 EXPORT_SYMBOL(scsi_prep_state_check
);
1198 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1200 struct scsi_device
*sdev
= q
->queuedata
;
1204 req
->errors
= DID_NO_CONNECT
<< 16;
1205 /* release the command and kill it */
1207 struct scsi_cmnd
*cmd
= req
->special
;
1208 scsi_release_buffers(cmd
);
1209 scsi_put_command(cmd
);
1210 req
->special
= NULL
;
1215 * If we defer, the elv_next_request() returns NULL, but the
1216 * queue must be restarted, so we plug here if no returning
1217 * command will automatically do that.
1219 if (sdev
->device_busy
== 0)
1223 req
->cmd_flags
|= REQ_DONTPREP
;
1228 EXPORT_SYMBOL(scsi_prep_return
);
1230 int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1232 struct scsi_device
*sdev
= q
->queuedata
;
1233 int ret
= BLKPREP_KILL
;
1235 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1236 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1237 return scsi_prep_return(q
, req
, ret
);
1241 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1244 * Called with the queue_lock held.
1246 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1247 struct scsi_device
*sdev
)
1249 if (sdev
->device_busy
>= sdev
->queue_depth
)
1251 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1253 * unblock after device_blocked iterates to zero
1255 if (--sdev
->device_blocked
== 0) {
1257 sdev_printk(KERN_INFO
, sdev
,
1258 "unblocking device at zero depth\n"));
1264 if (sdev
->device_blocked
)
1271 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1272 * return 0. We must end up running the queue again whenever 0 is
1273 * returned, else IO can hang.
1275 * Called with host_lock held.
1277 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1278 struct Scsi_Host
*shost
,
1279 struct scsi_device
*sdev
)
1281 if (scsi_host_in_recovery(shost
))
1283 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1285 * unblock after host_blocked iterates to zero
1287 if (--shost
->host_blocked
== 0) {
1289 printk("scsi%d unblocking host at zero depth\n",
1296 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1297 shost
->host_blocked
|| shost
->host_self_blocked
) {
1298 if (list_empty(&sdev
->starved_entry
))
1299 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1303 /* We're OK to process the command, so we can't be starved */
1304 if (!list_empty(&sdev
->starved_entry
))
1305 list_del_init(&sdev
->starved_entry
);
1311 * Kill a request for a dead device
1313 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1315 struct scsi_cmnd
*cmd
= req
->special
;
1316 struct scsi_device
*sdev
= cmd
->device
;
1317 struct Scsi_Host
*shost
= sdev
->host
;
1319 blkdev_dequeue_request(req
);
1321 if (unlikely(cmd
== NULL
)) {
1322 printk(KERN_CRIT
"impossible request in %s.\n",
1327 scsi_init_cmd_errh(cmd
);
1328 cmd
->result
= DID_NO_CONNECT
<< 16;
1329 atomic_inc(&cmd
->device
->iorequest_cnt
);
1332 * SCSI request completion path will do scsi_device_unbusy(),
1333 * bump busy counts. To bump the counters, we need to dance
1334 * with the locks as normal issue path does.
1336 sdev
->device_busy
++;
1337 spin_unlock(sdev
->request_queue
->queue_lock
);
1338 spin_lock(shost
->host_lock
);
1340 spin_unlock(shost
->host_lock
);
1341 spin_lock(sdev
->request_queue
->queue_lock
);
1346 static void scsi_softirq_done(struct request
*rq
)
1348 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1349 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1352 INIT_LIST_HEAD(&cmd
->eh_entry
);
1354 disposition
= scsi_decide_disposition(cmd
);
1355 if (disposition
!= SUCCESS
&&
1356 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1357 sdev_printk(KERN_ERR
, cmd
->device
,
1358 "timing out command, waited %lus\n",
1360 disposition
= SUCCESS
;
1363 scsi_log_completion(cmd
, disposition
);
1365 switch (disposition
) {
1367 scsi_finish_command(cmd
);
1370 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1372 case ADD_TO_MLQUEUE
:
1373 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1376 if (!scsi_eh_scmd_add(cmd
, 0))
1377 scsi_finish_command(cmd
);
1382 * Function: scsi_request_fn()
1384 * Purpose: Main strategy routine for SCSI.
1386 * Arguments: q - Pointer to actual queue.
1390 * Lock status: IO request lock assumed to be held when called.
1392 static void scsi_request_fn(struct request_queue
*q
)
1394 struct scsi_device
*sdev
= q
->queuedata
;
1395 struct Scsi_Host
*shost
;
1396 struct scsi_cmnd
*cmd
;
1397 struct request
*req
;
1400 printk("scsi: killing requests for dead queue\n");
1401 while ((req
= elv_next_request(q
)) != NULL
)
1402 scsi_kill_request(req
, q
);
1406 if(!get_device(&sdev
->sdev_gendev
))
1407 /* We must be tearing the block queue down already */
1411 * To start with, we keep looping until the queue is empty, or until
1412 * the host is no longer able to accept any more requests.
1415 while (!blk_queue_plugged(q
)) {
1418 * get next queueable request. We do this early to make sure
1419 * that the request is fully prepared even if we cannot
1422 req
= elv_next_request(q
);
1423 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1426 if (unlikely(!scsi_device_online(sdev
))) {
1427 sdev_printk(KERN_ERR
, sdev
,
1428 "rejecting I/O to offline device\n");
1429 scsi_kill_request(req
, q
);
1435 * Remove the request from the request list.
1437 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1438 blkdev_dequeue_request(req
);
1439 sdev
->device_busy
++;
1441 spin_unlock(q
->queue_lock
);
1443 if (unlikely(cmd
== NULL
)) {
1444 printk(KERN_CRIT
"impossible request in %s.\n"
1445 "please mail a stack trace to "
1446 "linux-scsi@vger.kernel.org\n",
1448 blk_dump_rq_flags(req
, "foo");
1451 spin_lock(shost
->host_lock
);
1453 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1455 if (scsi_target(sdev
)->single_lun
) {
1456 if (scsi_target(sdev
)->starget_sdev_user
&&
1457 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1459 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1464 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1465 * take the lock again.
1467 spin_unlock_irq(shost
->host_lock
);
1470 * Finally, initialize any error handling parameters, and set up
1471 * the timers for timeouts.
1473 scsi_init_cmd_errh(cmd
);
1476 * Dispatch the command to the low-level driver.
1478 rtn
= scsi_dispatch_cmd(cmd
);
1479 spin_lock_irq(q
->queue_lock
);
1481 /* we're refusing the command; because of
1482 * the way locks get dropped, we need to
1483 * check here if plugging is required */
1484 if(sdev
->device_busy
== 0)
1494 spin_unlock_irq(shost
->host_lock
);
1497 * lock q, handle tag, requeue req, and decrement device_busy. We
1498 * must return with queue_lock held.
1500 * Decrementing device_busy without checking it is OK, as all such
1501 * cases (host limits or settings) should run the queue at some
1504 spin_lock_irq(q
->queue_lock
);
1505 blk_requeue_request(q
, req
);
1506 sdev
->device_busy
--;
1507 if(sdev
->device_busy
== 0)
1510 /* must be careful here...if we trigger the ->remove() function
1511 * we cannot be holding the q lock */
1512 spin_unlock_irq(q
->queue_lock
);
1513 put_device(&sdev
->sdev_gendev
);
1514 spin_lock_irq(q
->queue_lock
);
1517 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1519 struct device
*host_dev
;
1520 u64 bounce_limit
= 0xffffffff;
1522 if (shost
->unchecked_isa_dma
)
1523 return BLK_BOUNCE_ISA
;
1525 * Platforms with virtual-DMA translation
1526 * hardware have no practical limit.
1528 if (!PCI_DMA_BUS_IS_PHYS
)
1529 return BLK_BOUNCE_ANY
;
1531 host_dev
= scsi_get_device(shost
);
1532 if (host_dev
&& host_dev
->dma_mask
)
1533 bounce_limit
= *host_dev
->dma_mask
;
1535 return bounce_limit
;
1537 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1539 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1540 request_fn_proc
*request_fn
)
1542 struct request_queue
*q
;
1544 q
= blk_init_queue(request_fn
, NULL
);
1549 * this limit is imposed by hardware restrictions
1551 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1554 * In the future, sg chaining support will be mandatory and this
1555 * ifdef can then go away. Right now we don't have all archs
1556 * converted, so better keep it safe.
1558 #ifdef ARCH_HAS_SG_CHAIN
1559 if (shost
->use_sg_chaining
)
1560 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_CHAIN_SEGMENTS
);
1562 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_SEGMENTS
);
1564 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_SEGMENTS
);
1567 blk_queue_max_sectors(q
, shost
->max_sectors
);
1568 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1569 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1571 if (!shost
->use_clustering
)
1572 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1575 * set a reasonable default alignment on word boundaries: the
1576 * host and device may alter it using
1577 * blk_queue_update_dma_alignment() later.
1579 blk_queue_dma_alignment(q
, 0x03);
1583 EXPORT_SYMBOL(__scsi_alloc_queue
);
1585 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1587 struct request_queue
*q
;
1589 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1593 blk_queue_prep_rq(q
, scsi_prep_fn
);
1594 blk_queue_softirq_done(q
, scsi_softirq_done
);
1598 void scsi_free_queue(struct request_queue
*q
)
1600 blk_cleanup_queue(q
);
1604 * Function: scsi_block_requests()
1606 * Purpose: Utility function used by low-level drivers to prevent further
1607 * commands from being queued to the device.
1609 * Arguments: shost - Host in question
1613 * Lock status: No locks are assumed held.
1615 * Notes: There is no timer nor any other means by which the requests
1616 * get unblocked other than the low-level driver calling
1617 * scsi_unblock_requests().
1619 void scsi_block_requests(struct Scsi_Host
*shost
)
1621 shost
->host_self_blocked
= 1;
1623 EXPORT_SYMBOL(scsi_block_requests
);
1626 * Function: scsi_unblock_requests()
1628 * Purpose: Utility function used by low-level drivers to allow further
1629 * commands from being queued to the device.
1631 * Arguments: shost - Host in question
1635 * Lock status: No locks are assumed held.
1637 * Notes: There is no timer nor any other means by which the requests
1638 * get unblocked other than the low-level driver calling
1639 * scsi_unblock_requests().
1641 * This is done as an API function so that changes to the
1642 * internals of the scsi mid-layer won't require wholesale
1643 * changes to drivers that use this feature.
1645 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1647 shost
->host_self_blocked
= 0;
1648 scsi_run_host_queues(shost
);
1650 EXPORT_SYMBOL(scsi_unblock_requests
);
1652 int __init
scsi_init_queue(void)
1656 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1657 sizeof(struct scsi_io_context
),
1659 if (!scsi_io_context_cache
) {
1660 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1664 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1665 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1666 int size
= sgp
->size
* sizeof(struct scatterlist
);
1668 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1669 SLAB_HWCACHE_ALIGN
, NULL
);
1671 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1675 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1678 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1686 void scsi_exit_queue(void)
1690 kmem_cache_destroy(scsi_io_context_cache
);
1692 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1693 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1694 mempool_destroy(sgp
->pool
);
1695 kmem_cache_destroy(sgp
->slab
);
1700 * scsi_mode_select - issue a mode select
1701 * @sdev: SCSI device to be queried
1702 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1703 * @sp: Save page bit (0 == don't save, 1 == save)
1704 * @modepage: mode page being requested
1705 * @buffer: request buffer (may not be smaller than eight bytes)
1706 * @len: length of request buffer.
1707 * @timeout: command timeout
1708 * @retries: number of retries before failing
1709 * @data: returns a structure abstracting the mode header data
1710 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1711 * must be SCSI_SENSE_BUFFERSIZE big.
1713 * Returns zero if successful; negative error number or scsi
1718 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1719 unsigned char *buffer
, int len
, int timeout
, int retries
,
1720 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1722 unsigned char cmd
[10];
1723 unsigned char *real_buffer
;
1726 memset(cmd
, 0, sizeof(cmd
));
1727 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1729 if (sdev
->use_10_for_ms
) {
1732 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1735 memcpy(real_buffer
+ 8, buffer
, len
);
1739 real_buffer
[2] = data
->medium_type
;
1740 real_buffer
[3] = data
->device_specific
;
1741 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1743 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1744 real_buffer
[7] = data
->block_descriptor_length
;
1746 cmd
[0] = MODE_SELECT_10
;
1750 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1754 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1757 memcpy(real_buffer
+ 4, buffer
, len
);
1760 real_buffer
[1] = data
->medium_type
;
1761 real_buffer
[2] = data
->device_specific
;
1762 real_buffer
[3] = data
->block_descriptor_length
;
1765 cmd
[0] = MODE_SELECT
;
1769 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1770 sshdr
, timeout
, retries
);
1774 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1777 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1778 * @sdev: SCSI device to be queried
1779 * @dbd: set if mode sense will allow block descriptors to be returned
1780 * @modepage: mode page being requested
1781 * @buffer: request buffer (may not be smaller than eight bytes)
1782 * @len: length of request buffer.
1783 * @timeout: command timeout
1784 * @retries: number of retries before failing
1785 * @data: returns a structure abstracting the mode header data
1786 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1787 * must be SCSI_SENSE_BUFFERSIZE big.
1789 * Returns zero if unsuccessful, or the header offset (either 4
1790 * or 8 depending on whether a six or ten byte command was
1791 * issued) if successful.
1794 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1795 unsigned char *buffer
, int len
, int timeout
, int retries
,
1796 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1798 unsigned char cmd
[12];
1802 struct scsi_sense_hdr my_sshdr
;
1804 memset(data
, 0, sizeof(*data
));
1805 memset(&cmd
[0], 0, 12);
1806 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1809 /* caller might not be interested in sense, but we need it */
1814 use_10_for_ms
= sdev
->use_10_for_ms
;
1816 if (use_10_for_ms
) {
1820 cmd
[0] = MODE_SENSE_10
;
1827 cmd
[0] = MODE_SENSE
;
1832 memset(buffer
, 0, len
);
1834 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1835 sshdr
, timeout
, retries
);
1837 /* This code looks awful: what it's doing is making sure an
1838 * ILLEGAL REQUEST sense return identifies the actual command
1839 * byte as the problem. MODE_SENSE commands can return
1840 * ILLEGAL REQUEST if the code page isn't supported */
1842 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1843 (driver_byte(result
) & DRIVER_SENSE
)) {
1844 if (scsi_sense_valid(sshdr
)) {
1845 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1846 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1848 * Invalid command operation code
1850 sdev
->use_10_for_ms
= 0;
1856 if(scsi_status_is_good(result
)) {
1857 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1858 (modepage
== 6 || modepage
== 8))) {
1859 /* Initio breakage? */
1862 data
->medium_type
= 0;
1863 data
->device_specific
= 0;
1865 data
->block_descriptor_length
= 0;
1866 } else if(use_10_for_ms
) {
1867 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1868 data
->medium_type
= buffer
[2];
1869 data
->device_specific
= buffer
[3];
1870 data
->longlba
= buffer
[4] & 0x01;
1871 data
->block_descriptor_length
= buffer
[6]*256
1874 data
->length
= buffer
[0] + 1;
1875 data
->medium_type
= buffer
[1];
1876 data
->device_specific
= buffer
[2];
1877 data
->block_descriptor_length
= buffer
[3];
1879 data
->header_length
= header_length
;
1884 EXPORT_SYMBOL(scsi_mode_sense
);
1887 * scsi_test_unit_ready - test if unit is ready
1888 * @sdev: scsi device to change the state of.
1889 * @timeout: command timeout
1890 * @retries: number of retries before failing
1891 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1892 * returning sense. Make sure that this is cleared before passing
1895 * Returns zero if unsuccessful or an error if TUR failed. For
1896 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1897 * translated to success, with the ->changed flag updated.
1900 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
1901 struct scsi_sense_hdr
*sshdr_external
)
1904 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1906 struct scsi_sense_hdr
*sshdr
;
1909 if (!sshdr_external
)
1910 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1912 sshdr
= sshdr_external
;
1914 /* try to eat the UNIT_ATTENTION if there are enough retries */
1916 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
1918 } while ((driver_byte(result
) & DRIVER_SENSE
) &&
1919 sshdr
&& sshdr
->sense_key
== UNIT_ATTENTION
&&
1923 /* could not allocate sense buffer, so can't process it */
1926 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1928 if ((scsi_sense_valid(sshdr
)) &&
1929 ((sshdr
->sense_key
== UNIT_ATTENTION
) ||
1930 (sshdr
->sense_key
== NOT_READY
))) {
1935 if (!sshdr_external
)
1939 EXPORT_SYMBOL(scsi_test_unit_ready
);
1942 * scsi_device_set_state - Take the given device through the device state model.
1943 * @sdev: scsi device to change the state of.
1944 * @state: state to change to.
1946 * Returns zero if unsuccessful or an error if the requested
1947 * transition is illegal.
1950 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
1952 enum scsi_device_state oldstate
= sdev
->sdev_state
;
1954 if (state
== oldstate
)
1959 /* There are no legal states that come back to
1960 * created. This is the manually initialised start
2034 sdev
->sdev_state
= state
;
2038 SCSI_LOG_ERROR_RECOVERY(1,
2039 sdev_printk(KERN_ERR
, sdev
,
2040 "Illegal state transition %s->%s\n",
2041 scsi_device_state_name(oldstate
),
2042 scsi_device_state_name(state
))
2046 EXPORT_SYMBOL(scsi_device_set_state
);
2049 * sdev_evt_emit - emit a single SCSI device uevent
2050 * @sdev: associated SCSI device
2051 * @evt: event to emit
2053 * Send a single uevent (scsi_event) to the associated scsi_device.
2055 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2060 switch (evt
->evt_type
) {
2061 case SDEV_EVT_MEDIA_CHANGE
:
2062 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2072 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2076 * sdev_evt_thread - send a uevent for each scsi event
2077 * @work: work struct for scsi_device
2079 * Dispatch queued events to their associated scsi_device kobjects
2082 void scsi_evt_thread(struct work_struct
*work
)
2084 struct scsi_device
*sdev
;
2085 LIST_HEAD(event_list
);
2087 sdev
= container_of(work
, struct scsi_device
, event_work
);
2090 struct scsi_event
*evt
;
2091 struct list_head
*this, *tmp
;
2092 unsigned long flags
;
2094 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2095 list_splice_init(&sdev
->event_list
, &event_list
);
2096 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2098 if (list_empty(&event_list
))
2101 list_for_each_safe(this, tmp
, &event_list
) {
2102 evt
= list_entry(this, struct scsi_event
, node
);
2103 list_del(&evt
->node
);
2104 scsi_evt_emit(sdev
, evt
);
2111 * sdev_evt_send - send asserted event to uevent thread
2112 * @sdev: scsi_device event occurred on
2113 * @evt: event to send
2115 * Assert scsi device event asynchronously.
2117 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2119 unsigned long flags
;
2121 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2126 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2127 list_add_tail(&evt
->node
, &sdev
->event_list
);
2128 schedule_work(&sdev
->event_work
);
2129 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2131 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2134 * sdev_evt_alloc - allocate a new scsi event
2135 * @evt_type: type of event to allocate
2136 * @gfpflags: GFP flags for allocation
2138 * Allocates and returns a new scsi_event.
2140 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2143 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2147 evt
->evt_type
= evt_type
;
2148 INIT_LIST_HEAD(&evt
->node
);
2150 /* evt_type-specific initialization, if any */
2152 case SDEV_EVT_MEDIA_CHANGE
:
2160 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2163 * sdev_evt_send_simple - send asserted event to uevent thread
2164 * @sdev: scsi_device event occurred on
2165 * @evt_type: type of event to send
2166 * @gfpflags: GFP flags for allocation
2168 * Assert scsi device event asynchronously, given an event type.
2170 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2171 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2173 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2175 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2180 sdev_evt_send(sdev
, evt
);
2182 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2185 * scsi_device_quiesce - Block user issued commands.
2186 * @sdev: scsi device to quiesce.
2188 * This works by trying to transition to the SDEV_QUIESCE state
2189 * (which must be a legal transition). When the device is in this
2190 * state, only special requests will be accepted, all others will
2191 * be deferred. Since special requests may also be requeued requests,
2192 * a successful return doesn't guarantee the device will be
2193 * totally quiescent.
2195 * Must be called with user context, may sleep.
2197 * Returns zero if unsuccessful or an error if not.
2200 scsi_device_quiesce(struct scsi_device
*sdev
)
2202 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2206 scsi_run_queue(sdev
->request_queue
);
2207 while (sdev
->device_busy
) {
2208 msleep_interruptible(200);
2209 scsi_run_queue(sdev
->request_queue
);
2213 EXPORT_SYMBOL(scsi_device_quiesce
);
2216 * scsi_device_resume - Restart user issued commands to a quiesced device.
2217 * @sdev: scsi device to resume.
2219 * Moves the device from quiesced back to running and restarts the
2222 * Must be called with user context, may sleep.
2225 scsi_device_resume(struct scsi_device
*sdev
)
2227 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2229 scsi_run_queue(sdev
->request_queue
);
2231 EXPORT_SYMBOL(scsi_device_resume
);
2234 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2236 scsi_device_quiesce(sdev
);
2240 scsi_target_quiesce(struct scsi_target
*starget
)
2242 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2244 EXPORT_SYMBOL(scsi_target_quiesce
);
2247 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2249 scsi_device_resume(sdev
);
2253 scsi_target_resume(struct scsi_target
*starget
)
2255 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2257 EXPORT_SYMBOL(scsi_target_resume
);
2260 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2261 * @sdev: device to block
2263 * Block request made by scsi lld's to temporarily stop all
2264 * scsi commands on the specified device. Called from interrupt
2265 * or normal process context.
2267 * Returns zero if successful or error if not
2270 * This routine transitions the device to the SDEV_BLOCK state
2271 * (which must be a legal transition). When the device is in this
2272 * state, all commands are deferred until the scsi lld reenables
2273 * the device with scsi_device_unblock or device_block_tmo fires.
2274 * This routine assumes the host_lock is held on entry.
2277 scsi_internal_device_block(struct scsi_device
*sdev
)
2279 struct request_queue
*q
= sdev
->request_queue
;
2280 unsigned long flags
;
2283 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2288 * The device has transitioned to SDEV_BLOCK. Stop the
2289 * block layer from calling the midlayer with this device's
2292 spin_lock_irqsave(q
->queue_lock
, flags
);
2294 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2298 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2301 * scsi_internal_device_unblock - resume a device after a block request
2302 * @sdev: device to resume
2304 * Called by scsi lld's or the midlayer to restart the device queue
2305 * for the previously suspended scsi device. Called from interrupt or
2306 * normal process context.
2308 * Returns zero if successful or error if not.
2311 * This routine transitions the device to the SDEV_RUNNING state
2312 * (which must be a legal transition) allowing the midlayer to
2313 * goose the queue for this device. This routine assumes the
2314 * host_lock is held upon entry.
2317 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2319 struct request_queue
*q
= sdev
->request_queue
;
2321 unsigned long flags
;
2324 * Try to transition the scsi device to SDEV_RUNNING
2325 * and goose the device queue if successful.
2327 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2331 spin_lock_irqsave(q
->queue_lock
, flags
);
2333 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2337 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2340 device_block(struct scsi_device
*sdev
, void *data
)
2342 scsi_internal_device_block(sdev
);
2346 target_block(struct device
*dev
, void *data
)
2348 if (scsi_is_target_device(dev
))
2349 starget_for_each_device(to_scsi_target(dev
), NULL
,
2355 scsi_target_block(struct device
*dev
)
2357 if (scsi_is_target_device(dev
))
2358 starget_for_each_device(to_scsi_target(dev
), NULL
,
2361 device_for_each_child(dev
, NULL
, target_block
);
2363 EXPORT_SYMBOL_GPL(scsi_target_block
);
2366 device_unblock(struct scsi_device
*sdev
, void *data
)
2368 scsi_internal_device_unblock(sdev
);
2372 target_unblock(struct device
*dev
, void *data
)
2374 if (scsi_is_target_device(dev
))
2375 starget_for_each_device(to_scsi_target(dev
), NULL
,
2381 scsi_target_unblock(struct device
*dev
)
2383 if (scsi_is_target_device(dev
))
2384 starget_for_each_device(to_scsi_target(dev
), NULL
,
2387 device_for_each_child(dev
, NULL
, target_unblock
);
2389 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2392 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2393 * @sgl: scatter-gather list
2394 * @sg_count: number of segments in sg
2395 * @offset: offset in bytes into sg, on return offset into the mapped area
2396 * @len: bytes to map, on return number of bytes mapped
2398 * Returns virtual address of the start of the mapped page
2400 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2401 size_t *offset
, size_t *len
)
2404 size_t sg_len
= 0, len_complete
= 0;
2405 struct scatterlist
*sg
;
2408 WARN_ON(!irqs_disabled());
2410 for_each_sg(sgl
, sg
, sg_count
, i
) {
2411 len_complete
= sg_len
; /* Complete sg-entries */
2412 sg_len
+= sg
->length
;
2413 if (sg_len
> *offset
)
2417 if (unlikely(i
== sg_count
)) {
2418 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2420 __FUNCTION__
, sg_len
, *offset
, sg_count
);
2425 /* Offset starting from the beginning of first page in this sg-entry */
2426 *offset
= *offset
- len_complete
+ sg
->offset
;
2428 /* Assumption: contiguous pages can be accessed as "page + i" */
2429 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2430 *offset
&= ~PAGE_MASK
;
2432 /* Bytes in this sg-entry from *offset to the end of the page */
2433 sg_len
= PAGE_SIZE
- *offset
;
2437 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2439 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2442 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2443 * @virt: virtual address to be unmapped
2445 void scsi_kunmap_atomic_sg(void *virt
)
2447 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2449 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);