2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/completion.h>
13 #include <linux/kernel.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/pci.h>
18 #include <linux/delay.h>
20 #include <scsi/scsi.h>
21 #include <scsi/scsi_dbg.h>
22 #include <scsi/scsi_device.h>
23 #include <scsi/scsi_driver.h>
24 #include <scsi/scsi_eh.h>
25 #include <scsi/scsi_host.h>
26 #include <scsi/scsi_request.h>
28 #include "scsi_priv.h"
29 #include "scsi_logging.h"
32 #define SG_MEMPOOL_NR (sizeof(scsi_sg_pools)/sizeof(struct scsi_host_sg_pool))
33 #define SG_MEMPOOL_SIZE 32
35 struct scsi_host_sg_pool
{
42 #if (SCSI_MAX_PHYS_SEGMENTS < 32)
43 #error SCSI_MAX_PHYS_SEGMENTS is too small
46 #define SP(x) { x, "sgpool-" #x }
47 static struct scsi_host_sg_pool scsi_sg_pools
[] = {
51 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
53 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
55 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
57 #if (SCSI_MAX_PHYS_SEGMENTS > 256)
58 #error SCSI_MAX_PHYS_SEGMENTS is too large
66 static void scsi_run_queue(struct request_queue
*q
);
69 * Function: scsi_unprep_request()
71 * Purpose: Remove all preparation done for a request, including its
72 * associated scsi_cmnd, so that it can be requeued.
74 * Arguments: req - request to unprepare
76 * Lock status: Assumed that no locks are held upon entry.
80 static void scsi_unprep_request(struct request
*req
)
82 struct scsi_cmnd
*cmd
= req
->special
;
84 req
->flags
&= ~REQ_DONTPREP
;
85 req
->special
= (req
->flags
& REQ_SPECIAL
) ? cmd
->sc_request
: NULL
;
87 scsi_put_command(cmd
);
91 * Function: scsi_queue_insert()
93 * Purpose: Insert a command in the midlevel queue.
95 * Arguments: cmd - command that we are adding to queue.
96 * reason - why we are inserting command to queue.
98 * Lock status: Assumed that lock is not held upon entry.
102 * Notes: We do this for one of two cases. Either the host is busy
103 * and it cannot accept any more commands for the time being,
104 * or the device returned QUEUE_FULL and can accept no more
106 * Notes: This could be called either from an interrupt context or a
107 * normal process context.
109 int scsi_queue_insert(struct scsi_cmnd
*cmd
, int reason
)
111 struct Scsi_Host
*host
= cmd
->device
->host
;
112 struct scsi_device
*device
= cmd
->device
;
113 struct request_queue
*q
= device
->request_queue
;
117 printk("Inserting command %p into mlqueue\n", cmd
));
120 * Set the appropriate busy bit for the device/host.
122 * If the host/device isn't busy, assume that something actually
123 * completed, and that we should be able to queue a command now.
125 * Note that the prior mid-layer assumption that any host could
126 * always queue at least one command is now broken. The mid-layer
127 * will implement a user specifiable stall (see
128 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
129 * if a command is requeued with no other commands outstanding
130 * either for the device or for the host.
132 if (reason
== SCSI_MLQUEUE_HOST_BUSY
)
133 host
->host_blocked
= host
->max_host_blocked
;
134 else if (reason
== SCSI_MLQUEUE_DEVICE_BUSY
)
135 device
->device_blocked
= device
->max_device_blocked
;
138 * Decrement the counters, since these commands are no longer
139 * active on the host/device.
141 scsi_device_unbusy(device
);
144 * Requeue this command. It will go before all other commands
145 * that are already in the queue.
147 * NOTE: there is magic here about the way the queue is plugged if
148 * we have no outstanding commands.
150 * Although we *don't* plug the queue, we call the request
151 * function. The SCSI request function detects the blocked condition
152 * and plugs the queue appropriately.
154 spin_lock_irqsave(q
->queue_lock
, flags
);
155 blk_requeue_request(q
, cmd
->request
);
156 spin_unlock_irqrestore(q
->queue_lock
, flags
);
164 * Function: scsi_do_req
166 * Purpose: Queue a SCSI request
168 * Arguments: sreq - command descriptor.
169 * cmnd - actual SCSI command to be performed.
170 * buffer - data buffer.
171 * bufflen - size of data buffer.
172 * done - completion function to be run.
173 * timeout - how long to let it run before timeout.
174 * retries - number of retries we allow.
176 * Lock status: No locks held upon entry.
180 * Notes: This function is only used for queueing requests for things
181 * like ioctls and character device requests - this is because
182 * we essentially just inject a request into the queue for the
185 * In order to support the scsi_device_quiesce function, we
186 * now inject requests on the *head* of the device queue
187 * rather than the tail.
189 void scsi_do_req(struct scsi_request
*sreq
, const void *cmnd
,
190 void *buffer
, unsigned bufflen
,
191 void (*done
)(struct scsi_cmnd
*),
192 int timeout
, int retries
)
195 * If the upper level driver is reusing these things, then
196 * we should release the low-level block now. Another one will
197 * be allocated later when this request is getting queued.
199 __scsi_release_request(sreq
);
202 * Our own function scsi_done (which marks the host as not busy,
203 * disables the timeout counter, etc) will be called by us or by the
204 * scsi_hosts[host].queuecommand() function needs to also call
205 * the completion function for the high level driver.
207 memcpy(sreq
->sr_cmnd
, cmnd
, sizeof(sreq
->sr_cmnd
));
208 sreq
->sr_bufflen
= bufflen
;
209 sreq
->sr_buffer
= buffer
;
210 sreq
->sr_allowed
= retries
;
211 sreq
->sr_done
= done
;
212 sreq
->sr_timeout_per_command
= timeout
;
214 if (sreq
->sr_cmd_len
== 0)
215 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
218 * head injection *required* here otherwise quiesce won't work
220 * Because users of this function are apt to reuse requests with no
221 * modification, we have to sanitise the request flags here
223 sreq
->sr_request
->flags
&= ~REQ_DONTPREP
;
224 blk_insert_request(sreq
->sr_device
->request_queue
, sreq
->sr_request
,
227 EXPORT_SYMBOL(scsi_do_req
);
230 * scsi_execute - insert request and wait for the result
233 * @data_direction: data direction
234 * @buffer: data buffer
235 * @bufflen: len of buffer
236 * @sense: optional sense buffer
237 * @timeout: request timeout in seconds
238 * @retries: number of times to retry request
239 * @flags: or into request flags;
241 * returns the req->errors value which is the the scsi_cmnd result
244 int scsi_execute(struct scsi_device
*sdev
, const unsigned char *cmd
,
245 int data_direction
, void *buffer
, unsigned bufflen
,
246 unsigned char *sense
, int timeout
, int retries
, int flags
)
249 int write
= (data_direction
== DMA_TO_DEVICE
);
250 int ret
= DRIVER_ERROR
<< 24;
252 req
= blk_get_request(sdev
->request_queue
, write
, __GFP_WAIT
);
254 if (bufflen
&& blk_rq_map_kern(sdev
->request_queue
, req
,
255 buffer
, bufflen
, __GFP_WAIT
))
258 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
259 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
262 req
->retries
= retries
;
263 req
->timeout
= timeout
;
264 req
->flags
|= flags
| REQ_BLOCK_PC
| REQ_SPECIAL
| REQ_QUIET
;
267 * head injection *required* here otherwise quiesce won't work
269 blk_execute_rq(req
->q
, NULL
, req
, 1);
273 blk_put_request(req
);
277 EXPORT_SYMBOL(scsi_execute
);
280 int scsi_execute_req(struct scsi_device
*sdev
, const unsigned char *cmd
,
281 int data_direction
, void *buffer
, unsigned bufflen
,
282 struct scsi_sense_hdr
*sshdr
, int timeout
, int retries
)
288 sense
= kmalloc(SCSI_SENSE_BUFFERSIZE
, GFP_NOIO
);
290 return DRIVER_ERROR
<< 24;
291 memset(sense
, 0, SCSI_SENSE_BUFFERSIZE
);
293 result
= scsi_execute(sdev
, cmd
, data_direction
, buffer
, bufflen
,
294 sense
, timeout
, retries
, 0);
296 scsi_normalize_sense(sense
, SCSI_SENSE_BUFFERSIZE
, sshdr
);
301 EXPORT_SYMBOL(scsi_execute_req
);
303 struct scsi_io_context
{
305 void (*done
)(void *data
, char *sense
, int result
, int resid
);
306 char sense
[SCSI_SENSE_BUFFERSIZE
];
309 static void scsi_end_async(struct request
*req
)
311 struct scsi_io_context
*sioc
= req
->end_io_data
;
314 sioc
->done(sioc
->data
, sioc
->sense
, req
->errors
, req
->data_len
);
317 __blk_put_request(req
->q
, req
);
320 static int scsi_merge_bio(struct request
*rq
, struct bio
*bio
)
322 struct request_queue
*q
= rq
->q
;
324 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
325 if (rq_data_dir(rq
) == WRITE
)
326 bio
->bi_rw
|= (1 << BIO_RW
);
327 blk_queue_bounce(q
, &bio
);
330 blk_rq_bio_prep(q
, rq
, bio
);
331 else if (!q
->back_merge_fn(q
, rq
, bio
))
334 rq
->biotail
->bi_next
= bio
;
336 rq
->hard_nr_sectors
+= bio_sectors(bio
);
337 rq
->nr_sectors
= rq
->hard_nr_sectors
;
343 static int scsi_bi_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
353 * scsi_req_map_sg - map a scatterlist into a request
354 * @rq: request to fill
356 * @nsegs: number of elements
357 * @bufflen: len of buffer
358 * @gfp: memory allocation flags
360 * scsi_req_map_sg maps a scatterlist into a request so that the
361 * request can be sent to the block layer. We do not trust the scatterlist
362 * sent to use, as some ULDs use that struct to only organize the pages.
364 static int scsi_req_map_sg(struct request
*rq
, struct scatterlist
*sgl
,
365 int nsegs
, unsigned bufflen
, gfp_t gfp
)
367 struct request_queue
*q
= rq
->q
;
368 int nr_pages
= (bufflen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
369 unsigned int data_len
= 0, len
, bytes
, off
;
371 struct bio
*bio
= NULL
;
372 int i
, err
, nr_vecs
= 0;
374 for (i
= 0; i
< nsegs
; i
++) {
381 bytes
= min_t(unsigned int, len
, PAGE_SIZE
- off
);
384 nr_vecs
= min_t(int, BIO_MAX_PAGES
, nr_pages
);
387 bio
= bio_alloc(gfp
, nr_vecs
);
392 bio
->bi_end_io
= scsi_bi_endio
;
395 if (bio_add_pc_page(q
, bio
, page
, bytes
, off
) !=
402 if (bio
->bi_vcnt
>= nr_vecs
) {
403 err
= scsi_merge_bio(rq
, bio
);
405 bio_endio(bio
, bio
->bi_size
, 0);
417 rq
->buffer
= rq
->data
= NULL
;
418 rq
->data_len
= data_len
;
422 while ((bio
= rq
->bio
) != NULL
) {
423 rq
->bio
= bio
->bi_next
;
425 * call endio instead of bio_put incase it was bounced
427 bio_endio(bio
, bio
->bi_size
, 0);
434 * scsi_execute_async - insert request
437 * @data_direction: data direction
438 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
439 * @bufflen: len of buffer
440 * @use_sg: if buffer is a scatterlist this is the number of elements
441 * @timeout: request timeout in seconds
442 * @retries: number of times to retry request
443 * @flags: or into request flags
445 int scsi_execute_async(struct scsi_device
*sdev
, const unsigned char *cmd
,
446 int data_direction
, void *buffer
, unsigned bufflen
,
447 int use_sg
, int timeout
, int retries
, void *privdata
,
448 void (*done
)(void *, char *, int, int), gfp_t gfp
)
451 struct scsi_io_context
*sioc
;
453 int write
= (data_direction
== DMA_TO_DEVICE
);
455 sioc
= kzalloc(sizeof(*sioc
), gfp
);
457 return DRIVER_ERROR
<< 24;
459 req
= blk_get_request(sdev
->request_queue
, write
, gfp
);
464 err
= scsi_req_map_sg(req
, buffer
, use_sg
, bufflen
, gfp
);
466 err
= blk_rq_map_kern(req
->q
, req
, buffer
, bufflen
, gfp
);
471 req
->cmd_len
= COMMAND_SIZE(cmd
[0]);
472 memcpy(req
->cmd
, cmd
, req
->cmd_len
);
473 req
->sense
= sioc
->sense
;
475 req
->timeout
= timeout
;
476 req
->retries
= retries
;
477 req
->flags
|= REQ_BLOCK_PC
| REQ_QUIET
;
478 req
->end_io_data
= sioc
;
480 sioc
->data
= privdata
;
483 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, scsi_end_async
);
487 blk_put_request(req
);
490 return DRIVER_ERROR
<< 24;
492 EXPORT_SYMBOL_GPL(scsi_execute_async
);
495 * Function: scsi_init_cmd_errh()
497 * Purpose: Initialize cmd fields related to error handling.
499 * Arguments: cmd - command that is ready to be queued.
503 * Notes: This function has the job of initializing a number of
504 * fields related to error handling. Typically this will
505 * be called once for each command, as required.
507 static int scsi_init_cmd_errh(struct scsi_cmnd
*cmd
)
509 cmd
->serial_number
= 0;
511 memset(cmd
->sense_buffer
, 0, sizeof cmd
->sense_buffer
);
513 if (cmd
->cmd_len
== 0)
514 cmd
->cmd_len
= COMMAND_SIZE(cmd
->cmnd
[0]);
517 * We need saved copies of a number of fields - this is because
518 * error handling may need to overwrite these with different values
519 * to run different commands, and once error handling is complete,
520 * we will need to restore these values prior to running the actual
523 cmd
->old_use_sg
= cmd
->use_sg
;
524 cmd
->old_cmd_len
= cmd
->cmd_len
;
525 cmd
->sc_old_data_direction
= cmd
->sc_data_direction
;
526 cmd
->old_underflow
= cmd
->underflow
;
527 memcpy(cmd
->data_cmnd
, cmd
->cmnd
, sizeof(cmd
->cmnd
));
528 cmd
->buffer
= cmd
->request_buffer
;
529 cmd
->bufflen
= cmd
->request_bufflen
;
535 * Function: scsi_setup_cmd_retry()
537 * Purpose: Restore the command state for a retry
539 * Arguments: cmd - command to be restored
543 * Notes: Immediately prior to retrying a command, we need
544 * to restore certain fields that we saved above.
546 void scsi_setup_cmd_retry(struct scsi_cmnd
*cmd
)
548 memcpy(cmd
->cmnd
, cmd
->data_cmnd
, sizeof(cmd
->data_cmnd
));
549 cmd
->request_buffer
= cmd
->buffer
;
550 cmd
->request_bufflen
= cmd
->bufflen
;
551 cmd
->use_sg
= cmd
->old_use_sg
;
552 cmd
->cmd_len
= cmd
->old_cmd_len
;
553 cmd
->sc_data_direction
= cmd
->sc_old_data_direction
;
554 cmd
->underflow
= cmd
->old_underflow
;
557 void scsi_device_unbusy(struct scsi_device
*sdev
)
559 struct Scsi_Host
*shost
= sdev
->host
;
562 spin_lock_irqsave(shost
->host_lock
, flags
);
564 if (unlikely(scsi_host_in_recovery(shost
) &&
566 scsi_eh_wakeup(shost
);
567 spin_unlock(shost
->host_lock
);
568 spin_lock(sdev
->request_queue
->queue_lock
);
570 spin_unlock_irqrestore(sdev
->request_queue
->queue_lock
, flags
);
574 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
575 * and call blk_run_queue for all the scsi_devices on the target -
576 * including current_sdev first.
578 * Called with *no* scsi locks held.
580 static void scsi_single_lun_run(struct scsi_device
*current_sdev
)
582 struct Scsi_Host
*shost
= current_sdev
->host
;
583 struct scsi_device
*sdev
, *tmp
;
584 struct scsi_target
*starget
= scsi_target(current_sdev
);
587 spin_lock_irqsave(shost
->host_lock
, flags
);
588 starget
->starget_sdev_user
= NULL
;
589 spin_unlock_irqrestore(shost
->host_lock
, flags
);
592 * Call blk_run_queue for all LUNs on the target, starting with
593 * current_sdev. We race with others (to set starget_sdev_user),
594 * but in most cases, we will be first. Ideally, each LU on the
595 * target would get some limited time or requests on the target.
597 blk_run_queue(current_sdev
->request_queue
);
599 spin_lock_irqsave(shost
->host_lock
, flags
);
600 if (starget
->starget_sdev_user
)
602 list_for_each_entry_safe(sdev
, tmp
, &starget
->devices
,
603 same_target_siblings
) {
604 if (sdev
== current_sdev
)
606 if (scsi_device_get(sdev
))
609 spin_unlock_irqrestore(shost
->host_lock
, flags
);
610 blk_run_queue(sdev
->request_queue
);
611 spin_lock_irqsave(shost
->host_lock
, flags
);
613 scsi_device_put(sdev
);
616 spin_unlock_irqrestore(shost
->host_lock
, flags
);
620 * Function: scsi_run_queue()
622 * Purpose: Select a proper request queue to serve next
624 * Arguments: q - last request's queue
628 * Notes: The previous command was completely finished, start
629 * a new one if possible.
631 static void scsi_run_queue(struct request_queue
*q
)
633 struct scsi_device
*sdev
= q
->queuedata
;
634 struct Scsi_Host
*shost
= sdev
->host
;
637 if (sdev
->single_lun
)
638 scsi_single_lun_run(sdev
);
640 spin_lock_irqsave(shost
->host_lock
, flags
);
641 while (!list_empty(&shost
->starved_list
) &&
642 !shost
->host_blocked
&& !shost
->host_self_blocked
&&
643 !((shost
->can_queue
> 0) &&
644 (shost
->host_busy
>= shost
->can_queue
))) {
646 * As long as shost is accepting commands and we have
647 * starved queues, call blk_run_queue. scsi_request_fn
648 * drops the queue_lock and can add us back to the
651 * host_lock protects the starved_list and starved_entry.
652 * scsi_request_fn must get the host_lock before checking
653 * or modifying starved_list or starved_entry.
655 sdev
= list_entry(shost
->starved_list
.next
,
656 struct scsi_device
, starved_entry
);
657 list_del_init(&sdev
->starved_entry
);
658 spin_unlock_irqrestore(shost
->host_lock
, flags
);
660 blk_run_queue(sdev
->request_queue
);
662 spin_lock_irqsave(shost
->host_lock
, flags
);
663 if (unlikely(!list_empty(&sdev
->starved_entry
)))
665 * sdev lost a race, and was put back on the
666 * starved list. This is unlikely but without this
667 * in theory we could loop forever.
671 spin_unlock_irqrestore(shost
->host_lock
, flags
);
677 * Function: scsi_requeue_command()
679 * Purpose: Handle post-processing of completed commands.
681 * Arguments: q - queue to operate on
682 * cmd - command that may need to be requeued.
686 * Notes: After command completion, there may be blocks left
687 * over which weren't finished by the previous command
688 * this can be for a number of reasons - the main one is
689 * I/O errors in the middle of the request, in which case
690 * we need to request the blocks that come after the bad
692 * Notes: Upon return, cmd is a stale pointer.
694 static void scsi_requeue_command(struct request_queue
*q
, struct scsi_cmnd
*cmd
)
696 struct request
*req
= cmd
->request
;
699 scsi_unprep_request(req
);
700 spin_lock_irqsave(q
->queue_lock
, flags
);
701 blk_requeue_request(q
, req
);
702 spin_unlock_irqrestore(q
->queue_lock
, flags
);
707 void scsi_next_command(struct scsi_cmnd
*cmd
)
709 struct scsi_device
*sdev
= cmd
->device
;
710 struct request_queue
*q
= sdev
->request_queue
;
712 /* need to hold a reference on the device before we let go of the cmd */
713 get_device(&sdev
->sdev_gendev
);
715 scsi_put_command(cmd
);
718 /* ok to remove device now */
719 put_device(&sdev
->sdev_gendev
);
722 void scsi_run_host_queues(struct Scsi_Host
*shost
)
724 struct scsi_device
*sdev
;
726 shost_for_each_device(sdev
, shost
)
727 scsi_run_queue(sdev
->request_queue
);
731 * Function: scsi_end_request()
733 * Purpose: Post-processing of completed commands (usually invoked at end
734 * of upper level post-processing and scsi_io_completion).
736 * Arguments: cmd - command that is complete.
737 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
738 * bytes - number of bytes of completed I/O
739 * requeue - indicates whether we should requeue leftovers.
741 * Lock status: Assumed that lock is not held upon entry.
743 * Returns: cmd if requeue required, NULL otherwise.
745 * Notes: This is called for block device requests in order to
746 * mark some number of sectors as complete.
748 * We are guaranteeing that the request queue will be goosed
749 * at some point during this call.
750 * Notes: If cmd was requeued, upon return it will be a stale pointer.
752 static struct scsi_cmnd
*scsi_end_request(struct scsi_cmnd
*cmd
, int uptodate
,
753 int bytes
, int requeue
)
755 request_queue_t
*q
= cmd
->device
->request_queue
;
756 struct request
*req
= cmd
->request
;
760 * If there are blocks left over at the end, set up the command
761 * to queue the remainder of them.
763 if (end_that_request_chunk(req
, uptodate
, bytes
)) {
764 int leftover
= (req
->hard_nr_sectors
<< 9);
766 if (blk_pc_request(req
))
767 leftover
= req
->data_len
;
769 /* kill remainder if no retrys */
770 if (!uptodate
&& blk_noretry_request(req
))
771 end_that_request_chunk(req
, 0, leftover
);
775 * Bleah. Leftovers again. Stick the
776 * leftovers in the front of the
777 * queue, and goose the queue again.
779 scsi_requeue_command(q
, cmd
);
786 add_disk_randomness(req
->rq_disk
);
788 spin_lock_irqsave(q
->queue_lock
, flags
);
789 if (blk_rq_tagged(req
))
790 blk_queue_end_tag(q
, req
);
791 end_that_request_last(req
);
792 spin_unlock_irqrestore(q
->queue_lock
, flags
);
795 * This will goose the queue request function at the end, so we don't
796 * need to worry about launching another command.
798 scsi_next_command(cmd
);
802 static struct scatterlist
*scsi_alloc_sgtable(struct scsi_cmnd
*cmd
, gfp_t gfp_mask
)
804 struct scsi_host_sg_pool
*sgp
;
805 struct scatterlist
*sgl
;
807 BUG_ON(!cmd
->use_sg
);
809 switch (cmd
->use_sg
) {
819 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
823 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
827 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
838 sgp
= scsi_sg_pools
+ cmd
->sglist_len
;
839 sgl
= mempool_alloc(sgp
->pool
, gfp_mask
);
843 static void scsi_free_sgtable(struct scatterlist
*sgl
, int index
)
845 struct scsi_host_sg_pool
*sgp
;
847 BUG_ON(index
>= SG_MEMPOOL_NR
);
849 sgp
= scsi_sg_pools
+ index
;
850 mempool_free(sgl
, sgp
->pool
);
854 * Function: scsi_release_buffers()
856 * Purpose: Completion processing for block device I/O requests.
858 * Arguments: cmd - command that we are bailing.
860 * Lock status: Assumed that no lock is held upon entry.
864 * Notes: In the event that an upper level driver rejects a
865 * command, we must release resources allocated during
866 * the __init_io() function. Primarily this would involve
867 * the scatter-gather table, and potentially any bounce
870 static void scsi_release_buffers(struct scsi_cmnd
*cmd
)
872 struct request
*req
= cmd
->request
;
875 * Free up any indirection buffers we allocated for DMA purposes.
878 scsi_free_sgtable(cmd
->request_buffer
, cmd
->sglist_len
);
879 else if (cmd
->request_buffer
!= req
->buffer
)
880 kfree(cmd
->request_buffer
);
883 * Zero these out. They now point to freed memory, and it is
884 * dangerous to hang onto the pointers.
888 cmd
->request_buffer
= NULL
;
889 cmd
->request_bufflen
= 0;
893 * Function: scsi_io_completion()
895 * Purpose: Completion processing for block device I/O requests.
897 * Arguments: cmd - command that is finished.
899 * Lock status: Assumed that no lock is held upon entry.
903 * Notes: This function is matched in terms of capabilities to
904 * the function that created the scatter-gather list.
905 * In other words, if there are no bounce buffers
906 * (the normal case for most drivers), we don't need
907 * the logic to deal with cleaning up afterwards.
909 * We must do one of several things here:
911 * a) Call scsi_end_request. This will finish off the
912 * specified number of sectors. If we are done, the
913 * command block will be released, and the queue
914 * function will be goosed. If we are not done, then
915 * scsi_end_request will directly goose the queue.
917 * b) We can just use scsi_requeue_command() here. This would
918 * be used if we just wanted to retry, for example.
920 void scsi_io_completion(struct scsi_cmnd
*cmd
, unsigned int good_bytes
,
921 unsigned int block_bytes
)
923 int result
= cmd
->result
;
924 int this_count
= cmd
->bufflen
;
925 request_queue_t
*q
= cmd
->device
->request_queue
;
926 struct request
*req
= cmd
->request
;
927 int clear_errors
= 1;
928 struct scsi_sense_hdr sshdr
;
930 int sense_deferred
= 0;
932 if (blk_complete_barrier_rq(q
, req
, good_bytes
>> 9))
936 * Free up any indirection buffers we allocated for DMA purposes.
937 * For the case of a READ, we need to copy the data out of the
938 * bounce buffer and into the real buffer.
941 scsi_free_sgtable(cmd
->buffer
, cmd
->sglist_len
);
942 else if (cmd
->buffer
!= req
->buffer
) {
943 if (rq_data_dir(req
) == READ
) {
945 char *to
= bio_kmap_irq(req
->bio
, &flags
);
946 memcpy(to
, cmd
->buffer
, cmd
->bufflen
);
947 bio_kunmap_irq(to
, &flags
);
953 sense_valid
= scsi_command_normalize_sense(cmd
, &sshdr
);
955 sense_deferred
= scsi_sense_is_deferred(&sshdr
);
957 if (blk_pc_request(req
)) { /* SG_IO ioctl from block level */
958 req
->errors
= result
;
961 if (sense_valid
&& req
->sense
) {
963 * SG_IO wants current and deferred errors
965 int len
= 8 + cmd
->sense_buffer
[7];
967 if (len
> SCSI_SENSE_BUFFERSIZE
)
968 len
= SCSI_SENSE_BUFFERSIZE
;
969 memcpy(req
->sense
, cmd
->sense_buffer
, len
);
970 req
->sense_len
= len
;
973 req
->data_len
= cmd
->resid
;
977 * Zero these out. They now point to freed memory, and it is
978 * dangerous to hang onto the pointers.
982 cmd
->request_buffer
= NULL
;
983 cmd
->request_bufflen
= 0;
986 * Next deal with any sectors which we were able to correctly
989 if (good_bytes
>= 0) {
990 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, %d bytes done.\n",
991 req
->nr_sectors
, good_bytes
));
992 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd
->use_sg
));
997 * If multiple sectors are requested in one buffer, then
998 * they will have been finished off by the first command.
999 * If not, then we have a multi-buffer command.
1001 * If block_bytes != 0, it means we had a medium error
1002 * of some sort, and that we want to mark some number of
1003 * sectors as not uptodate. Thus we want to inhibit
1004 * requeueing right here - we will requeue down below
1005 * when we handle the bad sectors.
1009 * If the command completed without error, then either
1010 * finish off the rest of the command, or start a new one.
1012 if (scsi_end_request(cmd
, 1, good_bytes
, result
== 0) == NULL
)
1016 * Now, if we were good little boys and girls, Santa left us a request
1017 * sense buffer. We can extract information from this, so we
1018 * can choose a block to remap, etc.
1020 if (sense_valid
&& !sense_deferred
) {
1021 switch (sshdr
.sense_key
) {
1022 case UNIT_ATTENTION
:
1023 if (cmd
->device
->removable
) {
1024 /* detected disc change. set a bit
1025 * and quietly refuse further access.
1027 cmd
->device
->changed
= 1;
1028 scsi_end_request(cmd
, 0,
1033 * Must have been a power glitch, or a
1034 * bus reset. Could not have been a
1035 * media change, so we just retry the
1036 * request and see what happens.
1038 scsi_requeue_command(q
, cmd
);
1042 case ILLEGAL_REQUEST
:
1044 * If we had an ILLEGAL REQUEST returned, then we may
1045 * have performed an unsupported command. The only
1046 * thing this should be would be a ten byte read where
1047 * only a six byte read was supported. Also, on a
1048 * system where READ CAPACITY failed, we may have read
1049 * past the end of the disk.
1051 if ((cmd
->device
->use_10_for_rw
&&
1052 sshdr
.asc
== 0x20 && sshdr
.ascq
== 0x00) &&
1053 (cmd
->cmnd
[0] == READ_10
||
1054 cmd
->cmnd
[0] == WRITE_10
)) {
1055 cmd
->device
->use_10_for_rw
= 0;
1057 * This will cause a retry with a 6-byte
1060 scsi_requeue_command(q
, cmd
);
1063 scsi_end_request(cmd
, 0, this_count
, 1);
1069 * If the device is in the process of becoming ready,
1072 if (sshdr
.asc
== 0x04 && sshdr
.ascq
== 0x01) {
1073 scsi_requeue_command(q
, cmd
);
1076 if (!(req
->flags
& REQ_QUIET
))
1077 scmd_printk(KERN_INFO
, cmd
,
1078 "Device not ready.\n");
1079 scsi_end_request(cmd
, 0, this_count
, 1);
1081 case VOLUME_OVERFLOW
:
1082 if (!(req
->flags
& REQ_QUIET
)) {
1083 scmd_printk(KERN_INFO
, cmd
,
1084 "Volume overflow, CDB: ");
1085 __scsi_print_command(cmd
->data_cmnd
);
1086 scsi_print_sense("", cmd
);
1088 scsi_end_request(cmd
, 0, block_bytes
, 1);
1093 } /* driver byte != 0 */
1094 if (host_byte(result
) == DID_RESET
) {
1096 * Third party bus reset or reset for error
1097 * recovery reasons. Just retry the request
1098 * and see what happens.
1100 scsi_requeue_command(q
, cmd
);
1104 if (!(req
->flags
& REQ_QUIET
)) {
1105 scmd_printk(KERN_INFO
, cmd
,
1106 "SCSI error: return code = 0x%x\n", result
);
1108 if (driver_byte(result
) & DRIVER_SENSE
)
1109 scsi_print_sense("", cmd
);
1112 * Mark a single buffer as not uptodate. Queue the remainder.
1113 * We sometimes get this cruft in the event that a medium error
1114 * isn't properly reported.
1116 block_bytes
= req
->hard_cur_sectors
<< 9;
1118 block_bytes
= req
->data_len
;
1119 scsi_end_request(cmd
, 0, block_bytes
, 1);
1122 EXPORT_SYMBOL(scsi_io_completion
);
1125 * Function: scsi_init_io()
1127 * Purpose: SCSI I/O initialize function.
1129 * Arguments: cmd - Command descriptor we wish to initialize
1131 * Returns: 0 on success
1132 * BLKPREP_DEFER if the failure is retryable
1133 * BLKPREP_KILL if the failure is fatal
1135 static int scsi_init_io(struct scsi_cmnd
*cmd
)
1137 struct request
*req
= cmd
->request
;
1138 struct scatterlist
*sgpnt
;
1142 * if this is a rq->data based REQ_BLOCK_PC, setup for a non-sg xfer
1144 if ((req
->flags
& REQ_BLOCK_PC
) && !req
->bio
) {
1145 cmd
->request_bufflen
= req
->data_len
;
1146 cmd
->request_buffer
= req
->data
;
1147 req
->buffer
= req
->data
;
1153 * we used to not use scatter-gather for single segment request,
1154 * but now we do (it makes highmem I/O easier to support without
1157 cmd
->use_sg
= req
->nr_phys_segments
;
1160 * if sg table allocation fails, requeue request later.
1162 sgpnt
= scsi_alloc_sgtable(cmd
, GFP_ATOMIC
);
1163 if (unlikely(!sgpnt
)) {
1164 scsi_unprep_request(req
);
1165 return BLKPREP_DEFER
;
1168 cmd
->request_buffer
= (char *) sgpnt
;
1169 cmd
->request_bufflen
= req
->nr_sectors
<< 9;
1170 if (blk_pc_request(req
))
1171 cmd
->request_bufflen
= req
->data_len
;
1175 * Next, walk the list, and fill in the addresses and sizes of
1178 count
= blk_rq_map_sg(req
->q
, req
, cmd
->request_buffer
);
1181 * mapped well, send it off
1183 if (likely(count
<= cmd
->use_sg
)) {
1184 cmd
->use_sg
= count
;
1188 printk(KERN_ERR
"Incorrect number of segments after building list\n");
1189 printk(KERN_ERR
"counted %d, received %d\n", count
, cmd
->use_sg
);
1190 printk(KERN_ERR
"req nr_sec %lu, cur_nr_sec %u\n", req
->nr_sectors
,
1191 req
->current_nr_sectors
);
1193 /* release the command and kill it */
1194 scsi_release_buffers(cmd
);
1195 scsi_put_command(cmd
);
1196 return BLKPREP_KILL
;
1199 static int scsi_prepare_flush_fn(request_queue_t
*q
, struct request
*rq
)
1201 struct scsi_device
*sdev
= q
->queuedata
;
1202 struct scsi_driver
*drv
;
1204 if (sdev
->sdev_state
== SDEV_RUNNING
) {
1205 drv
= *(struct scsi_driver
**) rq
->rq_disk
->private_data
;
1207 if (drv
->prepare_flush
)
1208 return drv
->prepare_flush(q
, rq
);
1214 static void scsi_end_flush_fn(request_queue_t
*q
, struct request
*rq
)
1216 struct scsi_device
*sdev
= q
->queuedata
;
1217 struct request
*flush_rq
= rq
->end_io_data
;
1218 struct scsi_driver
*drv
;
1220 if (flush_rq
->errors
) {
1221 printk("scsi: barrier error, disabling flush support\n");
1222 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1225 if (sdev
->sdev_state
== SDEV_RUNNING
) {
1226 drv
= *(struct scsi_driver
**) rq
->rq_disk
->private_data
;
1227 drv
->end_flush(q
, rq
);
1231 static int scsi_issue_flush_fn(request_queue_t
*q
, struct gendisk
*disk
,
1232 sector_t
*error_sector
)
1234 struct scsi_device
*sdev
= q
->queuedata
;
1235 struct scsi_driver
*drv
;
1237 if (sdev
->sdev_state
!= SDEV_RUNNING
)
1240 drv
= *(struct scsi_driver
**) disk
->private_data
;
1241 if (drv
->issue_flush
)
1242 return drv
->issue_flush(&sdev
->sdev_gendev
, error_sector
);
1247 static void scsi_generic_done(struct scsi_cmnd
*cmd
)
1249 BUG_ON(!blk_pc_request(cmd
->request
));
1251 * This will complete the whole command with uptodate=1 so
1252 * as far as the block layer is concerned the command completed
1253 * successfully. Since this is a REQ_BLOCK_PC command the
1254 * caller should check the request's errors value
1256 scsi_io_completion(cmd
, cmd
->bufflen
, 0);
1259 static int scsi_prep_fn(struct request_queue
*q
, struct request
*req
)
1261 struct scsi_device
*sdev
= q
->queuedata
;
1262 struct scsi_cmnd
*cmd
;
1263 int specials_only
= 0;
1266 * Just check to see if the device is online. If it isn't, we
1267 * refuse to process any commands. The device must be brought
1268 * online before trying any recovery commands
1270 if (unlikely(!scsi_device_online(sdev
))) {
1271 sdev_printk(KERN_ERR
, sdev
,
1272 "rejecting I/O to offline device\n");
1275 if (unlikely(sdev
->sdev_state
!= SDEV_RUNNING
)) {
1276 /* OK, we're not in a running state don't prep
1278 if (sdev
->sdev_state
== SDEV_DEL
) {
1279 /* Device is fully deleted, no commands
1280 * at all allowed down */
1281 sdev_printk(KERN_ERR
, sdev
,
1282 "rejecting I/O to dead device\n");
1285 /* OK, we only allow special commands (i.e. not
1286 * user initiated ones */
1287 specials_only
= sdev
->sdev_state
;
1291 * Find the actual device driver associated with this command.
1292 * The SPECIAL requests are things like character device or
1293 * ioctls, which did not originate from ll_rw_blk. Note that
1294 * the special field is also used to indicate the cmd for
1295 * the remainder of a partially fulfilled request that can
1296 * come up when there is a medium error. We have to treat
1297 * these two cases differently. We differentiate by looking
1298 * at request->cmd, as this tells us the real story.
1300 if (req
->flags
& REQ_SPECIAL
&& req
->special
) {
1301 struct scsi_request
*sreq
= req
->special
;
1303 if (sreq
->sr_magic
== SCSI_REQ_MAGIC
) {
1304 cmd
= scsi_get_command(sreq
->sr_device
, GFP_ATOMIC
);
1307 scsi_init_cmd_from_req(cmd
, sreq
);
1310 } else if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1312 if(unlikely(specials_only
) && !(req
->flags
& REQ_SPECIAL
)) {
1313 if(specials_only
== SDEV_QUIESCE
||
1314 specials_only
== SDEV_BLOCK
)
1317 sdev_printk(KERN_ERR
, sdev
,
1318 "rejecting I/O to device being removed\n");
1324 * Now try and find a command block that we can use.
1326 if (!req
->special
) {
1327 cmd
= scsi_get_command(sdev
, GFP_ATOMIC
);
1333 /* pull a tag out of the request if we have one */
1334 cmd
->tag
= req
->tag
;
1336 blk_dump_rq_flags(req
, "SCSI bad req");
1340 /* note the overloading of req->special. When the tag
1341 * is active it always means cmd. If the tag goes
1342 * back for re-queueing, it may be reset */
1347 * FIXME: drop the lock here because the functions below
1348 * expect to be called without the queue lock held. Also,
1349 * previously, we dequeued the request before dropping the
1350 * lock. We hope REQ_STARTED prevents anything untoward from
1353 if (req
->flags
& (REQ_CMD
| REQ_BLOCK_PC
)) {
1354 struct scsi_driver
*drv
;
1358 * This will do a couple of things:
1359 * 1) Fill in the actual SCSI command.
1360 * 2) Fill in any other upper-level specific fields
1363 * If this returns 0, it means that the request failed
1364 * (reading past end of disk, reading offline device,
1365 * etc). This won't actually talk to the device, but
1366 * some kinds of consistency checking may cause the
1367 * request to be rejected immediately.
1371 * This sets up the scatter-gather table (allocating if
1374 ret
= scsi_init_io(cmd
);
1376 /* For BLKPREP_KILL/DEFER the cmd was released */
1384 * Initialize the actual SCSI command for this request.
1387 drv
= *(struct scsi_driver
**)req
->rq_disk
->private_data
;
1388 if (unlikely(!drv
->init_command(cmd
))) {
1389 scsi_release_buffers(cmd
);
1390 scsi_put_command(cmd
);
1394 memcpy(cmd
->cmnd
, req
->cmd
, sizeof(cmd
->cmnd
));
1395 cmd
->cmd_len
= req
->cmd_len
;
1396 if (rq_data_dir(req
) == WRITE
)
1397 cmd
->sc_data_direction
= DMA_TO_DEVICE
;
1398 else if (req
->data_len
)
1399 cmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1401 cmd
->sc_data_direction
= DMA_NONE
;
1403 cmd
->transfersize
= req
->data_len
;
1404 cmd
->allowed
= req
->retries
;
1405 cmd
->timeout_per_command
= req
->timeout
;
1406 cmd
->done
= scsi_generic_done
;
1411 * The request is now prepped, no need to come back here
1413 req
->flags
|= REQ_DONTPREP
;
1417 /* If we defer, the elv_next_request() returns NULL, but the
1418 * queue must be restarted, so we plug here if no returning
1419 * command will automatically do that. */
1420 if (sdev
->device_busy
== 0)
1422 return BLKPREP_DEFER
;
1424 req
->errors
= DID_NO_CONNECT
<< 16;
1425 return BLKPREP_KILL
;
1429 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1432 * Called with the queue_lock held.
1434 static inline int scsi_dev_queue_ready(struct request_queue
*q
,
1435 struct scsi_device
*sdev
)
1437 if (sdev
->device_busy
>= sdev
->queue_depth
)
1439 if (sdev
->device_busy
== 0 && sdev
->device_blocked
) {
1441 * unblock after device_blocked iterates to zero
1443 if (--sdev
->device_blocked
== 0) {
1445 sdev_printk(KERN_INFO
, sdev
,
1446 "unblocking device at zero depth\n"));
1452 if (sdev
->device_blocked
)
1459 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1460 * return 0. We must end up running the queue again whenever 0 is
1461 * returned, else IO can hang.
1463 * Called with host_lock held.
1465 static inline int scsi_host_queue_ready(struct request_queue
*q
,
1466 struct Scsi_Host
*shost
,
1467 struct scsi_device
*sdev
)
1469 if (scsi_host_in_recovery(shost
))
1471 if (shost
->host_busy
== 0 && shost
->host_blocked
) {
1473 * unblock after host_blocked iterates to zero
1475 if (--shost
->host_blocked
== 0) {
1477 printk("scsi%d unblocking host at zero depth\n",
1484 if ((shost
->can_queue
> 0 && shost
->host_busy
>= shost
->can_queue
) ||
1485 shost
->host_blocked
|| shost
->host_self_blocked
) {
1486 if (list_empty(&sdev
->starved_entry
))
1487 list_add_tail(&sdev
->starved_entry
, &shost
->starved_list
);
1491 /* We're OK to process the command, so we can't be starved */
1492 if (!list_empty(&sdev
->starved_entry
))
1493 list_del_init(&sdev
->starved_entry
);
1499 * Kill a request for a dead device
1501 static void scsi_kill_request(struct request
*req
, request_queue_t
*q
)
1503 struct scsi_cmnd
*cmd
= req
->special
;
1505 blkdev_dequeue_request(req
);
1507 if (unlikely(cmd
== NULL
)) {
1508 printk(KERN_CRIT
"impossible request in %s.\n",
1513 scsi_init_cmd_errh(cmd
);
1514 cmd
->result
= DID_NO_CONNECT
<< 16;
1515 atomic_inc(&cmd
->device
->iorequest_cnt
);
1520 * Function: scsi_request_fn()
1522 * Purpose: Main strategy routine for SCSI.
1524 * Arguments: q - Pointer to actual queue.
1528 * Lock status: IO request lock assumed to be held when called.
1530 static void scsi_request_fn(struct request_queue
*q
)
1532 struct scsi_device
*sdev
= q
->queuedata
;
1533 struct Scsi_Host
*shost
;
1534 struct scsi_cmnd
*cmd
;
1535 struct request
*req
;
1538 printk("scsi: killing requests for dead queue\n");
1539 while ((req
= elv_next_request(q
)) != NULL
)
1540 scsi_kill_request(req
, q
);
1544 if(!get_device(&sdev
->sdev_gendev
))
1545 /* We must be tearing the block queue down already */
1549 * To start with, we keep looping until the queue is empty, or until
1550 * the host is no longer able to accept any more requests.
1553 while (!blk_queue_plugged(q
)) {
1556 * get next queueable request. We do this early to make sure
1557 * that the request is fully prepared even if we cannot
1560 req
= elv_next_request(q
);
1561 if (!req
|| !scsi_dev_queue_ready(q
, sdev
))
1564 if (unlikely(!scsi_device_online(sdev
))) {
1565 sdev_printk(KERN_ERR
, sdev
,
1566 "rejecting I/O to offline device\n");
1567 scsi_kill_request(req
, q
);
1573 * Remove the request from the request list.
1575 if (!(blk_queue_tagged(q
) && !blk_queue_start_tag(q
, req
)))
1576 blkdev_dequeue_request(req
);
1577 sdev
->device_busy
++;
1579 spin_unlock(q
->queue_lock
);
1581 if (unlikely(cmd
== NULL
)) {
1582 printk(KERN_CRIT
"impossible request in %s.\n"
1583 "please mail a stack trace to "
1584 "linux-scsi@vger.kernel.org",
1588 spin_lock(shost
->host_lock
);
1590 if (!scsi_host_queue_ready(q
, shost
, sdev
))
1592 if (sdev
->single_lun
) {
1593 if (scsi_target(sdev
)->starget_sdev_user
&&
1594 scsi_target(sdev
)->starget_sdev_user
!= sdev
)
1596 scsi_target(sdev
)->starget_sdev_user
= sdev
;
1601 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1602 * take the lock again.
1604 spin_unlock_irq(shost
->host_lock
);
1607 * Finally, initialize any error handling parameters, and set up
1608 * the timers for timeouts.
1610 scsi_init_cmd_errh(cmd
);
1613 * Dispatch the command to the low-level driver.
1615 rtn
= scsi_dispatch_cmd(cmd
);
1616 spin_lock_irq(q
->queue_lock
);
1618 /* we're refusing the command; because of
1619 * the way locks get dropped, we need to
1620 * check here if plugging is required */
1621 if(sdev
->device_busy
== 0)
1631 spin_unlock_irq(shost
->host_lock
);
1634 * lock q, handle tag, requeue req, and decrement device_busy. We
1635 * must return with queue_lock held.
1637 * Decrementing device_busy without checking it is OK, as all such
1638 * cases (host limits or settings) should run the queue at some
1641 spin_lock_irq(q
->queue_lock
);
1642 blk_requeue_request(q
, req
);
1643 sdev
->device_busy
--;
1644 if(sdev
->device_busy
== 0)
1647 /* must be careful here...if we trigger the ->remove() function
1648 * we cannot be holding the q lock */
1649 spin_unlock_irq(q
->queue_lock
);
1650 put_device(&sdev
->sdev_gendev
);
1651 spin_lock_irq(q
->queue_lock
);
1654 u64
scsi_calculate_bounce_limit(struct Scsi_Host
*shost
)
1656 struct device
*host_dev
;
1657 u64 bounce_limit
= 0xffffffff;
1659 if (shost
->unchecked_isa_dma
)
1660 return BLK_BOUNCE_ISA
;
1662 * Platforms with virtual-DMA translation
1663 * hardware have no practical limit.
1665 if (!PCI_DMA_BUS_IS_PHYS
)
1666 return BLK_BOUNCE_ANY
;
1668 host_dev
= scsi_get_device(shost
);
1669 if (host_dev
&& host_dev
->dma_mask
)
1670 bounce_limit
= *host_dev
->dma_mask
;
1672 return bounce_limit
;
1674 EXPORT_SYMBOL(scsi_calculate_bounce_limit
);
1676 struct request_queue
*scsi_alloc_queue(struct scsi_device
*sdev
)
1678 struct Scsi_Host
*shost
= sdev
->host
;
1679 struct request_queue
*q
;
1681 q
= blk_init_queue(scsi_request_fn
, NULL
);
1685 blk_queue_prep_rq(q
, scsi_prep_fn
);
1687 blk_queue_max_hw_segments(q
, shost
->sg_tablesize
);
1688 blk_queue_max_phys_segments(q
, SCSI_MAX_PHYS_SEGMENTS
);
1689 blk_queue_max_sectors(q
, shost
->max_sectors
);
1690 blk_queue_bounce_limit(q
, scsi_calculate_bounce_limit(shost
));
1691 blk_queue_segment_boundary(q
, shost
->dma_boundary
);
1692 blk_queue_issue_flush_fn(q
, scsi_issue_flush_fn
);
1695 * ordered tags are superior to flush ordering
1697 if (shost
->ordered_tag
)
1698 blk_queue_ordered(q
, QUEUE_ORDERED_TAG
);
1699 else if (shost
->ordered_flush
) {
1700 blk_queue_ordered(q
, QUEUE_ORDERED_FLUSH
);
1701 q
->prepare_flush_fn
= scsi_prepare_flush_fn
;
1702 q
->end_flush_fn
= scsi_end_flush_fn
;
1705 if (!shost
->use_clustering
)
1706 clear_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
1710 void scsi_free_queue(struct request_queue
*q
)
1712 blk_cleanup_queue(q
);
1716 * Function: scsi_block_requests()
1718 * Purpose: Utility function used by low-level drivers to prevent further
1719 * commands from being queued to the device.
1721 * Arguments: shost - Host in question
1725 * Lock status: No locks are assumed held.
1727 * Notes: There is no timer nor any other means by which the requests
1728 * get unblocked other than the low-level driver calling
1729 * scsi_unblock_requests().
1731 void scsi_block_requests(struct Scsi_Host
*shost
)
1733 shost
->host_self_blocked
= 1;
1735 EXPORT_SYMBOL(scsi_block_requests
);
1738 * Function: scsi_unblock_requests()
1740 * Purpose: Utility function used by low-level drivers to allow further
1741 * commands from being queued to the device.
1743 * Arguments: shost - Host in question
1747 * Lock status: No locks are assumed held.
1749 * Notes: There is no timer nor any other means by which the requests
1750 * get unblocked other than the low-level driver calling
1751 * scsi_unblock_requests().
1753 * This is done as an API function so that changes to the
1754 * internals of the scsi mid-layer won't require wholesale
1755 * changes to drivers that use this feature.
1757 void scsi_unblock_requests(struct Scsi_Host
*shost
)
1759 shost
->host_self_blocked
= 0;
1760 scsi_run_host_queues(shost
);
1762 EXPORT_SYMBOL(scsi_unblock_requests
);
1764 int __init
scsi_init_queue(void)
1768 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1769 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1770 int size
= sgp
->size
* sizeof(struct scatterlist
);
1772 sgp
->slab
= kmem_cache_create(sgp
->name
, size
, 0,
1773 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
1775 printk(KERN_ERR
"SCSI: can't init sg slab %s\n",
1779 sgp
->pool
= mempool_create(SG_MEMPOOL_SIZE
,
1780 mempool_alloc_slab
, mempool_free_slab
,
1783 printk(KERN_ERR
"SCSI: can't init sg mempool %s\n",
1791 void scsi_exit_queue(void)
1795 for (i
= 0; i
< SG_MEMPOOL_NR
; i
++) {
1796 struct scsi_host_sg_pool
*sgp
= scsi_sg_pools
+ i
;
1797 mempool_destroy(sgp
->pool
);
1798 kmem_cache_destroy(sgp
->slab
);
1802 * scsi_mode_sense - issue a mode sense, falling back from 10 to
1803 * six bytes if necessary.
1804 * @sdev: SCSI device to be queried
1805 * @dbd: set if mode sense will allow block descriptors to be returned
1806 * @modepage: mode page being requested
1807 * @buffer: request buffer (may not be smaller than eight bytes)
1808 * @len: length of request buffer.
1809 * @timeout: command timeout
1810 * @retries: number of retries before failing
1811 * @data: returns a structure abstracting the mode header data
1812 * @sense: place to put sense data (or NULL if no sense to be collected).
1813 * must be SCSI_SENSE_BUFFERSIZE big.
1815 * Returns zero if unsuccessful, or the header offset (either 4
1816 * or 8 depending on whether a six or ten byte command was
1817 * issued) if successful.
1820 scsi_mode_sense(struct scsi_device
*sdev
, int dbd
, int modepage
,
1821 unsigned char *buffer
, int len
, int timeout
, int retries
,
1822 struct scsi_mode_data
*data
, struct scsi_sense_hdr
*sshdr
) {
1823 unsigned char cmd
[12];
1827 struct scsi_sense_hdr my_sshdr
;
1829 memset(data
, 0, sizeof(*data
));
1830 memset(&cmd
[0], 0, 12);
1831 cmd
[1] = dbd
& 0x18; /* allows DBD and LLBA bits */
1834 /* caller might not be interested in sense, but we need it */
1839 use_10_for_ms
= sdev
->use_10_for_ms
;
1841 if (use_10_for_ms
) {
1845 cmd
[0] = MODE_SENSE_10
;
1852 cmd
[0] = MODE_SENSE
;
1857 memset(buffer
, 0, len
);
1859 result
= scsi_execute_req(sdev
, cmd
, DMA_FROM_DEVICE
, buffer
, len
,
1860 sshdr
, timeout
, retries
);
1862 /* This code looks awful: what it's doing is making sure an
1863 * ILLEGAL REQUEST sense return identifies the actual command
1864 * byte as the problem. MODE_SENSE commands can return
1865 * ILLEGAL REQUEST if the code page isn't supported */
1867 if (use_10_for_ms
&& !scsi_status_is_good(result
) &&
1868 (driver_byte(result
) & DRIVER_SENSE
)) {
1869 if (scsi_sense_valid(sshdr
)) {
1870 if ((sshdr
->sense_key
== ILLEGAL_REQUEST
) &&
1871 (sshdr
->asc
== 0x20) && (sshdr
->ascq
== 0)) {
1873 * Invalid command operation code
1875 sdev
->use_10_for_ms
= 0;
1881 if(scsi_status_is_good(result
)) {
1882 data
->header_length
= header_length
;
1884 data
->length
= buffer
[0]*256 + buffer
[1] + 2;
1885 data
->medium_type
= buffer
[2];
1886 data
->device_specific
= buffer
[3];
1887 data
->longlba
= buffer
[4] & 0x01;
1888 data
->block_descriptor_length
= buffer
[6]*256
1891 data
->length
= buffer
[0] + 1;
1892 data
->medium_type
= buffer
[1];
1893 data
->device_specific
= buffer
[2];
1894 data
->block_descriptor_length
= buffer
[3];
1900 EXPORT_SYMBOL(scsi_mode_sense
);
1903 scsi_test_unit_ready(struct scsi_device
*sdev
, int timeout
, int retries
)
1906 TEST_UNIT_READY
, 0, 0, 0, 0, 0,
1908 struct scsi_sense_hdr sshdr
;
1911 result
= scsi_execute_req(sdev
, cmd
, DMA_NONE
, NULL
, 0, &sshdr
,
1914 if ((driver_byte(result
) & DRIVER_SENSE
) && sdev
->removable
) {
1916 if ((scsi_sense_valid(&sshdr
)) &&
1917 ((sshdr
.sense_key
== UNIT_ATTENTION
) ||
1918 (sshdr
.sense_key
== NOT_READY
))) {
1925 EXPORT_SYMBOL(scsi_test_unit_ready
);
1928 * scsi_device_set_state - Take the given device through the device
1930 * @sdev: scsi device to change the state of.
1931 * @state: state to change to.
1933 * Returns zero if unsuccessful or an error if the requested
1934 * transition is illegal.
1937 scsi_device_set_state(struct scsi_device
*sdev
, enum scsi_device_state state
)
1939 enum scsi_device_state oldstate
= sdev
->sdev_state
;
1941 if (state
== oldstate
)
1946 /* There are no legal states that come back to
1947 * created. This is the manually initialised start
2017 sdev
->sdev_state
= state
;
2021 SCSI_LOG_ERROR_RECOVERY(1,
2022 sdev_printk(KERN_ERR
, sdev
,
2023 "Illegal state transition %s->%s\n",
2024 scsi_device_state_name(oldstate
),
2025 scsi_device_state_name(state
))
2029 EXPORT_SYMBOL(scsi_device_set_state
);
2032 * scsi_device_quiesce - Block user issued commands.
2033 * @sdev: scsi device to quiesce.
2035 * This works by trying to transition to the SDEV_QUIESCE state
2036 * (which must be a legal transition). When the device is in this
2037 * state, only special requests will be accepted, all others will
2038 * be deferred. Since special requests may also be requeued requests,
2039 * a successful return doesn't guarantee the device will be
2040 * totally quiescent.
2042 * Must be called with user context, may sleep.
2044 * Returns zero if unsuccessful or an error if not.
2047 scsi_device_quiesce(struct scsi_device
*sdev
)
2049 int err
= scsi_device_set_state(sdev
, SDEV_QUIESCE
);
2053 scsi_run_queue(sdev
->request_queue
);
2054 while (sdev
->device_busy
) {
2055 msleep_interruptible(200);
2056 scsi_run_queue(sdev
->request_queue
);
2060 EXPORT_SYMBOL(scsi_device_quiesce
);
2063 * scsi_device_resume - Restart user issued commands to a quiesced device.
2064 * @sdev: scsi device to resume.
2066 * Moves the device from quiesced back to running and restarts the
2069 * Must be called with user context, may sleep.
2072 scsi_device_resume(struct scsi_device
*sdev
)
2074 if(scsi_device_set_state(sdev
, SDEV_RUNNING
))
2076 scsi_run_queue(sdev
->request_queue
);
2078 EXPORT_SYMBOL(scsi_device_resume
);
2081 device_quiesce_fn(struct scsi_device
*sdev
, void *data
)
2083 scsi_device_quiesce(sdev
);
2087 scsi_target_quiesce(struct scsi_target
*starget
)
2089 starget_for_each_device(starget
, NULL
, device_quiesce_fn
);
2091 EXPORT_SYMBOL(scsi_target_quiesce
);
2094 device_resume_fn(struct scsi_device
*sdev
, void *data
)
2096 scsi_device_resume(sdev
);
2100 scsi_target_resume(struct scsi_target
*starget
)
2102 starget_for_each_device(starget
, NULL
, device_resume_fn
);
2104 EXPORT_SYMBOL(scsi_target_resume
);
2107 * scsi_internal_device_block - internal function to put a device
2108 * temporarily into the SDEV_BLOCK state
2109 * @sdev: device to block
2111 * Block request made by scsi lld's to temporarily stop all
2112 * scsi commands on the specified device. Called from interrupt
2113 * or normal process context.
2115 * Returns zero if successful or error if not
2118 * This routine transitions the device to the SDEV_BLOCK state
2119 * (which must be a legal transition). When the device is in this
2120 * state, all commands are deferred until the scsi lld reenables
2121 * the device with scsi_device_unblock or device_block_tmo fires.
2122 * This routine assumes the host_lock is held on entry.
2125 scsi_internal_device_block(struct scsi_device
*sdev
)
2127 request_queue_t
*q
= sdev
->request_queue
;
2128 unsigned long flags
;
2131 err
= scsi_device_set_state(sdev
, SDEV_BLOCK
);
2136 * The device has transitioned to SDEV_BLOCK. Stop the
2137 * block layer from calling the midlayer with this device's
2140 spin_lock_irqsave(q
->queue_lock
, flags
);
2142 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2146 EXPORT_SYMBOL_GPL(scsi_internal_device_block
);
2149 * scsi_internal_device_unblock - resume a device after a block request
2150 * @sdev: device to resume
2152 * Called by scsi lld's or the midlayer to restart the device queue
2153 * for the previously suspended scsi device. Called from interrupt or
2154 * normal process context.
2156 * Returns zero if successful or error if not.
2159 * This routine transitions the device to the SDEV_RUNNING state
2160 * (which must be a legal transition) allowing the midlayer to
2161 * goose the queue for this device. This routine assumes the
2162 * host_lock is held upon entry.
2165 scsi_internal_device_unblock(struct scsi_device
*sdev
)
2167 request_queue_t
*q
= sdev
->request_queue
;
2169 unsigned long flags
;
2172 * Try to transition the scsi device to SDEV_RUNNING
2173 * and goose the device queue if successful.
2175 err
= scsi_device_set_state(sdev
, SDEV_RUNNING
);
2179 spin_lock_irqsave(q
->queue_lock
, flags
);
2181 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2185 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock
);
2188 device_block(struct scsi_device
*sdev
, void *data
)
2190 scsi_internal_device_block(sdev
);
2194 target_block(struct device
*dev
, void *data
)
2196 if (scsi_is_target_device(dev
))
2197 starget_for_each_device(to_scsi_target(dev
), NULL
,
2203 scsi_target_block(struct device
*dev
)
2205 if (scsi_is_target_device(dev
))
2206 starget_for_each_device(to_scsi_target(dev
), NULL
,
2209 device_for_each_child(dev
, NULL
, target_block
);
2211 EXPORT_SYMBOL_GPL(scsi_target_block
);
2214 device_unblock(struct scsi_device
*sdev
, void *data
)
2216 scsi_internal_device_unblock(sdev
);
2220 target_unblock(struct device
*dev
, void *data
)
2222 if (scsi_is_target_device(dev
))
2223 starget_for_each_device(to_scsi_target(dev
), NULL
,
2229 scsi_target_unblock(struct device
*dev
)
2231 if (scsi_is_target_device(dev
))
2232 starget_for_each_device(to_scsi_target(dev
), NULL
,
2235 device_for_each_child(dev
, NULL
, target_unblock
);
2237 EXPORT_SYMBOL_GPL(scsi_target_unblock
);