2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/bitops.h>
12 #include <linux/blkdev.h>
13 #include <linux/completion.h>
14 #include <linux/kernel.h>
15 #include <linux/mempool.h>
16 #include <linux/slab.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
23 #include <scsi/scsi.h>
24 #include <scsi/scsi_cmnd.h>
25 #include <scsi/scsi_dbg.h>
26 #include <scsi/scsi_device.h>
27 #include <scsi/scsi_driver.h>
28 #include <scsi/scsi_eh.h>
29 #include <scsi/scsi_host.h>
31 #include "scsi_priv.h"
32 #include "scsi_logging.h"
35 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
36 #define SG_MEMPOOL_SIZE 2
38 struct scsi_host_sg_pool {
41 struct kmem_cache *slab;
45 #define SP(x) { x, "sgpool-" __stringify(x) }
46 #if (SCSI_MAX_SG_SEGMENTS < 32)
47 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
49 static struct scsi_host_sg_pool scsi_sg_pools[] = {
52 #if (SCSI_MAX_SG_SEGMENTS > 32)
54 #if (SCSI_MAX_SG_SEGMENTS > 64)
56 #if (SCSI_MAX_SG_SEGMENTS > 128)
58 #if (SCSI_MAX_SG_SEGMENTS > 256)
59 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
64 SP(SCSI_MAX_SG_SEGMENTS)
68 static struct kmem_cache *scsi_bidi_sdb_cache;
70 static void scsi_run_queue(struct request_queue *q);
73 * Function: scsi_unprep_request()
75 * Purpose: Remove all preparation done for a request, including its
76 * associated scsi_cmnd, so that it can be requeued.
78 * Arguments: req - request to unprepare
80 * Lock status: Assumed that no locks are held upon entry.
84 static void scsi_unprep_request(struct request *req)
86 struct scsi_cmnd *cmd = req->special;
88 req->cmd_flags &= ~REQ_DONTPREP;
91 scsi_put_command(cmd);
95 * Function: scsi_queue_insert()
97 * Purpose: Insert a command in the midlevel queue.
99 * Arguments: cmd - command that we are adding to queue.
100 * reason - why we are inserting command to queue.
102 * Lock status: Assumed that lock is not held upon entry.
106 * Notes: We do this for one of two cases. Either the host is busy
107 * and it cannot accept any more commands for the time being,
108 * or the device returned QUEUE_FULL and can accept no more
110 * Notes: This could be called either from an interrupt context or a
111 * normal process context.
113 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
115 struct Scsi_Host *host = cmd->device->host;
116 struct scsi_device *device = cmd->device;
117 struct request_queue *q = device->request_queue;
121 printk("Inserting command %p into mlqueue\n", cmd));
124 * Set the appropriate busy bit for the device/host.
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
136 if (reason == SCSI_MLQUEUE_HOST_BUSY)
137 host->host_blocked = host->max_host_blocked;
138 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
139 device->device_blocked = device->max_device_blocked;
142 * Decrement the counters, since these commands are no longer
143 * active on the host/device.
145 scsi_device_unbusy(device);
148 * Requeue this command. It will go before all other commands
149 * that are already in the queue.
151 * NOTE: there is magic here about the way the queue is plugged if
152 * we have no outstanding commands.
154 * Although we *don't* plug the queue, we call the request
155 * function. The SCSI request function detects the blocked condition
156 * and plugs the queue appropriately.
158 spin_lock_irqsave(q->queue_lock, flags);
159 blk_requeue_request(q, cmd->request);
160 spin_unlock_irqrestore(q->queue_lock, flags);
168 * scsi_execute - insert request and wait for the result
171 * @data_direction: data direction
172 * @buffer: data buffer
173 * @bufflen: len of buffer
174 * @sense: optional sense buffer
175 * @timeout: request timeout in seconds
176 * @retries: number of times to retry request
177 * @flags: or into request flags;
179 * returns the req->errors value which is the scsi_cmnd result
182 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
183 int data_direction, void *buffer, unsigned bufflen,
184 unsigned char *sense, int timeout, int retries, int flags)
187 int write = (data_direction == DMA_TO_DEVICE);
188 int ret = DRIVER_ERROR << 24;
190 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
192 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
193 buffer, bufflen, __GFP_WAIT))
196 req->cmd_len = COMMAND_SIZE(cmd[0]);
197 memcpy(req->cmd, cmd, req->cmd_len);
200 req->retries = retries;
201 req->timeout = timeout;
202 req->cmd_type = REQ_TYPE_BLOCK_PC;
203 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
206 * head injection *required* here otherwise quiesce won't work
208 blk_execute_rq(req->q, NULL, req, 1);
212 blk_put_request(req);
216 EXPORT_SYMBOL(scsi_execute);
219 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
220 int data_direction, void *buffer, unsigned bufflen,
221 struct scsi_sense_hdr *sshdr, int timeout, int retries)
227 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
229 return DRIVER_ERROR << 24;
231 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
232 sense, timeout, retries, 0);
234 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
239 EXPORT_SYMBOL(scsi_execute_req);
241 struct scsi_io_context {
243 void (*done)(void *data, char *sense, int result, int resid);
244 char sense[SCSI_SENSE_BUFFERSIZE];
247 static struct kmem_cache *scsi_io_context_cache;
249 static void scsi_end_async(struct request *req, int uptodate)
251 struct scsi_io_context *sioc = req->end_io_data;
254 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
256 kmem_cache_free(scsi_io_context_cache, sioc);
257 __blk_put_request(req->q, req);
260 static int scsi_merge_bio(struct request *rq, struct bio *bio)
262 struct request_queue *q = rq->q;
264 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
265 if (rq_data_dir(rq) == WRITE)
266 bio->bi_rw |= (1 << BIO_RW);
267 blk_queue_bounce(q, &bio);
269 return blk_rq_append_bio(q, rq, bio);
272 static void scsi_bi_endio(struct bio *bio, int error)
278 * scsi_req_map_sg - map a scatterlist into a request
279 * @rq: request to fill
281 * @nsegs: number of elements
282 * @bufflen: len of buffer
283 * @gfp: memory allocation flags
285 * scsi_req_map_sg maps a scatterlist into a request so that the
286 * request can be sent to the block layer. We do not trust the scatterlist
287 * sent to use, as some ULDs use that struct to only organize the pages.
289 static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
290 int nsegs, unsigned bufflen, gfp_t gfp)
292 struct request_queue *q = rq->q;
293 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
294 unsigned int data_len = bufflen, len, bytes, off;
295 struct scatterlist *sg;
297 struct bio *bio = NULL;
298 int i, err, nr_vecs = 0;
300 for_each_sg(sgl, sg, nsegs, i) {
305 while (len > 0 && data_len > 0) {
307 * sg sends a scatterlist that is larger than
308 * the data_len it wants transferred for certain
311 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
312 bytes = min(bytes, data_len);
315 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
318 bio = bio_alloc(gfp, nr_vecs);
323 bio->bi_end_io = scsi_bi_endio;
326 if (bio_add_pc_page(q, bio, page, bytes, off) !=
333 if (bio->bi_vcnt >= nr_vecs) {
334 err = scsi_merge_bio(rq, bio);
349 rq->buffer = rq->data = NULL;
350 rq->data_len = bufflen;
354 while ((bio = rq->bio) != NULL) {
355 rq->bio = bio->bi_next;
357 * call endio instead of bio_put incase it was bounced
366 * scsi_execute_async - insert request
369 * @cmd_len: length of scsi cdb
370 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
371 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
372 * @bufflen: len of buffer
373 * @use_sg: if buffer is a scatterlist this is the number of elements
374 * @timeout: request timeout in seconds
375 * @retries: number of times to retry request
376 * @privdata: data passed to done()
377 * @done: callback function when done
378 * @gfp: memory allocation flags
380 int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
381 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
382 int use_sg, int timeout, int retries, void *privdata,
383 void (*done)(void *, char *, int, int), gfp_t gfp)
386 struct scsi_io_context *sioc;
388 int write = (data_direction == DMA_TO_DEVICE);
390 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp);
392 return DRIVER_ERROR << 24;
394 req = blk_get_request(sdev->request_queue, write, gfp);
397 req->cmd_type = REQ_TYPE_BLOCK_PC;
398 req->cmd_flags |= REQ_QUIET;
401 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
403 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
408 req->cmd_len = cmd_len;
409 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
410 memcpy(req->cmd, cmd, req->cmd_len);
411 req->sense = sioc->sense;
413 req->timeout = timeout;
414 req->retries = retries;
415 req->end_io_data = sioc;
417 sioc->data = privdata;
420 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
424 blk_put_request(req);
426 kmem_cache_free(scsi_io_context_cache, sioc);
427 return DRIVER_ERROR << 24;
429 EXPORT_SYMBOL_GPL(scsi_execute_async);
432 * Function: scsi_init_cmd_errh()
434 * Purpose: Initialize cmd fields related to error handling.
436 * Arguments: cmd - command that is ready to be queued.
438 * Notes: This function has the job of initializing a number of
439 * fields related to error handling. Typically this will
440 * be called once for each command, as required.
442 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
444 cmd->serial_number = 0;
445 scsi_set_resid(cmd, 0);
446 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
447 if (cmd->cmd_len == 0)
448 cmd->cmd_len = scsi_command_size(cmd->cmnd);
451 void scsi_device_unbusy(struct scsi_device *sdev)
453 struct Scsi_Host *shost = sdev->host;
456 spin_lock_irqsave(shost->host_lock, flags);
458 if (unlikely(scsi_host_in_recovery(shost) &&
459 (shost->host_failed || shost->host_eh_scheduled)))
460 scsi_eh_wakeup(shost);
461 spin_unlock(shost->host_lock);
462 spin_lock(sdev->request_queue->queue_lock);
464 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
468 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
469 * and call blk_run_queue for all the scsi_devices on the target -
470 * including current_sdev first.
472 * Called with *no* scsi locks held.
474 static void scsi_single_lun_run(struct scsi_device *current_sdev)
476 struct Scsi_Host *shost = current_sdev->host;
477 struct scsi_device *sdev, *tmp;
478 struct scsi_target *starget = scsi_target(current_sdev);
481 spin_lock_irqsave(shost->host_lock, flags);
482 starget->starget_sdev_user = NULL;
483 spin_unlock_irqrestore(shost->host_lock, flags);
486 * Call blk_run_queue for all LUNs on the target, starting with
487 * current_sdev. We race with others (to set starget_sdev_user),
488 * but in most cases, we will be first. Ideally, each LU on the
489 * target would get some limited time or requests on the target.
491 blk_run_queue(current_sdev->request_queue);
493 spin_lock_irqsave(shost->host_lock, flags);
494 if (starget->starget_sdev_user)
496 list_for_each_entry_safe(sdev, tmp, &starget->devices,
497 same_target_siblings) {
498 if (sdev == current_sdev)
500 if (scsi_device_get(sdev))
503 spin_unlock_irqrestore(shost->host_lock, flags);
504 blk_run_queue(sdev->request_queue);
505 spin_lock_irqsave(shost->host_lock, flags);
507 scsi_device_put(sdev);
510 spin_unlock_irqrestore(shost->host_lock, flags);
514 * Function: scsi_run_queue()
516 * Purpose: Select a proper request queue to serve next
518 * Arguments: q - last request's queue
522 * Notes: The previous command was completely finished, start
523 * a new one if possible.
525 static void scsi_run_queue(struct request_queue *q)
527 struct scsi_device *sdev = q->queuedata;
528 struct Scsi_Host *shost = sdev->host;
531 if (scsi_target(sdev)->single_lun)
532 scsi_single_lun_run(sdev);
534 spin_lock_irqsave(shost->host_lock, flags);
535 while (!list_empty(&shost->starved_list) &&
536 !shost->host_blocked && !shost->host_self_blocked &&
537 !((shost->can_queue > 0) &&
538 (shost->host_busy >= shost->can_queue))) {
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);
695 static inline unsigned int scsi_sgtable_index(unsigned short nents)
699 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
704 index = get_count_order(nents) - 3;
709 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
711 struct scsi_host_sg_pool *sgp;
713 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
714 mempool_free(sgl, sgp->pool);
717 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
719 struct scsi_host_sg_pool *sgp;
721 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
722 return mempool_alloc(sgp->pool, gfp_mask);
725 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
732 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
733 gfp_mask, scsi_sg_alloc);
735 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
741 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
743 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
747 * Function: scsi_release_buffers()
749 * Purpose: Completion processing for block device I/O requests.
751 * Arguments: cmd - command that we are bailing.
753 * Lock status: Assumed that no lock is held upon entry.
757 * Notes: In the event that an upper level driver rejects a
758 * command, we must release resources allocated during
759 * the __init_io() function. Primarily this would involve
760 * the scatter-gather table, and potentially any bounce
763 void scsi_release_buffers(struct scsi_cmnd *cmd)
765 if (cmd->sdb.table.nents)
766 scsi_free_sgtable(&cmd->sdb);
768 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
770 if (scsi_bidi_cmnd(cmd)) {
771 struct scsi_data_buffer *bidi_sdb =
772 cmd->request->next_rq->special;
773 scsi_free_sgtable(bidi_sdb);
774 kmem_cache_free(scsi_bidi_sdb_cache, bidi_sdb);
775 cmd->request->next_rq->special = NULL;
778 EXPORT_SYMBOL(scsi_release_buffers);
781 * Bidi commands Must be complete as a whole, both sides at once.
782 * If part of the bytes were written and lld returned
783 * scsi_in()->resid and/or scsi_out()->resid this information will be left
784 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
785 * decide what to do with this information.
787 static void scsi_end_bidi_request(struct scsi_cmnd *cmd)
789 struct request *req = cmd->request;
790 unsigned int dlen = req->data_len;
791 unsigned int next_dlen = req->next_rq->data_len;
793 req->data_len = scsi_out(cmd)->resid;
794 req->next_rq->data_len = scsi_in(cmd)->resid;
796 /* The req and req->next_rq have not been completed */
797 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen));
799 scsi_release_buffers(cmd);
802 * This will goose the queue request function at the end, so we don't
803 * need to worry about launching another command.
805 scsi_next_command(cmd);
809 * Function: scsi_io_completion()
811 * Purpose: Completion processing for block device I/O requests.
813 * Arguments: cmd - command that is finished.
815 * Lock status: Assumed that no lock is held upon entry.
819 * Notes: This function is matched in terms of capabilities to
820 * the function that created the scatter-gather list.
821 * In other words, if there are no bounce buffers
822 * (the normal case for most drivers), we don't need
823 * the logic to deal with cleaning up afterwards.
825 * We must do one of several things here:
827 * a) Call scsi_end_request. This will finish off the
828 * specified number of sectors. If we are done, the
829 * command block will be released, and the queue
830 * function will be goosed. If we are not done, then
831 * scsi_end_request will directly goose the queue.
833 * b) We can just use scsi_requeue_command() here. This would
834 * be used if we just wanted to retry, for example.
836 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
838 int result = cmd->result;
839 int this_count = scsi_bufflen(cmd);
840 struct request_queue *q = cmd->device->request_queue;
841 struct request *req = cmd->request;
843 struct scsi_sense_hdr sshdr;
845 int sense_deferred = 0;
848 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
850 sense_deferred = scsi_sense_is_deferred(&sshdr);
853 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
854 req->errors = result;
856 if (sense_valid && req->sense) {
858 * SG_IO wants current and deferred errors
860 int len = 8 + cmd->sense_buffer[7];
862 if (len > SCSI_SENSE_BUFFERSIZE)
863 len = SCSI_SENSE_BUFFERSIZE;
864 memcpy(req->sense, cmd->sense_buffer, len);
865 req->sense_len = len;
870 if (scsi_bidi_cmnd(cmd)) {
871 /* will also release_buffers */
872 scsi_end_bidi_request(cmd);
875 req->data_len = scsi_get_resid(cmd);
878 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
879 scsi_release_buffers(cmd);
882 * Next deal with any sectors which we were able to correctly
885 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
887 req->nr_sectors, good_bytes));
889 /* A number of bytes were successfully read. If there
890 * are leftovers and there is some kind of error
891 * (result != 0), retry the rest.
893 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
896 /* good_bytes = 0, or (inclusive) there were leftovers and
897 * result = 0, so scsi_end_request couldn't retry.
899 if (sense_valid && !sense_deferred) {
900 switch (sshdr.sense_key) {
902 if (cmd->device->removable) {
903 /* Detected disc change. Set a bit
904 * and quietly refuse further access.
906 cmd->device->changed = 1;
907 scsi_end_request(cmd, -EIO, this_count, 1);
910 /* Must have been a power glitch, or a
911 * bus reset. Could not have been a
912 * media change, so we just retry the
913 * request and see what happens.
915 scsi_requeue_command(q, cmd);
919 case ILLEGAL_REQUEST:
920 /* If we had an ILLEGAL REQUEST returned, then
921 * we may have performed an unsupported
922 * command. The only thing this should be
923 * would be a ten byte read where only a six
924 * byte read was supported. Also, on a system
925 * where READ CAPACITY failed, we may have
926 * read past the end of the disk.
928 if ((cmd->device->use_10_for_rw &&
929 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
930 (cmd->cmnd[0] == READ_10 ||
931 cmd->cmnd[0] == WRITE_10)) {
932 cmd->device->use_10_for_rw = 0;
933 /* This will cause a retry with a
936 scsi_requeue_command(q, cmd);
939 scsi_end_request(cmd, -EIO, this_count, 1);
944 /* If the device is in the process of becoming
945 * ready, or has a temporary blockage, retry.
947 if (sshdr.asc == 0x04) {
948 switch (sshdr.ascq) {
949 case 0x01: /* becoming ready */
950 case 0x04: /* format in progress */
951 case 0x05: /* rebuild in progress */
952 case 0x06: /* recalculation in progress */
953 case 0x07: /* operation in progress */
954 case 0x08: /* Long write in progress */
955 case 0x09: /* self test in progress */
956 scsi_requeue_command(q, cmd);
962 if (!(req->cmd_flags & REQ_QUIET))
963 scsi_cmd_print_sense_hdr(cmd,
967 scsi_end_request(cmd, -EIO, this_count, 1);
969 case VOLUME_OVERFLOW:
970 if (!(req->cmd_flags & REQ_QUIET)) {
971 scmd_printk(KERN_INFO, cmd,
972 "Volume overflow, CDB: ");
973 __scsi_print_command(cmd->cmnd);
974 scsi_print_sense("", cmd);
976 /* See SSC3rXX or current. */
977 scsi_end_request(cmd, -EIO, this_count, 1);
983 if (host_byte(result) == DID_RESET) {
984 /* Third party bus reset or reset for error recovery
985 * reasons. Just retry the request and see what
988 scsi_requeue_command(q, cmd);
992 if (!(req->cmd_flags & REQ_QUIET)) {
993 scsi_print_result(cmd);
994 if (driver_byte(result) & DRIVER_SENSE)
995 scsi_print_sense("", cmd);
998 scsi_end_request(cmd, -EIO, this_count, !result);
1001 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1007 * If sg table allocation fails, requeue request later.
1009 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1011 return BLKPREP_DEFER;
1017 * Next, walk the list, and fill in the addresses and sizes of
1020 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1021 BUG_ON(count > sdb->table.nents);
1022 sdb->table.nents = count;
1023 if (blk_pc_request(req))
1024 sdb->length = req->data_len;
1026 sdb->length = req->nr_sectors << 9;
1031 * Function: scsi_init_io()
1033 * Purpose: SCSI I/O initialize function.
1035 * Arguments: cmd - Command descriptor we wish to initialize
1037 * Returns: 0 on success
1038 * BLKPREP_DEFER if the failure is retryable
1039 * BLKPREP_KILL if the failure is fatal
1041 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1043 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
1047 if (blk_bidi_rq(cmd->request)) {
1048 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1049 scsi_bidi_sdb_cache, GFP_ATOMIC);
1051 error = BLKPREP_DEFER;
1055 cmd->request->next_rq->special = bidi_sdb;
1056 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1065 scsi_release_buffers(cmd);
1066 if (error == BLKPREP_KILL)
1067 scsi_put_command(cmd);
1068 else /* BLKPREP_DEFER */
1069 scsi_unprep_request(cmd->request);
1073 EXPORT_SYMBOL(scsi_init_io);
1075 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1076 struct request *req)
1078 struct scsi_cmnd *cmd;
1080 if (!req->special) {
1081 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1089 /* pull a tag out of the request if we have one */
1090 cmd->tag = req->tag;
1093 cmd->cmnd = req->cmd;
1098 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1100 struct scsi_cmnd *cmd;
1101 int ret = scsi_prep_state_check(sdev, req);
1103 if (ret != BLKPREP_OK)
1106 cmd = scsi_get_cmd_from_req(sdev, req);
1108 return BLKPREP_DEFER;
1111 * BLOCK_PC requests may transfer data, in which case they must
1112 * a bio attached to them. Or they might contain a SCSI command
1113 * that does not transfer data, in which case they may optionally
1114 * submit a request without an attached bio.
1119 BUG_ON(!req->nr_phys_segments);
1121 ret = scsi_init_io(cmd, GFP_ATOMIC);
1125 BUG_ON(req->data_len);
1128 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1132 cmd->cmd_len = req->cmd_len;
1134 cmd->sc_data_direction = DMA_NONE;
1135 else if (rq_data_dir(req) == WRITE)
1136 cmd->sc_data_direction = DMA_TO_DEVICE;
1138 cmd->sc_data_direction = DMA_FROM_DEVICE;
1140 cmd->transfersize = req->data_len;
1141 cmd->allowed = req->retries;
1142 cmd->timeout_per_command = req->timeout;
1145 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1148 * Setup a REQ_TYPE_FS command. These are simple read/write request
1149 * from filesystems that still need to be translated to SCSI CDBs from
1152 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1154 struct scsi_cmnd *cmd;
1155 int ret = scsi_prep_state_check(sdev, req);
1157 if (ret != BLKPREP_OK)
1160 * Filesystem requests must transfer data.
1162 BUG_ON(!req->nr_phys_segments);
1164 cmd = scsi_get_cmd_from_req(sdev, req);
1166 return BLKPREP_DEFER;
1168 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1169 return scsi_init_io(cmd, GFP_ATOMIC);
1171 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1173 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1175 int ret = BLKPREP_OK;
1178 * If the device is not in running state we will reject some
1181 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1182 switch (sdev->sdev_state) {
1185 * If the device is offline we refuse to process any
1186 * commands. The device must be brought online
1187 * before trying any recovery commands.
1189 sdev_printk(KERN_ERR, sdev,
1190 "rejecting I/O to offline device\n");
1195 * If the device is fully deleted, we refuse to
1196 * process any commands as well.
1198 sdev_printk(KERN_ERR, sdev,
1199 "rejecting I/O to dead device\n");
1205 * If the devices is blocked we defer normal commands.
1207 if (!(req->cmd_flags & REQ_PREEMPT))
1208 ret = BLKPREP_DEFER;
1212 * For any other not fully online state we only allow
1213 * special commands. In particular any user initiated
1214 * command is not allowed.
1216 if (!(req->cmd_flags & REQ_PREEMPT))
1223 EXPORT_SYMBOL(scsi_prep_state_check);
1225 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1227 struct scsi_device *sdev = q->queuedata;
1231 req->errors = DID_NO_CONNECT << 16;
1232 /* release the command and kill it */
1234 struct scsi_cmnd *cmd = req->special;
1235 scsi_release_buffers(cmd);
1236 scsi_put_command(cmd);
1237 req->special = NULL;
1242 * If we defer, the elv_next_request() returns NULL, but the
1243 * queue must be restarted, so we plug here if no returning
1244 * command will automatically do that.
1246 if (sdev->device_busy == 0)
1250 req->cmd_flags |= REQ_DONTPREP;
1255 EXPORT_SYMBOL(scsi_prep_return);
1257 int scsi_prep_fn(struct request_queue *q, struct request *req)
1259 struct scsi_device *sdev = q->queuedata;
1260 int ret = BLKPREP_KILL;
1262 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1263 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1264 return scsi_prep_return(q, req, ret);
1268 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1271 * Called with the queue_lock held.
1273 static inline int scsi_dev_queue_ready(struct request_queue *q,
1274 struct scsi_device *sdev)
1276 if (sdev->device_busy >= sdev->queue_depth)
1278 if (sdev->device_busy == 0 && sdev->device_blocked) {
1280 * unblock after device_blocked iterates to zero
1282 if (--sdev->device_blocked == 0) {
1284 sdev_printk(KERN_INFO, sdev,
1285 "unblocking device at zero depth\n"));
1291 if (sdev->device_blocked)
1298 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1299 * return 0. We must end up running the queue again whenever 0 is
1300 * returned, else IO can hang.
1302 * Called with host_lock held.
1304 static inline int scsi_host_queue_ready(struct request_queue *q,
1305 struct Scsi_Host *shost,
1306 struct scsi_device *sdev)
1308 if (scsi_host_in_recovery(shost))
1310 if (shost->host_busy == 0 && shost->host_blocked) {
1312 * unblock after host_blocked iterates to zero
1314 if (--shost->host_blocked == 0) {
1316 printk("scsi%d unblocking host at zero depth\n",
1323 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1324 shost->host_blocked || shost->host_self_blocked) {
1325 if (list_empty(&sdev->starved_entry))
1326 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1330 /* We're OK to process the command, so we can't be starved */
1331 if (!list_empty(&sdev->starved_entry))
1332 list_del_init(&sdev->starved_entry);
1338 * Kill a request for a dead device
1340 static void scsi_kill_request(struct request *req, struct request_queue *q)
1342 struct scsi_cmnd *cmd = req->special;
1343 struct scsi_device *sdev = cmd->device;
1344 struct Scsi_Host *shost = sdev->host;
1346 blkdev_dequeue_request(req);
1348 if (unlikely(cmd == NULL)) {
1349 printk(KERN_CRIT "impossible request in %s.\n",
1354 scsi_init_cmd_errh(cmd);
1355 cmd->result = DID_NO_CONNECT << 16;
1356 atomic_inc(&cmd->device->iorequest_cnt);
1359 * SCSI request completion path will do scsi_device_unbusy(),
1360 * bump busy counts. To bump the counters, we need to dance
1361 * with the locks as normal issue path does.
1363 sdev->device_busy++;
1364 spin_unlock(sdev->request_queue->queue_lock);
1365 spin_lock(shost->host_lock);
1367 spin_unlock(shost->host_lock);
1368 spin_lock(sdev->request_queue->queue_lock);
1373 static void scsi_softirq_done(struct request *rq)
1375 struct scsi_cmnd *cmd = rq->completion_data;
1376 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
1379 INIT_LIST_HEAD(&cmd->eh_entry);
1381 disposition = scsi_decide_disposition(cmd);
1382 if (disposition != SUCCESS &&
1383 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1384 sdev_printk(KERN_ERR, cmd->device,
1385 "timing out command, waited %lus\n",
1387 disposition = SUCCESS;
1390 scsi_log_completion(cmd, disposition);
1392 switch (disposition) {
1394 scsi_finish_command(cmd);
1397 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1399 case ADD_TO_MLQUEUE:
1400 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1403 if (!scsi_eh_scmd_add(cmd, 0))
1404 scsi_finish_command(cmd);
1409 * Function: scsi_request_fn()
1411 * Purpose: Main strategy routine for SCSI.
1413 * Arguments: q - Pointer to actual queue.
1417 * Lock status: IO request lock assumed to be held when called.
1419 static void scsi_request_fn(struct request_queue *q)
1421 struct scsi_device *sdev = q->queuedata;
1422 struct Scsi_Host *shost;
1423 struct scsi_cmnd *cmd;
1424 struct request *req;
1427 printk("scsi: killing requests for dead queue\n");
1428 while ((req = elv_next_request(q)) != NULL)
1429 scsi_kill_request(req, q);
1433 if(!get_device(&sdev->sdev_gendev))
1434 /* We must be tearing the block queue down already */
1438 * To start with, we keep looping until the queue is empty, or until
1439 * the host is no longer able to accept any more requests.
1442 while (!blk_queue_plugged(q)) {
1445 * get next queueable request. We do this early to make sure
1446 * that the request is fully prepared even if we cannot
1449 req = elv_next_request(q);
1450 if (!req || !scsi_dev_queue_ready(q, sdev))
1453 if (unlikely(!scsi_device_online(sdev))) {
1454 sdev_printk(KERN_ERR, sdev,
1455 "rejecting I/O to offline device\n");
1456 scsi_kill_request(req, q);
1462 * Remove the request from the request list.
1464 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1465 blkdev_dequeue_request(req);
1466 sdev->device_busy++;
1468 spin_unlock(q->queue_lock);
1470 if (unlikely(cmd == NULL)) {
1471 printk(KERN_CRIT "impossible request in %s.\n"
1472 "please mail a stack trace to "
1473 "linux-scsi@vger.kernel.org\n",
1475 blk_dump_rq_flags(req, "foo");
1478 spin_lock(shost->host_lock);
1480 if (!scsi_host_queue_ready(q, shost, sdev))
1482 if (scsi_target(sdev)->single_lun) {
1483 if (scsi_target(sdev)->starget_sdev_user &&
1484 scsi_target(sdev)->starget_sdev_user != sdev)
1486 scsi_target(sdev)->starget_sdev_user = sdev;
1491 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1492 * take the lock again.
1494 spin_unlock_irq(shost->host_lock);
1497 * Finally, initialize any error handling parameters, and set up
1498 * the timers for timeouts.
1500 scsi_init_cmd_errh(cmd);
1503 * Dispatch the command to the low-level driver.
1505 rtn = scsi_dispatch_cmd(cmd);
1506 spin_lock_irq(q->queue_lock);
1508 /* we're refusing the command; because of
1509 * the way locks get dropped, we need to
1510 * check here if plugging is required */
1511 if(sdev->device_busy == 0)
1521 spin_unlock_irq(shost->host_lock);
1524 * lock q, handle tag, requeue req, and decrement device_busy. We
1525 * must return with queue_lock held.
1527 * Decrementing device_busy without checking it is OK, as all such
1528 * cases (host limits or settings) should run the queue at some
1531 spin_lock_irq(q->queue_lock);
1532 blk_requeue_request(q, req);
1533 sdev->device_busy--;
1534 if(sdev->device_busy == 0)
1537 /* must be careful here...if we trigger the ->remove() function
1538 * we cannot be holding the q lock */
1539 spin_unlock_irq(q->queue_lock);
1540 put_device(&sdev->sdev_gendev);
1541 spin_lock_irq(q->queue_lock);
1544 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1546 struct device *host_dev;
1547 u64 bounce_limit = 0xffffffff;
1549 if (shost->unchecked_isa_dma)
1550 return BLK_BOUNCE_ISA;
1552 * Platforms with virtual-DMA translation
1553 * hardware have no practical limit.
1555 if (!PCI_DMA_BUS_IS_PHYS)
1556 return BLK_BOUNCE_ANY;
1558 host_dev = scsi_get_device(shost);
1559 if (host_dev && host_dev->dma_mask)
1560 bounce_limit = *host_dev->dma_mask;
1562 return bounce_limit;
1564 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1566 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1567 request_fn_proc *request_fn)
1569 struct request_queue *q;
1570 struct device *dev = shost->shost_gendev.parent;
1572 q = blk_init_queue(request_fn, NULL);
1577 * this limit is imposed by hardware restrictions
1579 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1580 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1582 blk_queue_max_sectors(q, shost->max_sectors);
1583 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1584 blk_queue_segment_boundary(q, shost->dma_boundary);
1585 dma_set_seg_boundary(dev, shost->dma_boundary);
1587 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1589 if (!shost->use_clustering)
1590 clear_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
1593 * set a reasonable default alignment on word boundaries: the
1594 * host and device may alter it using
1595 * blk_queue_update_dma_alignment() later.
1597 blk_queue_dma_alignment(q, 0x03);
1601 EXPORT_SYMBOL(__scsi_alloc_queue);
1603 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1605 struct request_queue *q;
1607 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1611 blk_queue_prep_rq(q, scsi_prep_fn);
1612 blk_queue_softirq_done(q, scsi_softirq_done);
1616 void scsi_free_queue(struct request_queue *q)
1618 blk_cleanup_queue(q);
1622 * Function: scsi_block_requests()
1624 * Purpose: Utility function used by low-level drivers to prevent further
1625 * commands from being queued to the device.
1627 * Arguments: shost - Host in question
1631 * Lock status: No locks are assumed held.
1633 * Notes: There is no timer nor any other means by which the requests
1634 * get unblocked other than the low-level driver calling
1635 * scsi_unblock_requests().
1637 void scsi_block_requests(struct Scsi_Host *shost)
1639 shost->host_self_blocked = 1;
1641 EXPORT_SYMBOL(scsi_block_requests);
1644 * Function: scsi_unblock_requests()
1646 * Purpose: Utility function used by low-level drivers to allow further
1647 * commands from being queued to the device.
1649 * Arguments: shost - Host in question
1653 * Lock status: No locks are assumed held.
1655 * Notes: There is no timer nor any other means by which the requests
1656 * get unblocked other than the low-level driver calling
1657 * scsi_unblock_requests().
1659 * This is done as an API function so that changes to the
1660 * internals of the scsi mid-layer won't require wholesale
1661 * changes to drivers that use this feature.
1663 void scsi_unblock_requests(struct Scsi_Host *shost)
1665 shost->host_self_blocked = 0;
1666 scsi_run_host_queues(shost);
1668 EXPORT_SYMBOL(scsi_unblock_requests);
1670 int __init scsi_init_queue(void)
1674 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1675 sizeof(struct scsi_io_context),
1677 if (!scsi_io_context_cache) {
1678 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1682 scsi_bidi_sdb_cache = kmem_cache_create("scsi_bidi_sdb",
1683 sizeof(struct scsi_data_buffer),
1685 if (!scsi_bidi_sdb_cache) {
1686 printk(KERN_ERR "SCSI: can't init scsi bidi sdb cache\n");
1687 goto cleanup_io_context;
1690 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1691 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1692 int size = sgp->size * sizeof(struct scatterlist);
1694 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1695 SLAB_HWCACHE_ALIGN, NULL);
1697 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1699 goto cleanup_bidi_sdb;
1702 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1705 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1707 goto cleanup_bidi_sdb;
1714 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1715 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1717 mempool_destroy(sgp->pool);
1719 kmem_cache_destroy(sgp->slab);
1721 kmem_cache_destroy(scsi_bidi_sdb_cache);
1723 kmem_cache_destroy(scsi_io_context_cache);
1728 void scsi_exit_queue(void)
1732 kmem_cache_destroy(scsi_io_context_cache);
1733 kmem_cache_destroy(scsi_bidi_sdb_cache);
1735 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1736 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1737 mempool_destroy(sgp->pool);
1738 kmem_cache_destroy(sgp->slab);
1743 * scsi_mode_select - issue a mode select
1744 * @sdev: SCSI device to be queried
1745 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1746 * @sp: Save page bit (0 == don't save, 1 == save)
1747 * @modepage: mode page being requested
1748 * @buffer: request buffer (may not be smaller than eight bytes)
1749 * @len: length of request buffer.
1750 * @timeout: command timeout
1751 * @retries: number of retries before failing
1752 * @data: returns a structure abstracting the mode header data
1753 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1754 * must be SCSI_SENSE_BUFFERSIZE big.
1756 * Returns zero if successful; negative error number or scsi
1761 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1762 unsigned char *buffer, int len, int timeout, int retries,
1763 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1765 unsigned char cmd[10];
1766 unsigned char *real_buffer;
1769 memset(cmd, 0, sizeof(cmd));
1770 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1772 if (sdev->use_10_for_ms) {
1775 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1778 memcpy(real_buffer + 8, buffer, len);
1782 real_buffer[2] = data->medium_type;
1783 real_buffer[3] = data->device_specific;
1784 real_buffer[4] = data->longlba ? 0x01 : 0;
1786 real_buffer[6] = data->block_descriptor_length >> 8;
1787 real_buffer[7] = data->block_descriptor_length;
1789 cmd[0] = MODE_SELECT_10;
1793 if (len > 255 || data->block_descriptor_length > 255 ||
1797 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1800 memcpy(real_buffer + 4, buffer, len);
1803 real_buffer[1] = data->medium_type;
1804 real_buffer[2] = data->device_specific;
1805 real_buffer[3] = data->block_descriptor_length;
1808 cmd[0] = MODE_SELECT;
1812 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1813 sshdr, timeout, retries);
1817 EXPORT_SYMBOL_GPL(scsi_mode_select);
1820 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1821 * @sdev: SCSI device to be queried
1822 * @dbd: set if mode sense will allow block descriptors to be returned
1823 * @modepage: mode page being requested
1824 * @buffer: request buffer (may not be smaller than eight bytes)
1825 * @len: length of request buffer.
1826 * @timeout: command timeout
1827 * @retries: number of retries before failing
1828 * @data: returns a structure abstracting the mode header data
1829 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1830 * must be SCSI_SENSE_BUFFERSIZE big.
1832 * Returns zero if unsuccessful, or the header offset (either 4
1833 * or 8 depending on whether a six or ten byte command was
1834 * issued) if successful.
1837 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1838 unsigned char *buffer, int len, int timeout, int retries,
1839 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1841 unsigned char cmd[12];
1845 struct scsi_sense_hdr my_sshdr;
1847 memset(data, 0, sizeof(*data));
1848 memset(&cmd[0], 0, 12);
1849 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1852 /* caller might not be interested in sense, but we need it */
1857 use_10_for_ms = sdev->use_10_for_ms;
1859 if (use_10_for_ms) {
1863 cmd[0] = MODE_SENSE_10;
1870 cmd[0] = MODE_SENSE;
1875 memset(buffer, 0, len);
1877 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1878 sshdr, timeout, retries);
1880 /* This code looks awful: what it's doing is making sure an
1881 * ILLEGAL REQUEST sense return identifies the actual command
1882 * byte as the problem. MODE_SENSE commands can return
1883 * ILLEGAL REQUEST if the code page isn't supported */
1885 if (use_10_for_ms && !scsi_status_is_good(result) &&
1886 (driver_byte(result) & DRIVER_SENSE)) {
1887 if (scsi_sense_valid(sshdr)) {
1888 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1889 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1891 * Invalid command operation code
1893 sdev->use_10_for_ms = 0;
1899 if(scsi_status_is_good(result)) {
1900 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1901 (modepage == 6 || modepage == 8))) {
1902 /* Initio breakage? */
1905 data->medium_type = 0;
1906 data->device_specific = 0;
1908 data->block_descriptor_length = 0;
1909 } else if(use_10_for_ms) {
1910 data->length = buffer[0]*256 + buffer[1] + 2;
1911 data->medium_type = buffer[2];
1912 data->device_specific = buffer[3];
1913 data->longlba = buffer[4] & 0x01;
1914 data->block_descriptor_length = buffer[6]*256
1917 data->length = buffer[0] + 1;
1918 data->medium_type = buffer[1];
1919 data->device_specific = buffer[2];
1920 data->block_descriptor_length = buffer[3];
1922 data->header_length = header_length;
1927 EXPORT_SYMBOL(scsi_mode_sense);
1930 * scsi_test_unit_ready - test if unit is ready
1931 * @sdev: scsi device to change the state of.
1932 * @timeout: command timeout
1933 * @retries: number of retries before failing
1934 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1935 * returning sense. Make sure that this is cleared before passing
1938 * Returns zero if unsuccessful or an error if TUR failed. For
1939 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1940 * translated to success, with the ->changed flag updated.
1943 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1944 struct scsi_sense_hdr *sshdr_external)
1947 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1949 struct scsi_sense_hdr *sshdr;
1952 if (!sshdr_external)
1953 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1955 sshdr = sshdr_external;
1957 /* try to eat the UNIT_ATTENTION if there are enough retries */
1959 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
1961 } while ((driver_byte(result) & DRIVER_SENSE) &&
1962 sshdr && sshdr->sense_key == UNIT_ATTENTION &&
1966 /* could not allocate sense buffer, so can't process it */
1969 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
1971 if ((scsi_sense_valid(sshdr)) &&
1972 ((sshdr->sense_key == UNIT_ATTENTION) ||
1973 (sshdr->sense_key == NOT_READY))) {
1978 if (!sshdr_external)
1982 EXPORT_SYMBOL(scsi_test_unit_ready);
1985 * scsi_device_set_state - Take the given device through the device state model.
1986 * @sdev: scsi device to change the state of.
1987 * @state: state to change to.
1989 * Returns zero if unsuccessful or an error if the requested
1990 * transition is illegal.
1993 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
1995 enum scsi_device_state oldstate = sdev->sdev_state;
1997 if (state == oldstate)
2002 /* There are no legal states that come back to
2003 * created. This is the manually initialised start
2077 sdev->sdev_state = state;
2081 SCSI_LOG_ERROR_RECOVERY(1,
2082 sdev_printk(KERN_ERR, sdev,
2083 "Illegal state transition %s->%s\n",
2084 scsi_device_state_name(oldstate),
2085 scsi_device_state_name(state))
2089 EXPORT_SYMBOL(scsi_device_set_state);
2092 * sdev_evt_emit - emit a single SCSI device uevent
2093 * @sdev: associated SCSI device
2094 * @evt: event to emit
2096 * Send a single uevent (scsi_event) to the associated scsi_device.
2098 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2103 switch (evt->evt_type) {
2104 case SDEV_EVT_MEDIA_CHANGE:
2105 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2115 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2119 * sdev_evt_thread - send a uevent for each scsi event
2120 * @work: work struct for scsi_device
2122 * Dispatch queued events to their associated scsi_device kobjects
2125 void scsi_evt_thread(struct work_struct *work)
2127 struct scsi_device *sdev;
2128 LIST_HEAD(event_list);
2130 sdev = container_of(work, struct scsi_device, event_work);
2133 struct scsi_event *evt;
2134 struct list_head *this, *tmp;
2135 unsigned long flags;
2137 spin_lock_irqsave(&sdev->list_lock, flags);
2138 list_splice_init(&sdev->event_list, &event_list);
2139 spin_unlock_irqrestore(&sdev->list_lock, flags);
2141 if (list_empty(&event_list))
2144 list_for_each_safe(this, tmp, &event_list) {
2145 evt = list_entry(this, struct scsi_event, node);
2146 list_del(&evt->node);
2147 scsi_evt_emit(sdev, evt);
2154 * sdev_evt_send - send asserted event to uevent thread
2155 * @sdev: scsi_device event occurred on
2156 * @evt: event to send
2158 * Assert scsi device event asynchronously.
2160 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2162 unsigned long flags;
2165 /* FIXME: currently this check eliminates all media change events
2166 * for polled devices. Need to update to discriminate between AN
2167 * and polled events */
2168 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2174 spin_lock_irqsave(&sdev->list_lock, flags);
2175 list_add_tail(&evt->node, &sdev->event_list);
2176 schedule_work(&sdev->event_work);
2177 spin_unlock_irqrestore(&sdev->list_lock, flags);
2179 EXPORT_SYMBOL_GPL(sdev_evt_send);
2182 * sdev_evt_alloc - allocate a new scsi event
2183 * @evt_type: type of event to allocate
2184 * @gfpflags: GFP flags for allocation
2186 * Allocates and returns a new scsi_event.
2188 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2191 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2195 evt->evt_type = evt_type;
2196 INIT_LIST_HEAD(&evt->node);
2198 /* evt_type-specific initialization, if any */
2200 case SDEV_EVT_MEDIA_CHANGE:
2208 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2211 * sdev_evt_send_simple - send asserted event to uevent thread
2212 * @sdev: scsi_device event occurred on
2213 * @evt_type: type of event to send
2214 * @gfpflags: GFP flags for allocation
2216 * Assert scsi device event asynchronously, given an event type.
2218 void sdev_evt_send_simple(struct scsi_device *sdev,
2219 enum scsi_device_event evt_type, gfp_t gfpflags)
2221 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2223 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2228 sdev_evt_send(sdev, evt);
2230 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2233 * scsi_device_quiesce - Block user issued commands.
2234 * @sdev: scsi device to quiesce.
2236 * This works by trying to transition to the SDEV_QUIESCE state
2237 * (which must be a legal transition). When the device is in this
2238 * state, only special requests will be accepted, all others will
2239 * be deferred. Since special requests may also be requeued requests,
2240 * a successful return doesn't guarantee the device will be
2241 * totally quiescent.
2243 * Must be called with user context, may sleep.
2245 * Returns zero if unsuccessful or an error if not.
2248 scsi_device_quiesce(struct scsi_device *sdev)
2250 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2254 scsi_run_queue(sdev->request_queue);
2255 while (sdev->device_busy) {
2256 msleep_interruptible(200);
2257 scsi_run_queue(sdev->request_queue);
2261 EXPORT_SYMBOL(scsi_device_quiesce);
2264 * scsi_device_resume - Restart user issued commands to a quiesced device.
2265 * @sdev: scsi device to resume.
2267 * Moves the device from quiesced back to running and restarts the
2270 * Must be called with user context, may sleep.
2273 scsi_device_resume(struct scsi_device *sdev)
2275 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2277 scsi_run_queue(sdev->request_queue);
2279 EXPORT_SYMBOL(scsi_device_resume);
2282 device_quiesce_fn(struct scsi_device *sdev, void *data)
2284 scsi_device_quiesce(sdev);
2288 scsi_target_quiesce(struct scsi_target *starget)
2290 starget_for_each_device(starget, NULL, device_quiesce_fn);
2292 EXPORT_SYMBOL(scsi_target_quiesce);
2295 device_resume_fn(struct scsi_device *sdev, void *data)
2297 scsi_device_resume(sdev);
2301 scsi_target_resume(struct scsi_target *starget)
2303 starget_for_each_device(starget, NULL, device_resume_fn);
2305 EXPORT_SYMBOL(scsi_target_resume);
2308 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2309 * @sdev: device to block
2311 * Block request made by scsi lld's to temporarily stop all
2312 * scsi commands on the specified device. Called from interrupt
2313 * or normal process context.
2315 * Returns zero if successful or error if not
2318 * This routine transitions the device to the SDEV_BLOCK state
2319 * (which must be a legal transition). When the device is in this
2320 * state, all commands are deferred until the scsi lld reenables
2321 * the device with scsi_device_unblock or device_block_tmo fires.
2322 * This routine assumes the host_lock is held on entry.
2325 scsi_internal_device_block(struct scsi_device *sdev)
2327 struct request_queue *q = sdev->request_queue;
2328 unsigned long flags;
2331 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2336 * The device has transitioned to SDEV_BLOCK. Stop the
2337 * block layer from calling the midlayer with this device's
2340 spin_lock_irqsave(q->queue_lock, flags);
2342 spin_unlock_irqrestore(q->queue_lock, flags);
2346 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2349 * scsi_internal_device_unblock - resume a device after a block request
2350 * @sdev: device to resume
2352 * Called by scsi lld's or the midlayer to restart the device queue
2353 * for the previously suspended scsi device. Called from interrupt or
2354 * normal process context.
2356 * Returns zero if successful or error if not.
2359 * This routine transitions the device to the SDEV_RUNNING state
2360 * (which must be a legal transition) allowing the midlayer to
2361 * goose the queue for this device. This routine assumes the
2362 * host_lock is held upon entry.
2365 scsi_internal_device_unblock(struct scsi_device *sdev)
2367 struct request_queue *q = sdev->request_queue;
2369 unsigned long flags;
2372 * Try to transition the scsi device to SDEV_RUNNING
2373 * and goose the device queue if successful.
2375 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2379 spin_lock_irqsave(q->queue_lock, flags);
2381 spin_unlock_irqrestore(q->queue_lock, flags);
2385 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2388 device_block(struct scsi_device *sdev, void *data)
2390 scsi_internal_device_block(sdev);
2394 target_block(struct device *dev, void *data)
2396 if (scsi_is_target_device(dev))
2397 starget_for_each_device(to_scsi_target(dev), NULL,
2403 scsi_target_block(struct device *dev)
2405 if (scsi_is_target_device(dev))
2406 starget_for_each_device(to_scsi_target(dev), NULL,
2409 device_for_each_child(dev, NULL, target_block);
2411 EXPORT_SYMBOL_GPL(scsi_target_block);
2414 device_unblock(struct scsi_device *sdev, void *data)
2416 scsi_internal_device_unblock(sdev);
2420 target_unblock(struct device *dev, void *data)
2422 if (scsi_is_target_device(dev))
2423 starget_for_each_device(to_scsi_target(dev), NULL,
2429 scsi_target_unblock(struct device *dev)
2431 if (scsi_is_target_device(dev))
2432 starget_for_each_device(to_scsi_target(dev), NULL,
2435 device_for_each_child(dev, NULL, target_unblock);
2437 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2440 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2441 * @sgl: scatter-gather list
2442 * @sg_count: number of segments in sg
2443 * @offset: offset in bytes into sg, on return offset into the mapped area
2444 * @len: bytes to map, on return number of bytes mapped
2446 * Returns virtual address of the start of the mapped page
2448 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2449 size_t *offset, size_t *len)
2452 size_t sg_len = 0, len_complete = 0;
2453 struct scatterlist *sg;
2456 WARN_ON(!irqs_disabled());
2458 for_each_sg(sgl, sg, sg_count, i) {
2459 len_complete = sg_len; /* Complete sg-entries */
2460 sg_len += sg->length;
2461 if (sg_len > *offset)
2465 if (unlikely(i == sg_count)) {
2466 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2468 __FUNCTION__, sg_len, *offset, sg_count);
2473 /* Offset starting from the beginning of first page in this sg-entry */
2474 *offset = *offset - len_complete + sg->offset;
2476 /* Assumption: contiguous pages can be accessed as "page + i" */
2477 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2478 *offset &= ~PAGE_MASK;
2480 /* Bytes in this sg-entry from *offset to the end of the page */
2481 sg_len = PAGE_SIZE - *offset;
2485 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2487 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2490 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2491 * @virt: virtual address to be unmapped
2493 void scsi_kunmap_atomic_sg(void *virt)
2495 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2497 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);