2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/completion.h>
13 #include <linux/kernel.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/pci.h>
18 #include <linux/delay.h>
19 #include <linux/hardirq.h>
20 #include <linux/scatterlist.h>
22 #include <scsi/scsi.h>
23 #include <scsi/scsi_cmnd.h>
24 #include <scsi/scsi_dbg.h>
25 #include <scsi/scsi_device.h>
26 #include <scsi/scsi_driver.h>
27 #include <scsi/scsi_eh.h>
28 #include <scsi/scsi_host.h>
30 #include "scsi_priv.h"
31 #include "scsi_logging.h"
34 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
35 #define SG_MEMPOOL_SIZE 2
38 * The maximum number of SG segments that we will put inside a scatterlist
39 * (unless chaining is used). Should ideally fit inside a single page, to
40 * avoid a higher order allocation.
42 #define SCSI_MAX_SG_SEGMENTS 128
44 struct scsi_host_sg_pool {
47 struct kmem_cache *slab;
51 #define SP(x) { x, "sgpool-" #x }
52 static struct scsi_host_sg_pool scsi_sg_pools[] = {
55 #if (SCSI_MAX_SG_SEGMENTS > 16)
57 #if (SCSI_MAX_SG_SEGMENTS > 32)
59 #if (SCSI_MAX_SG_SEGMENTS > 64)
67 static void scsi_run_queue(struct request_queue *q);
70 * Function: scsi_unprep_request()
72 * Purpose: Remove all preparation done for a request, including its
73 * associated scsi_cmnd, so that it can be requeued.
75 * Arguments: req - request to unprepare
77 * Lock status: Assumed that no locks are held upon entry.
81 static void scsi_unprep_request(struct request *req)
83 struct scsi_cmnd *cmd = req->special;
85 req->cmd_flags &= ~REQ_DONTPREP;
88 scsi_put_command(cmd);
92 * Function: scsi_queue_insert()
94 * Purpose: Insert a command in the midlevel queue.
96 * Arguments: cmd - command that we are adding to queue.
97 * reason - why we are inserting command to queue.
99 * Lock status: Assumed that lock is not held upon entry.
103 * Notes: We do this for one of two cases. Either the host is busy
104 * and it cannot accept any more commands for the time being,
105 * or the device returned QUEUE_FULL and can accept no more
107 * Notes: This could be called either from an interrupt context or a
108 * normal process context.
110 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
112 struct Scsi_Host *host = cmd->device->host;
113 struct scsi_device *device = cmd->device;
114 struct request_queue *q = device->request_queue;
118 printk("Inserting command %p into mlqueue\n", cmd));
121 * Set the appropriate busy bit for the device/host.
123 * If the host/device isn't busy, assume that something actually
124 * completed, and that we should be able to queue a command now.
126 * Note that the prior mid-layer assumption that any host could
127 * always queue at least one command is now broken. The mid-layer
128 * will implement a user specifiable stall (see
129 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
130 * if a command is requeued with no other commands outstanding
131 * either for the device or for the host.
133 if (reason == SCSI_MLQUEUE_HOST_BUSY)
134 host->host_blocked = host->max_host_blocked;
135 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
136 device->device_blocked = device->max_device_blocked;
139 * Decrement the counters, since these commands are no longer
140 * active on the host/device.
142 scsi_device_unbusy(device);
145 * Requeue this command. It will go before all other commands
146 * that are already in the queue.
148 * NOTE: there is magic here about the way the queue is plugged if
149 * we have no outstanding commands.
151 * Although we *don't* plug the queue, we call the request
152 * function. The SCSI request function detects the blocked condition
153 * and plugs the queue appropriately.
155 spin_lock_irqsave(q->queue_lock, flags);
156 blk_requeue_request(q, cmd->request);
157 spin_unlock_irqrestore(q->queue_lock, flags);
165 * scsi_execute - insert request and wait for the result
168 * @data_direction: data direction
169 * @buffer: data buffer
170 * @bufflen: len of buffer
171 * @sense: optional sense buffer
172 * @timeout: request timeout in seconds
173 * @retries: number of times to retry request
174 * @flags: or into request flags;
176 * returns the req->errors value which is the scsi_cmnd result
179 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
180 int data_direction, void *buffer, unsigned bufflen,
181 unsigned char *sense, int timeout, int retries, int flags)
184 int write = (data_direction == DMA_TO_DEVICE);
185 int ret = DRIVER_ERROR << 24;
187 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
189 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
190 buffer, bufflen, __GFP_WAIT))
193 req->cmd_len = COMMAND_SIZE(cmd[0]);
194 memcpy(req->cmd, cmd, req->cmd_len);
197 req->retries = retries;
198 req->timeout = timeout;
199 req->cmd_type = REQ_TYPE_BLOCK_PC;
200 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
203 * head injection *required* here otherwise quiesce won't work
205 blk_execute_rq(req->q, NULL, req, 1);
209 blk_put_request(req);
213 EXPORT_SYMBOL(scsi_execute);
216 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
217 int data_direction, void *buffer, unsigned bufflen,
218 struct scsi_sense_hdr *sshdr, int timeout, int retries)
224 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
226 return DRIVER_ERROR << 24;
228 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
229 sense, timeout, retries, 0);
231 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
236 EXPORT_SYMBOL(scsi_execute_req);
238 struct scsi_io_context {
240 void (*done)(void *data, char *sense, int result, int resid);
241 char sense[SCSI_SENSE_BUFFERSIZE];
244 static struct kmem_cache *scsi_io_context_cache;
246 static void scsi_end_async(struct request *req, int uptodate)
248 struct scsi_io_context *sioc = req->end_io_data;
251 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
253 kmem_cache_free(scsi_io_context_cache, sioc);
254 __blk_put_request(req->q, req);
257 static int scsi_merge_bio(struct request *rq, struct bio *bio)
259 struct request_queue *q = rq->q;
261 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
262 if (rq_data_dir(rq) == WRITE)
263 bio->bi_rw |= (1 << BIO_RW);
264 blk_queue_bounce(q, &bio);
266 return blk_rq_append_bio(q, rq, bio);
269 static void scsi_bi_endio(struct bio *bio, int error)
275 * scsi_req_map_sg - map a scatterlist into a request
276 * @rq: request to fill
278 * @nsegs: number of elements
279 * @bufflen: len of buffer
280 * @gfp: memory allocation flags
282 * scsi_req_map_sg maps a scatterlist into a request so that the
283 * request can be sent to the block layer. We do not trust the scatterlist
284 * sent to use, as some ULDs use that struct to only organize the pages.
286 static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
287 int nsegs, unsigned bufflen, gfp_t gfp)
289 struct request_queue *q = rq->q;
290 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
291 unsigned int data_len = bufflen, len, bytes, off;
292 struct scatterlist *sg;
294 struct bio *bio = NULL;
295 int i, err, nr_vecs = 0;
297 for_each_sg(sgl, sg, nsegs, i) {
303 while (len > 0 && data_len > 0) {
305 * sg sends a scatterlist that is larger than
306 * the data_len it wants transferred for certain
309 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
310 bytes = min(bytes, data_len);
313 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
316 bio = bio_alloc(gfp, nr_vecs);
321 bio->bi_end_io = scsi_bi_endio;
324 if (bio_add_pc_page(q, bio, page, bytes, off) !=
331 if (bio->bi_vcnt >= nr_vecs) {
332 err = scsi_merge_bio(rq, bio);
347 rq->buffer = rq->data = NULL;
348 rq->data_len = bufflen;
352 while ((bio = rq->bio) != NULL) {
353 rq->bio = bio->bi_next;
355 * call endio instead of bio_put incase it was bounced
364 * scsi_execute_async - insert request
367 * @cmd_len: length of scsi cdb
368 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
369 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
370 * @bufflen: len of buffer
371 * @use_sg: if buffer is a scatterlist this is the number of elements
372 * @timeout: request timeout in seconds
373 * @retries: number of times to retry request
374 * @privdata: data passed to done()
375 * @done: callback function when done
376 * @gfp: memory allocation flags
378 int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
379 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
380 int use_sg, int timeout, int retries, void *privdata,
381 void (*done)(void *, char *, int, int), gfp_t gfp)
384 struct scsi_io_context *sioc;
386 int write = (data_direction == DMA_TO_DEVICE);
388 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp);
390 return DRIVER_ERROR << 24;
392 req = blk_get_request(sdev->request_queue, write, gfp);
395 req->cmd_type = REQ_TYPE_BLOCK_PC;
396 req->cmd_flags |= REQ_QUIET;
399 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
401 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
406 req->cmd_len = cmd_len;
407 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
408 memcpy(req->cmd, cmd, req->cmd_len);
409 req->sense = sioc->sense;
411 req->timeout = timeout;
412 req->retries = retries;
413 req->end_io_data = sioc;
415 sioc->data = privdata;
418 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
422 blk_put_request(req);
424 kmem_cache_free(scsi_io_context_cache, sioc);
425 return DRIVER_ERROR << 24;
427 EXPORT_SYMBOL_GPL(scsi_execute_async);
430 * Function: scsi_init_cmd_errh()
432 * Purpose: Initialize cmd fields related to error handling.
434 * Arguments: cmd - command that is ready to be queued.
436 * Notes: This function has the job of initializing a number of
437 * fields related to error handling. Typically this will
438 * be called once for each command, as required.
440 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
442 cmd->serial_number = 0;
444 memset(cmd->sense_buffer, 0, sizeof cmd->sense_buffer);
445 if (cmd->cmd_len == 0)
446 cmd->cmd_len = COMMAND_SIZE(cmd->cmnd[0]);
449 void scsi_device_unbusy(struct scsi_device *sdev)
451 struct Scsi_Host *shost = sdev->host;
454 spin_lock_irqsave(shost->host_lock, flags);
456 if (unlikely(scsi_host_in_recovery(shost) &&
457 (shost->host_failed || shost->host_eh_scheduled)))
458 scsi_eh_wakeup(shost);
459 spin_unlock(shost->host_lock);
460 spin_lock(sdev->request_queue->queue_lock);
462 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
466 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
467 * and call blk_run_queue for all the scsi_devices on the target -
468 * including current_sdev first.
470 * Called with *no* scsi locks held.
472 static void scsi_single_lun_run(struct scsi_device *current_sdev)
474 struct Scsi_Host *shost = current_sdev->host;
475 struct scsi_device *sdev, *tmp;
476 struct scsi_target *starget = scsi_target(current_sdev);
479 spin_lock_irqsave(shost->host_lock, flags);
480 starget->starget_sdev_user = NULL;
481 spin_unlock_irqrestore(shost->host_lock, flags);
484 * Call blk_run_queue for all LUNs on the target, starting with
485 * current_sdev. We race with others (to set starget_sdev_user),
486 * but in most cases, we will be first. Ideally, each LU on the
487 * target would get some limited time or requests on the target.
489 blk_run_queue(current_sdev->request_queue);
491 spin_lock_irqsave(shost->host_lock, flags);
492 if (starget->starget_sdev_user)
494 list_for_each_entry_safe(sdev, tmp, &starget->devices,
495 same_target_siblings) {
496 if (sdev == current_sdev)
498 if (scsi_device_get(sdev))
501 spin_unlock_irqrestore(shost->host_lock, flags);
502 blk_run_queue(sdev->request_queue);
503 spin_lock_irqsave(shost->host_lock, flags);
505 scsi_device_put(sdev);
508 spin_unlock_irqrestore(shost->host_lock, flags);
512 * Function: scsi_run_queue()
514 * Purpose: Select a proper request queue to serve next
516 * Arguments: q - last request's queue
520 * Notes: The previous command was completely finished, start
521 * a new one if possible.
523 static void scsi_run_queue(struct request_queue *q)
525 struct scsi_device *sdev = q->queuedata;
526 struct Scsi_Host *shost = sdev->host;
529 if (scsi_target(sdev)->single_lun)
530 scsi_single_lun_run(sdev);
532 spin_lock_irqsave(shost->host_lock, flags);
533 while (!list_empty(&shost->starved_list) &&
534 !shost->host_blocked && !shost->host_self_blocked &&
535 !((shost->can_queue > 0) &&
536 (shost->host_busy >= shost->can_queue))) {
538 * As long as shost is accepting commands and we have
539 * starved queues, call blk_run_queue. scsi_request_fn
540 * drops the queue_lock and can add us back to the
543 * host_lock protects the starved_list and starved_entry.
544 * scsi_request_fn must get the host_lock before checking
545 * or modifying starved_list or starved_entry.
547 sdev = list_entry(shost->starved_list.next,
548 struct scsi_device, starved_entry);
549 list_del_init(&sdev->starved_entry);
550 spin_unlock_irqrestore(shost->host_lock, flags);
553 if (test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
554 !test_and_set_bit(QUEUE_FLAG_REENTER,
555 &sdev->request_queue->queue_flags)) {
556 blk_run_queue(sdev->request_queue);
557 clear_bit(QUEUE_FLAG_REENTER,
558 &sdev->request_queue->queue_flags);
560 blk_run_queue(sdev->request_queue);
562 spin_lock_irqsave(shost->host_lock, flags);
563 if (unlikely(!list_empty(&sdev->starved_entry)))
565 * sdev lost a race, and was put back on the
566 * starved list. This is unlikely but without this
567 * in theory we could loop forever.
571 spin_unlock_irqrestore(shost->host_lock, flags);
577 * Function: scsi_requeue_command()
579 * Purpose: Handle post-processing of completed commands.
581 * Arguments: q - queue to operate on
582 * cmd - command that may need to be requeued.
586 * Notes: After command completion, there may be blocks left
587 * over which weren't finished by the previous command
588 * this can be for a number of reasons - the main one is
589 * I/O errors in the middle of the request, in which case
590 * we need to request the blocks that come after the bad
592 * Notes: Upon return, cmd is a stale pointer.
594 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
596 struct request *req = cmd->request;
599 scsi_unprep_request(req);
600 spin_lock_irqsave(q->queue_lock, flags);
601 blk_requeue_request(q, req);
602 spin_unlock_irqrestore(q->queue_lock, flags);
607 void scsi_next_command(struct scsi_cmnd *cmd)
609 struct scsi_device *sdev = cmd->device;
610 struct request_queue *q = sdev->request_queue;
612 /* need to hold a reference on the device before we let go of the cmd */
613 get_device(&sdev->sdev_gendev);
615 scsi_put_command(cmd);
618 /* ok to remove device now */
619 put_device(&sdev->sdev_gendev);
622 void scsi_run_host_queues(struct Scsi_Host *shost)
624 struct scsi_device *sdev;
626 shost_for_each_device(sdev, shost)
627 scsi_run_queue(sdev->request_queue);
631 * Function: scsi_end_request()
633 * Purpose: Post-processing of completed commands (usually invoked at end
634 * of upper level post-processing and scsi_io_completion).
636 * Arguments: cmd - command that is complete.
637 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
638 * bytes - number of bytes of completed I/O
639 * requeue - indicates whether we should requeue leftovers.
641 * Lock status: Assumed that lock is not held upon entry.
643 * Returns: cmd if requeue required, NULL otherwise.
645 * Notes: This is called for block device requests in order to
646 * mark some number of sectors as complete.
648 * We are guaranteeing that the request queue will be goosed
649 * at some point during this call.
650 * Notes: If cmd was requeued, upon return it will be a stale pointer.
652 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int uptodate,
653 int bytes, int requeue)
655 struct request_queue *q = cmd->device->request_queue;
656 struct request *req = cmd->request;
660 * If there are blocks left over at the end, set up the command
661 * to queue the remainder of them.
663 if (end_that_request_chunk(req, uptodate, bytes)) {
664 int leftover = (req->hard_nr_sectors << 9);
666 if (blk_pc_request(req))
667 leftover = req->data_len;
669 /* kill remainder if no retrys */
670 if (!uptodate && blk_noretry_request(req))
671 end_that_request_chunk(req, 0, leftover);
675 * Bleah. Leftovers again. Stick the
676 * leftovers in the front of the
677 * queue, and goose the queue again.
679 scsi_requeue_command(q, cmd);
686 add_disk_randomness(req->rq_disk);
688 spin_lock_irqsave(q->queue_lock, flags);
689 if (blk_rq_tagged(req))
690 blk_queue_end_tag(q, req);
691 end_that_request_last(req, uptodate);
692 spin_unlock_irqrestore(q->queue_lock, flags);
695 * This will goose the queue request function at the end, so we don't
696 * need to worry about launching another command.
698 scsi_next_command(cmd);
703 * Like SCSI_MAX_SG_SEGMENTS, but for archs that have sg chaining. This limit
704 * is totally arbitrary, a setting of 2048 will get you at least 8mb ios.
706 #define SCSI_MAX_SG_CHAIN_SEGMENTS 2048
708 static inline unsigned int scsi_sgtable_index(unsigned short nents)
719 #if (SCSI_MAX_SG_SEGMENTS > 16)
723 #if (SCSI_MAX_SG_SEGMENTS > 32)
727 #if (SCSI_MAX_SG_SEGMENTS > 64)
735 printk(KERN_ERR "scsi: bad segment count=%d\n", nents);
742 struct scatterlist *scsi_alloc_sgtable(struct scsi_cmnd *cmd, gfp_t gfp_mask)
744 struct scsi_host_sg_pool *sgp;
745 struct scatterlist *sgl, *prev, *ret;
749 BUG_ON(!cmd->use_sg);
755 if (this > SCSI_MAX_SG_SEGMENTS) {
756 this = SCSI_MAX_SG_SEGMENTS - 1;
757 index = SG_MEMPOOL_NR - 1;
759 index = scsi_sgtable_index(this);
763 sgp = scsi_sg_pools + index;
765 sgl = mempool_alloc(sgp->pool, gfp_mask);
769 sg_init_table(sgl, sgp->size);
772 * first loop through, set initial index and return value
778 * chain previous sglist, if any. we know the previous
779 * sglist must be the biggest one, or we would not have
780 * ended up doing another loop.
783 sg_chain(prev, SCSI_MAX_SG_SEGMENTS, sgl);
786 * if we have nothing left, mark the last segment as
790 sg_mark_end(&sgl[this - 1]);
793 * don't allow subsequent mempool allocs to sleep, it would
794 * violate the mempool principle.
796 gfp_mask &= ~__GFP_WAIT;
797 gfp_mask |= __GFP_HIGH;
802 * ->use_sg may get modified after dma mapping has potentially
803 * shrunk the number of segments, so keep a copy of it for free.
805 cmd->__use_sg = cmd->use_sg;
810 * Free entries chained off ret. Since we were trying to
811 * allocate another sglist, we know that all entries are of
814 sgp = scsi_sg_pools + SG_MEMPOOL_NR - 1;
816 ret = &ret[SCSI_MAX_SG_SEGMENTS - 1];
818 while ((sgl = sg_chain_ptr(ret)) != NULL) {
819 ret = &sgl[SCSI_MAX_SG_SEGMENTS - 1];
820 mempool_free(sgl, sgp->pool);
823 mempool_free(prev, sgp->pool);
828 EXPORT_SYMBOL(scsi_alloc_sgtable);
830 void scsi_free_sgtable(struct scsi_cmnd *cmd)
832 struct scatterlist *sgl = cmd->request_buffer;
833 struct scsi_host_sg_pool *sgp;
836 * if this is the biggest size sglist, check if we have
837 * chained parts we need to free
839 if (cmd->__use_sg > SCSI_MAX_SG_SEGMENTS) {
840 unsigned short this, left;
841 struct scatterlist *next;
844 left = cmd->__use_sg - (SCSI_MAX_SG_SEGMENTS - 1);
845 next = sg_chain_ptr(&sgl[SCSI_MAX_SG_SEGMENTS - 1]);
846 while (left && next) {
849 if (this > SCSI_MAX_SG_SEGMENTS) {
850 this = SCSI_MAX_SG_SEGMENTS - 1;
851 index = SG_MEMPOOL_NR - 1;
853 index = scsi_sgtable_index(this);
857 sgp = scsi_sg_pools + index;
860 next = sg_chain_ptr(&sgl[sgp->size - 1]);
862 mempool_free(sgl, sgp->pool);
866 * Restore original, will be freed below
868 sgl = cmd->request_buffer;
869 sgp = scsi_sg_pools + SG_MEMPOOL_NR - 1;
871 sgp = scsi_sg_pools + scsi_sgtable_index(cmd->__use_sg);
873 mempool_free(sgl, sgp->pool);
876 EXPORT_SYMBOL(scsi_free_sgtable);
879 * Function: scsi_release_buffers()
881 * Purpose: Completion processing for block device I/O requests.
883 * Arguments: cmd - command that we are bailing.
885 * Lock status: Assumed that no lock is held upon entry.
889 * Notes: In the event that an upper level driver rejects a
890 * command, we must release resources allocated during
891 * the __init_io() function. Primarily this would involve
892 * the scatter-gather table, and potentially any bounce
895 static void scsi_release_buffers(struct scsi_cmnd *cmd)
898 scsi_free_sgtable(cmd);
901 * Zero these out. They now point to freed memory, and it is
902 * dangerous to hang onto the pointers.
904 cmd->request_buffer = NULL;
905 cmd->request_bufflen = 0;
909 * Function: scsi_io_completion()
911 * Purpose: Completion processing for block device I/O requests.
913 * Arguments: cmd - command that is finished.
915 * Lock status: Assumed that no lock is held upon entry.
919 * Notes: This function is matched in terms of capabilities to
920 * the function that created the scatter-gather list.
921 * In other words, if there are no bounce buffers
922 * (the normal case for most drivers), we don't need
923 * the logic to deal with cleaning up afterwards.
925 * We must do one of several things here:
927 * a) Call scsi_end_request. This will finish off the
928 * specified number of sectors. If we are done, the
929 * command block will be released, and the queue
930 * function will be goosed. If we are not done, then
931 * scsi_end_request will directly goose the queue.
933 * b) We can just use scsi_requeue_command() here. This would
934 * be used if we just wanted to retry, for example.
936 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
938 int result = cmd->result;
939 int this_count = cmd->request_bufflen;
940 struct request_queue *q = cmd->device->request_queue;
941 struct request *req = cmd->request;
942 int clear_errors = 1;
943 struct scsi_sense_hdr sshdr;
945 int sense_deferred = 0;
947 scsi_release_buffers(cmd);
950 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
952 sense_deferred = scsi_sense_is_deferred(&sshdr);
955 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
956 req->errors = result;
959 if (sense_valid && req->sense) {
961 * SG_IO wants current and deferred errors
963 int len = 8 + cmd->sense_buffer[7];
965 if (len > SCSI_SENSE_BUFFERSIZE)
966 len = SCSI_SENSE_BUFFERSIZE;
967 memcpy(req->sense, cmd->sense_buffer, len);
968 req->sense_len = len;
971 req->data_len = cmd->resid;
975 * Next deal with any sectors which we were able to correctly
978 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
980 req->nr_sectors, good_bytes));
981 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd->use_sg));
986 /* A number of bytes were successfully read. If there
987 * are leftovers and there is some kind of error
988 * (result != 0), retry the rest.
990 if (scsi_end_request(cmd, 1, good_bytes, result == 0) == NULL)
993 /* good_bytes = 0, or (inclusive) there were leftovers and
994 * result = 0, so scsi_end_request couldn't retry.
996 if (sense_valid && !sense_deferred) {
997 switch (sshdr.sense_key) {
999 if (cmd->device->removable) {
1000 /* Detected disc change. Set a bit
1001 * and quietly refuse further access.
1003 cmd->device->changed = 1;
1004 scsi_end_request(cmd, 0, this_count, 1);
1007 /* Must have been a power glitch, or a
1008 * bus reset. Could not have been a
1009 * media change, so we just retry the
1010 * request and see what happens.
1012 scsi_requeue_command(q, cmd);
1016 case ILLEGAL_REQUEST:
1017 /* If we had an ILLEGAL REQUEST returned, then
1018 * we may have performed an unsupported
1019 * command. The only thing this should be
1020 * would be a ten byte read where only a six
1021 * byte read was supported. Also, on a system
1022 * where READ CAPACITY failed, we may have
1023 * read past the end of the disk.
1025 if ((cmd->device->use_10_for_rw &&
1026 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
1027 (cmd->cmnd[0] == READ_10 ||
1028 cmd->cmnd[0] == WRITE_10)) {
1029 cmd->device->use_10_for_rw = 0;
1030 /* This will cause a retry with a
1033 scsi_requeue_command(q, cmd);
1036 scsi_end_request(cmd, 0, this_count, 1);
1041 /* If the device is in the process of becoming
1042 * ready, or has a temporary blockage, retry.
1044 if (sshdr.asc == 0x04) {
1045 switch (sshdr.ascq) {
1046 case 0x01: /* becoming ready */
1047 case 0x04: /* format in progress */
1048 case 0x05: /* rebuild in progress */
1049 case 0x06: /* recalculation in progress */
1050 case 0x07: /* operation in progress */
1051 case 0x08: /* Long write in progress */
1052 case 0x09: /* self test in progress */
1053 scsi_requeue_command(q, cmd);
1059 if (!(req->cmd_flags & REQ_QUIET))
1060 scsi_cmd_print_sense_hdr(cmd,
1064 scsi_end_request(cmd, 0, this_count, 1);
1066 case VOLUME_OVERFLOW:
1067 if (!(req->cmd_flags & REQ_QUIET)) {
1068 scmd_printk(KERN_INFO, cmd,
1069 "Volume overflow, CDB: ");
1070 __scsi_print_command(cmd->cmnd);
1071 scsi_print_sense("", cmd);
1073 /* See SSC3rXX or current. */
1074 scsi_end_request(cmd, 0, this_count, 1);
1080 if (host_byte(result) == DID_RESET) {
1081 /* Third party bus reset or reset for error recovery
1082 * reasons. Just retry the request and see what
1085 scsi_requeue_command(q, cmd);
1089 if (!(req->cmd_flags & REQ_QUIET)) {
1090 scsi_print_result(cmd);
1091 if (driver_byte(result) & DRIVER_SENSE)
1092 scsi_print_sense("", cmd);
1095 scsi_end_request(cmd, 0, this_count, !result);
1099 * Function: scsi_init_io()
1101 * Purpose: SCSI I/O initialize function.
1103 * Arguments: cmd - Command descriptor we wish to initialize
1105 * Returns: 0 on success
1106 * BLKPREP_DEFER if the failure is retryable
1107 * BLKPREP_KILL if the failure is fatal
1109 static int scsi_init_io(struct scsi_cmnd *cmd)
1111 struct request *req = cmd->request;
1115 * We used to not use scatter-gather for single segment request,
1116 * but now we do (it makes highmem I/O easier to support without
1119 cmd->use_sg = req->nr_phys_segments;
1122 * If sg table allocation fails, requeue request later.
1124 cmd->request_buffer = scsi_alloc_sgtable(cmd, GFP_ATOMIC);
1125 if (unlikely(!cmd->request_buffer)) {
1126 scsi_unprep_request(req);
1127 return BLKPREP_DEFER;
1131 if (blk_pc_request(req))
1132 cmd->request_bufflen = req->data_len;
1134 cmd->request_bufflen = req->nr_sectors << 9;
1137 * Next, walk the list, and fill in the addresses and sizes of
1140 count = blk_rq_map_sg(req->q, req, cmd->request_buffer);
1141 if (likely(count <= cmd->use_sg)) {
1142 cmd->use_sg = count;
1146 printk(KERN_ERR "Incorrect number of segments after building list\n");
1147 printk(KERN_ERR "counted %d, received %d\n", count, cmd->use_sg);
1148 printk(KERN_ERR "req nr_sec %lu, cur_nr_sec %u\n", req->nr_sectors,
1149 req->current_nr_sectors);
1151 return BLKPREP_KILL;
1154 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1155 struct request *req)
1157 struct scsi_cmnd *cmd;
1159 if (!req->special) {
1160 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1168 /* pull a tag out of the request if we have one */
1169 cmd->tag = req->tag;
1175 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1177 struct scsi_cmnd *cmd;
1178 int ret = scsi_prep_state_check(sdev, req);
1180 if (ret != BLKPREP_OK)
1183 cmd = scsi_get_cmd_from_req(sdev, req);
1185 return BLKPREP_DEFER;
1188 * BLOCK_PC requests may transfer data, in which case they must
1189 * a bio attached to them. Or they might contain a SCSI command
1190 * that does not transfer data, in which case they may optionally
1191 * submit a request without an attached bio.
1196 BUG_ON(!req->nr_phys_segments);
1198 ret = scsi_init_io(cmd);
1202 BUG_ON(req->data_len);
1205 cmd->request_bufflen = 0;
1206 cmd->request_buffer = NULL;
1211 BUILD_BUG_ON(sizeof(req->cmd) > sizeof(cmd->cmnd));
1212 memcpy(cmd->cmnd, req->cmd, sizeof(cmd->cmnd));
1213 cmd->cmd_len = req->cmd_len;
1215 cmd->sc_data_direction = DMA_NONE;
1216 else if (rq_data_dir(req) == WRITE)
1217 cmd->sc_data_direction = DMA_TO_DEVICE;
1219 cmd->sc_data_direction = DMA_FROM_DEVICE;
1221 cmd->transfersize = req->data_len;
1222 cmd->allowed = req->retries;
1223 cmd->timeout_per_command = req->timeout;
1226 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1229 * Setup a REQ_TYPE_FS command. These are simple read/write request
1230 * from filesystems that still need to be translated to SCSI CDBs from
1233 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1235 struct scsi_cmnd *cmd;
1236 int ret = scsi_prep_state_check(sdev, req);
1238 if (ret != BLKPREP_OK)
1241 * Filesystem requests must transfer data.
1243 BUG_ON(!req->nr_phys_segments);
1245 cmd = scsi_get_cmd_from_req(sdev, req);
1247 return BLKPREP_DEFER;
1249 return scsi_init_io(cmd);
1251 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1253 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1255 int ret = BLKPREP_OK;
1258 * If the device is not in running state we will reject some
1261 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1262 switch (sdev->sdev_state) {
1265 * If the device is offline we refuse to process any
1266 * commands. The device must be brought online
1267 * before trying any recovery commands.
1269 sdev_printk(KERN_ERR, sdev,
1270 "rejecting I/O to offline device\n");
1275 * If the device is fully deleted, we refuse to
1276 * process any commands as well.
1278 sdev_printk(KERN_ERR, sdev,
1279 "rejecting I/O to dead device\n");
1285 * If the devices is blocked we defer normal commands.
1287 if (!(req->cmd_flags & REQ_PREEMPT))
1288 ret = BLKPREP_DEFER;
1292 * For any other not fully online state we only allow
1293 * special commands. In particular any user initiated
1294 * command is not allowed.
1296 if (!(req->cmd_flags & REQ_PREEMPT))
1303 EXPORT_SYMBOL(scsi_prep_state_check);
1305 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1307 struct scsi_device *sdev = q->queuedata;
1311 req->errors = DID_NO_CONNECT << 16;
1312 /* release the command and kill it */
1314 struct scsi_cmnd *cmd = req->special;
1315 scsi_release_buffers(cmd);
1316 scsi_put_command(cmd);
1317 req->special = NULL;
1322 * If we defer, the elv_next_request() returns NULL, but the
1323 * queue must be restarted, so we plug here if no returning
1324 * command will automatically do that.
1326 if (sdev->device_busy == 0)
1330 req->cmd_flags |= REQ_DONTPREP;
1335 EXPORT_SYMBOL(scsi_prep_return);
1337 int scsi_prep_fn(struct request_queue *q, struct request *req)
1339 struct scsi_device *sdev = q->queuedata;
1340 int ret = BLKPREP_KILL;
1342 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1343 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1344 return scsi_prep_return(q, req, ret);
1348 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1351 * Called with the queue_lock held.
1353 static inline int scsi_dev_queue_ready(struct request_queue *q,
1354 struct scsi_device *sdev)
1356 if (sdev->device_busy >= sdev->queue_depth)
1358 if (sdev->device_busy == 0 && sdev->device_blocked) {
1360 * unblock after device_blocked iterates to zero
1362 if (--sdev->device_blocked == 0) {
1364 sdev_printk(KERN_INFO, sdev,
1365 "unblocking device at zero depth\n"));
1371 if (sdev->device_blocked)
1378 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1379 * return 0. We must end up running the queue again whenever 0 is
1380 * returned, else IO can hang.
1382 * Called with host_lock held.
1384 static inline int scsi_host_queue_ready(struct request_queue *q,
1385 struct Scsi_Host *shost,
1386 struct scsi_device *sdev)
1388 if (scsi_host_in_recovery(shost))
1390 if (shost->host_busy == 0 && shost->host_blocked) {
1392 * unblock after host_blocked iterates to zero
1394 if (--shost->host_blocked == 0) {
1396 printk("scsi%d unblocking host at zero depth\n",
1403 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1404 shost->host_blocked || shost->host_self_blocked) {
1405 if (list_empty(&sdev->starved_entry))
1406 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1410 /* We're OK to process the command, so we can't be starved */
1411 if (!list_empty(&sdev->starved_entry))
1412 list_del_init(&sdev->starved_entry);
1418 * Kill a request for a dead device
1420 static void scsi_kill_request(struct request *req, struct request_queue *q)
1422 struct scsi_cmnd *cmd = req->special;
1423 struct scsi_device *sdev = cmd->device;
1424 struct Scsi_Host *shost = sdev->host;
1426 blkdev_dequeue_request(req);
1428 if (unlikely(cmd == NULL)) {
1429 printk(KERN_CRIT "impossible request in %s.\n",
1434 scsi_init_cmd_errh(cmd);
1435 cmd->result = DID_NO_CONNECT << 16;
1436 atomic_inc(&cmd->device->iorequest_cnt);
1439 * SCSI request completion path will do scsi_device_unbusy(),
1440 * bump busy counts. To bump the counters, we need to dance
1441 * with the locks as normal issue path does.
1443 sdev->device_busy++;
1444 spin_unlock(sdev->request_queue->queue_lock);
1445 spin_lock(shost->host_lock);
1447 spin_unlock(shost->host_lock);
1448 spin_lock(sdev->request_queue->queue_lock);
1453 static void scsi_softirq_done(struct request *rq)
1455 struct scsi_cmnd *cmd = rq->completion_data;
1456 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
1459 INIT_LIST_HEAD(&cmd->eh_entry);
1461 disposition = scsi_decide_disposition(cmd);
1462 if (disposition != SUCCESS &&
1463 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1464 sdev_printk(KERN_ERR, cmd->device,
1465 "timing out command, waited %lus\n",
1467 disposition = SUCCESS;
1470 scsi_log_completion(cmd, disposition);
1472 switch (disposition) {
1474 scsi_finish_command(cmd);
1477 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1479 case ADD_TO_MLQUEUE:
1480 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1483 if (!scsi_eh_scmd_add(cmd, 0))
1484 scsi_finish_command(cmd);
1489 * Function: scsi_request_fn()
1491 * Purpose: Main strategy routine for SCSI.
1493 * Arguments: q - Pointer to actual queue.
1497 * Lock status: IO request lock assumed to be held when called.
1499 static void scsi_request_fn(struct request_queue *q)
1501 struct scsi_device *sdev = q->queuedata;
1502 struct Scsi_Host *shost;
1503 struct scsi_cmnd *cmd;
1504 struct request *req;
1507 printk("scsi: killing requests for dead queue\n");
1508 while ((req = elv_next_request(q)) != NULL)
1509 scsi_kill_request(req, q);
1513 if(!get_device(&sdev->sdev_gendev))
1514 /* We must be tearing the block queue down already */
1518 * To start with, we keep looping until the queue is empty, or until
1519 * the host is no longer able to accept any more requests.
1522 while (!blk_queue_plugged(q)) {
1525 * get next queueable request. We do this early to make sure
1526 * that the request is fully prepared even if we cannot
1529 req = elv_next_request(q);
1530 if (!req || !scsi_dev_queue_ready(q, sdev))
1533 if (unlikely(!scsi_device_online(sdev))) {
1534 sdev_printk(KERN_ERR, sdev,
1535 "rejecting I/O to offline device\n");
1536 scsi_kill_request(req, q);
1542 * Remove the request from the request list.
1544 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1545 blkdev_dequeue_request(req);
1546 sdev->device_busy++;
1548 spin_unlock(q->queue_lock);
1550 if (unlikely(cmd == NULL)) {
1551 printk(KERN_CRIT "impossible request in %s.\n"
1552 "please mail a stack trace to "
1553 "linux-scsi@vger.kernel.org\n",
1555 blk_dump_rq_flags(req, "foo");
1558 spin_lock(shost->host_lock);
1560 if (!scsi_host_queue_ready(q, shost, sdev))
1562 if (scsi_target(sdev)->single_lun) {
1563 if (scsi_target(sdev)->starget_sdev_user &&
1564 scsi_target(sdev)->starget_sdev_user != sdev)
1566 scsi_target(sdev)->starget_sdev_user = sdev;
1571 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1572 * take the lock again.
1574 spin_unlock_irq(shost->host_lock);
1577 * Finally, initialize any error handling parameters, and set up
1578 * the timers for timeouts.
1580 scsi_init_cmd_errh(cmd);
1583 * Dispatch the command to the low-level driver.
1585 rtn = scsi_dispatch_cmd(cmd);
1586 spin_lock_irq(q->queue_lock);
1588 /* we're refusing the command; because of
1589 * the way locks get dropped, we need to
1590 * check here if plugging is required */
1591 if(sdev->device_busy == 0)
1601 spin_unlock_irq(shost->host_lock);
1604 * lock q, handle tag, requeue req, and decrement device_busy. We
1605 * must return with queue_lock held.
1607 * Decrementing device_busy without checking it is OK, as all such
1608 * cases (host limits or settings) should run the queue at some
1611 spin_lock_irq(q->queue_lock);
1612 blk_requeue_request(q, req);
1613 sdev->device_busy--;
1614 if(sdev->device_busy == 0)
1617 /* must be careful here...if we trigger the ->remove() function
1618 * we cannot be holding the q lock */
1619 spin_unlock_irq(q->queue_lock);
1620 put_device(&sdev->sdev_gendev);
1621 spin_lock_irq(q->queue_lock);
1624 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1626 struct device *host_dev;
1627 u64 bounce_limit = 0xffffffff;
1629 if (shost->unchecked_isa_dma)
1630 return BLK_BOUNCE_ISA;
1632 * Platforms with virtual-DMA translation
1633 * hardware have no practical limit.
1635 if (!PCI_DMA_BUS_IS_PHYS)
1636 return BLK_BOUNCE_ANY;
1638 host_dev = scsi_get_device(shost);
1639 if (host_dev && host_dev->dma_mask)
1640 bounce_limit = *host_dev->dma_mask;
1642 return bounce_limit;
1644 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1646 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1647 request_fn_proc *request_fn)
1649 struct request_queue *q;
1651 q = blk_init_queue(request_fn, NULL);
1656 * this limit is imposed by hardware restrictions
1658 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1661 * In the future, sg chaining support will be mandatory and this
1662 * ifdef can then go away. Right now we don't have all archs
1663 * converted, so better keep it safe.
1665 #ifdef ARCH_HAS_SG_CHAIN
1666 if (shost->use_sg_chaining)
1667 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1669 blk_queue_max_phys_segments(q, SCSI_MAX_SG_SEGMENTS);
1671 blk_queue_max_phys_segments(q, SCSI_MAX_SG_SEGMENTS);
1674 blk_queue_max_sectors(q, shost->max_sectors);
1675 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1676 blk_queue_segment_boundary(q, shost->dma_boundary);
1678 if (!shost->use_clustering)
1679 clear_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
1682 EXPORT_SYMBOL(__scsi_alloc_queue);
1684 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1686 struct request_queue *q;
1688 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1692 blk_queue_prep_rq(q, scsi_prep_fn);
1693 blk_queue_softirq_done(q, scsi_softirq_done);
1697 void scsi_free_queue(struct request_queue *q)
1699 blk_cleanup_queue(q);
1703 * Function: scsi_block_requests()
1705 * Purpose: Utility function used by low-level drivers to prevent further
1706 * commands from being queued to the device.
1708 * Arguments: shost - Host in question
1712 * Lock status: No locks are assumed held.
1714 * Notes: There is no timer nor any other means by which the requests
1715 * get unblocked other than the low-level driver calling
1716 * scsi_unblock_requests().
1718 void scsi_block_requests(struct Scsi_Host *shost)
1720 shost->host_self_blocked = 1;
1722 EXPORT_SYMBOL(scsi_block_requests);
1725 * Function: scsi_unblock_requests()
1727 * Purpose: Utility function used by low-level drivers to allow further
1728 * commands from being queued to the device.
1730 * Arguments: shost - Host in question
1734 * Lock status: No locks are assumed held.
1736 * Notes: There is no timer nor any other means by which the requests
1737 * get unblocked other than the low-level driver calling
1738 * scsi_unblock_requests().
1740 * This is done as an API function so that changes to the
1741 * internals of the scsi mid-layer won't require wholesale
1742 * changes to drivers that use this feature.
1744 void scsi_unblock_requests(struct Scsi_Host *shost)
1746 shost->host_self_blocked = 0;
1747 scsi_run_host_queues(shost);
1749 EXPORT_SYMBOL(scsi_unblock_requests);
1751 int __init scsi_init_queue(void)
1755 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1756 sizeof(struct scsi_io_context),
1758 if (!scsi_io_context_cache) {
1759 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1763 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1764 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1765 int size = sgp->size * sizeof(struct scatterlist);
1767 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1768 SLAB_HWCACHE_ALIGN, NULL);
1770 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1774 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1777 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1785 void scsi_exit_queue(void)
1789 kmem_cache_destroy(scsi_io_context_cache);
1791 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1792 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1793 mempool_destroy(sgp->pool);
1794 kmem_cache_destroy(sgp->slab);
1799 * scsi_mode_select - issue a mode select
1800 * @sdev: SCSI device to be queried
1801 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1802 * @sp: Save page bit (0 == don't save, 1 == save)
1803 * @modepage: mode page being requested
1804 * @buffer: request buffer (may not be smaller than eight bytes)
1805 * @len: length of request buffer.
1806 * @timeout: command timeout
1807 * @retries: number of retries before failing
1808 * @data: returns a structure abstracting the mode header data
1809 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1810 * must be SCSI_SENSE_BUFFERSIZE big.
1812 * Returns zero if successful; negative error number or scsi
1817 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1818 unsigned char *buffer, int len, int timeout, int retries,
1819 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1821 unsigned char cmd[10];
1822 unsigned char *real_buffer;
1825 memset(cmd, 0, sizeof(cmd));
1826 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1828 if (sdev->use_10_for_ms) {
1831 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1834 memcpy(real_buffer + 8, buffer, len);
1838 real_buffer[2] = data->medium_type;
1839 real_buffer[3] = data->device_specific;
1840 real_buffer[4] = data->longlba ? 0x01 : 0;
1842 real_buffer[6] = data->block_descriptor_length >> 8;
1843 real_buffer[7] = data->block_descriptor_length;
1845 cmd[0] = MODE_SELECT_10;
1849 if (len > 255 || data->block_descriptor_length > 255 ||
1853 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1856 memcpy(real_buffer + 4, buffer, len);
1859 real_buffer[1] = data->medium_type;
1860 real_buffer[2] = data->device_specific;
1861 real_buffer[3] = data->block_descriptor_length;
1864 cmd[0] = MODE_SELECT;
1868 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1869 sshdr, timeout, retries);
1873 EXPORT_SYMBOL_GPL(scsi_mode_select);
1876 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1877 * @sdev: SCSI device to be queried
1878 * @dbd: set if mode sense will allow block descriptors to be returned
1879 * @modepage: mode page being requested
1880 * @buffer: request buffer (may not be smaller than eight bytes)
1881 * @len: length of request buffer.
1882 * @timeout: command timeout
1883 * @retries: number of retries before failing
1884 * @data: returns a structure abstracting the mode header data
1885 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1886 * must be SCSI_SENSE_BUFFERSIZE big.
1888 * Returns zero if unsuccessful, or the header offset (either 4
1889 * or 8 depending on whether a six or ten byte command was
1890 * issued) if successful.
1893 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1894 unsigned char *buffer, int len, int timeout, int retries,
1895 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1897 unsigned char cmd[12];
1901 struct scsi_sense_hdr my_sshdr;
1903 memset(data, 0, sizeof(*data));
1904 memset(&cmd[0], 0, 12);
1905 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1908 /* caller might not be interested in sense, but we need it */
1913 use_10_for_ms = sdev->use_10_for_ms;
1915 if (use_10_for_ms) {
1919 cmd[0] = MODE_SENSE_10;
1926 cmd[0] = MODE_SENSE;
1931 memset(buffer, 0, len);
1933 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1934 sshdr, timeout, retries);
1936 /* This code looks awful: what it's doing is making sure an
1937 * ILLEGAL REQUEST sense return identifies the actual command
1938 * byte as the problem. MODE_SENSE commands can return
1939 * ILLEGAL REQUEST if the code page isn't supported */
1941 if (use_10_for_ms && !scsi_status_is_good(result) &&
1942 (driver_byte(result) & DRIVER_SENSE)) {
1943 if (scsi_sense_valid(sshdr)) {
1944 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1945 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1947 * Invalid command operation code
1949 sdev->use_10_for_ms = 0;
1955 if(scsi_status_is_good(result)) {
1956 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1957 (modepage == 6 || modepage == 8))) {
1958 /* Initio breakage? */
1961 data->medium_type = 0;
1962 data->device_specific = 0;
1964 data->block_descriptor_length = 0;
1965 } else if(use_10_for_ms) {
1966 data->length = buffer[0]*256 + buffer[1] + 2;
1967 data->medium_type = buffer[2];
1968 data->device_specific = buffer[3];
1969 data->longlba = buffer[4] & 0x01;
1970 data->block_descriptor_length = buffer[6]*256
1973 data->length = buffer[0] + 1;
1974 data->medium_type = buffer[1];
1975 data->device_specific = buffer[2];
1976 data->block_descriptor_length = buffer[3];
1978 data->header_length = header_length;
1983 EXPORT_SYMBOL(scsi_mode_sense);
1986 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries)
1989 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1991 struct scsi_sense_hdr sshdr;
1994 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, &sshdr,
1997 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
1999 if ((scsi_sense_valid(&sshdr)) &&
2000 ((sshdr.sense_key == UNIT_ATTENTION) ||
2001 (sshdr.sense_key == NOT_READY))) {
2008 EXPORT_SYMBOL(scsi_test_unit_ready);
2011 * scsi_device_set_state - Take the given device through the device state model.
2012 * @sdev: scsi device to change the state of.
2013 * @state: state to change to.
2015 * Returns zero if unsuccessful or an error if the requested
2016 * transition is illegal.
2019 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2021 enum scsi_device_state oldstate = sdev->sdev_state;
2023 if (state == oldstate)
2028 /* There are no legal states that come back to
2029 * created. This is the manually initialised start
2103 sdev->sdev_state = state;
2107 SCSI_LOG_ERROR_RECOVERY(1,
2108 sdev_printk(KERN_ERR, sdev,
2109 "Illegal state transition %s->%s\n",
2110 scsi_device_state_name(oldstate),
2111 scsi_device_state_name(state))
2115 EXPORT_SYMBOL(scsi_device_set_state);
2118 * sdev_evt_emit - emit a single SCSI device uevent
2119 * @sdev: associated SCSI device
2120 * @evt: event to emit
2122 * Send a single uevent (scsi_event) to the associated scsi_device.
2124 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2129 switch (evt->evt_type) {
2130 case SDEV_EVT_MEDIA_CHANGE:
2131 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2141 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2145 * sdev_evt_thread - send a uevent for each scsi event
2146 * @work: work struct for scsi_device
2148 * Dispatch queued events to their associated scsi_device kobjects
2151 void scsi_evt_thread(struct work_struct *work)
2153 struct scsi_device *sdev;
2154 LIST_HEAD(event_list);
2156 sdev = container_of(work, struct scsi_device, event_work);
2159 struct scsi_event *evt;
2160 struct list_head *this, *tmp;
2161 unsigned long flags;
2163 spin_lock_irqsave(&sdev->list_lock, flags);
2164 list_splice_init(&sdev->event_list, &event_list);
2165 spin_unlock_irqrestore(&sdev->list_lock, flags);
2167 if (list_empty(&event_list))
2170 list_for_each_safe(this, tmp, &event_list) {
2171 evt = list_entry(this, struct scsi_event, node);
2172 list_del(&evt->node);
2173 scsi_evt_emit(sdev, evt);
2180 * sdev_evt_send - send asserted event to uevent thread
2181 * @sdev: scsi_device event occurred on
2182 * @evt: event to send
2184 * Assert scsi device event asynchronously.
2186 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2188 unsigned long flags;
2190 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2195 spin_lock_irqsave(&sdev->list_lock, flags);
2196 list_add_tail(&evt->node, &sdev->event_list);
2197 schedule_work(&sdev->event_work);
2198 spin_unlock_irqrestore(&sdev->list_lock, flags);
2200 EXPORT_SYMBOL_GPL(sdev_evt_send);
2203 * sdev_evt_alloc - allocate a new scsi event
2204 * @evt_type: type of event to allocate
2205 * @gfpflags: GFP flags for allocation
2207 * Allocates and returns a new scsi_event.
2209 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2212 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2216 evt->evt_type = evt_type;
2217 INIT_LIST_HEAD(&evt->node);
2219 /* evt_type-specific initialization, if any */
2221 case SDEV_EVT_MEDIA_CHANGE:
2229 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2232 * sdev_evt_send_simple - send asserted event to uevent thread
2233 * @sdev: scsi_device event occurred on
2234 * @evt_type: type of event to send
2235 * @gfpflags: GFP flags for allocation
2237 * Assert scsi device event asynchronously, given an event type.
2239 void sdev_evt_send_simple(struct scsi_device *sdev,
2240 enum scsi_device_event evt_type, gfp_t gfpflags)
2242 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2244 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2249 sdev_evt_send(sdev, evt);
2251 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2254 * scsi_device_quiesce - Block user issued commands.
2255 * @sdev: scsi device to quiesce.
2257 * This works by trying to transition to the SDEV_QUIESCE state
2258 * (which must be a legal transition). When the device is in this
2259 * state, only special requests will be accepted, all others will
2260 * be deferred. Since special requests may also be requeued requests,
2261 * a successful return doesn't guarantee the device will be
2262 * totally quiescent.
2264 * Must be called with user context, may sleep.
2266 * Returns zero if unsuccessful or an error if not.
2269 scsi_device_quiesce(struct scsi_device *sdev)
2271 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2275 scsi_run_queue(sdev->request_queue);
2276 while (sdev->device_busy) {
2277 msleep_interruptible(200);
2278 scsi_run_queue(sdev->request_queue);
2282 EXPORT_SYMBOL(scsi_device_quiesce);
2285 * scsi_device_resume - Restart user issued commands to a quiesced device.
2286 * @sdev: scsi device to resume.
2288 * Moves the device from quiesced back to running and restarts the
2291 * Must be called with user context, may sleep.
2294 scsi_device_resume(struct scsi_device *sdev)
2296 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2298 scsi_run_queue(sdev->request_queue);
2300 EXPORT_SYMBOL(scsi_device_resume);
2303 device_quiesce_fn(struct scsi_device *sdev, void *data)
2305 scsi_device_quiesce(sdev);
2309 scsi_target_quiesce(struct scsi_target *starget)
2311 starget_for_each_device(starget, NULL, device_quiesce_fn);
2313 EXPORT_SYMBOL(scsi_target_quiesce);
2316 device_resume_fn(struct scsi_device *sdev, void *data)
2318 scsi_device_resume(sdev);
2322 scsi_target_resume(struct scsi_target *starget)
2324 starget_for_each_device(starget, NULL, device_resume_fn);
2326 EXPORT_SYMBOL(scsi_target_resume);
2329 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2330 * @sdev: device to block
2332 * Block request made by scsi lld's to temporarily stop all
2333 * scsi commands on the specified device. Called from interrupt
2334 * or normal process context.
2336 * Returns zero if successful or error if not
2339 * This routine transitions the device to the SDEV_BLOCK state
2340 * (which must be a legal transition). When the device is in this
2341 * state, all commands are deferred until the scsi lld reenables
2342 * the device with scsi_device_unblock or device_block_tmo fires.
2343 * This routine assumes the host_lock is held on entry.
2346 scsi_internal_device_block(struct scsi_device *sdev)
2348 struct request_queue *q = sdev->request_queue;
2349 unsigned long flags;
2352 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2357 * The device has transitioned to SDEV_BLOCK. Stop the
2358 * block layer from calling the midlayer with this device's
2361 spin_lock_irqsave(q->queue_lock, flags);
2363 spin_unlock_irqrestore(q->queue_lock, flags);
2367 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2370 * scsi_internal_device_unblock - resume a device after a block request
2371 * @sdev: device to resume
2373 * Called by scsi lld's or the midlayer to restart the device queue
2374 * for the previously suspended scsi device. Called from interrupt or
2375 * normal process context.
2377 * Returns zero if successful or error if not.
2380 * This routine transitions the device to the SDEV_RUNNING state
2381 * (which must be a legal transition) allowing the midlayer to
2382 * goose the queue for this device. This routine assumes the
2383 * host_lock is held upon entry.
2386 scsi_internal_device_unblock(struct scsi_device *sdev)
2388 struct request_queue *q = sdev->request_queue;
2390 unsigned long flags;
2393 * Try to transition the scsi device to SDEV_RUNNING
2394 * and goose the device queue if successful.
2396 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2400 spin_lock_irqsave(q->queue_lock, flags);
2402 spin_unlock_irqrestore(q->queue_lock, flags);
2406 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2409 device_block(struct scsi_device *sdev, void *data)
2411 scsi_internal_device_block(sdev);
2415 target_block(struct device *dev, void *data)
2417 if (scsi_is_target_device(dev))
2418 starget_for_each_device(to_scsi_target(dev), NULL,
2424 scsi_target_block(struct device *dev)
2426 if (scsi_is_target_device(dev))
2427 starget_for_each_device(to_scsi_target(dev), NULL,
2430 device_for_each_child(dev, NULL, target_block);
2432 EXPORT_SYMBOL_GPL(scsi_target_block);
2435 device_unblock(struct scsi_device *sdev, void *data)
2437 scsi_internal_device_unblock(sdev);
2441 target_unblock(struct device *dev, void *data)
2443 if (scsi_is_target_device(dev))
2444 starget_for_each_device(to_scsi_target(dev), NULL,
2450 scsi_target_unblock(struct device *dev)
2452 if (scsi_is_target_device(dev))
2453 starget_for_each_device(to_scsi_target(dev), NULL,
2456 device_for_each_child(dev, NULL, target_unblock);
2458 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2461 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2462 * @sgl: scatter-gather list
2463 * @sg_count: number of segments in sg
2464 * @offset: offset in bytes into sg, on return offset into the mapped area
2465 * @len: bytes to map, on return number of bytes mapped
2467 * Returns virtual address of the start of the mapped page
2469 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2470 size_t *offset, size_t *len)
2473 size_t sg_len = 0, len_complete = 0;
2474 struct scatterlist *sg;
2477 WARN_ON(!irqs_disabled());
2479 for_each_sg(sgl, sg, sg_count, i) {
2480 len_complete = sg_len; /* Complete sg-entries */
2481 sg_len += sg->length;
2482 if (sg_len > *offset)
2486 if (unlikely(i == sg_count)) {
2487 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2489 __FUNCTION__, sg_len, *offset, sg_count);
2494 /* Offset starting from the beginning of first page in this sg-entry */
2495 *offset = *offset - len_complete + sg->offset;
2497 /* Assumption: contiguous pages can be accessed as "page + i" */
2498 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2499 *offset &= ~PAGE_MASK;
2501 /* Bytes in this sg-entry from *offset to the end of the page */
2502 sg_len = PAGE_SIZE - *offset;
2506 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2508 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2511 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2512 * @virt: virtual address to be unmapped
2514 void scsi_kunmap_atomic_sg(void *virt)
2516 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2518 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);