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
) {
305 while (len
> 0 && data_len
> 0) {
307 * sg sends a scatterlist that is larger than
308 * the data_len it wants transferred for certain
311 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
312 bytes
= min(bytes
, data_len
);
315 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
318 bio
= bio_alloc(gfp
, nr_vecs
);
323 bio
->bi_end_io
= scsi_bi_endio
;
326 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
333 if (bio
->bi_vcnt
>= nr_vecs
) {
334 err
= scsi_merge_bio(rq
, bio
);
349 rq
->buffer
= rq
->data
= NULL
;
350 rq
->data_len
= bufflen
;
354 while ((bio
= rq
->bio
) != NULL
) {
355 rq
->bio
= bio
->bi_next
;
357 * call endio instead of bio_put incase it was bounced
366 * scsi_execute_async - insert request
369 * @cmd_len: length of scsi cdb
370 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
371 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
372 * @bufflen: len of buffer
373 * @use_sg: if buffer is a scatterlist this is the number of elements
374 * @timeout: request timeout in seconds
375 * @retries: number of times to retry request
376 * @privdata: data passed to done()
377 * @done: callback function when done
378 * @gfp: memory allocation flags
380 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
381 int cmd_len
, int data_direction
, void *buffer
, unsigned bufflen
,
382 int use_sg
, int timeout
, int retries
, void *privdata
,
383 void (*done
)(void *, char *, int, int), gfp_t gfp
)
386 struct scsi_io_context
*sioc
;
388 int write
= (data_direction
== DMA_TO_DEVICE
);
390 sioc
= kmem_cache_zalloc(scsi_io_context_cache
, gfp
);
392 return DRIVER_ERROR
<< 24;
394 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
397 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
398 req
->cmd_flags
|= REQ_QUIET
;
401 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
403 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
408 req
->cmd_len
= cmd_len
;
409 memset(req
->cmd
, 0, BLK_MAX_CDB
); /* ATAPI hates garbage after CDB */
410 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
411 req
->sense
= sioc
->sense
;
413 req
->timeout
= timeout
;
414 req
->retries
= retries
;
415 req
->end_io_data
= sioc
;
417 sioc
->data
= privdata
;
420 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
424 blk_put_request(req
);
426 kmem_cache_free(scsi_io_context_cache
, sioc
);
427 return DRIVER_ERROR
<< 24;
429 EXPORT_SYMBOL_GPL(scsi_execute_async
);
432 * Function: scsi_init_cmd_errh()
434 * Purpose: Initialize cmd fields related to error handling.
436 * Arguments: cmd - command that is ready to be queued.
438 * Notes: This function has the job of initializing a number of
439 * fields related to error handling. Typically this will
440 * be called once for each command, as required.
442 static void scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
444 cmd
->serial_number
= 0;
445 scsi_set_resid(cmd
, 0);
446 memset(cmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
447 if (cmd
->cmd_len
== 0)
448 cmd
->cmd_len
= scsi_command_size(cmd
->cmnd
);
451 void scsi_device_unbusy(struct scsi_device
*sdev
)
453 struct Scsi_Host
*shost
= sdev
->host
;
456 spin_lock_irqsave(shost
->host_lock
, flags
);
458 if (unlikely(scsi_host_in_recovery(shost
) &&
459 (shost
->host_failed
|| shost
->host_eh_scheduled
)))
460 scsi_eh_wakeup(shost
);
461 spin_unlock(shost
->host_lock
);
462 spin_lock(sdev
->request_queue
->queue_lock
);
464 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
468 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
469 * and call blk_run_queue for all the scsi_devices on the target -
470 * including current_sdev first.
472 * Called with *no* scsi locks held.
474 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
476 struct Scsi_Host
*shost
= current_sdev
->host
;
477 struct scsi_device
*sdev
, *tmp
;
478 struct scsi_target
*starget
= scsi_target(current_sdev
);
481 spin_lock_irqsave(shost
->host_lock
, flags
);
482 starget
->starget_sdev_user
= NULL
;
483 spin_unlock_irqrestore(shost
->host_lock
, flags
);
486 * Call blk_run_queue for all LUNs on the target, starting with
487 * current_sdev. We race with others (to set starget_sdev_user),
488 * but in most cases, we will be first. Ideally, each LU on the
489 * target would get some limited time or requests on the target.
491 blk_run_queue(current_sdev
->request_queue
);
493 spin_lock_irqsave(shost
->host_lock
, flags
);
494 if (starget
->starget_sdev_user
)
496 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
497 same_target_siblings
) {
498 if (sdev
== current_sdev
)
500 if (scsi_device_get(sdev
))
503 spin_unlock_irqrestore(shost
->host_lock
, flags
);
504 blk_run_queue(sdev
->request_queue
);
505 spin_lock_irqsave(shost
->host_lock
, flags
);
507 scsi_device_put(sdev
);
510 spin_unlock_irqrestore(shost
->host_lock
, flags
);
514 * Function: scsi_run_queue()
516 * Purpose: Select a proper request queue to serve next
518 * Arguments: q - last request's queue
522 * Notes: The previous command was completely finished, start
523 * a new one if possible.
525 static void scsi_run_queue(struct request_queue
*q
)
527 struct scsi_device
*sdev
= q
->queuedata
;
528 struct Scsi_Host
*shost
= sdev
->host
;
531 if (scsi_target(sdev
)->single_lun
)
532 scsi_single_lun_run(sdev
);
534 spin_lock_irqsave(shost
->host_lock
, flags
);
535 while (!list_empty(&shost
->starved_list
) &&
536 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
537 !((shost
->can_queue
> 0) &&
538 (shost
->host_busy
>= shost
->can_queue
))) {
543 * As long as shost is accepting commands and we have
544 * starved queues, call blk_run_queue. scsi_request_fn
545 * drops the queue_lock and can add us back to the
548 * host_lock protects the starved_list and starved_entry.
549 * scsi_request_fn must get the host_lock before checking
550 * or modifying starved_list or starved_entry.
552 sdev
= list_entry(shost
->starved_list
.next
,
553 struct scsi_device
, starved_entry
);
554 list_del_init(&sdev
->starved_entry
);
555 spin_unlock(shost
->host_lock
);
557 spin_lock(sdev
->request_queue
->queue_lock
);
558 flagset
= test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
) &&
559 !test_bit(QUEUE_FLAG_REENTER
,
560 &sdev
->request_queue
->queue_flags
);
562 queue_flag_set(QUEUE_FLAG_REENTER
, sdev
->request_queue
);
563 __blk_run_queue(sdev
->request_queue
);
565 queue_flag_clear(QUEUE_FLAG_REENTER
, sdev
->request_queue
);
566 spin_unlock(sdev
->request_queue
->queue_lock
);
568 spin_lock(shost
->host_lock
);
569 if (unlikely(!list_empty(&sdev
->starved_entry
)))
571 * sdev lost a race, and was put back on the
572 * starved list. This is unlikely but without this
573 * in theory we could loop forever.
577 spin_unlock_irqrestore(shost
->host_lock
, flags
);
583 * Function: scsi_requeue_command()
585 * Purpose: Handle post-processing of completed commands.
587 * Arguments: q - queue to operate on
588 * cmd - command that may need to be requeued.
592 * Notes: After command completion, there may be blocks left
593 * over which weren't finished by the previous command
594 * this can be for a number of reasons - the main one is
595 * I/O errors in the middle of the request, in which case
596 * we need to request the blocks that come after the bad
598 * Notes: Upon return, cmd is a stale pointer.
600 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
602 struct request
*req
= cmd
->request
;
605 scsi_unprep_request(req
);
606 spin_lock_irqsave(q
->queue_lock
, flags
);
607 blk_requeue_request(q
, req
);
608 spin_unlock_irqrestore(q
->queue_lock
, flags
);
613 void scsi_next_command(struct scsi_cmnd
*cmd
)
615 struct scsi_device
*sdev
= cmd
->device
;
616 struct request_queue
*q
= sdev
->request_queue
;
618 /* need to hold a reference on the device before we let go of the cmd */
619 get_device(&sdev
->sdev_gendev
);
621 scsi_put_command(cmd
);
624 /* ok to remove device now */
625 put_device(&sdev
->sdev_gendev
);
628 void scsi_run_host_queues(struct Scsi_Host
*shost
)
630 struct scsi_device
*sdev
;
632 shost_for_each_device(sdev
, shost
)
633 scsi_run_queue(sdev
->request_queue
);
637 * Function: scsi_end_request()
639 * Purpose: Post-processing of completed commands (usually invoked at end
640 * of upper level post-processing and scsi_io_completion).
642 * Arguments: cmd - command that is complete.
643 * error - 0 if I/O indicates success, < 0 for I/O error.
644 * bytes - number of bytes of completed I/O
645 * requeue - indicates whether we should requeue leftovers.
647 * Lock status: Assumed that lock is not held upon entry.
649 * Returns: cmd if requeue required, NULL otherwise.
651 * Notes: This is called for block device requests in order to
652 * mark some number of sectors as complete.
654 * We are guaranteeing that the request queue will be goosed
655 * at some point during this call.
656 * Notes: If cmd was requeued, upon return it will be a stale pointer.
658 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int error
,
659 int bytes
, int requeue
)
661 struct request_queue
*q
= cmd
->device
->request_queue
;
662 struct request
*req
= cmd
->request
;
665 * If there are blocks left over at the end, set up the command
666 * to queue the remainder of them.
668 if (blk_end_request(req
, error
, bytes
)) {
669 int leftover
= (req
->hard_nr_sectors
<< 9);
671 if (blk_pc_request(req
))
672 leftover
= req
->data_len
;
674 /* kill remainder if no retrys */
675 if (error
&& blk_noretry_request(req
))
676 blk_end_request(req
, error
, leftover
);
680 * Bleah. Leftovers again. Stick the
681 * leftovers in the front of the
682 * queue, and goose the queue again.
684 scsi_requeue_command(q
, cmd
);
692 * This will goose the queue request function at the end, so we don't
693 * need to worry about launching another command.
695 scsi_next_command(cmd
);
699 static inline unsigned int scsi_sgtable_index(unsigned short nents
)
703 BUG_ON(nents
> SCSI_MAX_SG_SEGMENTS
);
708 index
= get_count_order(nents
) - 3;
713 static void scsi_sg_free(struct scatterlist
*sgl
, unsigned int nents
)
715 struct scsi_host_sg_pool
*sgp
;
717 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
718 mempool_free(sgl
, sgp
->pool
);
721 static struct scatterlist
*scsi_sg_alloc(unsigned int nents
, gfp_t gfp_mask
)
723 struct scsi_host_sg_pool
*sgp
;
725 sgp
= scsi_sg_pools
+ scsi_sgtable_index(nents
);
726 return mempool_alloc(sgp
->pool
, gfp_mask
);
729 static int scsi_alloc_sgtable(struct scsi_data_buffer
*sdb
, int nents
,
736 ret
= __sg_alloc_table(&sdb
->table
, nents
, SCSI_MAX_SG_SEGMENTS
,
737 gfp_mask
, scsi_sg_alloc
);
739 __sg_free_table(&sdb
->table
, SCSI_MAX_SG_SEGMENTS
,
745 static void scsi_free_sgtable(struct scsi_data_buffer
*sdb
)
747 __sg_free_table(&sdb
->table
, SCSI_MAX_SG_SEGMENTS
, scsi_sg_free
);
751 * Function: scsi_release_buffers()
753 * Purpose: Completion processing for block device I/O requests.
755 * Arguments: cmd - command that we are bailing.
757 * Lock status: Assumed that no lock is held upon entry.
761 * Notes: In the event that an upper level driver rejects a
762 * command, we must release resources allocated during
763 * the __init_io() function. Primarily this would involve
764 * the scatter-gather table, and potentially any bounce
767 void scsi_release_buffers(struct scsi_cmnd
*cmd
)
769 if (cmd
->sdb
.table
.nents
)
770 scsi_free_sgtable(&cmd
->sdb
);
772 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
774 if (scsi_bidi_cmnd(cmd
)) {
775 struct scsi_data_buffer
*bidi_sdb
=
776 cmd
->request
->next_rq
->special
;
777 scsi_free_sgtable(bidi_sdb
);
778 kmem_cache_free(scsi_bidi_sdb_cache
, bidi_sdb
);
779 cmd
->request
->next_rq
->special
= NULL
;
782 EXPORT_SYMBOL(scsi_release_buffers
);
785 * Bidi commands Must be complete as a whole, both sides at once.
786 * If part of the bytes were written and lld returned
787 * scsi_in()->resid and/or scsi_out()->resid this information will be left
788 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
789 * decide what to do with this information.
791 static void scsi_end_bidi_request(struct scsi_cmnd
*cmd
)
793 struct request
*req
= cmd
->request
;
794 unsigned int dlen
= req
->data_len
;
795 unsigned int next_dlen
= req
->next_rq
->data_len
;
797 req
->data_len
= scsi_out(cmd
)->resid
;
798 req
->next_rq
->data_len
= scsi_in(cmd
)->resid
;
800 /* The req and req->next_rq have not been completed */
801 BUG_ON(blk_end_bidi_request(req
, 0, dlen
, next_dlen
));
803 scsi_release_buffers(cmd
);
806 * This will goose the queue request function at the end, so we don't
807 * need to worry about launching another command.
809 scsi_next_command(cmd
);
813 * Function: scsi_io_completion()
815 * Purpose: Completion processing for block device I/O requests.
817 * Arguments: cmd - command that is finished.
819 * Lock status: Assumed that no lock is held upon entry.
823 * Notes: This function is matched in terms of capabilities to
824 * the function that created the scatter-gather list.
825 * In other words, if there are no bounce buffers
826 * (the normal case for most drivers), we don't need
827 * the logic to deal with cleaning up afterwards.
829 * We must do one of several things here:
831 * a) Call scsi_end_request. This will finish off the
832 * specified number of sectors. If we are done, the
833 * command block will be released, and the queue
834 * function will be goosed. If we are not done, then
835 * scsi_end_request will directly goose the queue.
837 * b) We can just use scsi_requeue_command() here. This would
838 * be used if we just wanted to retry, for example.
840 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
)
842 int result
= cmd
->result
;
843 int this_count
= scsi_bufflen(cmd
);
844 struct request_queue
*q
= cmd
->device
->request_queue
;
845 struct request
*req
= cmd
->request
;
847 struct scsi_sense_hdr sshdr
;
849 int sense_deferred
= 0;
852 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
854 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
857 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
858 req
->errors
= result
;
860 if (sense_valid
&& req
->sense
) {
862 * SG_IO wants current and deferred errors
864 int len
= 8 + cmd
->sense_buffer
[7];
866 if (len
> SCSI_SENSE_BUFFERSIZE
)
867 len
= SCSI_SENSE_BUFFERSIZE
;
868 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
869 req
->sense_len
= len
;
874 if (scsi_bidi_cmnd(cmd
)) {
875 /* will also release_buffers */
876 scsi_end_bidi_request(cmd
);
879 req
->data_len
= scsi_get_resid(cmd
);
882 BUG_ON(blk_bidi_rq(req
)); /* bidi not support for !blk_pc_request yet */
883 scsi_release_buffers(cmd
);
886 * Next deal with any sectors which we were able to correctly
889 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
891 req
->nr_sectors
, good_bytes
));
893 /* A number of bytes were successfully read. If there
894 * are leftovers and there is some kind of error
895 * (result != 0), retry the rest.
897 if (scsi_end_request(cmd
, error
, good_bytes
, result
== 0) == NULL
)
900 /* good_bytes = 0, or (inclusive) there were leftovers and
901 * result = 0, so scsi_end_request couldn't retry.
903 if (sense_valid
&& !sense_deferred
) {
904 switch (sshdr
.sense_key
) {
906 if (cmd
->device
->removable
) {
907 /* Detected disc change. Set a bit
908 * and quietly refuse further access.
910 cmd
->device
->changed
= 1;
911 scsi_end_request(cmd
, -EIO
, this_count
, 1);
914 /* Must have been a power glitch, or a
915 * bus reset. Could not have been a
916 * media change, so we just retry the
917 * request and see what happens.
919 scsi_requeue_command(q
, cmd
);
923 case ILLEGAL_REQUEST
:
924 /* If we had an ILLEGAL REQUEST returned, then
925 * we may have performed an unsupported
926 * command. The only thing this should be
927 * would be a ten byte read where only a six
928 * byte read was supported. Also, on a system
929 * where READ CAPACITY failed, we may have
930 * read past the end of the disk.
932 if ((cmd
->device
->use_10_for_rw
&&
933 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
934 (cmd
->cmnd
[0] == READ_10
||
935 cmd
->cmnd
[0] == WRITE_10
)) {
936 cmd
->device
->use_10_for_rw
= 0;
937 /* This will cause a retry with a
940 scsi_requeue_command(q
, cmd
);
943 scsi_end_request(cmd
, -EIO
, this_count
, 1);
948 /* If the device is in the process of becoming
949 * ready, or has a temporary blockage, retry.
951 if (sshdr
.asc
== 0x04) {
952 switch (sshdr
.ascq
) {
953 case 0x01: /* becoming ready */
954 case 0x04: /* format in progress */
955 case 0x05: /* rebuild in progress */
956 case 0x06: /* recalculation in progress */
957 case 0x07: /* operation in progress */
958 case 0x08: /* Long write in progress */
959 case 0x09: /* self test in progress */
960 scsi_requeue_command(q
, cmd
);
966 if (!(req
->cmd_flags
& REQ_QUIET
))
967 scsi_cmd_print_sense_hdr(cmd
,
971 scsi_end_request(cmd
, -EIO
, this_count
, 1);
973 case VOLUME_OVERFLOW
:
974 if (!(req
->cmd_flags
& REQ_QUIET
)) {
975 scmd_printk(KERN_INFO
, cmd
,
976 "Volume overflow, CDB: ");
977 __scsi_print_command(cmd
->cmnd
);
978 scsi_print_sense("", cmd
);
980 /* See SSC3rXX or current. */
981 scsi_end_request(cmd
, -EIO
, this_count
, 1);
987 if (host_byte(result
) == DID_RESET
) {
988 /* Third party bus reset or reset for error recovery
989 * reasons. Just retry the request and see what
992 scsi_requeue_command(q
, cmd
);
996 if (!(req
->cmd_flags
& REQ_QUIET
)) {
997 scsi_print_result(cmd
);
998 if (driver_byte(result
) & DRIVER_SENSE
)
999 scsi_print_sense("", cmd
);
1002 scsi_end_request(cmd
, -EIO
, this_count
, !result
);
1005 static int scsi_init_sgtable(struct request
*req
, struct scsi_data_buffer
*sdb
,
1011 * If sg table allocation fails, requeue request later.
1013 if (unlikely(scsi_alloc_sgtable(sdb
, req
->nr_phys_segments
,
1015 return BLKPREP_DEFER
;
1021 * Next, walk the list, and fill in the addresses and sizes of
1024 count
= blk_rq_map_sg(req
->q
, req
, sdb
->table
.sgl
);
1025 BUG_ON(count
> sdb
->table
.nents
);
1026 sdb
->table
.nents
= count
;
1027 if (blk_pc_request(req
))
1028 sdb
->length
= req
->data_len
;
1030 sdb
->length
= req
->nr_sectors
<< 9;
1035 * Function: scsi_init_io()
1037 * Purpose: SCSI I/O initialize function.
1039 * Arguments: cmd - Command descriptor we wish to initialize
1041 * Returns: 0 on success
1042 * BLKPREP_DEFER if the failure is retryable
1043 * BLKPREP_KILL if the failure is fatal
1045 int scsi_init_io(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
1047 int error
= scsi_init_sgtable(cmd
->request
, &cmd
->sdb
, gfp_mask
);
1051 if (blk_bidi_rq(cmd
->request
)) {
1052 struct scsi_data_buffer
*bidi_sdb
= kmem_cache_zalloc(
1053 scsi_bidi_sdb_cache
, GFP_ATOMIC
);
1055 error
= BLKPREP_DEFER
;
1059 cmd
->request
->next_rq
->special
= bidi_sdb
;
1060 error
= scsi_init_sgtable(cmd
->request
->next_rq
, bidi_sdb
,
1069 scsi_release_buffers(cmd
);
1070 if (error
== BLKPREP_KILL
)
1071 scsi_put_command(cmd
);
1072 else /* BLKPREP_DEFER */
1073 scsi_unprep_request(cmd
->request
);
1077 EXPORT_SYMBOL(scsi_init_io
);
1079 static struct scsi_cmnd
*scsi_get_cmd_from_req(struct scsi_device
*sdev
,
1080 struct request
*req
)
1082 struct scsi_cmnd
*cmd
;
1084 if (!req
->special
) {
1085 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1093 /* pull a tag out of the request if we have one */
1094 cmd
->tag
= req
->tag
;
1097 cmd
->cmnd
= req
->cmd
;
1102 int scsi_setup_blk_pc_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1104 struct scsi_cmnd
*cmd
;
1105 int ret
= scsi_prep_state_check(sdev
, req
);
1107 if (ret
!= BLKPREP_OK
)
1110 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1112 return BLKPREP_DEFER
;
1115 * BLOCK_PC requests may transfer data, in which case they must
1116 * a bio attached to them. Or they might contain a SCSI command
1117 * that does not transfer data, in which case they may optionally
1118 * submit a request without an attached bio.
1123 BUG_ON(!req
->nr_phys_segments
);
1125 ret
= scsi_init_io(cmd
, GFP_ATOMIC
);
1129 BUG_ON(req
->data_len
);
1132 memset(&cmd
->sdb
, 0, sizeof(cmd
->sdb
));
1136 cmd
->cmd_len
= req
->cmd_len
;
1138 cmd
->sc_data_direction
= DMA_NONE
;
1139 else if (rq_data_dir(req
) == WRITE
)
1140 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1142 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1144 cmd
->transfersize
= req
->data_len
;
1145 cmd
->allowed
= req
->retries
;
1146 cmd
->timeout_per_command
= req
->timeout
;
1149 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd
);
1152 * Setup a REQ_TYPE_FS command. These are simple read/write request
1153 * from filesystems that still need to be translated to SCSI CDBs from
1156 int scsi_setup_fs_cmnd(struct scsi_device
*sdev
, struct request
*req
)
1158 struct scsi_cmnd
*cmd
;
1159 int ret
= scsi_prep_state_check(sdev
, req
);
1161 if (ret
!= BLKPREP_OK
)
1164 * Filesystem requests must transfer data.
1166 BUG_ON(!req
->nr_phys_segments
);
1168 cmd
= scsi_get_cmd_from_req(sdev
, req
);
1170 return BLKPREP_DEFER
;
1172 memset(cmd
->cmnd
, 0, BLK_MAX_CDB
);
1173 return scsi_init_io(cmd
, GFP_ATOMIC
);
1175 EXPORT_SYMBOL(scsi_setup_fs_cmnd
);
1177 int scsi_prep_state_check(struct scsi_device
*sdev
, struct request
*req
)
1179 int ret
= BLKPREP_OK
;
1182 * If the device is not in running state we will reject some
1185 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1186 switch (sdev
->sdev_state
) {
1189 * If the device is offline we refuse to process any
1190 * commands. The device must be brought online
1191 * before trying any recovery commands.
1193 sdev_printk(KERN_ERR
, sdev
,
1194 "rejecting I/O to offline device\n");
1199 * If the device is fully deleted, we refuse to
1200 * process any commands as well.
1202 sdev_printk(KERN_ERR
, sdev
,
1203 "rejecting I/O to dead device\n");
1209 * If the devices is blocked we defer normal commands.
1211 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1212 ret
= BLKPREP_DEFER
;
1216 * For any other not fully online state we only allow
1217 * special commands. In particular any user initiated
1218 * command is not allowed.
1220 if (!(req
->cmd_flags
& REQ_PREEMPT
))
1227 EXPORT_SYMBOL(scsi_prep_state_check
);
1229 int scsi_prep_return(struct request_queue
*q
, struct request
*req
, int ret
)
1231 struct scsi_device
*sdev
= q
->queuedata
;
1235 req
->errors
= DID_NO_CONNECT
<< 16;
1236 /* release the command and kill it */
1238 struct scsi_cmnd
*cmd
= req
->special
;
1239 scsi_release_buffers(cmd
);
1240 scsi_put_command(cmd
);
1241 req
->special
= NULL
;
1246 * If we defer, the elv_next_request() returns NULL, but the
1247 * queue must be restarted, so we plug here if no returning
1248 * command will automatically do that.
1250 if (sdev
->device_busy
== 0)
1254 req
->cmd_flags
|= REQ_DONTPREP
;
1259 EXPORT_SYMBOL(scsi_prep_return
);
1261 int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1263 struct scsi_device
*sdev
= q
->queuedata
;
1264 int ret
= BLKPREP_KILL
;
1266 if (req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1267 ret
= scsi_setup_blk_pc_cmnd(sdev
, req
);
1268 return scsi_prep_return(q
, req
, ret
);
1272 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1275 * Called with the queue_lock held.
1277 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1278 struct scsi_device
*sdev
)
1280 if (sdev
->device_busy
>= sdev
->queue_depth
)
1282 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1284 * unblock after device_blocked iterates to zero
1286 if (--sdev
->device_blocked
== 0) {
1288 sdev_printk(KERN_INFO
, sdev
,
1289 "unblocking device at zero depth\n"));
1295 if (sdev
->device_blocked
)
1302 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1303 * return 0. We must end up running the queue again whenever 0 is
1304 * returned, else IO can hang.
1306 * Called with host_lock held.
1308 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1309 struct Scsi_Host
*shost
,
1310 struct scsi_device
*sdev
)
1312 if (scsi_host_in_recovery(shost
))
1314 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1316 * unblock after host_blocked iterates to zero
1318 if (--shost
->host_blocked
== 0) {
1320 printk("scsi%d unblocking host at zero depth\n",
1327 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1328 shost
->host_blocked
|| shost
->host_self_blocked
) {
1329 if (list_empty(&sdev
->starved_entry
))
1330 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1334 /* We're OK to process the command, so we can't be starved */
1335 if (!list_empty(&sdev
->starved_entry
))
1336 list_del_init(&sdev
->starved_entry
);
1342 * Kill a request for a dead device
1344 static void scsi_kill_request(struct request
*req
, struct request_queue
*q
)
1346 struct scsi_cmnd
*cmd
= req
->special
;
1347 struct scsi_device
*sdev
= cmd
->device
;
1348 struct Scsi_Host
*shost
= sdev
->host
;
1350 blkdev_dequeue_request(req
);
1352 if (unlikely(cmd
== NULL
)) {
1353 printk(KERN_CRIT
"impossible request in %s.\n",
1358 scsi_init_cmd_errh(cmd
);
1359 cmd
->result
= DID_NO_CONNECT
<< 16;
1360 atomic_inc(&cmd
->device
->iorequest_cnt
);
1363 * SCSI request completion path will do scsi_device_unbusy(),
1364 * bump busy counts. To bump the counters, we need to dance
1365 * with the locks as normal issue path does.
1367 sdev
->device_busy
++;
1368 spin_unlock(sdev
->request_queue
->queue_lock
);
1369 spin_lock(shost
->host_lock
);
1371 spin_unlock(shost
->host_lock
);
1372 spin_lock(sdev
->request_queue
->queue_lock
);
1377 static void scsi_softirq_done(struct request
*rq
)
1379 struct scsi_cmnd
*cmd
= rq
->completion_data
;
1380 unsigned long wait_for
= (cmd
->allowed
+ 1) * cmd
->timeout_per_command
;
1383 INIT_LIST_HEAD(&cmd
->eh_entry
);
1385 disposition
= scsi_decide_disposition(cmd
);
1386 if (disposition
!= SUCCESS
&&
1387 time_before(cmd
->jiffies_at_alloc
+ wait_for
, jiffies
)) {
1388 sdev_printk(KERN_ERR
, cmd
->device
,
1389 "timing out command, waited %lus\n",
1391 disposition
= SUCCESS
;
1394 scsi_log_completion(cmd
, disposition
);
1396 switch (disposition
) {
1398 scsi_finish_command(cmd
);
1401 scsi_queue_insert(cmd
, SCSI_MLQUEUE_EH_RETRY
);
1403 case ADD_TO_MLQUEUE
:
1404 scsi_queue_insert(cmd
, SCSI_MLQUEUE_DEVICE_BUSY
);
1407 if (!scsi_eh_scmd_add(cmd
, 0))
1408 scsi_finish_command(cmd
);
1413 * Function: scsi_request_fn()
1415 * Purpose: Main strategy routine for SCSI.
1417 * Arguments: q - Pointer to actual queue.
1421 * Lock status: IO request lock assumed to be held when called.
1423 static void scsi_request_fn(struct request_queue
*q
)
1425 struct scsi_device
*sdev
= q
->queuedata
;
1426 struct Scsi_Host
*shost
;
1427 struct scsi_cmnd
*cmd
;
1428 struct request
*req
;
1431 printk("scsi: killing requests for dead queue\n");
1432 while ((req
= elv_next_request(q
)) != NULL
)
1433 scsi_kill_request(req
, q
);
1437 if(!get_device(&sdev
->sdev_gendev
))
1438 /* We must be tearing the block queue down already */
1442 * To start with, we keep looping until the queue is empty, or until
1443 * the host is no longer able to accept any more requests.
1446 while (!blk_queue_plugged(q
)) {
1449 * get next queueable request. We do this early to make sure
1450 * that the request is fully prepared even if we cannot
1453 req
= elv_next_request(q
);
1454 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1457 if (unlikely(!scsi_device_online(sdev
))) {
1458 sdev_printk(KERN_ERR
, sdev
,
1459 "rejecting I/O to offline device\n");
1460 scsi_kill_request(req
, q
);
1466 * Remove the request from the request list.
1468 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1469 blkdev_dequeue_request(req
);
1470 sdev
->device_busy
++;
1472 spin_unlock(q
->queue_lock
);
1474 if (unlikely(cmd
== NULL
)) {
1475 printk(KERN_CRIT
"impossible request in %s.\n"
1476 "please mail a stack trace to "
1477 "linux-scsi@vger.kernel.org\n",
1479 blk_dump_rq_flags(req
, "foo");
1482 spin_lock(shost
->host_lock
);
1484 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1486 if (scsi_target(sdev
)->single_lun
) {
1487 if (scsi_target(sdev
)->starget_sdev_user
&&
1488 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1490 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1495 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1496 * take the lock again.
1498 spin_unlock_irq(shost
->host_lock
);
1501 * Finally, initialize any error handling parameters, and set up
1502 * the timers for timeouts.
1504 scsi_init_cmd_errh(cmd
);
1507 * Dispatch the command to the low-level driver.
1509 rtn
= scsi_dispatch_cmd(cmd
);
1510 spin_lock_irq(q
->queue_lock
);
1512 /* we're refusing the command; because of
1513 * the way locks get dropped, we need to
1514 * check here if plugging is required */
1515 if(sdev
->device_busy
== 0)
1525 spin_unlock_irq(shost
->host_lock
);
1528 * lock q, handle tag, requeue req, and decrement device_busy. We
1529 * must return with queue_lock held.
1531 * Decrementing device_busy without checking it is OK, as all such
1532 * cases (host limits or settings) should run the queue at some
1535 spin_lock_irq(q
->queue_lock
);
1536 blk_requeue_request(q
, req
);
1537 sdev
->device_busy
--;
1538 if(sdev
->device_busy
== 0)
1541 /* must be careful here...if we trigger the ->remove() function
1542 * we cannot be holding the q lock */
1543 spin_unlock_irq(q
->queue_lock
);
1544 put_device(&sdev
->sdev_gendev
);
1545 spin_lock_irq(q
->queue_lock
);
1548 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1550 struct device
*host_dev
;
1551 u64 bounce_limit
= 0xffffffff;
1553 if (shost
->unchecked_isa_dma
)
1554 return BLK_BOUNCE_ISA
;
1556 * Platforms with virtual-DMA translation
1557 * hardware have no practical limit.
1559 if (!PCI_DMA_BUS_IS_PHYS
)
1560 return BLK_BOUNCE_ANY
;
1562 host_dev
= scsi_get_device(shost
);
1563 if (host_dev
&& host_dev
->dma_mask
)
1564 bounce_limit
= *host_dev
->dma_mask
;
1566 return bounce_limit
;
1568 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1570 struct request_queue
*__scsi_alloc_queue(struct Scsi_Host
*shost
,
1571 request_fn_proc
*request_fn
)
1573 struct request_queue
*q
;
1574 struct device
*dev
= shost
->shost_gendev
.parent
;
1576 q
= blk_init_queue(request_fn
, NULL
);
1581 * this limit is imposed by hardware restrictions
1583 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1584 blk_queue_max_phys_segments(q
, SCSI_MAX_SG_CHAIN_SEGMENTS
);
1586 blk_queue_max_sectors(q
, shost
->max_sectors
);
1587 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1588 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1589 dma_set_seg_boundary(dev
, shost
->dma_boundary
);
1591 blk_queue_max_segment_size(q
, dma_get_max_seg_size(dev
));
1593 /* New queue, no concurrency on queue_flags */
1594 if (!shost
->use_clustering
)
1595 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER
, q
);
1598 * set a reasonable default alignment on word boundaries: the
1599 * host and device may alter it using
1600 * blk_queue_update_dma_alignment() later.
1602 blk_queue_dma_alignment(q
, 0x03);
1606 EXPORT_SYMBOL(__scsi_alloc_queue
);
1608 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1610 struct request_queue
*q
;
1612 q
= __scsi_alloc_queue(sdev
->host
, scsi_request_fn
);
1616 blk_queue_prep_rq(q
, scsi_prep_fn
);
1617 blk_queue_softirq_done(q
, scsi_softirq_done
);
1621 void scsi_free_queue(struct request_queue
*q
)
1623 blk_cleanup_queue(q
);
1627 * Function: scsi_block_requests()
1629 * Purpose: Utility function used by low-level drivers to prevent further
1630 * commands from being queued to the device.
1632 * Arguments: shost - Host in question
1636 * Lock status: No locks are assumed held.
1638 * Notes: There is no timer nor any other means by which the requests
1639 * get unblocked other than the low-level driver calling
1640 * scsi_unblock_requests().
1642 void scsi_block_requests(struct Scsi_Host
*shost
)
1644 shost
->host_self_blocked
= 1;
1646 EXPORT_SYMBOL(scsi_block_requests
);
1649 * Function: scsi_unblock_requests()
1651 * Purpose: Utility function used by low-level drivers to allow further
1652 * commands from being queued to the device.
1654 * Arguments: shost - Host in question
1658 * Lock status: No locks are assumed held.
1660 * Notes: There is no timer nor any other means by which the requests
1661 * get unblocked other than the low-level driver calling
1662 * scsi_unblock_requests().
1664 * This is done as an API function so that changes to the
1665 * internals of the scsi mid-layer won't require wholesale
1666 * changes to drivers that use this feature.
1668 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1670 shost
->host_self_blocked
= 0;
1671 scsi_run_host_queues(shost
);
1673 EXPORT_SYMBOL(scsi_unblock_requests
);
1675 int __init
scsi_init_queue(void)
1679 scsi_io_context_cache
= kmem_cache_create("scsi_io_context",
1680 sizeof(struct scsi_io_context
),
1682 if (!scsi_io_context_cache
) {
1683 printk(KERN_ERR
"SCSI: can't init scsi io context cache\n");
1687 scsi_bidi_sdb_cache
= kmem_cache_create("scsi_bidi_sdb",
1688 sizeof(struct scsi_data_buffer
),
1690 if (!scsi_bidi_sdb_cache
) {
1691 printk(KERN_ERR
"SCSI: can't init scsi bidi sdb cache\n");
1692 goto cleanup_io_context
;
1695 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1696 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1697 int size
= sgp
->size
* sizeof(struct scatterlist
);
1699 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1700 SLAB_HWCACHE_ALIGN
, NULL
);
1702 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1704 goto cleanup_bidi_sdb
;
1707 sgp
->pool
= mempool_create_slab_pool(SG_MEMPOOL_SIZE
,
1710 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1712 goto cleanup_bidi_sdb
;
1719 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1720 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1722 mempool_destroy(sgp
->pool
);
1724 kmem_cache_destroy(sgp
->slab
);
1726 kmem_cache_destroy(scsi_bidi_sdb_cache
);
1728 kmem_cache_destroy(scsi_io_context_cache
);
1733 void scsi_exit_queue(void)
1737 kmem_cache_destroy(scsi_io_context_cache
);
1738 kmem_cache_destroy(scsi_bidi_sdb_cache
);
1740 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1741 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1742 mempool_destroy(sgp
->pool
);
1743 kmem_cache_destroy(sgp
->slab
);
1748 * scsi_mode_select - issue a mode select
1749 * @sdev: SCSI device to be queried
1750 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1751 * @sp: Save page bit (0 == don't save, 1 == save)
1752 * @modepage: mode page being requested
1753 * @buffer: request buffer (may not be smaller than eight bytes)
1754 * @len: length of request buffer.
1755 * @timeout: command timeout
1756 * @retries: number of retries before failing
1757 * @data: returns a structure abstracting the mode header data
1758 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1759 * must be SCSI_SENSE_BUFFERSIZE big.
1761 * Returns zero if successful; negative error number or scsi
1766 scsi_mode_select(struct scsi_device
*sdev
, int pf
, int sp
, int modepage
,
1767 unsigned char *buffer
, int len
, int timeout
, int retries
,
1768 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1770 unsigned char cmd
[10];
1771 unsigned char *real_buffer
;
1774 memset(cmd
, 0, sizeof(cmd
));
1775 cmd
[1] = (pf
? 0x10 : 0) | (sp
? 0x01 : 0);
1777 if (sdev
->use_10_for_ms
) {
1780 real_buffer
= kmalloc(8 + len
, GFP_KERNEL
);
1783 memcpy(real_buffer
+ 8, buffer
, len
);
1787 real_buffer
[2] = data
->medium_type
;
1788 real_buffer
[3] = data
->device_specific
;
1789 real_buffer
[4] = data
->longlba
? 0x01 : 0;
1791 real_buffer
[6] = data
->block_descriptor_length
>> 8;
1792 real_buffer
[7] = data
->block_descriptor_length
;
1794 cmd
[0] = MODE_SELECT_10
;
1798 if (len
> 255 || data
->block_descriptor_length
> 255 ||
1802 real_buffer
= kmalloc(4 + len
, GFP_KERNEL
);
1805 memcpy(real_buffer
+ 4, buffer
, len
);
1808 real_buffer
[1] = data
->medium_type
;
1809 real_buffer
[2] = data
->device_specific
;
1810 real_buffer
[3] = data
->block_descriptor_length
;
1813 cmd
[0] = MODE_SELECT
;
1817 ret
= scsi_execute_req(sdev
, cmd
, DMA_TO_DEVICE
, real_buffer
, len
,
1818 sshdr
, timeout
, retries
);
1822 EXPORT_SYMBOL_GPL(scsi_mode_select
);
1825 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1826 * @sdev: SCSI device to be queried
1827 * @dbd: set if mode sense will allow block descriptors to be returned
1828 * @modepage: mode page being requested
1829 * @buffer: request buffer (may not be smaller than eight bytes)
1830 * @len: length of request buffer.
1831 * @timeout: command timeout
1832 * @retries: number of retries before failing
1833 * @data: returns a structure abstracting the mode header data
1834 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1835 * must be SCSI_SENSE_BUFFERSIZE big.
1837 * Returns zero if unsuccessful, or the header offset (either 4
1838 * or 8 depending on whether a six or ten byte command was
1839 * issued) if successful.
1842 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1843 unsigned char *buffer
, int len
, int timeout
, int retries
,
1844 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
)
1846 unsigned char cmd
[12];
1850 struct scsi_sense_hdr my_sshdr
;
1852 memset(data
, 0, sizeof(*data
));
1853 memset(&cmd
[0], 0, 12);
1854 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1857 /* caller might not be interested in sense, but we need it */
1862 use_10_for_ms
= sdev
->use_10_for_ms
;
1864 if (use_10_for_ms
) {
1868 cmd
[0] = MODE_SENSE_10
;
1875 cmd
[0] = MODE_SENSE
;
1880 memset(buffer
, 0, len
);
1882 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1883 sshdr
, timeout
, retries
);
1885 /* This code looks awful: what it's doing is making sure an
1886 * ILLEGAL REQUEST sense return identifies the actual command
1887 * byte as the problem. MODE_SENSE commands can return
1888 * ILLEGAL REQUEST if the code page isn't supported */
1890 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1891 (driver_byte(result
) & DRIVER_SENSE
)) {
1892 if (scsi_sense_valid(sshdr
)) {
1893 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1894 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1896 * Invalid command operation code
1898 sdev
->use_10_for_ms
= 0;
1904 if(scsi_status_is_good(result
)) {
1905 if (unlikely(buffer
[0] == 0x86 && buffer
[1] == 0x0b &&
1906 (modepage
== 6 || modepage
== 8))) {
1907 /* Initio breakage? */
1910 data
->medium_type
= 0;
1911 data
->device_specific
= 0;
1913 data
->block_descriptor_length
= 0;
1914 } else if(use_10_for_ms
) {
1915 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1916 data
->medium_type
= buffer
[2];
1917 data
->device_specific
= buffer
[3];
1918 data
->longlba
= buffer
[4] & 0x01;
1919 data
->block_descriptor_length
= buffer
[6]*256
1922 data
->length
= buffer
[0] + 1;
1923 data
->medium_type
= buffer
[1];
1924 data
->device_specific
= buffer
[2];
1925 data
->block_descriptor_length
= buffer
[3];
1927 data
->header_length
= header_length
;
1932 EXPORT_SYMBOL(scsi_mode_sense
);
1935 * scsi_test_unit_ready - test if unit is ready
1936 * @sdev: scsi device to change the state of.
1937 * @timeout: command timeout
1938 * @retries: number of retries before failing
1939 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1940 * returning sense. Make sure that this is cleared before passing
1943 * Returns zero if unsuccessful or an error if TUR failed. For
1944 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1945 * translated to success, with the ->changed flag updated.
1948 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
,
1949 struct scsi_sense_hdr
*sshdr_external
)
1952 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1954 struct scsi_sense_hdr
*sshdr
;
1957 if (!sshdr_external
)
1958 sshdr
= kzalloc(sizeof(*sshdr
), GFP_KERNEL
);
1960 sshdr
= sshdr_external
;
1962 /* try to eat the UNIT_ATTENTION if there are enough retries */
1964 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, sshdr
,
1966 } while ((driver_byte(result
) & DRIVER_SENSE
) &&
1967 sshdr
&& sshdr
->sense_key
== UNIT_ATTENTION
&&
1971 /* could not allocate sense buffer, so can't process it */
1974 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1976 if ((scsi_sense_valid(sshdr
)) &&
1977 ((sshdr
->sense_key
== UNIT_ATTENTION
) ||
1978 (sshdr
->sense_key
== NOT_READY
))) {
1983 if (!sshdr_external
)
1987 EXPORT_SYMBOL(scsi_test_unit_ready
);
1990 * scsi_device_set_state - Take the given device through the device state model.
1991 * @sdev: scsi device to change the state of.
1992 * @state: state to change to.
1994 * Returns zero if unsuccessful or an error if the requested
1995 * transition is illegal.
1998 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
2000 enum scsi_device_state oldstate
= sdev
->sdev_state
;
2002 if (state
== oldstate
)
2007 /* There are no legal states that come back to
2008 * created. This is the manually initialised start
2082 sdev
->sdev_state
= state
;
2086 SCSI_LOG_ERROR_RECOVERY(1,
2087 sdev_printk(KERN_ERR
, sdev
,
2088 "Illegal state transition %s->%s\n",
2089 scsi_device_state_name(oldstate
),
2090 scsi_device_state_name(state
))
2094 EXPORT_SYMBOL(scsi_device_set_state
);
2097 * sdev_evt_emit - emit a single SCSI device uevent
2098 * @sdev: associated SCSI device
2099 * @evt: event to emit
2101 * Send a single uevent (scsi_event) to the associated scsi_device.
2103 static void scsi_evt_emit(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2108 switch (evt
->evt_type
) {
2109 case SDEV_EVT_MEDIA_CHANGE
:
2110 envp
[idx
++] = "SDEV_MEDIA_CHANGE=1";
2120 kobject_uevent_env(&sdev
->sdev_gendev
.kobj
, KOBJ_CHANGE
, envp
);
2124 * sdev_evt_thread - send a uevent for each scsi event
2125 * @work: work struct for scsi_device
2127 * Dispatch queued events to their associated scsi_device kobjects
2130 void scsi_evt_thread(struct work_struct
*work
)
2132 struct scsi_device
*sdev
;
2133 LIST_HEAD(event_list
);
2135 sdev
= container_of(work
, struct scsi_device
, event_work
);
2138 struct scsi_event
*evt
;
2139 struct list_head
*this, *tmp
;
2140 unsigned long flags
;
2142 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2143 list_splice_init(&sdev
->event_list
, &event_list
);
2144 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2146 if (list_empty(&event_list
))
2149 list_for_each_safe(this, tmp
, &event_list
) {
2150 evt
= list_entry(this, struct scsi_event
, node
);
2151 list_del(&evt
->node
);
2152 scsi_evt_emit(sdev
, evt
);
2159 * sdev_evt_send - send asserted event to uevent thread
2160 * @sdev: scsi_device event occurred on
2161 * @evt: event to send
2163 * Assert scsi device event asynchronously.
2165 void sdev_evt_send(struct scsi_device
*sdev
, struct scsi_event
*evt
)
2167 unsigned long flags
;
2170 /* FIXME: currently this check eliminates all media change events
2171 * for polled devices. Need to update to discriminate between AN
2172 * and polled events */
2173 if (!test_bit(evt
->evt_type
, sdev
->supported_events
)) {
2179 spin_lock_irqsave(&sdev
->list_lock
, flags
);
2180 list_add_tail(&evt
->node
, &sdev
->event_list
);
2181 schedule_work(&sdev
->event_work
);
2182 spin_unlock_irqrestore(&sdev
->list_lock
, flags
);
2184 EXPORT_SYMBOL_GPL(sdev_evt_send
);
2187 * sdev_evt_alloc - allocate a new scsi event
2188 * @evt_type: type of event to allocate
2189 * @gfpflags: GFP flags for allocation
2191 * Allocates and returns a new scsi_event.
2193 struct scsi_event
*sdev_evt_alloc(enum scsi_device_event evt_type
,
2196 struct scsi_event
*evt
= kzalloc(sizeof(struct scsi_event
), gfpflags
);
2200 evt
->evt_type
= evt_type
;
2201 INIT_LIST_HEAD(&evt
->node
);
2203 /* evt_type-specific initialization, if any */
2205 case SDEV_EVT_MEDIA_CHANGE
:
2213 EXPORT_SYMBOL_GPL(sdev_evt_alloc
);
2216 * sdev_evt_send_simple - send asserted event to uevent thread
2217 * @sdev: scsi_device event occurred on
2218 * @evt_type: type of event to send
2219 * @gfpflags: GFP flags for allocation
2221 * Assert scsi device event asynchronously, given an event type.
2223 void sdev_evt_send_simple(struct scsi_device
*sdev
,
2224 enum scsi_device_event evt_type
, gfp_t gfpflags
)
2226 struct scsi_event
*evt
= sdev_evt_alloc(evt_type
, gfpflags
);
2228 sdev_printk(KERN_ERR
, sdev
, "event %d eaten due to OOM\n",
2233 sdev_evt_send(sdev
, evt
);
2235 EXPORT_SYMBOL_GPL(sdev_evt_send_simple
);
2238 * scsi_device_quiesce - Block user issued commands.
2239 * @sdev: scsi device to quiesce.
2241 * This works by trying to transition to the SDEV_QUIESCE state
2242 * (which must be a legal transition). When the device is in this
2243 * state, only special requests will be accepted, all others will
2244 * be deferred. Since special requests may also be requeued requests,
2245 * a successful return doesn't guarantee the device will be
2246 * totally quiescent.
2248 * Must be called with user context, may sleep.
2250 * Returns zero if unsuccessful or an error if not.
2253 scsi_device_quiesce(struct scsi_device
*sdev
)
2255 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2259 scsi_run_queue(sdev
->request_queue
);
2260 while (sdev
->device_busy
) {
2261 msleep_interruptible(200);
2262 scsi_run_queue(sdev
->request_queue
);
2266 EXPORT_SYMBOL(scsi_device_quiesce
);
2269 * scsi_device_resume - Restart user issued commands to a quiesced device.
2270 * @sdev: scsi device to resume.
2272 * Moves the device from quiesced back to running and restarts the
2275 * Must be called with user context, may sleep.
2278 scsi_device_resume(struct scsi_device
*sdev
)
2280 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2282 scsi_run_queue(sdev
->request_queue
);
2284 EXPORT_SYMBOL(scsi_device_resume
);
2287 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2289 scsi_device_quiesce(sdev
);
2293 scsi_target_quiesce(struct scsi_target
*starget
)
2295 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2297 EXPORT_SYMBOL(scsi_target_quiesce
);
2300 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2302 scsi_device_resume(sdev
);
2306 scsi_target_resume(struct scsi_target
*starget
)
2308 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2310 EXPORT_SYMBOL(scsi_target_resume
);
2313 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2314 * @sdev: device to block
2316 * Block request made by scsi lld's to temporarily stop all
2317 * scsi commands on the specified device. Called from interrupt
2318 * or normal process context.
2320 * Returns zero if successful or error if not
2323 * This routine transitions the device to the SDEV_BLOCK state
2324 * (which must be a legal transition). When the device is in this
2325 * state, all commands are deferred until the scsi lld reenables
2326 * the device with scsi_device_unblock or device_block_tmo fires.
2327 * This routine assumes the host_lock is held on entry.
2330 scsi_internal_device_block(struct scsi_device
*sdev
)
2332 struct request_queue
*q
= sdev
->request_queue
;
2333 unsigned long flags
;
2336 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2341 * The device has transitioned to SDEV_BLOCK. Stop the
2342 * block layer from calling the midlayer with this device's
2345 spin_lock_irqsave(q
->queue_lock
, flags
);
2347 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2351 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2354 * scsi_internal_device_unblock - resume a device after a block request
2355 * @sdev: device to resume
2357 * Called by scsi lld's or the midlayer to restart the device queue
2358 * for the previously suspended scsi device. Called from interrupt or
2359 * normal process context.
2361 * Returns zero if successful or error if not.
2364 * This routine transitions the device to the SDEV_RUNNING state
2365 * (which must be a legal transition) allowing the midlayer to
2366 * goose the queue for this device. This routine assumes the
2367 * host_lock is held upon entry.
2370 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2372 struct request_queue
*q
= sdev
->request_queue
;
2374 unsigned long flags
;
2377 * Try to transition the scsi device to SDEV_RUNNING
2378 * and goose the device queue if successful.
2380 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2384 spin_lock_irqsave(q
->queue_lock
, flags
);
2386 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2390 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2393 device_block(struct scsi_device
*sdev
, void *data
)
2395 scsi_internal_device_block(sdev
);
2399 target_block(struct device
*dev
, void *data
)
2401 if (scsi_is_target_device(dev
))
2402 starget_for_each_device(to_scsi_target(dev
), NULL
,
2408 scsi_target_block(struct device
*dev
)
2410 if (scsi_is_target_device(dev
))
2411 starget_for_each_device(to_scsi_target(dev
), NULL
,
2414 device_for_each_child(dev
, NULL
, target_block
);
2416 EXPORT_SYMBOL_GPL(scsi_target_block
);
2419 device_unblock(struct scsi_device
*sdev
, void *data
)
2421 scsi_internal_device_unblock(sdev
);
2425 target_unblock(struct device
*dev
, void *data
)
2427 if (scsi_is_target_device(dev
))
2428 starget_for_each_device(to_scsi_target(dev
), NULL
,
2434 scsi_target_unblock(struct device
*dev
)
2436 if (scsi_is_target_device(dev
))
2437 starget_for_each_device(to_scsi_target(dev
), NULL
,
2440 device_for_each_child(dev
, NULL
, target_unblock
);
2442 EXPORT_SYMBOL_GPL(scsi_target_unblock
);
2445 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2446 * @sgl: scatter-gather list
2447 * @sg_count: number of segments in sg
2448 * @offset: offset in bytes into sg, on return offset into the mapped area
2449 * @len: bytes to map, on return number of bytes mapped
2451 * Returns virtual address of the start of the mapped page
2453 void *scsi_kmap_atomic_sg(struct scatterlist
*sgl
, int sg_count
,
2454 size_t *offset
, size_t *len
)
2457 size_t sg_len
= 0, len_complete
= 0;
2458 struct scatterlist
*sg
;
2461 WARN_ON(!irqs_disabled());
2463 for_each_sg(sgl
, sg
, sg_count
, i
) {
2464 len_complete
= sg_len
; /* Complete sg-entries */
2465 sg_len
+= sg
->length
;
2466 if (sg_len
> *offset
)
2470 if (unlikely(i
== sg_count
)) {
2471 printk(KERN_ERR
"%s: Bytes in sg: %zu, requested offset %zu, "
2473 __FUNCTION__
, sg_len
, *offset
, sg_count
);
2478 /* Offset starting from the beginning of first page in this sg-entry */
2479 *offset
= *offset
- len_complete
+ sg
->offset
;
2481 /* Assumption: contiguous pages can be accessed as "page + i" */
2482 page
= nth_page(sg_page(sg
), (*offset
>> PAGE_SHIFT
));
2483 *offset
&= ~PAGE_MASK
;
2485 /* Bytes in this sg-entry from *offset to the end of the page */
2486 sg_len
= PAGE_SIZE
- *offset
;
2490 return kmap_atomic(page
, KM_BIO_SRC_IRQ
);
2492 EXPORT_SYMBOL(scsi_kmap_atomic_sg
);
2495 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2496 * @virt: virtual address to be unmapped
2498 void scsi_kunmap_atomic_sg(void *virt
)
2500 kunmap_atomic(virt
, KM_BIO_SRC_IRQ
);
2502 EXPORT_SYMBOL(scsi_kunmap_atomic_sg
);