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 static struct kmem_cache
*scsi_bidi_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 * Function: scsi_queue_insert()
97 * Purpose: Insert a command in the midlevel queue.
99 * Arguments: cmd - command that we are adding to queue.
100 * reason - why we are inserting command to queue.
102 * Lock status: Assumed that lock is not held upon entry.
106 * Notes: We do this for one of two cases. Either the host is busy
107 * and it cannot accept any more commands for the time being,
108 * or the device returned QUEUE_FULL and can accept no more
110 * Notes: This could be called either from an interrupt context or a
111 * normal process context.
113 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
115 struct Scsi_Host
*host
= cmd
->device
->host
;
116 struct scsi_device
*device
= cmd
->device
;
117 struct request_queue
*q
= device
->request_queue
;
121 printk("Inserting command %p into mlqueue\n", cmd
));
124 * Set the appropriate busy bit for the device/host.
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
136 if (reason
== SCSI_MLQUEUE_HOST_BUSY
)
137 host
->host_blocked
= host
->max_host_blocked
;
138 else if (reason
== SCSI_MLQUEUE_DEVICE_BUSY
)
139 device
->device_blocked
= device
->max_device_blocked
;
142 * Decrement the counters, since these commands are no longer
143 * active on the host/device.
145 scsi_device_unbusy(device
);
148 * Requeue this command. It will go before all other commands
149 * that are already in the queue.
151 * NOTE: there is magic here about the way the queue is plugged if
152 * we have no outstanding commands.
154 * Although we *don't* plug the queue, we call the request
155 * function. The SCSI request function detects the blocked condition
156 * and plugs the queue appropriately.
158 spin_lock_irqsave(q
->queue_lock
, flags
);
159 blk_requeue_request(q
, cmd
->request
);
160 spin_unlock_irqrestore(q
->queue_lock
, flags
);
168 * scsi_execute - insert request and wait for the result
171 * @data_direction: data direction
172 * @buffer: data buffer
173 * @bufflen: len of buffer
174 * @sense: optional sense buffer
175 * @timeout: request timeout in seconds
176 * @retries: number of times to retry request
177 * @flags: or into request flags;
179 * returns the req->errors value which is the scsi_cmnd result
182 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
183 int data_direction
, void *buffer
, unsigned bufflen
,
184 unsigned char *sense
, int timeout
, int retries
, int flags
)
187 int write
= (data_direction
== DMA_TO_DEVICE
);
188 int ret
= DRIVER_ERROR
<< 24;
190 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
192 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
193 buffer
, bufflen
, __GFP_WAIT
))
196 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
197 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
200 req
->retries
= retries
;
201 req
->timeout
= timeout
;
202 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
203 req
->cmd_flags
|= flags
| REQ_QUIET
| REQ_PREEMPT
;
206 * head injection *required* here otherwise quiesce won't work
208 blk_execute_rq(req
->q
, NULL
, req
, 1);
212 blk_put_request(req
);
216 EXPORT_SYMBOL(scsi_execute
);
219 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
220 int data_direction
, void *buffer
, unsigned bufflen
,
221 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
227 sense
= kzalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
229 return DRIVER_ERROR
<< 24;
231 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
232 sense
, timeout
, retries
, 0);
234 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
239 EXPORT_SYMBOL(scsi_execute_req
);
241 struct scsi_io_context
{
243 void (*done
)(void *data
, char *sense
, int result
, int resid
);
244 char sense
[SCSI_SENSE_BUFFERSIZE
];
247 static struct kmem_cache
*scsi_io_context_cache
;
249 static void scsi_end_async(struct request
*req
, int uptodate
)
251 struct scsi_io_context
*sioc
= req
->end_io_data
;
254 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
256 kmem_cache_free(scsi_io_context_cache
, sioc
);
257 __blk_put_request(req
->q
, req
);
260 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
262 struct request_queue
*q
= rq
->q
;
264 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
265 if (rq_data_dir(rq
) == WRITE
)
266 bio
->bi_rw
|= (1 << BIO_RW
);
267 blk_queue_bounce(q
, &bio
);
269 return blk_rq_append_bio(q
, rq
, bio
);
272 static void scsi_bi_endio(struct bio
*bio
, int error
)
278 * scsi_req_map_sg - map a scatterlist into a request
279 * @rq: request to fill
281 * @nsegs: number of elements
282 * @bufflen: len of buffer
283 * @gfp: memory allocation flags
285 * scsi_req_map_sg maps a scatterlist into a request so that the
286 * request can be sent to the block layer. We do not trust the scatterlist
287 * sent to use, as some ULDs use that struct to only organize the pages.
289 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
290 int nsegs
, unsigned bufflen
, gfp_t gfp
)
292 struct request_queue
*q
= rq
->q
;
293 int nr_pages
= (bufflen
+ sgl
[0].offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
294 unsigned int data_len
= bufflen
, len
, bytes
, off
;
295 struct scatterlist
*sg
;
297 struct bio
*bio
= NULL
;
298 int i
, err
, nr_vecs
= 0;
300 for_each_sg(sgl
, sg
, nsegs
, i
) {
306 while (len
> 0 && data_len
> 0) {
308 * sg sends a scatterlist that is larger than
309 * the data_len it wants transferred for certain
312 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
313 bytes
= min(bytes
, data_len
);
316 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
319 bio
= bio_alloc(gfp
, nr_vecs
);
324 bio
->bi_end_io
= scsi_bi_endio
;
327 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
334 if (bio
->bi_vcnt
>= nr_vecs
) {
335 err
= scsi_merge_bio(rq
, bio
);
350 rq
->buffer
= rq
->data
= NULL
;
351 rq
->data_len
= bufflen
;
355 while ((bio
= rq
->bio
) != NULL
) {
356 rq
->bio
= bio
->bi_next
;
358 * call endio instead of bio_put incase it was bounced
367 * scsi_execute_async - insert request
370 * @cmd_len: length of scsi cdb
371 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
372 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
373 * @bufflen: len of buffer
374 * @use_sg: if buffer is a scatterlist this is the number of elements
375 * @timeout: request timeout in seconds
376 * @retries: number of times to retry request
377 * @privdata: data passed to done()
378 * @done: callback function when done
379 * @gfp: memory allocation flags
381 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
382 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
383 int use_sg
, int timeout
, int retries
, void *privdata
,
384 void (*done
)(void *, char *, int, int), gfp_t gfp
)
387 struct scsi_io_context
*sioc
;
389 int write
= (data_direction
== DMA_TO_DEVICE
);
391 sioc
= kmem_cache_zalloc(scsi_io_context_cache
, gfp
);
393 return DRIVER_ERROR
<< 24;
395 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
398 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
399 req
->cmd_flags
|= REQ_QUIET
;
402 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
404 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
409 req
->cmd_len
= cmd_len
;
410 memset(req
->cmd
, 0, BLK_MAX_CDB
); /* ATAPI hates garbage after CDB */
411 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
412 req
->sense
= sioc
->sense
;
414 req
->timeout
= timeout
;
415 req
->retries
= retries
;
416 req
->end_io_data
= sioc
;
418 sioc
->data
= privdata
;
421 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
425 blk_put_request(req
);
427 kmem_cache_free(scsi_io_context_cache
, sioc
);
428 return DRIVER_ERROR
<< 24;
430 EXPORT_SYMBOL_GPL(scsi_execute_async
);
433 * Function: scsi_init_cmd_errh()
435 * Purpose: Initialize cmd fields related to error handling.
437 * Arguments: cmd - command that is ready to be queued.
439 * Notes: This function has the job of initializing a number of
440 * fields related to error handling. Typically this will
441 * be called once for each command, as required.
443 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
445 cmd
->serial_number
= 0;
446 scsi_set_resid(cmd
, 0);
447 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
448 if (cmd
->cmd_len
== 0)
449 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
452 void scsi_device_unbusy(struct scsi_device
*sdev
)
454 struct Scsi_Host
*shost
= sdev
->host
;
457 spin_lock_irqsave(shost
->host_lock
, flags
);
459 if (unlikely(scsi_host_in_recovery(shost
) &&
460 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
461 scsi_eh_wakeup(shost
);
462 spin_unlock(shost
->host_lock
);
463 spin_lock(sdev
->request_queue
->queue_lock
);
465 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
469 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
470 * and call blk_run_queue for all the scsi_devices on the target -
471 * including current_sdev first.
473 * Called with *no* scsi locks held.
475 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
477 struct Scsi_Host
*shost
= current_sdev
->host
;
478 struct scsi_device
*sdev
, *tmp
;
479 struct scsi_target
*starget
= scsi_target(current_sdev
);
482 spin_lock_irqsave(shost
->host_lock
, flags
);
483 starget
->starget_sdev_user
= NULL
;
484 spin_unlock_irqrestore(shost
->host_lock
, flags
);
487 * Call blk_run_queue for all LUNs on the target, starting with
488 * current_sdev. We race with others (to set starget_sdev_user),
489 * but in most cases, we will be first. Ideally, each LU on the
490 * target would get some limited time or requests on the target.
492 blk_run_queue(current_sdev
->request_queue
);
494 spin_lock_irqsave(shost
->host_lock
, flags
);
495 if (starget
->starget_sdev_user
)
497 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
498 same_target_siblings
) {
499 if (sdev
== current_sdev
)
501 if (scsi_device_get(sdev
))
504 spin_unlock_irqrestore(shost
->host_lock
, flags
);
505 blk_run_queue(sdev
->request_queue
);
506 spin_lock_irqsave(shost
->host_lock
, flags
);
508 scsi_device_put(sdev
);
511 spin_unlock_irqrestore(shost
->host_lock
, flags
);
515 * Function: scsi_run_queue()
517 * Purpose: Select a proper request queue to serve next
519 * Arguments: q - last request's queue
523 * Notes: The previous command was completely finished, start
524 * a new one if possible.
526 static void scsi_run_queue(struct request_queue
*q
)
528 struct scsi_device
*sdev
= q
->queuedata
;
529 struct Scsi_Host
*shost
= sdev
->host
;
532 if (scsi_target(sdev
)->single_lun
)
533 scsi_single_lun_run(sdev
);
535 spin_lock_irqsave(shost
->host_lock
, flags
);
536 while (!list_empty(&shost
->starved_list
) &&
537 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
538 !((shost
->can_queue
> 0) &&
539 (shost
->host_busy
>= shost
->can_queue
))) {
541 * As long as shost is accepting commands and we have
542 * starved queues, call blk_run_queue. scsi_request_fn
543 * drops the queue_lock and can add us back to the
546 * host_lock protects the starved_list and starved_entry.
547 * scsi_request_fn must get the host_lock before checking
548 * or modifying starved_list or starved_entry.
550 sdev
= list_entry(shost
->starved_list
.next
,
551 struct scsi_device
, starved_entry
);
552 list_del_init(&sdev
->starved_entry
);
553 spin_unlock_irqrestore(shost
->host_lock
, flags
);
556 if (test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
) &&
557 !test_and_set_bit(QUEUE_FLAG_REENTER
,
558 &sdev
->request_queue
->queue_flags
)) {
559 blk_run_queue(sdev
->request_queue
);
560 clear_bit(QUEUE_FLAG_REENTER
,
561 &sdev
->request_queue
->queue_flags
);
563 blk_run_queue(sdev
->request_queue
);
565 spin_lock_irqsave(shost
->host_lock
, flags
);
566 if (unlikely(!list_empty(&sdev
->starved_entry
)))
568 * sdev lost a race, and was put back on the
569 * starved list. This is unlikely but without this
570 * in theory we could loop forever.
574 spin_unlock_irqrestore(shost
->host_lock
, flags
);
580 * Function: scsi_requeue_command()
582 * Purpose: Handle post-processing of completed commands.
584 * Arguments: q - queue to operate on
585 * cmd - command that may need to be requeued.
589 * Notes: After command completion, there may be blocks left
590 * over which weren't finished by the previous command
591 * this can be for a number of reasons - the main one is
592 * I/O errors in the middle of the request, in which case
593 * we need to request the blocks that come after the bad
595 * Notes: Upon return, cmd is a stale pointer.
597 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
599 struct request
*req
= cmd
->request
;
602 scsi_unprep_request(req
);
603 spin_lock_irqsave(q
->queue_lock
, flags
);
604 blk_requeue_request(q
, req
);
605 spin_unlock_irqrestore(q
->queue_lock
, flags
);
610 void scsi_next_command(struct scsi_cmnd
*cmd
)
612 struct scsi_device
*sdev
= cmd
->device
;
613 struct request_queue
*q
= sdev
->request_queue
;
615 /* need to hold a reference on the device before we let go of the cmd */
616 get_device(&sdev
->sdev_gendev
);
618 scsi_put_command(cmd
);
621 /* ok to remove device now */
622 put_device(&sdev
->sdev_gendev
);
625 void scsi_run_host_queues(struct Scsi_Host
*shost
)
627 struct scsi_device
*sdev
;
629 shost_for_each_device(sdev
, shost
)
630 scsi_run_queue(sdev
->request_queue
);
634 * Function: scsi_end_request()
636 * Purpose: Post-processing of completed commands (usually invoked at end
637 * of upper level post-processing and scsi_io_completion).
639 * Arguments: cmd - command that is complete.
640 * error - 0 if I/O indicates success, < 0 for I/O error.
641 * bytes - number of bytes of completed I/O
642 * requeue - indicates whether we should requeue leftovers.
644 * Lock status: Assumed that lock is not held upon entry.
646 * Returns: cmd if requeue required, NULL otherwise.
648 * Notes: This is called for block device requests in order to
649 * mark some number of sectors as complete.
651 * We are guaranteeing that the request queue will be goosed
652 * at some point during this call.
653 * Notes: If cmd was requeued, upon return it will be a stale pointer.
655 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int error
,
656 int bytes
, int requeue
)
658 struct request_queue
*q
= cmd
->device
->request_queue
;
659 struct request
*req
= cmd
->request
;
662 * If there are blocks left over at the end, set up the command
663 * to queue the remainder of them.
665 if (blk_end_request(req
, error
, bytes
)) {
666 int leftover
= (req
->hard_nr_sectors
<< 9);
668 if (blk_pc_request(req
))
669 leftover
= req
->data_len
;
671 /* kill remainder if no retrys */
672 if (error
&& blk_noretry_request(req
))
673 blk_end_request(req
, error
, leftover
);
677 * Bleah. Leftovers again. Stick the
678 * leftovers in the front of the
679 * queue, and goose the queue again.
681 scsi_requeue_command(q
, cmd
);
689 * This will goose the queue request function at the end, so we don't
690 * need to worry about launching another command.
692 scsi_next_command(cmd
);
696 static inline unsigned int scsi_sgtable_index(unsigned short nents
)
700 BUG_ON(nents
> SCSI_MAX_SG_SEGMENTS
);
705 index
= get_count_order(nents
) - 3;
710 static void scsi_sg_free(struct scatterlist
*sgl
, unsigned int nents
)
712 struct scsi_host_sg_pool
*sgp
;
714 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
715 mempool_free(sgl
, sgp
->pool
);
718 static struct scatterlist
*scsi_sg_alloc(unsigned int nents
, gfp_t gfp_mask
)
720 struct scsi_host_sg_pool
*sgp
;
722 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
723 return mempool_alloc(sgp
->pool
, gfp_mask
);
726 static int scsi_alloc_sgtable(struct scsi_data_buffer
*sdb
, int nents
,
733 ret
= __sg_alloc_table(&sdb
->table
, nents
, SCSI_MAX_SG_SEGMENTS
,
734 gfp_mask
, scsi_sg_alloc
);
736 __sg_free_table(&sdb
->table
, SCSI_MAX_SG_SEGMENTS
,
742 static void scsi_free_sgtable(struct scsi_data_buffer
*sdb
)
744 __sg_free_table(&sdb
->table
, SCSI_MAX_SG_SEGMENTS
, scsi_sg_free
);
748 * Function: scsi_release_buffers()
750 * Purpose: Completion processing for block device I/O requests.
752 * Arguments: cmd - command that we are bailing.
754 * Lock status: Assumed that no lock is held upon entry.
758 * Notes: In the event that an upper level driver rejects a
759 * command, we must release resources allocated during
760 * the __init_io() function. Primarily this would involve
761 * the scatter-gather table, and potentially any bounce
764 void scsi_release_buffers(struct scsi_cmnd
*cmd
)
766 if (cmd
->sdb
.table
.nents
)
767 scsi_free_sgtable(&cmd
->sdb
);
769 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
771 if (scsi_bidi_cmnd(cmd
)) {
772 struct scsi_data_buffer
*bidi_sdb
=
773 cmd
->request
->next_rq
->special
;
774 scsi_free_sgtable(bidi_sdb
);
775 kmem_cache_free(scsi_bidi_sdb_cache
, bidi_sdb
);
776 cmd
->request
->next_rq
->special
= NULL
;
779 EXPORT_SYMBOL(scsi_release_buffers
);
782 * Bidi commands Must be complete as a whole, both sides at once.
783 * If part of the bytes were written and lld returned
784 * scsi_in()->resid and/or scsi_out()->resid this information will be left
785 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
786 * decide what to do with this information.
788 void scsi_end_bidi_request(struct scsi_cmnd
*cmd
)
790 struct request
*req
= cmd
->request
;
791 unsigned int dlen
= req
->data_len
;
792 unsigned int next_dlen
= req
->next_rq
->data_len
;
794 req
->data_len
= scsi_out(cmd
)->resid
;
795 req
->next_rq
->data_len
= scsi_in(cmd
)->resid
;
797 /* The req and req->next_rq have not been completed */
798 BUG_ON(blk_end_bidi_request(req
, 0, dlen
, next_dlen
));
800 scsi_release_buffers(cmd
);
803 * This will goose the queue request function at the end, so we don't
804 * need to worry about launching another command.
806 scsi_next_command(cmd
);
810 * Function: scsi_io_completion()
812 * Purpose: Completion processing for block device I/O requests.
814 * Arguments: cmd - command that is finished.
816 * Lock status: Assumed that no lock is held upon entry.
820 * Notes: This function is matched in terms of capabilities to
821 * the function that created the scatter-gather list.
822 * In other words, if there are no bounce buffers
823 * (the normal case for most drivers), we don't need
824 * the logic to deal with cleaning up afterwards.
826 * We must do one of several things here:
828 * a) Call scsi_end_request. This will finish off the
829 * specified number of sectors. If we are done, the
830 * command block will be released, and the queue
831 * function will be goosed. If we are not done, then
832 * scsi_end_request will directly goose the queue.
834 * b) We can just use scsi_requeue_command() here. This would
835 * be used if we just wanted to retry, for example.
837 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
839 int result
= cmd
->result
;
840 int this_count
= scsi_bufflen(cmd
);
841 struct request_queue
*q
= cmd
->device
->request_queue
;
842 struct request
*req
= cmd
->request
;
843 int clear_errors
= 1;
844 struct scsi_sense_hdr sshdr
;
846 int sense_deferred
= 0;
849 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
851 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
854 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
855 req
->errors
= result
;
858 if (sense_valid
&& req
->sense
) {
860 * SG_IO wants current and deferred errors
862 int len
= 8 + cmd
->sense_buffer
[7];
864 if (len
> SCSI_SENSE_BUFFERSIZE
)
865 len
= SCSI_SENSE_BUFFERSIZE
;
866 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
867 req
->sense_len
= len
;
870 if (scsi_bidi_cmnd(cmd
)) {
871 /* will also release_buffers */
872 scsi_end_bidi_request(cmd
);
875 req
->data_len
= scsi_get_resid(cmd
);
878 BUG_ON(blk_bidi_rq(req
)); /* bidi not support for !blk_pc_request yet */
879 scsi_release_buffers(cmd
);
882 * Next deal with any sectors which we were able to correctly
885 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
887 req
->nr_sectors
, good_bytes
));
892 /* A number of bytes were successfully read. If there
893 * are leftovers and there is some kind of error
894 * (result != 0), retry the rest.
896 if (scsi_end_request(cmd
, 0, good_bytes
, result
== 0) == NULL
)
899 /* good_bytes = 0, or (inclusive) there were leftovers and
900 * result = 0, so scsi_end_request couldn't retry.
902 if (sense_valid
&& !sense_deferred
) {
903 switch (sshdr
.sense_key
) {
905 if (cmd
->device
->removable
) {
906 /* Detected disc change. Set a bit
907 * and quietly refuse further access.
909 cmd
->device
->changed
= 1;
910 scsi_end_request(cmd
, -EIO
, this_count
, 1);
913 /* Must have been a power glitch, or a
914 * bus reset. Could not have been a
915 * media change, so we just retry the
916 * request and see what happens.
918 scsi_requeue_command(q
, cmd
);
922 case ILLEGAL_REQUEST
:
923 /* If we had an ILLEGAL REQUEST returned, then
924 * we may have performed an unsupported
925 * command. The only thing this should be
926 * would be a ten byte read where only a six
927 * byte read was supported. Also, on a system
928 * where READ CAPACITY failed, we may have
929 * read past the end of the disk.
931 if ((cmd
->device
->use_10_for_rw
&&
932 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
933 (cmd
->cmnd
[0] == READ_10
||
934 cmd
->cmnd
[0] == WRITE_10
)) {
935 cmd
->device
->use_10_for_rw
= 0;
936 /* This will cause a retry with a
939 scsi_requeue_command(q
, cmd
);
942 scsi_end_request(cmd
, -EIO
, this_count
, 1);
947 /* If the device is in the process of becoming
948 * ready, or has a temporary blockage, retry.
950 if (sshdr
.asc
== 0x04) {
951 switch (sshdr
.ascq
) {
952 case 0x01: /* becoming ready */
953 case 0x04: /* format in progress */
954 case 0x05: /* rebuild in progress */
955 case 0x06: /* recalculation in progress */
956 case 0x07: /* operation in progress */
957 case 0x08: /* Long write in progress */
958 case 0x09: /* self test in progress */
959 scsi_requeue_command(q
, cmd
);
965 if (!(req
->cmd_flags
& REQ_QUIET
))
966 scsi_cmd_print_sense_hdr(cmd
,
970 scsi_end_request(cmd
, -EIO
, this_count
, 1);
972 case VOLUME_OVERFLOW
:
973 if (!(req
->cmd_flags
& REQ_QUIET
)) {
974 scmd_printk(KERN_INFO
, cmd
,
975 "Volume overflow, CDB: ");
976 __scsi_print_command(cmd
->cmnd
);
977 scsi_print_sense("", cmd
);
979 /* See SSC3rXX or current. */
980 scsi_end_request(cmd
, -EIO
, this_count
, 1);
986 if (host_byte(result
) == DID_RESET
) {
987 /* Third party bus reset or reset for error recovery
988 * reasons. Just retry the request and see what
991 scsi_requeue_command(q
, cmd
);
995 if (!(req
->cmd_flags
& REQ_QUIET
)) {
996 scsi_print_result(cmd
);
997 if (driver_byte(result
) & DRIVER_SENSE
)
998 scsi_print_sense("", cmd
);
1001 scsi_end_request(cmd
, -EIO
, this_count
, !result
);
1004 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
1010 * If sg table allocation fails, requeue request later.
1012 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
1014 return BLKPREP_DEFER
;
1018 if (blk_pc_request(req
))
1019 sdb
->length
= req
->data_len
;
1021 sdb
->length
= req
->nr_sectors
<< 9;
1024 * Next, walk the list, and fill in the addresses and sizes of
1027 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1028 BUG_ON(count
> sdb
->table
.nents
);
1029 sdb
->table
.nents
= count
;
1034 * Function: scsi_init_io()
1036 * Purpose: SCSI I/O initialize function.
1038 * Arguments: cmd - Command descriptor we wish to initialize
1040 * Returns: 0 on success
1041 * BLKPREP_DEFER if the failure is retryable
1042 * BLKPREP_KILL if the failure is fatal
1044 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
1046 int error
= scsi_init_sgtable(cmd
->request
, &cmd
->sdb
, gfp_mask
);
1050 if (blk_bidi_rq(cmd
->request
)) {
1051 struct scsi_data_buffer
*bidi_sdb
= kmem_cache_zalloc(
1052 scsi_bidi_sdb_cache
, GFP_ATOMIC
);
1054 error
= BLKPREP_DEFER
;
1058 cmd
->request
->next_rq
->special
= bidi_sdb
;
1059 error
= scsi_init_sgtable(cmd
->request
->next_rq
, bidi_sdb
,
1068 scsi_release_buffers(cmd
);
1069 if (error
== BLKPREP_KILL
)
1070 scsi_put_command(cmd
);
1071 else /* BLKPREP_DEFER */
1072 scsi_unprep_request(cmd
->request
);
1076 EXPORT_SYMBOL(scsi_init_io
);
1078 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1079 struct request
*req
)
1081 struct scsi_cmnd
*cmd
;
1083 if (!req
->special
) {
1084 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1092 /* pull a tag out of the request if we have one */
1093 cmd
->tag
= req
->tag
;
1099 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1101 struct scsi_cmnd
*cmd
;
1102 int ret
= scsi_prep_state_check(sdev
, req
);
1104 if (ret
!= BLKPREP_OK
)
1107 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1109 return BLKPREP_DEFER
;
1112 * BLOCK_PC requests may transfer data, in which case they must
1113 * a bio attached to them. Or they might contain a SCSI command
1114 * that does not transfer data, in which case they may optionally
1115 * submit a request without an attached bio.
1120 BUG_ON(!req
->nr_phys_segments
);
1122 ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1126 BUG_ON(req
->data_len
);
1129 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1133 BUILD_BUG_ON(sizeof(req
->cmd
) > sizeof(cmd
->cmnd
));
1134 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1135 cmd
->cmd_len
= req
->cmd_len
;
1137 cmd
->sc_data_direction
= DMA_NONE
;
1138 else if (rq_data_dir(req
) == WRITE
)
1139 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1141 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1143 cmd
->transfersize
= req
->data_len
;
1144 cmd
->allowed
= req
->retries
;
1145 cmd
->timeout_per_command
= req
->timeout
;
1148 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1151 * Setup a REQ_TYPE_FS command. These are simple read/write request
1152 * from filesystems that still need to be translated to SCSI CDBs from
1155 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1157 struct scsi_cmnd
*cmd
;
1158 int ret
= scsi_prep_state_check(sdev
, req
);
1160 if (ret
!= BLKPREP_OK
)
1163 * Filesystem requests must transfer data.
1165 BUG_ON(!req
->nr_phys_segments
);
1167 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1169 return BLKPREP_DEFER
;
1171 return scsi_init_io(cmd
, GFP_ATOMIC
);
1173 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1175 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1177 int ret
= BLKPREP_OK
;
1180 * If the device is not in running state we will reject some
1183 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1184 switch (sdev
->sdev_state
) {
1187 * If the device is offline we refuse to process any
1188 * commands. The device must be brought online
1189 * before trying any recovery commands.
1191 sdev_printk(KERN_ERR
, sdev
,
1192 "rejecting I/O to offline device\n");
1197 * If the device is fully deleted, we refuse to
1198 * process any commands as well.
1200 sdev_printk(KERN_ERR
, sdev
,
1201 "rejecting I/O to dead device\n");
1207 * If the devices is blocked we defer normal commands.
1209 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1210 ret
= BLKPREP_DEFER
;
1214 * For any other not fully online state we only allow
1215 * special commands. In particular any user initiated
1216 * command is not allowed.
1218 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1225 EXPORT_SYMBOL(scsi_prep_state_check
);
1227 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1229 struct scsi_device
*sdev
= q
->queuedata
;
1233 req
->errors
= DID_NO_CONNECT
<< 16;
1234 /* release the command and kill it */
1236 struct scsi_cmnd
*cmd
= req
->special
;
1237 scsi_release_buffers(cmd
);
1238 scsi_put_command(cmd
);
1239 req
->special
= NULL
;
1244 * If we defer, the elv_next_request() returns NULL, but the
1245 * queue must be restarted, so we plug here if no returning
1246 * command will automatically do that.
1248 if (sdev
->device_busy
== 0)
1252 req
->cmd_flags
|= REQ_DONTPREP
;
1257 EXPORT_SYMBOL(scsi_prep_return
);
1259 int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1261 struct scsi_device
*sdev
= q
->queuedata
;
1262 int ret
= BLKPREP_KILL
;
1264 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1265 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1266 return scsi_prep_return(q
, req
, ret
);
1270 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1273 * Called with the queue_lock held.
1275 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1276 struct scsi_device
*sdev
)
1278 if (sdev
->device_busy
>= sdev
->queue_depth
)
1280 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1282 * unblock after device_blocked iterates to zero
1284 if (--sdev
->device_blocked
== 0) {
1286 sdev_printk(KERN_INFO
, sdev
,
1287 "unblocking device at zero depth\n"));
1293 if (sdev
->device_blocked
)
1300 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1301 * return 0. We must end up running the queue again whenever 0 is
1302 * returned, else IO can hang.
1304 * Called with host_lock held.
1306 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1307 struct Scsi_Host
*shost
,
1308 struct scsi_device
*sdev
)
1310 if (scsi_host_in_recovery(shost
))
1312 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1314 * unblock after host_blocked iterates to zero
1316 if (--shost
->host_blocked
== 0) {
1318 printk("scsi%d unblocking host at zero depth\n",
1325 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1326 shost
->host_blocked
|| shost
->host_self_blocked
) {
1327 if (list_empty(&sdev
->starved_entry
))
1328 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1332 /* We're OK to process the command, so we can't be starved */
1333 if (!list_empty(&sdev
->starved_entry
))
1334 list_del_init(&sdev
->starved_entry
);
1340 * Kill a request for a dead device
1342 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1344 struct scsi_cmnd
*cmd
= req
->special
;
1345 struct scsi_device
*sdev
= cmd
->device
;
1346 struct Scsi_Host
*shost
= sdev
->host
;
1348 blkdev_dequeue_request(req
);
1350 if (unlikely(cmd
== NULL
)) {
1351 printk(KERN_CRIT
"impossible request in %s.\n",
1356 scsi_init_cmd_errh(cmd
);
1357 cmd
->result
= DID_NO_CONNECT
<< 16;
1358 atomic_inc(&cmd
->device
->iorequest_cnt
);
1361 * SCSI request completion path will do scsi_device_unbusy(),
1362 * bump busy counts. To bump the counters, we need to dance
1363 * with the locks as normal issue path does.
1365 sdev
->device_busy
++;
1366 spin_unlock(sdev
->request_queue
->queue_lock
);
1367 spin_lock(shost
->host_lock
);
1369 spin_unlock(shost
->host_lock
);
1370 spin_lock(sdev
->request_queue
->queue_lock
);
1375 static void scsi_softirq_done(struct request
*rq
)
1377 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1378 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1381 INIT_LIST_HEAD(&cmd
->eh_entry
);
1383 disposition
= scsi_decide_disposition(cmd
);
1384 if (disposition
!= SUCCESS
&&
1385 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1386 sdev_printk(KERN_ERR
, cmd
->device
,
1387 "timing out command, waited %lus\n",
1389 disposition
= SUCCESS
;
1392 scsi_log_completion(cmd
, disposition
);
1394 switch (disposition
) {
1396 scsi_finish_command(cmd
);
1399 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1401 case ADD_TO_MLQUEUE
:
1402 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1405 if (!scsi_eh_scmd_add(cmd
, 0))
1406 scsi_finish_command(cmd
);
1411 * Function: scsi_request_fn()
1413 * Purpose: Main strategy routine for SCSI.
1415 * Arguments: q - Pointer to actual queue.
1419 * Lock status: IO request lock assumed to be held when called.
1421 static void scsi_request_fn(struct request_queue
*q
)
1423 struct scsi_device
*sdev
= q
->queuedata
;
1424 struct Scsi_Host
*shost
;
1425 struct scsi_cmnd
*cmd
;
1426 struct request
*req
;
1429 printk("scsi: killing requests for dead queue\n");
1430 while ((req
= elv_next_request(q
)) != NULL
)
1431 scsi_kill_request(req
, q
);
1435 if(!get_device(&sdev
->sdev_gendev
))
1436 /* We must be tearing the block queue down already */
1440 * To start with, we keep looping until the queue is empty, or until
1441 * the host is no longer able to accept any more requests.
1444 while (!blk_queue_plugged(q
)) {
1447 * get next queueable request. We do this early to make sure
1448 * that the request is fully prepared even if we cannot
1451 req
= elv_next_request(q
);
1452 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1455 if (unlikely(!scsi_device_online(sdev
))) {
1456 sdev_printk(KERN_ERR
, sdev
,
1457 "rejecting I/O to offline device\n");
1458 scsi_kill_request(req
, q
);
1464 * Remove the request from the request list.
1466 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1467 blkdev_dequeue_request(req
);
1468 sdev
->device_busy
++;
1470 spin_unlock(q
->queue_lock
);
1472 if (unlikely(cmd
== NULL
)) {
1473 printk(KERN_CRIT
"impossible request in %s.\n"
1474 "please mail a stack trace to "
1475 "linux-scsi@vger.kernel.org\n",
1477 blk_dump_rq_flags(req
, "foo");
1480 spin_lock(shost
->host_lock
);
1482 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1484 if (scsi_target(sdev
)->single_lun
) {
1485 if (scsi_target(sdev
)->starget_sdev_user
&&
1486 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1488 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1493 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1494 * take the lock again.
1496 spin_unlock_irq(shost
->host_lock
);
1499 * Finally, initialize any error handling parameters, and set up
1500 * the timers for timeouts.
1502 scsi_init_cmd_errh(cmd
);
1505 * Dispatch the command to the low-level driver.
1507 rtn
= scsi_dispatch_cmd(cmd
);
1508 spin_lock_irq(q
->queue_lock
);
1510 /* we're refusing the command; because of
1511 * the way locks get dropped, we need to
1512 * check here if plugging is required */
1513 if(sdev
->device_busy
== 0)
1523 spin_unlock_irq(shost
->host_lock
);
1526 * lock q, handle tag, requeue req, and decrement device_busy. We
1527 * must return with queue_lock held.
1529 * Decrementing device_busy without checking it is OK, as all such
1530 * cases (host limits or settings) should run the queue at some
1533 spin_lock_irq(q
->queue_lock
);
1534 blk_requeue_request(q
, req
);
1535 sdev
->device_busy
--;
1536 if(sdev
->device_busy
== 0)
1539 /* must be careful here...if we trigger the ->remove() function
1540 * we cannot be holding the q lock */
1541 spin_unlock_irq(q
->queue_lock
);
1542 put_device(&sdev
->sdev_gendev
);
1543 spin_lock_irq(q
->queue_lock
);
1546 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1548 struct device
*host_dev
;
1549 u64 bounce_limit
= 0xffffffff;
1551 if (shost
->unchecked_isa_dma
)
1552 return BLK_BOUNCE_ISA
;
1554 * Platforms with virtual-DMA translation
1555 * hardware have no practical limit.
1557 if (!PCI_DMA_BUS_IS_PHYS
)
1558 return BLK_BOUNCE_ANY
;
1560 host_dev
= scsi_get_device(shost
);
1561 if (host_dev
&& host_dev
->dma_mask
)
1562 bounce_limit
= *host_dev
->dma_mask
;
1564 return bounce_limit
;
1566 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1568 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1569 request_fn_proc
*request_fn
)
1571 struct request_queue
*q
;
1573 q
= blk_init_queue(request_fn
, NULL
);
1578 * this limit is imposed by hardware restrictions
1580 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1581 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_CHAIN_SEGMENTS
);
1583 blk_queue_max_sectors(q
, shost
->max_sectors
);
1584 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1585 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1587 if (!shost
->use_clustering
)
1588 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1591 * set a reasonable default alignment on word boundaries: the
1592 * host and device may alter it using
1593 * blk_queue_update_dma_alignment() later.
1595 blk_queue_dma_alignment(q
, 0x03);
1599 EXPORT_SYMBOL(__scsi_alloc_queue
);
1601 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1603 struct request_queue
*q
;
1605 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1609 blk_queue_prep_rq(q
, scsi_prep_fn
);
1610 blk_queue_softirq_done(q
, scsi_softirq_done
);
1614 void scsi_free_queue(struct request_queue
*q
)
1616 blk_cleanup_queue(q
);
1620 * Function: scsi_block_requests()
1622 * Purpose: Utility function used by low-level drivers to prevent further
1623 * commands from being queued to the device.
1625 * Arguments: shost - Host in question
1629 * Lock status: No locks are assumed held.
1631 * Notes: There is no timer nor any other means by which the requests
1632 * get unblocked other than the low-level driver calling
1633 * scsi_unblock_requests().
1635 void scsi_block_requests(struct Scsi_Host
*shost
)
1637 shost
->host_self_blocked
= 1;
1639 EXPORT_SYMBOL(scsi_block_requests
);
1642 * Function: scsi_unblock_requests()
1644 * Purpose: Utility function used by low-level drivers to allow further
1645 * commands from being queued to the device.
1647 * Arguments: shost - Host in question
1651 * Lock status: No locks are assumed held.
1653 * Notes: There is no timer nor any other means by which the requests
1654 * get unblocked other than the low-level driver calling
1655 * scsi_unblock_requests().
1657 * This is done as an API function so that changes to the
1658 * internals of the scsi mid-layer won't require wholesale
1659 * changes to drivers that use this feature.
1661 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1663 shost
->host_self_blocked
= 0;
1664 scsi_run_host_queues(shost
);
1666 EXPORT_SYMBOL(scsi_unblock_requests
);
1668 int __init
scsi_init_queue(void)
1672 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1673 sizeof(struct scsi_io_context
),
1675 if (!scsi_io_context_cache
) {
1676 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1680 scsi_bidi_sdb_cache
= kmem_cache_create("scsi_bidi_sdb",
1681 sizeof(struct scsi_data_buffer
),
1683 if (!scsi_bidi_sdb_cache
) {
1684 printk(KERN_ERR
"SCSI: can't init scsi bidi sdb cache\n");
1685 goto cleanup_io_context
;
1688 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1689 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1690 int size
= sgp
->size
* sizeof(struct scatterlist
);
1692 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1693 SLAB_HWCACHE_ALIGN
, NULL
);
1695 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1697 goto cleanup_bidi_sdb
;
1700 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1703 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1705 goto cleanup_bidi_sdb
;
1712 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1713 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1715 mempool_destroy(sgp
->pool
);
1717 kmem_cache_destroy(sgp
->slab
);
1719 kmem_cache_destroy(scsi_bidi_sdb_cache
);
1721 kmem_cache_destroy(scsi_io_context_cache
);
1726 void scsi_exit_queue(void)
1730 kmem_cache_destroy(scsi_io_context_cache
);
1731 kmem_cache_destroy(scsi_bidi_sdb_cache
);
1733 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1734 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1735 mempool_destroy(sgp
->pool
);
1736 kmem_cache_destroy(sgp
->slab
);
1741 * scsi_mode_select - issue a mode select
1742 * @sdev: SCSI device to be queried
1743 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1744 * @sp: Save page bit (0 == don't save, 1 == save)
1745 * @modepage: mode page being requested
1746 * @buffer: request buffer (may not be smaller than eight bytes)
1747 * @len: length of request buffer.
1748 * @timeout: command timeout
1749 * @retries: number of retries before failing
1750 * @data: returns a structure abstracting the mode header data
1751 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1752 * must be SCSI_SENSE_BUFFERSIZE big.
1754 * Returns zero if successful; negative error number or scsi
1759 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1760 unsigned char *buffer
, int len
, int timeout
, int retries
,
1761 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1763 unsigned char cmd
[10];
1764 unsigned char *real_buffer
;
1767 memset(cmd
, 0, sizeof(cmd
));
1768 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1770 if (sdev
->use_10_for_ms
) {
1773 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1776 memcpy(real_buffer
+ 8, buffer
, len
);
1780 real_buffer
[2] = data
->medium_type
;
1781 real_buffer
[3] = data
->device_specific
;
1782 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1784 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1785 real_buffer
[7] = data
->block_descriptor_length
;
1787 cmd
[0] = MODE_SELECT_10
;
1791 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1795 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1798 memcpy(real_buffer
+ 4, buffer
, len
);
1801 real_buffer
[1] = data
->medium_type
;
1802 real_buffer
[2] = data
->device_specific
;
1803 real_buffer
[3] = data
->block_descriptor_length
;
1806 cmd
[0] = MODE_SELECT
;
1810 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1811 sshdr
, timeout
, retries
);
1815 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1818 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1819 * @sdev: SCSI device to be queried
1820 * @dbd: set if mode sense will allow block descriptors to be returned
1821 * @modepage: mode page being requested
1822 * @buffer: request buffer (may not be smaller than eight bytes)
1823 * @len: length of request buffer.
1824 * @timeout: command timeout
1825 * @retries: number of retries before failing
1826 * @data: returns a structure abstracting the mode header data
1827 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1828 * must be SCSI_SENSE_BUFFERSIZE big.
1830 * Returns zero if unsuccessful, or the header offset (either 4
1831 * or 8 depending on whether a six or ten byte command was
1832 * issued) if successful.
1835 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1836 unsigned char *buffer
, int len
, int timeout
, int retries
,
1837 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1839 unsigned char cmd
[12];
1843 struct scsi_sense_hdr my_sshdr
;
1845 memset(data
, 0, sizeof(*data
));
1846 memset(&cmd
[0], 0, 12);
1847 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1850 /* caller might not be interested in sense, but we need it */
1855 use_10_for_ms
= sdev
->use_10_for_ms
;
1857 if (use_10_for_ms
) {
1861 cmd
[0] = MODE_SENSE_10
;
1868 cmd
[0] = MODE_SENSE
;
1873 memset(buffer
, 0, len
);
1875 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1876 sshdr
, timeout
, retries
);
1878 /* This code looks awful: what it's doing is making sure an
1879 * ILLEGAL REQUEST sense return identifies the actual command
1880 * byte as the problem. MODE_SENSE commands can return
1881 * ILLEGAL REQUEST if the code page isn't supported */
1883 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1884 (driver_byte(result
) & DRIVER_SENSE
)) {
1885 if (scsi_sense_valid(sshdr
)) {
1886 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1887 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1889 * Invalid command operation code
1891 sdev
->use_10_for_ms
= 0;
1897 if(scsi_status_is_good(result
)) {
1898 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1899 (modepage
== 6 || modepage
== 8))) {
1900 /* Initio breakage? */
1903 data
->medium_type
= 0;
1904 data
->device_specific
= 0;
1906 data
->block_descriptor_length
= 0;
1907 } else if(use_10_for_ms
) {
1908 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1909 data
->medium_type
= buffer
[2];
1910 data
->device_specific
= buffer
[3];
1911 data
->longlba
= buffer
[4] & 0x01;
1912 data
->block_descriptor_length
= buffer
[6]*256
1915 data
->length
= buffer
[0] + 1;
1916 data
->medium_type
= buffer
[1];
1917 data
->device_specific
= buffer
[2];
1918 data
->block_descriptor_length
= buffer
[3];
1920 data
->header_length
= header_length
;
1925 EXPORT_SYMBOL(scsi_mode_sense
);
1928 * scsi_test_unit_ready - test if unit is ready
1929 * @sdev: scsi device to change the state of.
1930 * @timeout: command timeout
1931 * @retries: number of retries before failing
1932 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1933 * returning sense. Make sure that this is cleared before passing
1936 * Returns zero if unsuccessful or an error if TUR failed. For
1937 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1938 * translated to success, with the ->changed flag updated.
1941 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
1942 struct scsi_sense_hdr
*sshdr_external
)
1945 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1947 struct scsi_sense_hdr
*sshdr
;
1950 if (!sshdr_external
)
1951 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1953 sshdr
= sshdr_external
;
1955 /* try to eat the UNIT_ATTENTION if there are enough retries */
1957 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
1959 } while ((driver_byte(result
) & DRIVER_SENSE
) &&
1960 sshdr
&& sshdr
->sense_key
== UNIT_ATTENTION
&&
1964 /* could not allocate sense buffer, so can't process it */
1967 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1969 if ((scsi_sense_valid(sshdr
)) &&
1970 ((sshdr
->sense_key
== UNIT_ATTENTION
) ||
1971 (sshdr
->sense_key
== NOT_READY
))) {
1976 if (!sshdr_external
)
1980 EXPORT_SYMBOL(scsi_test_unit_ready
);
1983 * scsi_device_set_state - Take the given device through the device state model.
1984 * @sdev: scsi device to change the state of.
1985 * @state: state to change to.
1987 * Returns zero if unsuccessful or an error if the requested
1988 * transition is illegal.
1991 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
1993 enum scsi_device_state oldstate
= sdev
->sdev_state
;
1995 if (state
== oldstate
)
2000 /* There are no legal states that come back to
2001 * created. This is the manually initialised start
2075 sdev
->sdev_state
= state
;
2079 SCSI_LOG_ERROR_RECOVERY(1,
2080 sdev_printk(KERN_ERR
, sdev
,
2081 "Illegal state transition %s->%s\n",
2082 scsi_device_state_name(oldstate
),
2083 scsi_device_state_name(state
))
2087 EXPORT_SYMBOL(scsi_device_set_state
);
2090 * sdev_evt_emit - emit a single SCSI device uevent
2091 * @sdev: associated SCSI device
2092 * @evt: event to emit
2094 * Send a single uevent (scsi_event) to the associated scsi_device.
2096 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2101 switch (evt
->evt_type
) {
2102 case SDEV_EVT_MEDIA_CHANGE
:
2103 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2113 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2117 * sdev_evt_thread - send a uevent for each scsi event
2118 * @work: work struct for scsi_device
2120 * Dispatch queued events to their associated scsi_device kobjects
2123 void scsi_evt_thread(struct work_struct
*work
)
2125 struct scsi_device
*sdev
;
2126 LIST_HEAD(event_list
);
2128 sdev
= container_of(work
, struct scsi_device
, event_work
);
2131 struct scsi_event
*evt
;
2132 struct list_head
*this, *tmp
;
2133 unsigned long flags
;
2135 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2136 list_splice_init(&sdev
->event_list
, &event_list
);
2137 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2139 if (list_empty(&event_list
))
2142 list_for_each_safe(this, tmp
, &event_list
) {
2143 evt
= list_entry(this, struct scsi_event
, node
);
2144 list_del(&evt
->node
);
2145 scsi_evt_emit(sdev
, evt
);
2152 * sdev_evt_send - send asserted event to uevent thread
2153 * @sdev: scsi_device event occurred on
2154 * @evt: event to send
2156 * Assert scsi device event asynchronously.
2158 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2160 unsigned long flags
;
2162 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2167 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2168 list_add_tail(&evt
->node
, &sdev
->event_list
);
2169 schedule_work(&sdev
->event_work
);
2170 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2172 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2175 * sdev_evt_alloc - allocate a new scsi event
2176 * @evt_type: type of event to allocate
2177 * @gfpflags: GFP flags for allocation
2179 * Allocates and returns a new scsi_event.
2181 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2184 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2188 evt
->evt_type
= evt_type
;
2189 INIT_LIST_HEAD(&evt
->node
);
2191 /* evt_type-specific initialization, if any */
2193 case SDEV_EVT_MEDIA_CHANGE
:
2201 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2204 * sdev_evt_send_simple - send asserted event to uevent thread
2205 * @sdev: scsi_device event occurred on
2206 * @evt_type: type of event to send
2207 * @gfpflags: GFP flags for allocation
2209 * Assert scsi device event asynchronously, given an event type.
2211 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2212 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2214 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2216 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2221 sdev_evt_send(sdev
, evt
);
2223 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2226 * scsi_device_quiesce - Block user issued commands.
2227 * @sdev: scsi device to quiesce.
2229 * This works by trying to transition to the SDEV_QUIESCE state
2230 * (which must be a legal transition). When the device is in this
2231 * state, only special requests will be accepted, all others will
2232 * be deferred. Since special requests may also be requeued requests,
2233 * a successful return doesn't guarantee the device will be
2234 * totally quiescent.
2236 * Must be called with user context, may sleep.
2238 * Returns zero if unsuccessful or an error if not.
2241 scsi_device_quiesce(struct scsi_device
*sdev
)
2243 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2247 scsi_run_queue(sdev
->request_queue
);
2248 while (sdev
->device_busy
) {
2249 msleep_interruptible(200);
2250 scsi_run_queue(sdev
->request_queue
);
2254 EXPORT_SYMBOL(scsi_device_quiesce
);
2257 * scsi_device_resume - Restart user issued commands to a quiesced device.
2258 * @sdev: scsi device to resume.
2260 * Moves the device from quiesced back to running and restarts the
2263 * Must be called with user context, may sleep.
2266 scsi_device_resume(struct scsi_device
*sdev
)
2268 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2270 scsi_run_queue(sdev
->request_queue
);
2272 EXPORT_SYMBOL(scsi_device_resume
);
2275 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2277 scsi_device_quiesce(sdev
);
2281 scsi_target_quiesce(struct scsi_target
*starget
)
2283 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2285 EXPORT_SYMBOL(scsi_target_quiesce
);
2288 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2290 scsi_device_resume(sdev
);
2294 scsi_target_resume(struct scsi_target
*starget
)
2296 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2298 EXPORT_SYMBOL(scsi_target_resume
);
2301 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2302 * @sdev: device to block
2304 * Block request made by scsi lld's to temporarily stop all
2305 * scsi commands on the specified device. Called from interrupt
2306 * or normal process context.
2308 * Returns zero if successful or error if not
2311 * This routine transitions the device to the SDEV_BLOCK state
2312 * (which must be a legal transition). When the device is in this
2313 * state, all commands are deferred until the scsi lld reenables
2314 * the device with scsi_device_unblock or device_block_tmo fires.
2315 * This routine assumes the host_lock is held on entry.
2318 scsi_internal_device_block(struct scsi_device
*sdev
)
2320 struct request_queue
*q
= sdev
->request_queue
;
2321 unsigned long flags
;
2324 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2329 * The device has transitioned to SDEV_BLOCK. Stop the
2330 * block layer from calling the midlayer with this device's
2333 spin_lock_irqsave(q
->queue_lock
, flags
);
2335 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2339 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2342 * scsi_internal_device_unblock - resume a device after a block request
2343 * @sdev: device to resume
2345 * Called by scsi lld's or the midlayer to restart the device queue
2346 * for the previously suspended scsi device. Called from interrupt or
2347 * normal process context.
2349 * Returns zero if successful or error if not.
2352 * This routine transitions the device to the SDEV_RUNNING state
2353 * (which must be a legal transition) allowing the midlayer to
2354 * goose the queue for this device. This routine assumes the
2355 * host_lock is held upon entry.
2358 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2360 struct request_queue
*q
= sdev
->request_queue
;
2362 unsigned long flags
;
2365 * Try to transition the scsi device to SDEV_RUNNING
2366 * and goose the device queue if successful.
2368 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2372 spin_lock_irqsave(q
->queue_lock
, flags
);
2374 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2378 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2381 device_block(struct scsi_device
*sdev
, void *data
)
2383 scsi_internal_device_block(sdev
);
2387 target_block(struct device
*dev
, void *data
)
2389 if (scsi_is_target_device(dev
))
2390 starget_for_each_device(to_scsi_target(dev
), NULL
,
2396 scsi_target_block(struct device
*dev
)
2398 if (scsi_is_target_device(dev
))
2399 starget_for_each_device(to_scsi_target(dev
), NULL
,
2402 device_for_each_child(dev
, NULL
, target_block
);
2404 EXPORT_SYMBOL_GPL(scsi_target_block
);
2407 device_unblock(struct scsi_device
*sdev
, void *data
)
2409 scsi_internal_device_unblock(sdev
);
2413 target_unblock(struct device
*dev
, void *data
)
2415 if (scsi_is_target_device(dev
))
2416 starget_for_each_device(to_scsi_target(dev
), NULL
,
2422 scsi_target_unblock(struct device
*dev
)
2424 if (scsi_is_target_device(dev
))
2425 starget_for_each_device(to_scsi_target(dev
), NULL
,
2428 device_for_each_child(dev
, NULL
, target_unblock
);
2430 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2433 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2434 * @sgl: scatter-gather list
2435 * @sg_count: number of segments in sg
2436 * @offset: offset in bytes into sg, on return offset into the mapped area
2437 * @len: bytes to map, on return number of bytes mapped
2439 * Returns virtual address of the start of the mapped page
2441 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2442 size_t *offset
, size_t *len
)
2445 size_t sg_len
= 0, len_complete
= 0;
2446 struct scatterlist
*sg
;
2449 WARN_ON(!irqs_disabled());
2451 for_each_sg(sgl
, sg
, sg_count
, i
) {
2452 len_complete
= sg_len
; /* Complete sg-entries */
2453 sg_len
+= sg
->length
;
2454 if (sg_len
> *offset
)
2458 if (unlikely(i
== sg_count
)) {
2459 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2461 __FUNCTION__
, sg_len
, *offset
, sg_count
);
2466 /* Offset starting from the beginning of first page in this sg-entry */
2467 *offset
= *offset
- len_complete
+ sg
->offset
;
2469 /* Assumption: contiguous pages can be accessed as "page + i" */
2470 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2471 *offset
&= ~PAGE_MASK
;
2473 /* Bytes in this sg-entry from *offset to the end of the page */
2474 sg_len
= PAGE_SIZE
- *offset
;
2478 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2480 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2483 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2484 * @virt: virtual address to be unmapped
2486 void scsi_kunmap_atomic_sg(void *virt
)
2488 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2490 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);