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 struct kmem_cache *scsi_bidi_sdb_cache;
69 static void scsi_run_queue(struct request_queue *q);
72 * Function: scsi_unprep_request()
74 * Purpose: Remove all preparation done for a request, including its
75 * associated scsi_cmnd, so that it can be requeued.
77 * Arguments: req - request to unprepare
79 * Lock status: Assumed that no locks are held upon entry.
83 static void scsi_unprep_request(struct request *req)
85 struct scsi_cmnd *cmd = req->special;
87 req->cmd_flags &= ~REQ_DONTPREP;
90 scsi_put_command(cmd);
94 * Function: scsi_queue_insert()
96 * Purpose: Insert a command in the midlevel queue.
98 * Arguments: cmd - command that we are adding to queue.
99 * reason - why we are inserting command to queue.
101 * Lock status: Assumed that lock is not held upon entry.
105 * Notes: We do this for one of two cases. Either the host is busy
106 * and it cannot accept any more commands for the time being,
107 * or the device returned QUEUE_FULL and can accept no more
109 * Notes: This could be called either from an interrupt context or a
110 * normal process context.
112 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
114 struct Scsi_Host *host = cmd->device->host;
115 struct scsi_device *device = cmd->device;
116 struct request_queue *q = device->request_queue;
120 printk("Inserting command %p into mlqueue\n", cmd));
123 * Set the appropriate busy bit for the device/host.
125 * If the host/device isn't busy, assume that something actually
126 * completed, and that we should be able to queue a command now.
128 * Note that the prior mid-layer assumption that any host could
129 * always queue at least one command is now broken. The mid-layer
130 * will implement a user specifiable stall (see
131 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
132 * if a command is requeued with no other commands outstanding
133 * either for the device or for the host.
135 if (reason == SCSI_MLQUEUE_HOST_BUSY)
136 host->host_blocked = host->max_host_blocked;
137 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
138 device->device_blocked = device->max_device_blocked;
141 * Decrement the counters, since these commands are no longer
142 * active on the host/device.
144 scsi_device_unbusy(device);
147 * Requeue this command. It will go before all other commands
148 * that are already in the queue.
150 * NOTE: there is magic here about the way the queue is plugged if
151 * we have no outstanding commands.
153 * Although we *don't* plug the queue, we call the request
154 * function. The SCSI request function detects the blocked condition
155 * and plugs the queue appropriately.
157 spin_lock_irqsave(q->queue_lock, flags);
158 blk_requeue_request(q, cmd->request);
159 spin_unlock_irqrestore(q->queue_lock, flags);
167 * scsi_execute - insert request and wait for the result
170 * @data_direction: data direction
171 * @buffer: data buffer
172 * @bufflen: len of buffer
173 * @sense: optional sense buffer
174 * @timeout: request timeout in seconds
175 * @retries: number of times to retry request
176 * @flags: or into request flags;
178 * returns the req->errors value which is the scsi_cmnd result
181 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
182 int data_direction, void *buffer, unsigned bufflen,
183 unsigned char *sense, int timeout, int retries, int flags)
186 int write = (data_direction == DMA_TO_DEVICE);
187 int ret = DRIVER_ERROR << 24;
189 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
191 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
192 buffer, bufflen, __GFP_WAIT))
195 req->cmd_len = COMMAND_SIZE(cmd[0]);
196 memcpy(req->cmd, cmd, req->cmd_len);
199 req->retries = retries;
200 req->timeout = timeout;
201 req->cmd_type = REQ_TYPE_BLOCK_PC;
202 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
205 * head injection *required* here otherwise quiesce won't work
207 blk_execute_rq(req->q, NULL, req, 1);
211 blk_put_request(req);
215 EXPORT_SYMBOL(scsi_execute);
218 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
219 int data_direction, void *buffer, unsigned bufflen,
220 struct scsi_sense_hdr *sshdr, int timeout, int retries)
226 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
228 return DRIVER_ERROR << 24;
230 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
231 sense, timeout, retries, 0);
233 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
238 EXPORT_SYMBOL(scsi_execute_req);
240 struct scsi_io_context {
242 void (*done)(void *data, char *sense, int result, int resid);
243 char sense[SCSI_SENSE_BUFFERSIZE];
246 static struct kmem_cache *scsi_io_context_cache;
248 static void scsi_end_async(struct request *req, int uptodate)
250 struct scsi_io_context *sioc = req->end_io_data;
253 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
255 kmem_cache_free(scsi_io_context_cache, sioc);
256 __blk_put_request(req->q, req);
259 static int scsi_merge_bio(struct request *rq, struct bio *bio)
261 struct request_queue *q = rq->q;
263 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
264 if (rq_data_dir(rq) == WRITE)
265 bio->bi_rw |= (1 << BIO_RW);
266 blk_queue_bounce(q, &bio);
268 return blk_rq_append_bio(q, rq, bio);
271 static void scsi_bi_endio(struct bio *bio, int error)
277 * scsi_req_map_sg - map a scatterlist into a request
278 * @rq: request to fill
280 * @nsegs: number of elements
281 * @bufflen: len of buffer
282 * @gfp: memory allocation flags
284 * scsi_req_map_sg maps a scatterlist into a request so that the
285 * request can be sent to the block layer. We do not trust the scatterlist
286 * sent to use, as some ULDs use that struct to only organize the pages.
288 static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
289 int nsegs, unsigned bufflen, gfp_t gfp)
291 struct request_queue *q = rq->q;
292 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
293 unsigned int data_len = bufflen, len, bytes, off;
294 struct scatterlist *sg;
296 struct bio *bio = NULL;
297 int i, err, nr_vecs = 0;
299 for_each_sg(sgl, sg, nsegs, i) {
305 while (len > 0 && data_len > 0) {
307 * sg sends a scatterlist that is larger than
308 * the data_len it wants transferred for certain
311 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
312 bytes = min(bytes, data_len);
315 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
318 bio = bio_alloc(gfp, nr_vecs);
323 bio->bi_end_io = scsi_bi_endio;
326 if (bio_add_pc_page(q, bio, page, bytes, off) !=
333 if (bio->bi_vcnt >= nr_vecs) {
334 err = scsi_merge_bio(rq, bio);
349 rq->buffer = rq->data = NULL;
350 rq->data_len = bufflen;
354 while ((bio = rq->bio) != NULL) {
355 rq->bio = bio->bi_next;
357 * call endio instead of bio_put incase it was bounced
366 * scsi_execute_async - insert request
369 * @cmd_len: length of scsi cdb
370 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
371 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
372 * @bufflen: len of buffer
373 * @use_sg: if buffer is a scatterlist this is the number of elements
374 * @timeout: request timeout in seconds
375 * @retries: number of times to retry request
376 * @privdata: data passed to done()
377 * @done: callback function when done
378 * @gfp: memory allocation flags
380 int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
381 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
382 int use_sg, int timeout, int retries, void *privdata,
383 void (*done)(void *, char *, int, int), gfp_t gfp)
386 struct scsi_io_context *sioc;
388 int write = (data_direction == DMA_TO_DEVICE);
390 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp);
392 return DRIVER_ERROR << 24;
394 req = blk_get_request(sdev->request_queue, write, gfp);
397 req->cmd_type = REQ_TYPE_BLOCK_PC;
398 req->cmd_flags |= REQ_QUIET;
401 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
403 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
408 req->cmd_len = cmd_len;
409 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
410 memcpy(req->cmd, cmd, req->cmd_len);
411 req->sense = sioc->sense;
413 req->timeout = timeout;
414 req->retries = retries;
415 req->end_io_data = sioc;
417 sioc->data = privdata;
420 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
424 blk_put_request(req);
426 kmem_cache_free(scsi_io_context_cache, sioc);
427 return DRIVER_ERROR << 24;
429 EXPORT_SYMBOL_GPL(scsi_execute_async);
432 * Function: scsi_init_cmd_errh()
434 * Purpose: Initialize cmd fields related to error handling.
436 * Arguments: cmd - command that is ready to be queued.
438 * Notes: This function has the job of initializing a number of
439 * fields related to error handling. Typically this will
440 * be called once for each command, as required.
442 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
444 cmd->serial_number = 0;
445 scsi_set_resid(cmd, 0);
446 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
447 if (cmd->cmd_len == 0)
448 cmd->cmd_len = COMMAND_SIZE(cmd->cmnd[0]);
451 void scsi_device_unbusy(struct scsi_device *sdev)
453 struct Scsi_Host *shost = sdev->host;
456 spin_lock_irqsave(shost->host_lock, flags);
458 if (unlikely(scsi_host_in_recovery(shost) &&
459 (shost->host_failed || shost->host_eh_scheduled)))
460 scsi_eh_wakeup(shost);
461 spin_unlock(shost->host_lock);
462 spin_lock(sdev->request_queue->queue_lock);
464 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
468 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
469 * and call blk_run_queue for all the scsi_devices on the target -
470 * including current_sdev first.
472 * Called with *no* scsi locks held.
474 static void scsi_single_lun_run(struct scsi_device *current_sdev)
476 struct Scsi_Host *shost = current_sdev->host;
477 struct scsi_device *sdev, *tmp;
478 struct scsi_target *starget = scsi_target(current_sdev);
481 spin_lock_irqsave(shost->host_lock, flags);
482 starget->starget_sdev_user = NULL;
483 spin_unlock_irqrestore(shost->host_lock, flags);
486 * Call blk_run_queue for all LUNs on the target, starting with
487 * current_sdev. We race with others (to set starget_sdev_user),
488 * but in most cases, we will be first. Ideally, each LU on the
489 * target would get some limited time or requests on the target.
491 blk_run_queue(current_sdev->request_queue);
493 spin_lock_irqsave(shost->host_lock, flags);
494 if (starget->starget_sdev_user)
496 list_for_each_entry_safe(sdev, tmp, &starget->devices,
497 same_target_siblings) {
498 if (sdev == current_sdev)
500 if (scsi_device_get(sdev))
503 spin_unlock_irqrestore(shost->host_lock, flags);
504 blk_run_queue(sdev->request_queue);
505 spin_lock_irqsave(shost->host_lock, flags);
507 scsi_device_put(sdev);
510 spin_unlock_irqrestore(shost->host_lock, flags);
514 * Function: scsi_run_queue()
516 * Purpose: Select a proper request queue to serve next
518 * Arguments: q - last request's queue
522 * Notes: The previous command was completely finished, start
523 * a new one if possible.
525 static void scsi_run_queue(struct request_queue *q)
527 struct scsi_device *sdev = q->queuedata;
528 struct Scsi_Host *shost = sdev->host;
531 if (scsi_target(sdev)->single_lun)
532 scsi_single_lun_run(sdev);
534 spin_lock_irqsave(shost->host_lock, flags);
535 while (!list_empty(&shost->starved_list) &&
536 !shost->host_blocked && !shost->host_self_blocked &&
537 !((shost->can_queue > 0) &&
538 (shost->host_busy >= shost->can_queue))) {
540 * As long as shost is accepting commands and we have
541 * starved queues, call blk_run_queue. scsi_request_fn
542 * drops the queue_lock and can add us back to the
545 * host_lock protects the starved_list and starved_entry.
546 * scsi_request_fn must get the host_lock before checking
547 * or modifying starved_list or starved_entry.
549 sdev = list_entry(shost->starved_list.next,
550 struct scsi_device, starved_entry);
551 list_del_init(&sdev->starved_entry);
552 spin_unlock_irqrestore(shost->host_lock, flags);
555 if (test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
556 !test_and_set_bit(QUEUE_FLAG_REENTER,
557 &sdev->request_queue->queue_flags)) {
558 blk_run_queue(sdev->request_queue);
559 clear_bit(QUEUE_FLAG_REENTER,
560 &sdev->request_queue->queue_flags);
562 blk_run_queue(sdev->request_queue);
564 spin_lock_irqsave(shost->host_lock, flags);
565 if (unlikely(!list_empty(&sdev->starved_entry)))
567 * sdev lost a race, and was put back on the
568 * starved list. This is unlikely but without this
569 * in theory we could loop forever.
573 spin_unlock_irqrestore(shost->host_lock, flags);
579 * Function: scsi_requeue_command()
581 * Purpose: Handle post-processing of completed commands.
583 * Arguments: q - queue to operate on
584 * cmd - command that may need to be requeued.
588 * Notes: After command completion, there may be blocks left
589 * over which weren't finished by the previous command
590 * this can be for a number of reasons - the main one is
591 * I/O errors in the middle of the request, in which case
592 * we need to request the blocks that come after the bad
594 * Notes: Upon return, cmd is a stale pointer.
596 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
598 struct request *req = cmd->request;
601 scsi_unprep_request(req);
602 spin_lock_irqsave(q->queue_lock, flags);
603 blk_requeue_request(q, req);
604 spin_unlock_irqrestore(q->queue_lock, flags);
609 void scsi_next_command(struct scsi_cmnd *cmd)
611 struct scsi_device *sdev = cmd->device;
612 struct request_queue *q = sdev->request_queue;
614 /* need to hold a reference on the device before we let go of the cmd */
615 get_device(&sdev->sdev_gendev);
617 scsi_put_command(cmd);
620 /* ok to remove device now */
621 put_device(&sdev->sdev_gendev);
624 void scsi_run_host_queues(struct Scsi_Host *shost)
626 struct scsi_device *sdev;
628 shost_for_each_device(sdev, shost)
629 scsi_run_queue(sdev->request_queue);
633 * Function: scsi_end_request()
635 * Purpose: Post-processing of completed commands (usually invoked at end
636 * of upper level post-processing and scsi_io_completion).
638 * Arguments: cmd - command that is complete.
639 * error - 0 if I/O indicates success, < 0 for I/O error.
640 * bytes - number of bytes of completed I/O
641 * requeue - indicates whether we should requeue leftovers.
643 * Lock status: Assumed that lock is not held upon entry.
645 * Returns: cmd if requeue required, NULL otherwise.
647 * Notes: This is called for block device requests in order to
648 * mark some number of sectors as complete.
650 * We are guaranteeing that the request queue will be goosed
651 * at some point during this call.
652 * Notes: If cmd was requeued, upon return it will be a stale pointer.
654 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
655 int bytes, int requeue)
657 struct request_queue *q = cmd->device->request_queue;
658 struct request *req = cmd->request;
661 * If there are blocks left over at the end, set up the command
662 * to queue the remainder of them.
664 if (blk_end_request(req, error, bytes)) {
665 int leftover = (req->hard_nr_sectors << 9);
667 if (blk_pc_request(req))
668 leftover = req->data_len;
670 /* kill remainder if no retrys */
671 if (error && blk_noretry_request(req))
672 blk_end_request(req, error, leftover);
676 * Bleah. Leftovers again. Stick the
677 * leftovers in the front of the
678 * queue, and goose the queue again.
680 scsi_requeue_command(q, cmd);
688 * This will goose the queue request function at the end, so we don't
689 * need to worry about launching another command.
691 scsi_next_command(cmd);
696 * Like SCSI_MAX_SG_SEGMENTS, but for archs that have sg chaining. This limit
697 * is totally arbitrary, a setting of 2048 will get you at least 8mb ios.
699 #define SCSI_MAX_SG_CHAIN_SEGMENTS 2048
701 static inline unsigned int scsi_sgtable_index(unsigned short nents)
712 #if (SCSI_MAX_SG_SEGMENTS > 16)
716 #if (SCSI_MAX_SG_SEGMENTS > 32)
720 #if (SCSI_MAX_SG_SEGMENTS > 64)
728 printk(KERN_ERR "scsi: bad segment count=%d\n", nents);
735 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
737 struct scsi_host_sg_pool *sgp;
739 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
740 mempool_free(sgl, sgp->pool);
743 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
745 struct scsi_host_sg_pool *sgp;
747 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
748 return mempool_alloc(sgp->pool, gfp_mask);
751 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
758 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
759 gfp_mask, scsi_sg_alloc);
761 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
767 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
769 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
773 * Function: scsi_release_buffers()
775 * Purpose: Completion processing for block device I/O requests.
777 * Arguments: cmd - command that we are bailing.
779 * Lock status: Assumed that no lock is held upon entry.
783 * Notes: In the event that an upper level driver rejects a
784 * command, we must release resources allocated during
785 * the __init_io() function. Primarily this would involve
786 * the scatter-gather table, and potentially any bounce
789 void scsi_release_buffers(struct scsi_cmnd *cmd)
791 if (cmd->sdb.table.nents)
792 scsi_free_sgtable(&cmd->sdb);
794 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
796 if (scsi_bidi_cmnd(cmd)) {
797 struct scsi_data_buffer *bidi_sdb =
798 cmd->request->next_rq->special;
799 scsi_free_sgtable(bidi_sdb);
800 kmem_cache_free(scsi_bidi_sdb_cache, bidi_sdb);
801 cmd->request->next_rq->special = NULL;
804 EXPORT_SYMBOL(scsi_release_buffers);
807 * Bidi commands Must be complete as a whole, both sides at once.
808 * If part of the bytes were written and lld returned
809 * scsi_in()->resid and/or scsi_out()->resid this information will be left
810 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
811 * decide what to do with this information.
813 void scsi_end_bidi_request(struct scsi_cmnd *cmd)
815 struct request *req = cmd->request;
816 unsigned int dlen = req->data_len;
817 unsigned int next_dlen = req->next_rq->data_len;
819 req->data_len = scsi_out(cmd)->resid;
820 req->next_rq->data_len = scsi_in(cmd)->resid;
822 /* The req and req->next_rq have not been completed */
823 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen));
825 scsi_release_buffers(cmd);
828 * This will goose the queue request function at the end, so we don't
829 * need to worry about launching another command.
831 scsi_next_command(cmd);
835 * Function: scsi_io_completion()
837 * Purpose: Completion processing for block device I/O requests.
839 * Arguments: cmd - command that is finished.
841 * Lock status: Assumed that no lock is held upon entry.
845 * Notes: This function is matched in terms of capabilities to
846 * the function that created the scatter-gather list.
847 * In other words, if there are no bounce buffers
848 * (the normal case for most drivers), we don't need
849 * the logic to deal with cleaning up afterwards.
851 * We must do one of several things here:
853 * a) Call scsi_end_request. This will finish off the
854 * specified number of sectors. If we are done, the
855 * command block will be released, and the queue
856 * function will be goosed. If we are not done, then
857 * scsi_end_request will directly goose the queue.
859 * b) We can just use scsi_requeue_command() here. This would
860 * be used if we just wanted to retry, for example.
862 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
864 int result = cmd->result;
865 int this_count = scsi_bufflen(cmd);
866 struct request_queue *q = cmd->device->request_queue;
867 struct request *req = cmd->request;
868 int clear_errors = 1;
869 struct scsi_sense_hdr sshdr;
871 int sense_deferred = 0;
874 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
876 sense_deferred = scsi_sense_is_deferred(&sshdr);
879 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
880 req->errors = result;
883 if (sense_valid && req->sense) {
885 * SG_IO wants current and deferred errors
887 int len = 8 + cmd->sense_buffer[7];
889 if (len > SCSI_SENSE_BUFFERSIZE)
890 len = SCSI_SENSE_BUFFERSIZE;
891 memcpy(req->sense, cmd->sense_buffer, len);
892 req->sense_len = len;
895 if (scsi_bidi_cmnd(cmd)) {
896 /* will also release_buffers */
897 scsi_end_bidi_request(cmd);
900 req->data_len = scsi_get_resid(cmd);
903 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
904 scsi_release_buffers(cmd);
907 * Next deal with any sectors which we were able to correctly
910 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
912 req->nr_sectors, good_bytes));
917 /* A number of bytes were successfully read. If there
918 * are leftovers and there is some kind of error
919 * (result != 0), retry the rest.
921 if (scsi_end_request(cmd, 0, good_bytes, result == 0) == NULL)
924 /* good_bytes = 0, or (inclusive) there were leftovers and
925 * result = 0, so scsi_end_request couldn't retry.
927 if (sense_valid && !sense_deferred) {
928 switch (sshdr.sense_key) {
930 if (cmd->device->removable) {
931 /* Detected disc change. Set a bit
932 * and quietly refuse further access.
934 cmd->device->changed = 1;
935 scsi_end_request(cmd, -EIO, this_count, 1);
938 /* Must have been a power glitch, or a
939 * bus reset. Could not have been a
940 * media change, so we just retry the
941 * request and see what happens.
943 scsi_requeue_command(q, cmd);
947 case ILLEGAL_REQUEST:
948 /* If we had an ILLEGAL REQUEST returned, then
949 * we may have performed an unsupported
950 * command. The only thing this should be
951 * would be a ten byte read where only a six
952 * byte read was supported. Also, on a system
953 * where READ CAPACITY failed, we may have
954 * read past the end of the disk.
956 if ((cmd->device->use_10_for_rw &&
957 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
958 (cmd->cmnd[0] == READ_10 ||
959 cmd->cmnd[0] == WRITE_10)) {
960 cmd->device->use_10_for_rw = 0;
961 /* This will cause a retry with a
964 scsi_requeue_command(q, cmd);
967 scsi_end_request(cmd, -EIO, this_count, 1);
972 /* If the device is in the process of becoming
973 * ready, or has a temporary blockage, retry.
975 if (sshdr.asc == 0x04) {
976 switch (sshdr.ascq) {
977 case 0x01: /* becoming ready */
978 case 0x04: /* format in progress */
979 case 0x05: /* rebuild in progress */
980 case 0x06: /* recalculation in progress */
981 case 0x07: /* operation in progress */
982 case 0x08: /* Long write in progress */
983 case 0x09: /* self test in progress */
984 scsi_requeue_command(q, cmd);
990 if (!(req->cmd_flags & REQ_QUIET))
991 scsi_cmd_print_sense_hdr(cmd,
995 scsi_end_request(cmd, -EIO, this_count, 1);
997 case VOLUME_OVERFLOW:
998 if (!(req->cmd_flags & REQ_QUIET)) {
999 scmd_printk(KERN_INFO, cmd,
1000 "Volume overflow, CDB: ");
1001 __scsi_print_command(cmd->cmnd);
1002 scsi_print_sense("", cmd);
1004 /* See SSC3rXX or current. */
1005 scsi_end_request(cmd, -EIO, this_count, 1);
1011 if (host_byte(result) == DID_RESET) {
1012 /* Third party bus reset or reset for error recovery
1013 * reasons. Just retry the request and see what
1016 scsi_requeue_command(q, cmd);
1020 if (!(req->cmd_flags & REQ_QUIET)) {
1021 scsi_print_result(cmd);
1022 if (driver_byte(result) & DRIVER_SENSE)
1023 scsi_print_sense("", cmd);
1026 scsi_end_request(cmd, -EIO, this_count, !result);
1029 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1035 * If sg table allocation fails, requeue request later.
1037 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1039 return BLKPREP_DEFER;
1043 if (blk_pc_request(req))
1044 sdb->length = req->data_len;
1046 sdb->length = req->nr_sectors << 9;
1049 * Next, walk the list, and fill in the addresses and sizes of
1052 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1053 BUG_ON(count > sdb->table.nents);
1054 sdb->table.nents = count;
1059 * Function: scsi_init_io()
1061 * Purpose: SCSI I/O initialize function.
1063 * Arguments: cmd - Command descriptor we wish to initialize
1065 * Returns: 0 on success
1066 * BLKPREP_DEFER if the failure is retryable
1067 * BLKPREP_KILL if the failure is fatal
1069 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1071 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
1075 if (blk_bidi_rq(cmd->request)) {
1076 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1077 scsi_bidi_sdb_cache, GFP_ATOMIC);
1079 error = BLKPREP_DEFER;
1083 cmd->request->next_rq->special = bidi_sdb;
1084 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1093 scsi_release_buffers(cmd);
1094 if (error == BLKPREP_KILL)
1095 scsi_put_command(cmd);
1096 else /* BLKPREP_DEFER */
1097 scsi_unprep_request(cmd->request);
1101 EXPORT_SYMBOL(scsi_init_io);
1103 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1104 struct request *req)
1106 struct scsi_cmnd *cmd;
1108 if (!req->special) {
1109 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1117 /* pull a tag out of the request if we have one */
1118 cmd->tag = req->tag;
1124 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1126 struct scsi_cmnd *cmd;
1127 int ret = scsi_prep_state_check(sdev, req);
1129 if (ret != BLKPREP_OK)
1132 cmd = scsi_get_cmd_from_req(sdev, req);
1134 return BLKPREP_DEFER;
1137 * BLOCK_PC requests may transfer data, in which case they must
1138 * a bio attached to them. Or they might contain a SCSI command
1139 * that does not transfer data, in which case they may optionally
1140 * submit a request without an attached bio.
1145 BUG_ON(!req->nr_phys_segments);
1147 ret = scsi_init_io(cmd, GFP_ATOMIC);
1151 BUG_ON(req->data_len);
1154 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1158 BUILD_BUG_ON(sizeof(req->cmd) > sizeof(cmd->cmnd));
1159 memcpy(cmd->cmnd, req->cmd, sizeof(cmd->cmnd));
1160 cmd->cmd_len = req->cmd_len;
1162 cmd->sc_data_direction = DMA_NONE;
1163 else if (rq_data_dir(req) == WRITE)
1164 cmd->sc_data_direction = DMA_TO_DEVICE;
1166 cmd->sc_data_direction = DMA_FROM_DEVICE;
1168 cmd->transfersize = req->data_len;
1169 cmd->allowed = req->retries;
1170 cmd->timeout_per_command = req->timeout;
1173 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1176 * Setup a REQ_TYPE_FS command. These are simple read/write request
1177 * from filesystems that still need to be translated to SCSI CDBs from
1180 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1182 struct scsi_cmnd *cmd;
1183 int ret = scsi_prep_state_check(sdev, req);
1185 if (ret != BLKPREP_OK)
1188 * Filesystem requests must transfer data.
1190 BUG_ON(!req->nr_phys_segments);
1192 cmd = scsi_get_cmd_from_req(sdev, req);
1194 return BLKPREP_DEFER;
1196 return scsi_init_io(cmd, GFP_ATOMIC);
1198 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1200 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1202 int ret = BLKPREP_OK;
1205 * If the device is not in running state we will reject some
1208 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1209 switch (sdev->sdev_state) {
1212 * If the device is offline we refuse to process any
1213 * commands. The device must be brought online
1214 * before trying any recovery commands.
1216 sdev_printk(KERN_ERR, sdev,
1217 "rejecting I/O to offline device\n");
1222 * If the device is fully deleted, we refuse to
1223 * process any commands as well.
1225 sdev_printk(KERN_ERR, sdev,
1226 "rejecting I/O to dead device\n");
1232 * If the devices is blocked we defer normal commands.
1234 if (!(req->cmd_flags & REQ_PREEMPT))
1235 ret = BLKPREP_DEFER;
1239 * For any other not fully online state we only allow
1240 * special commands. In particular any user initiated
1241 * command is not allowed.
1243 if (!(req->cmd_flags & REQ_PREEMPT))
1250 EXPORT_SYMBOL(scsi_prep_state_check);
1252 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1254 struct scsi_device *sdev = q->queuedata;
1258 req->errors = DID_NO_CONNECT << 16;
1259 /* release the command and kill it */
1261 struct scsi_cmnd *cmd = req->special;
1262 scsi_release_buffers(cmd);
1263 scsi_put_command(cmd);
1264 req->special = NULL;
1269 * If we defer, the elv_next_request() returns NULL, but the
1270 * queue must be restarted, so we plug here if no returning
1271 * command will automatically do that.
1273 if (sdev->device_busy == 0)
1277 req->cmd_flags |= REQ_DONTPREP;
1282 EXPORT_SYMBOL(scsi_prep_return);
1284 int scsi_prep_fn(struct request_queue *q, struct request *req)
1286 struct scsi_device *sdev = q->queuedata;
1287 int ret = BLKPREP_KILL;
1289 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1290 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1291 return scsi_prep_return(q, req, ret);
1295 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1298 * Called with the queue_lock held.
1300 static inline int scsi_dev_queue_ready(struct request_queue *q,
1301 struct scsi_device *sdev)
1303 if (sdev->device_busy >= sdev->queue_depth)
1305 if (sdev->device_busy == 0 && sdev->device_blocked) {
1307 * unblock after device_blocked iterates to zero
1309 if (--sdev->device_blocked == 0) {
1311 sdev_printk(KERN_INFO, sdev,
1312 "unblocking device at zero depth\n"));
1318 if (sdev->device_blocked)
1325 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1326 * return 0. We must end up running the queue again whenever 0 is
1327 * returned, else IO can hang.
1329 * Called with host_lock held.
1331 static inline int scsi_host_queue_ready(struct request_queue *q,
1332 struct Scsi_Host *shost,
1333 struct scsi_device *sdev)
1335 if (scsi_host_in_recovery(shost))
1337 if (shost->host_busy == 0 && shost->host_blocked) {
1339 * unblock after host_blocked iterates to zero
1341 if (--shost->host_blocked == 0) {
1343 printk("scsi%d unblocking host at zero depth\n",
1350 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1351 shost->host_blocked || shost->host_self_blocked) {
1352 if (list_empty(&sdev->starved_entry))
1353 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1357 /* We're OK to process the command, so we can't be starved */
1358 if (!list_empty(&sdev->starved_entry))
1359 list_del_init(&sdev->starved_entry);
1365 * Kill a request for a dead device
1367 static void scsi_kill_request(struct request *req, struct request_queue *q)
1369 struct scsi_cmnd *cmd = req->special;
1370 struct scsi_device *sdev = cmd->device;
1371 struct Scsi_Host *shost = sdev->host;
1373 blkdev_dequeue_request(req);
1375 if (unlikely(cmd == NULL)) {
1376 printk(KERN_CRIT "impossible request in %s.\n",
1381 scsi_init_cmd_errh(cmd);
1382 cmd->result = DID_NO_CONNECT << 16;
1383 atomic_inc(&cmd->device->iorequest_cnt);
1386 * SCSI request completion path will do scsi_device_unbusy(),
1387 * bump busy counts. To bump the counters, we need to dance
1388 * with the locks as normal issue path does.
1390 sdev->device_busy++;
1391 spin_unlock(sdev->request_queue->queue_lock);
1392 spin_lock(shost->host_lock);
1394 spin_unlock(shost->host_lock);
1395 spin_lock(sdev->request_queue->queue_lock);
1400 static void scsi_softirq_done(struct request *rq)
1402 struct scsi_cmnd *cmd = rq->completion_data;
1403 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
1406 INIT_LIST_HEAD(&cmd->eh_entry);
1408 disposition = scsi_decide_disposition(cmd);
1409 if (disposition != SUCCESS &&
1410 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1411 sdev_printk(KERN_ERR, cmd->device,
1412 "timing out command, waited %lus\n",
1414 disposition = SUCCESS;
1417 scsi_log_completion(cmd, disposition);
1419 switch (disposition) {
1421 scsi_finish_command(cmd);
1424 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1426 case ADD_TO_MLQUEUE:
1427 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1430 if (!scsi_eh_scmd_add(cmd, 0))
1431 scsi_finish_command(cmd);
1436 * Function: scsi_request_fn()
1438 * Purpose: Main strategy routine for SCSI.
1440 * Arguments: q - Pointer to actual queue.
1444 * Lock status: IO request lock assumed to be held when called.
1446 static void scsi_request_fn(struct request_queue *q)
1448 struct scsi_device *sdev = q->queuedata;
1449 struct Scsi_Host *shost;
1450 struct scsi_cmnd *cmd;
1451 struct request *req;
1454 printk("scsi: killing requests for dead queue\n");
1455 while ((req = elv_next_request(q)) != NULL)
1456 scsi_kill_request(req, q);
1460 if(!get_device(&sdev->sdev_gendev))
1461 /* We must be tearing the block queue down already */
1465 * To start with, we keep looping until the queue is empty, or until
1466 * the host is no longer able to accept any more requests.
1469 while (!blk_queue_plugged(q)) {
1472 * get next queueable request. We do this early to make sure
1473 * that the request is fully prepared even if we cannot
1476 req = elv_next_request(q);
1477 if (!req || !scsi_dev_queue_ready(q, sdev))
1480 if (unlikely(!scsi_device_online(sdev))) {
1481 sdev_printk(KERN_ERR, sdev,
1482 "rejecting I/O to offline device\n");
1483 scsi_kill_request(req, q);
1489 * Remove the request from the request list.
1491 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1492 blkdev_dequeue_request(req);
1493 sdev->device_busy++;
1495 spin_unlock(q->queue_lock);
1497 if (unlikely(cmd == NULL)) {
1498 printk(KERN_CRIT "impossible request in %s.\n"
1499 "please mail a stack trace to "
1500 "linux-scsi@vger.kernel.org\n",
1502 blk_dump_rq_flags(req, "foo");
1505 spin_lock(shost->host_lock);
1507 if (!scsi_host_queue_ready(q, shost, sdev))
1509 if (scsi_target(sdev)->single_lun) {
1510 if (scsi_target(sdev)->starget_sdev_user &&
1511 scsi_target(sdev)->starget_sdev_user != sdev)
1513 scsi_target(sdev)->starget_sdev_user = sdev;
1518 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1519 * take the lock again.
1521 spin_unlock_irq(shost->host_lock);
1524 * Finally, initialize any error handling parameters, and set up
1525 * the timers for timeouts.
1527 scsi_init_cmd_errh(cmd);
1530 * Dispatch the command to the low-level driver.
1532 rtn = scsi_dispatch_cmd(cmd);
1533 spin_lock_irq(q->queue_lock);
1535 /* we're refusing the command; because of
1536 * the way locks get dropped, we need to
1537 * check here if plugging is required */
1538 if(sdev->device_busy == 0)
1548 spin_unlock_irq(shost->host_lock);
1551 * lock q, handle tag, requeue req, and decrement device_busy. We
1552 * must return with queue_lock held.
1554 * Decrementing device_busy without checking it is OK, as all such
1555 * cases (host limits or settings) should run the queue at some
1558 spin_lock_irq(q->queue_lock);
1559 blk_requeue_request(q, req);
1560 sdev->device_busy--;
1561 if(sdev->device_busy == 0)
1564 /* must be careful here...if we trigger the ->remove() function
1565 * we cannot be holding the q lock */
1566 spin_unlock_irq(q->queue_lock);
1567 put_device(&sdev->sdev_gendev);
1568 spin_lock_irq(q->queue_lock);
1571 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1573 struct device *host_dev;
1574 u64 bounce_limit = 0xffffffff;
1576 if (shost->unchecked_isa_dma)
1577 return BLK_BOUNCE_ISA;
1579 * Platforms with virtual-DMA translation
1580 * hardware have no practical limit.
1582 if (!PCI_DMA_BUS_IS_PHYS)
1583 return BLK_BOUNCE_ANY;
1585 host_dev = scsi_get_device(shost);
1586 if (host_dev && host_dev->dma_mask)
1587 bounce_limit = *host_dev->dma_mask;
1589 return bounce_limit;
1591 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1593 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1594 request_fn_proc *request_fn)
1596 struct request_queue *q;
1598 q = blk_init_queue(request_fn, NULL);
1603 * this limit is imposed by hardware restrictions
1605 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1608 * In the future, sg chaining support will be mandatory and this
1609 * ifdef can then go away. Right now we don't have all archs
1610 * converted, so better keep it safe.
1612 #ifdef ARCH_HAS_SG_CHAIN
1613 if (shost->use_sg_chaining)
1614 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1616 blk_queue_max_phys_segments(q, SCSI_MAX_SG_SEGMENTS);
1618 blk_queue_max_phys_segments(q, SCSI_MAX_SG_SEGMENTS);
1621 blk_queue_max_sectors(q, shost->max_sectors);
1622 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1623 blk_queue_segment_boundary(q, shost->dma_boundary);
1625 if (!shost->use_clustering)
1626 clear_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
1629 * set a reasonable default alignment on word boundaries: the
1630 * host and device may alter it using
1631 * blk_queue_update_dma_alignment() later.
1633 blk_queue_dma_alignment(q, 0x03);
1637 EXPORT_SYMBOL(__scsi_alloc_queue);
1639 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1641 struct request_queue *q;
1643 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1647 blk_queue_prep_rq(q, scsi_prep_fn);
1648 blk_queue_softirq_done(q, scsi_softirq_done);
1652 void scsi_free_queue(struct request_queue *q)
1654 blk_cleanup_queue(q);
1658 * Function: scsi_block_requests()
1660 * Purpose: Utility function used by low-level drivers to prevent further
1661 * commands from being queued to the device.
1663 * Arguments: shost - Host in question
1667 * Lock status: No locks are assumed held.
1669 * Notes: There is no timer nor any other means by which the requests
1670 * get unblocked other than the low-level driver calling
1671 * scsi_unblock_requests().
1673 void scsi_block_requests(struct Scsi_Host *shost)
1675 shost->host_self_blocked = 1;
1677 EXPORT_SYMBOL(scsi_block_requests);
1680 * Function: scsi_unblock_requests()
1682 * Purpose: Utility function used by low-level drivers to allow further
1683 * commands from being queued to the device.
1685 * Arguments: shost - Host in question
1689 * Lock status: No locks are assumed held.
1691 * Notes: There is no timer nor any other means by which the requests
1692 * get unblocked other than the low-level driver calling
1693 * scsi_unblock_requests().
1695 * This is done as an API function so that changes to the
1696 * internals of the scsi mid-layer won't require wholesale
1697 * changes to drivers that use this feature.
1699 void scsi_unblock_requests(struct Scsi_Host *shost)
1701 shost->host_self_blocked = 0;
1702 scsi_run_host_queues(shost);
1704 EXPORT_SYMBOL(scsi_unblock_requests);
1706 int __init scsi_init_queue(void)
1710 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1711 sizeof(struct scsi_io_context),
1713 if (!scsi_io_context_cache) {
1714 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1718 scsi_bidi_sdb_cache = kmem_cache_create("scsi_bidi_sdb",
1719 sizeof(struct scsi_data_buffer),
1721 if (!scsi_bidi_sdb_cache) {
1722 printk(KERN_ERR "SCSI: can't init scsi bidi sdb cache\n");
1726 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1727 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1728 int size = sgp->size * sizeof(struct scatterlist);
1730 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1731 SLAB_HWCACHE_ALIGN, NULL);
1733 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1737 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1740 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1748 void scsi_exit_queue(void)
1752 kmem_cache_destroy(scsi_io_context_cache);
1754 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1755 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1756 mempool_destroy(sgp->pool);
1757 kmem_cache_destroy(sgp->slab);
1762 * scsi_mode_select - issue a mode select
1763 * @sdev: SCSI device to be queried
1764 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1765 * @sp: Save page bit (0 == don't save, 1 == save)
1766 * @modepage: mode page being requested
1767 * @buffer: request buffer (may not be smaller than eight bytes)
1768 * @len: length of request buffer.
1769 * @timeout: command timeout
1770 * @retries: number of retries before failing
1771 * @data: returns a structure abstracting the mode header data
1772 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1773 * must be SCSI_SENSE_BUFFERSIZE big.
1775 * Returns zero if successful; negative error number or scsi
1780 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1781 unsigned char *buffer, int len, int timeout, int retries,
1782 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1784 unsigned char cmd[10];
1785 unsigned char *real_buffer;
1788 memset(cmd, 0, sizeof(cmd));
1789 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1791 if (sdev->use_10_for_ms) {
1794 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1797 memcpy(real_buffer + 8, buffer, len);
1801 real_buffer[2] = data->medium_type;
1802 real_buffer[3] = data->device_specific;
1803 real_buffer[4] = data->longlba ? 0x01 : 0;
1805 real_buffer[6] = data->block_descriptor_length >> 8;
1806 real_buffer[7] = data->block_descriptor_length;
1808 cmd[0] = MODE_SELECT_10;
1812 if (len > 255 || data->block_descriptor_length > 255 ||
1816 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1819 memcpy(real_buffer + 4, buffer, len);
1822 real_buffer[1] = data->medium_type;
1823 real_buffer[2] = data->device_specific;
1824 real_buffer[3] = data->block_descriptor_length;
1827 cmd[0] = MODE_SELECT;
1831 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1832 sshdr, timeout, retries);
1836 EXPORT_SYMBOL_GPL(scsi_mode_select);
1839 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1840 * @sdev: SCSI device to be queried
1841 * @dbd: set if mode sense will allow block descriptors to be returned
1842 * @modepage: mode page being requested
1843 * @buffer: request buffer (may not be smaller than eight bytes)
1844 * @len: length of request buffer.
1845 * @timeout: command timeout
1846 * @retries: number of retries before failing
1847 * @data: returns a structure abstracting the mode header data
1848 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1849 * must be SCSI_SENSE_BUFFERSIZE big.
1851 * Returns zero if unsuccessful, or the header offset (either 4
1852 * or 8 depending on whether a six or ten byte command was
1853 * issued) if successful.
1856 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1857 unsigned char *buffer, int len, int timeout, int retries,
1858 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1860 unsigned char cmd[12];
1864 struct scsi_sense_hdr my_sshdr;
1866 memset(data, 0, sizeof(*data));
1867 memset(&cmd[0], 0, 12);
1868 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1871 /* caller might not be interested in sense, but we need it */
1876 use_10_for_ms = sdev->use_10_for_ms;
1878 if (use_10_for_ms) {
1882 cmd[0] = MODE_SENSE_10;
1889 cmd[0] = MODE_SENSE;
1894 memset(buffer, 0, len);
1896 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1897 sshdr, timeout, retries);
1899 /* This code looks awful: what it's doing is making sure an
1900 * ILLEGAL REQUEST sense return identifies the actual command
1901 * byte as the problem. MODE_SENSE commands can return
1902 * ILLEGAL REQUEST if the code page isn't supported */
1904 if (use_10_for_ms && !scsi_status_is_good(result) &&
1905 (driver_byte(result) & DRIVER_SENSE)) {
1906 if (scsi_sense_valid(sshdr)) {
1907 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1908 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1910 * Invalid command operation code
1912 sdev->use_10_for_ms = 0;
1918 if(scsi_status_is_good(result)) {
1919 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1920 (modepage == 6 || modepage == 8))) {
1921 /* Initio breakage? */
1924 data->medium_type = 0;
1925 data->device_specific = 0;
1927 data->block_descriptor_length = 0;
1928 } else if(use_10_for_ms) {
1929 data->length = buffer[0]*256 + buffer[1] + 2;
1930 data->medium_type = buffer[2];
1931 data->device_specific = buffer[3];
1932 data->longlba = buffer[4] & 0x01;
1933 data->block_descriptor_length = buffer[6]*256
1936 data->length = buffer[0] + 1;
1937 data->medium_type = buffer[1];
1938 data->device_specific = buffer[2];
1939 data->block_descriptor_length = buffer[3];
1941 data->header_length = header_length;
1946 EXPORT_SYMBOL(scsi_mode_sense);
1949 * scsi_test_unit_ready - test if unit is ready
1950 * @sdev: scsi device to change the state of.
1951 * @timeout: command timeout
1952 * @retries: number of retries before failing
1953 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1954 * returning sense. Make sure that this is cleared before passing
1957 * Returns zero if unsuccessful or an error if TUR failed. For
1958 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1959 * translated to success, with the ->changed flag updated.
1962 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1963 struct scsi_sense_hdr *sshdr_external)
1966 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1968 struct scsi_sense_hdr *sshdr;
1971 if (!sshdr_external)
1972 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1974 sshdr = sshdr_external;
1976 /* try to eat the UNIT_ATTENTION if there are enough retries */
1978 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
1980 } while ((driver_byte(result) & DRIVER_SENSE) &&
1981 sshdr && sshdr->sense_key == UNIT_ATTENTION &&
1985 /* could not allocate sense buffer, so can't process it */
1988 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
1990 if ((scsi_sense_valid(sshdr)) &&
1991 ((sshdr->sense_key == UNIT_ATTENTION) ||
1992 (sshdr->sense_key == NOT_READY))) {
1997 if (!sshdr_external)
2001 EXPORT_SYMBOL(scsi_test_unit_ready);
2004 * scsi_device_set_state - Take the given device through the device state model.
2005 * @sdev: scsi device to change the state of.
2006 * @state: state to change to.
2008 * Returns zero if unsuccessful or an error if the requested
2009 * transition is illegal.
2012 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2014 enum scsi_device_state oldstate = sdev->sdev_state;
2016 if (state == oldstate)
2021 /* There are no legal states that come back to
2022 * created. This is the manually initialised start
2096 sdev->sdev_state = state;
2100 SCSI_LOG_ERROR_RECOVERY(1,
2101 sdev_printk(KERN_ERR, sdev,
2102 "Illegal state transition %s->%s\n",
2103 scsi_device_state_name(oldstate),
2104 scsi_device_state_name(state))
2108 EXPORT_SYMBOL(scsi_device_set_state);
2111 * sdev_evt_emit - emit a single SCSI device uevent
2112 * @sdev: associated SCSI device
2113 * @evt: event to emit
2115 * Send a single uevent (scsi_event) to the associated scsi_device.
2117 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2122 switch (evt->evt_type) {
2123 case SDEV_EVT_MEDIA_CHANGE:
2124 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2134 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2138 * sdev_evt_thread - send a uevent for each scsi event
2139 * @work: work struct for scsi_device
2141 * Dispatch queued events to their associated scsi_device kobjects
2144 void scsi_evt_thread(struct work_struct *work)
2146 struct scsi_device *sdev;
2147 LIST_HEAD(event_list);
2149 sdev = container_of(work, struct scsi_device, event_work);
2152 struct scsi_event *evt;
2153 struct list_head *this, *tmp;
2154 unsigned long flags;
2156 spin_lock_irqsave(&sdev->list_lock, flags);
2157 list_splice_init(&sdev->event_list, &event_list);
2158 spin_unlock_irqrestore(&sdev->list_lock, flags);
2160 if (list_empty(&event_list))
2163 list_for_each_safe(this, tmp, &event_list) {
2164 evt = list_entry(this, struct scsi_event, node);
2165 list_del(&evt->node);
2166 scsi_evt_emit(sdev, evt);
2173 * sdev_evt_send - send asserted event to uevent thread
2174 * @sdev: scsi_device event occurred on
2175 * @evt: event to send
2177 * Assert scsi device event asynchronously.
2179 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2181 unsigned long flags;
2183 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2188 spin_lock_irqsave(&sdev->list_lock, flags);
2189 list_add_tail(&evt->node, &sdev->event_list);
2190 schedule_work(&sdev->event_work);
2191 spin_unlock_irqrestore(&sdev->list_lock, flags);
2193 EXPORT_SYMBOL_GPL(sdev_evt_send);
2196 * sdev_evt_alloc - allocate a new scsi event
2197 * @evt_type: type of event to allocate
2198 * @gfpflags: GFP flags for allocation
2200 * Allocates and returns a new scsi_event.
2202 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2205 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2209 evt->evt_type = evt_type;
2210 INIT_LIST_HEAD(&evt->node);
2212 /* evt_type-specific initialization, if any */
2214 case SDEV_EVT_MEDIA_CHANGE:
2222 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2225 * sdev_evt_send_simple - send asserted event to uevent thread
2226 * @sdev: scsi_device event occurred on
2227 * @evt_type: type of event to send
2228 * @gfpflags: GFP flags for allocation
2230 * Assert scsi device event asynchronously, given an event type.
2232 void sdev_evt_send_simple(struct scsi_device *sdev,
2233 enum scsi_device_event evt_type, gfp_t gfpflags)
2235 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2237 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2242 sdev_evt_send(sdev, evt);
2244 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2247 * scsi_device_quiesce - Block user issued commands.
2248 * @sdev: scsi device to quiesce.
2250 * This works by trying to transition to the SDEV_QUIESCE state
2251 * (which must be a legal transition). When the device is in this
2252 * state, only special requests will be accepted, all others will
2253 * be deferred. Since special requests may also be requeued requests,
2254 * a successful return doesn't guarantee the device will be
2255 * totally quiescent.
2257 * Must be called with user context, may sleep.
2259 * Returns zero if unsuccessful or an error if not.
2262 scsi_device_quiesce(struct scsi_device *sdev)
2264 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2268 scsi_run_queue(sdev->request_queue);
2269 while (sdev->device_busy) {
2270 msleep_interruptible(200);
2271 scsi_run_queue(sdev->request_queue);
2275 EXPORT_SYMBOL(scsi_device_quiesce);
2278 * scsi_device_resume - Restart user issued commands to a quiesced device.
2279 * @sdev: scsi device to resume.
2281 * Moves the device from quiesced back to running and restarts the
2284 * Must be called with user context, may sleep.
2287 scsi_device_resume(struct scsi_device *sdev)
2289 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2291 scsi_run_queue(sdev->request_queue);
2293 EXPORT_SYMBOL(scsi_device_resume);
2296 device_quiesce_fn(struct scsi_device *sdev, void *data)
2298 scsi_device_quiesce(sdev);
2302 scsi_target_quiesce(struct scsi_target *starget)
2304 starget_for_each_device(starget, NULL, device_quiesce_fn);
2306 EXPORT_SYMBOL(scsi_target_quiesce);
2309 device_resume_fn(struct scsi_device *sdev, void *data)
2311 scsi_device_resume(sdev);
2315 scsi_target_resume(struct scsi_target *starget)
2317 starget_for_each_device(starget, NULL, device_resume_fn);
2319 EXPORT_SYMBOL(scsi_target_resume);
2322 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2323 * @sdev: device to block
2325 * Block request made by scsi lld's to temporarily stop all
2326 * scsi commands on the specified device. Called from interrupt
2327 * or normal process context.
2329 * Returns zero if successful or error if not
2332 * This routine transitions the device to the SDEV_BLOCK state
2333 * (which must be a legal transition). When the device is in this
2334 * state, all commands are deferred until the scsi lld reenables
2335 * the device with scsi_device_unblock or device_block_tmo fires.
2336 * This routine assumes the host_lock is held on entry.
2339 scsi_internal_device_block(struct scsi_device *sdev)
2341 struct request_queue *q = sdev->request_queue;
2342 unsigned long flags;
2345 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2350 * The device has transitioned to SDEV_BLOCK. Stop the
2351 * block layer from calling the midlayer with this device's
2354 spin_lock_irqsave(q->queue_lock, flags);
2356 spin_unlock_irqrestore(q->queue_lock, flags);
2360 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2363 * scsi_internal_device_unblock - resume a device after a block request
2364 * @sdev: device to resume
2366 * Called by scsi lld's or the midlayer to restart the device queue
2367 * for the previously suspended scsi device. Called from interrupt or
2368 * normal process context.
2370 * Returns zero if successful or error if not.
2373 * This routine transitions the device to the SDEV_RUNNING state
2374 * (which must be a legal transition) allowing the midlayer to
2375 * goose the queue for this device. This routine assumes the
2376 * host_lock is held upon entry.
2379 scsi_internal_device_unblock(struct scsi_device *sdev)
2381 struct request_queue *q = sdev->request_queue;
2383 unsigned long flags;
2386 * Try to transition the scsi device to SDEV_RUNNING
2387 * and goose the device queue if successful.
2389 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2393 spin_lock_irqsave(q->queue_lock, flags);
2395 spin_unlock_irqrestore(q->queue_lock, flags);
2399 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2402 device_block(struct scsi_device *sdev, void *data)
2404 scsi_internal_device_block(sdev);
2408 target_block(struct device *dev, void *data)
2410 if (scsi_is_target_device(dev))
2411 starget_for_each_device(to_scsi_target(dev), NULL,
2417 scsi_target_block(struct device *dev)
2419 if (scsi_is_target_device(dev))
2420 starget_for_each_device(to_scsi_target(dev), NULL,
2423 device_for_each_child(dev, NULL, target_block);
2425 EXPORT_SYMBOL_GPL(scsi_target_block);
2428 device_unblock(struct scsi_device *sdev, void *data)
2430 scsi_internal_device_unblock(sdev);
2434 target_unblock(struct device *dev, void *data)
2436 if (scsi_is_target_device(dev))
2437 starget_for_each_device(to_scsi_target(dev), NULL,
2443 scsi_target_unblock(struct device *dev)
2445 if (scsi_is_target_device(dev))
2446 starget_for_each_device(to_scsi_target(dev), NULL,
2449 device_for_each_child(dev, NULL, target_unblock);
2451 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2454 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2455 * @sgl: scatter-gather list
2456 * @sg_count: number of segments in sg
2457 * @offset: offset in bytes into sg, on return offset into the mapped area
2458 * @len: bytes to map, on return number of bytes mapped
2460 * Returns virtual address of the start of the mapped page
2462 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2463 size_t *offset, size_t *len)
2466 size_t sg_len = 0, len_complete = 0;
2467 struct scatterlist *sg;
2470 WARN_ON(!irqs_disabled());
2472 for_each_sg(sgl, sg, sg_count, i) {
2473 len_complete = sg_len; /* Complete sg-entries */
2474 sg_len += sg->length;
2475 if (sg_len > *offset)
2479 if (unlikely(i == sg_count)) {
2480 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2482 __FUNCTION__, sg_len, *offset, sg_count);
2487 /* Offset starting from the beginning of first page in this sg-entry */
2488 *offset = *offset - len_complete + sg->offset;
2490 /* Assumption: contiguous pages can be accessed as "page + i" */
2491 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2492 *offset &= ~PAGE_MASK;
2494 /* Bytes in this sg-entry from *offset to the end of the page */
2495 sg_len = PAGE_SIZE - *offset;
2499 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2501 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2504 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2505 * @virt: virtual address to be unmapped
2507 void scsi_kunmap_atomic_sg(void *virt)
2509 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2511 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);