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 struct kmem_cache *scsi_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);
211 * Some devices (USB mass-storage in particular) may transfer
212 * garbage data together with a residue indicating that the data
213 * is invalid. Prevent the garbage from being misinterpreted
214 * and prevent security leaks by zeroing out the excess data.
216 if (unlikely(req->data_len > 0 && req->data_len <= bufflen))
217 memset(buffer + (bufflen - req->data_len), 0, req->data_len);
221 blk_put_request(req);
225 EXPORT_SYMBOL(scsi_execute);
228 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
229 int data_direction, void *buffer, unsigned bufflen,
230 struct scsi_sense_hdr *sshdr, int timeout, int retries)
236 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
238 return DRIVER_ERROR << 24;
240 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
241 sense, timeout, retries, 0);
243 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
248 EXPORT_SYMBOL(scsi_execute_req);
250 struct scsi_io_context {
252 void (*done)(void *data, char *sense, int result, int resid);
253 char sense[SCSI_SENSE_BUFFERSIZE];
256 static struct kmem_cache *scsi_io_context_cache;
258 static void scsi_end_async(struct request *req, int uptodate)
260 struct scsi_io_context *sioc = req->end_io_data;
263 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
265 kmem_cache_free(scsi_io_context_cache, sioc);
266 __blk_put_request(req->q, req);
269 static int scsi_merge_bio(struct request *rq, struct bio *bio)
271 struct request_queue *q = rq->q;
273 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
274 if (rq_data_dir(rq) == WRITE)
275 bio->bi_rw |= (1 << BIO_RW);
276 blk_queue_bounce(q, &bio);
278 return blk_rq_append_bio(q, rq, bio);
281 static void scsi_bi_endio(struct bio *bio, int error)
287 * scsi_req_map_sg - map a scatterlist into a request
288 * @rq: request to fill
290 * @nsegs: number of elements
291 * @bufflen: len of buffer
292 * @gfp: memory allocation flags
294 * scsi_req_map_sg maps a scatterlist into a request so that the
295 * request can be sent to the block layer. We do not trust the scatterlist
296 * sent to use, as some ULDs use that struct to only organize the pages.
298 static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
299 int nsegs, unsigned bufflen, gfp_t gfp)
301 struct request_queue *q = rq->q;
302 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
303 unsigned int data_len = bufflen, len, bytes, off;
304 struct scatterlist *sg;
306 struct bio *bio = NULL;
307 int i, err, nr_vecs = 0;
309 for_each_sg(sgl, sg, nsegs, i) {
314 while (len > 0 && data_len > 0) {
316 * sg sends a scatterlist that is larger than
317 * the data_len it wants transferred for certain
320 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
321 bytes = min(bytes, data_len);
324 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
327 bio = bio_alloc(gfp, nr_vecs);
332 bio->bi_end_io = scsi_bi_endio;
335 if (bio_add_pc_page(q, bio, page, bytes, off) !=
342 if (bio->bi_vcnt >= nr_vecs) {
343 err = scsi_merge_bio(rq, bio);
358 rq->buffer = rq->data = NULL;
359 rq->data_len = bufflen;
363 while ((bio = rq->bio) != NULL) {
364 rq->bio = bio->bi_next;
366 * call endio instead of bio_put incase it was bounced
375 * scsi_execute_async - insert request
378 * @cmd_len: length of scsi cdb
379 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
380 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
381 * @bufflen: len of buffer
382 * @use_sg: if buffer is a scatterlist this is the number of elements
383 * @timeout: request timeout in seconds
384 * @retries: number of times to retry request
385 * @privdata: data passed to done()
386 * @done: callback function when done
387 * @gfp: memory allocation flags
389 int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
390 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
391 int use_sg, int timeout, int retries, void *privdata,
392 void (*done)(void *, char *, int, int), gfp_t gfp)
395 struct scsi_io_context *sioc;
397 int write = (data_direction == DMA_TO_DEVICE);
399 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp);
401 return DRIVER_ERROR << 24;
403 req = blk_get_request(sdev->request_queue, write, gfp);
406 req->cmd_type = REQ_TYPE_BLOCK_PC;
407 req->cmd_flags |= REQ_QUIET;
410 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
412 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
417 req->cmd_len = cmd_len;
418 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
419 memcpy(req->cmd, cmd, req->cmd_len);
420 req->sense = sioc->sense;
422 req->timeout = timeout;
423 req->retries = retries;
424 req->end_io_data = sioc;
426 sioc->data = privdata;
429 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
433 blk_put_request(req);
435 kmem_cache_free(scsi_io_context_cache, sioc);
436 return DRIVER_ERROR << 24;
438 EXPORT_SYMBOL_GPL(scsi_execute_async);
441 * Function: scsi_init_cmd_errh()
443 * Purpose: Initialize cmd fields related to error handling.
445 * Arguments: cmd - command that is ready to be queued.
447 * Notes: This function has the job of initializing a number of
448 * fields related to error handling. Typically this will
449 * be called once for each command, as required.
451 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
453 cmd->serial_number = 0;
454 scsi_set_resid(cmd, 0);
455 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
456 if (cmd->cmd_len == 0)
457 cmd->cmd_len = scsi_command_size(cmd->cmnd);
460 void scsi_device_unbusy(struct scsi_device *sdev)
462 struct Scsi_Host *shost = sdev->host;
465 spin_lock_irqsave(shost->host_lock, flags);
467 if (unlikely(scsi_host_in_recovery(shost) &&
468 (shost->host_failed || shost->host_eh_scheduled)))
469 scsi_eh_wakeup(shost);
470 spin_unlock(shost->host_lock);
471 spin_lock(sdev->request_queue->queue_lock);
473 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
477 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
478 * and call blk_run_queue for all the scsi_devices on the target -
479 * including current_sdev first.
481 * Called with *no* scsi locks held.
483 static void scsi_single_lun_run(struct scsi_device *current_sdev)
485 struct Scsi_Host *shost = current_sdev->host;
486 struct scsi_device *sdev, *tmp;
487 struct scsi_target *starget = scsi_target(current_sdev);
490 spin_lock_irqsave(shost->host_lock, flags);
491 starget->starget_sdev_user = NULL;
492 spin_unlock_irqrestore(shost->host_lock, flags);
495 * Call blk_run_queue for all LUNs on the target, starting with
496 * current_sdev. We race with others (to set starget_sdev_user),
497 * but in most cases, we will be first. Ideally, each LU on the
498 * target would get some limited time or requests on the target.
500 blk_run_queue(current_sdev->request_queue);
502 spin_lock_irqsave(shost->host_lock, flags);
503 if (starget->starget_sdev_user)
505 list_for_each_entry_safe(sdev, tmp, &starget->devices,
506 same_target_siblings) {
507 if (sdev == current_sdev)
509 if (scsi_device_get(sdev))
512 spin_unlock_irqrestore(shost->host_lock, flags);
513 blk_run_queue(sdev->request_queue);
514 spin_lock_irqsave(shost->host_lock, flags);
516 scsi_device_put(sdev);
519 spin_unlock_irqrestore(shost->host_lock, flags);
523 * Function: scsi_run_queue()
525 * Purpose: Select a proper request queue to serve next
527 * Arguments: q - last request's queue
531 * Notes: The previous command was completely finished, start
532 * a new one if possible.
534 static void scsi_run_queue(struct request_queue *q)
536 struct scsi_device *sdev = q->queuedata;
537 struct Scsi_Host *shost = sdev->host;
540 if (scsi_target(sdev)->single_lun)
541 scsi_single_lun_run(sdev);
543 spin_lock_irqsave(shost->host_lock, flags);
544 while (!list_empty(&shost->starved_list) &&
545 !shost->host_blocked && !shost->host_self_blocked &&
546 !((shost->can_queue > 0) &&
547 (shost->host_busy >= shost->can_queue))) {
552 * As long as shost is accepting commands and we have
553 * starved queues, call blk_run_queue. scsi_request_fn
554 * drops the queue_lock and can add us back to the
557 * host_lock protects the starved_list and starved_entry.
558 * scsi_request_fn must get the host_lock before checking
559 * or modifying starved_list or starved_entry.
561 sdev = list_entry(shost->starved_list.next,
562 struct scsi_device, starved_entry);
563 list_del_init(&sdev->starved_entry);
564 spin_unlock(shost->host_lock);
566 spin_lock(sdev->request_queue->queue_lock);
567 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
568 !test_bit(QUEUE_FLAG_REENTER,
569 &sdev->request_queue->queue_flags);
571 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
572 __blk_run_queue(sdev->request_queue);
574 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
575 spin_unlock(sdev->request_queue->queue_lock);
577 spin_lock(shost->host_lock);
578 if (unlikely(!list_empty(&sdev->starved_entry)))
580 * sdev lost a race, and was put back on the
581 * starved list. This is unlikely but without this
582 * in theory we could loop forever.
586 spin_unlock_irqrestore(shost->host_lock, flags);
592 * Function: scsi_requeue_command()
594 * Purpose: Handle post-processing of completed commands.
596 * Arguments: q - queue to operate on
597 * cmd - command that may need to be requeued.
601 * Notes: After command completion, there may be blocks left
602 * over which weren't finished by the previous command
603 * this can be for a number of reasons - the main one is
604 * I/O errors in the middle of the request, in which case
605 * we need to request the blocks that come after the bad
607 * Notes: Upon return, cmd is a stale pointer.
609 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
611 struct request *req = cmd->request;
614 scsi_unprep_request(req);
615 spin_lock_irqsave(q->queue_lock, flags);
616 blk_requeue_request(q, req);
617 spin_unlock_irqrestore(q->queue_lock, flags);
622 void scsi_next_command(struct scsi_cmnd *cmd)
624 struct scsi_device *sdev = cmd->device;
625 struct request_queue *q = sdev->request_queue;
627 /* need to hold a reference on the device before we let go of the cmd */
628 get_device(&sdev->sdev_gendev);
630 scsi_put_command(cmd);
633 /* ok to remove device now */
634 put_device(&sdev->sdev_gendev);
637 void scsi_run_host_queues(struct Scsi_Host *shost)
639 struct scsi_device *sdev;
641 shost_for_each_device(sdev, shost)
642 scsi_run_queue(sdev->request_queue);
646 * Function: scsi_end_request()
648 * Purpose: Post-processing of completed commands (usually invoked at end
649 * of upper level post-processing and scsi_io_completion).
651 * Arguments: cmd - command that is complete.
652 * error - 0 if I/O indicates success, < 0 for I/O error.
653 * bytes - number of bytes of completed I/O
654 * requeue - indicates whether we should requeue leftovers.
656 * Lock status: Assumed that lock is not held upon entry.
658 * Returns: cmd if requeue required, NULL otherwise.
660 * Notes: This is called for block device requests in order to
661 * mark some number of sectors as complete.
663 * We are guaranteeing that the request queue will be goosed
664 * at some point during this call.
665 * Notes: If cmd was requeued, upon return it will be a stale pointer.
667 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
668 int bytes, int requeue)
670 struct request_queue *q = cmd->device->request_queue;
671 struct request *req = cmd->request;
674 * If there are blocks left over at the end, set up the command
675 * to queue the remainder of them.
677 if (blk_end_request(req, error, bytes)) {
678 int leftover = (req->hard_nr_sectors << 9);
680 if (blk_pc_request(req))
681 leftover = req->data_len;
683 /* kill remainder if no retrys */
684 if (error && blk_noretry_request(req))
685 blk_end_request(req, error, leftover);
689 * Bleah. Leftovers again. Stick the
690 * leftovers in the front of the
691 * queue, and goose the queue again.
693 scsi_requeue_command(q, cmd);
701 * This will goose the queue request function at the end, so we don't
702 * need to worry about launching another command.
704 scsi_next_command(cmd);
708 static inline unsigned int scsi_sgtable_index(unsigned short nents)
712 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
717 index = get_count_order(nents) - 3;
722 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
724 struct scsi_host_sg_pool *sgp;
726 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
727 mempool_free(sgl, sgp->pool);
730 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
732 struct scsi_host_sg_pool *sgp;
734 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
735 return mempool_alloc(sgp->pool, gfp_mask);
738 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
745 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
746 gfp_mask, scsi_sg_alloc);
748 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
754 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
756 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
760 * Function: scsi_release_buffers()
762 * Purpose: Completion processing for block device I/O requests.
764 * Arguments: cmd - command that we are bailing.
766 * Lock status: Assumed that no lock is held upon entry.
770 * Notes: In the event that an upper level driver rejects a
771 * command, we must release resources allocated during
772 * the __init_io() function. Primarily this would involve
773 * the scatter-gather table, and potentially any bounce
776 void scsi_release_buffers(struct scsi_cmnd *cmd)
778 if (cmd->sdb.table.nents)
779 scsi_free_sgtable(&cmd->sdb);
781 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
783 if (scsi_bidi_cmnd(cmd)) {
784 struct scsi_data_buffer *bidi_sdb =
785 cmd->request->next_rq->special;
786 scsi_free_sgtable(bidi_sdb);
787 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
788 cmd->request->next_rq->special = NULL;
791 if (scsi_prot_sg_count(cmd))
792 scsi_free_sgtable(cmd->prot_sdb);
794 EXPORT_SYMBOL(scsi_release_buffers);
797 * Bidi commands Must be complete as a whole, both sides at once.
798 * If part of the bytes were written and lld returned
799 * scsi_in()->resid and/or scsi_out()->resid this information will be left
800 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
801 * decide what to do with this information.
803 static void scsi_end_bidi_request(struct scsi_cmnd *cmd)
805 struct request *req = cmd->request;
806 unsigned int dlen = req->data_len;
807 unsigned int next_dlen = req->next_rq->data_len;
809 req->data_len = scsi_out(cmd)->resid;
810 req->next_rq->data_len = scsi_in(cmd)->resid;
812 /* The req and req->next_rq have not been completed */
813 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen));
815 scsi_release_buffers(cmd);
818 * This will goose the queue request function at the end, so we don't
819 * need to worry about launching another command.
821 scsi_next_command(cmd);
825 * Function: scsi_io_completion()
827 * Purpose: Completion processing for block device I/O requests.
829 * Arguments: cmd - command that is finished.
831 * Lock status: Assumed that no lock is held upon entry.
835 * Notes: This function is matched in terms of capabilities to
836 * the function that created the scatter-gather list.
837 * In other words, if there are no bounce buffers
838 * (the normal case for most drivers), we don't need
839 * the logic to deal with cleaning up afterwards.
841 * We must do one of several things here:
843 * a) Call scsi_end_request. This will finish off the
844 * specified number of sectors. If we are done, the
845 * command block will be released, and the queue
846 * function will be goosed. If we are not done, then
847 * scsi_end_request will directly goose the queue.
849 * b) We can just use scsi_requeue_command() here. This would
850 * be used if we just wanted to retry, for example.
852 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
854 int result = cmd->result;
856 struct request_queue *q = cmd->device->request_queue;
857 struct request *req = cmd->request;
859 struct scsi_sense_hdr sshdr;
861 int sense_deferred = 0;
864 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
866 sense_deferred = scsi_sense_is_deferred(&sshdr);
869 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
870 req->errors = result;
872 if (sense_valid && req->sense) {
874 * SG_IO wants current and deferred errors
876 int len = 8 + cmd->sense_buffer[7];
878 if (len > SCSI_SENSE_BUFFERSIZE)
879 len = SCSI_SENSE_BUFFERSIZE;
880 memcpy(req->sense, cmd->sense_buffer, len);
881 req->sense_len = len;
886 if (scsi_bidi_cmnd(cmd)) {
887 /* will also release_buffers */
888 scsi_end_bidi_request(cmd);
891 req->data_len = scsi_get_resid(cmd);
894 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
895 scsi_release_buffers(cmd);
898 * Next deal with any sectors which we were able to correctly
901 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
903 req->nr_sectors, good_bytes));
905 /* A number of bytes were successfully read. If there
906 * are leftovers and there is some kind of error
907 * (result != 0), retry the rest.
909 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
911 this_count = blk_rq_bytes(req);
913 /* good_bytes = 0, or (inclusive) there were leftovers and
914 * result = 0, so scsi_end_request couldn't retry.
916 if (sense_valid && !sense_deferred) {
917 switch (sshdr.sense_key) {
919 if (cmd->device->removable) {
920 /* Detected disc change. Set a bit
921 * and quietly refuse further access.
923 cmd->device->changed = 1;
924 scsi_end_request(cmd, -EIO, this_count, 1);
927 /* Must have been a power glitch, or a
928 * bus reset. Could not have been a
929 * media change, so we just retry the
930 * request and see what happens.
932 scsi_requeue_command(q, cmd);
936 case ILLEGAL_REQUEST:
937 /* If we had an ILLEGAL REQUEST returned, then
938 * we may have performed an unsupported
939 * command. The only thing this should be
940 * would be a ten byte read where only a six
941 * byte read was supported. Also, on a system
942 * where READ CAPACITY failed, we may have
943 * read past the end of the disk.
945 if ((cmd->device->use_10_for_rw &&
946 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
947 (cmd->cmnd[0] == READ_10 ||
948 cmd->cmnd[0] == WRITE_10)) {
949 cmd->device->use_10_for_rw = 0;
950 /* This will cause a retry with a
953 scsi_requeue_command(q, cmd);
954 } else if (sshdr.asc == 0x10) /* DIX */
955 scsi_end_request(cmd, -EIO, this_count, 0);
957 scsi_end_request(cmd, -EIO, this_count, 1);
959 case ABORTED_COMMAND:
960 if (sshdr.asc == 0x10) { /* DIF */
961 scsi_end_request(cmd, -EIO, this_count, 0);
966 /* If the device is in the process of becoming
967 * ready, or has a temporary blockage, retry.
969 if (sshdr.asc == 0x04) {
970 switch (sshdr.ascq) {
971 case 0x01: /* becoming ready */
972 case 0x04: /* format in progress */
973 case 0x05: /* rebuild in progress */
974 case 0x06: /* recalculation in progress */
975 case 0x07: /* operation in progress */
976 case 0x08: /* Long write in progress */
977 case 0x09: /* self test in progress */
978 scsi_requeue_command(q, cmd);
984 if (!(req->cmd_flags & REQ_QUIET))
985 scsi_cmd_print_sense_hdr(cmd,
989 scsi_end_request(cmd, -EIO, this_count, 1);
991 case VOLUME_OVERFLOW:
992 if (!(req->cmd_flags & REQ_QUIET)) {
993 scmd_printk(KERN_INFO, cmd,
994 "Volume overflow, CDB: ");
995 __scsi_print_command(cmd->cmnd);
996 scsi_print_sense("", cmd);
998 /* See SSC3rXX or current. */
999 scsi_end_request(cmd, -EIO, this_count, 1);
1005 if (host_byte(result) == DID_RESET) {
1006 /* Third party bus reset or reset for error recovery
1007 * reasons. Just retry the request and see what
1010 scsi_requeue_command(q, cmd);
1014 if (!(req->cmd_flags & REQ_QUIET)) {
1015 scsi_print_result(cmd);
1016 if (driver_byte(result) & DRIVER_SENSE)
1017 scsi_print_sense("", cmd);
1020 scsi_end_request(cmd, -EIO, this_count, !result);
1023 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1029 * If sg table allocation fails, requeue request later.
1031 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1033 return BLKPREP_DEFER;
1039 * Next, walk the list, and fill in the addresses and sizes of
1042 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1043 BUG_ON(count > sdb->table.nents);
1044 sdb->table.nents = count;
1045 if (blk_pc_request(req))
1046 sdb->length = req->data_len;
1048 sdb->length = req->nr_sectors << 9;
1053 * Function: scsi_init_io()
1055 * Purpose: SCSI I/O initialize function.
1057 * Arguments: cmd - Command descriptor we wish to initialize
1059 * Returns: 0 on success
1060 * BLKPREP_DEFER if the failure is retryable
1061 * BLKPREP_KILL if the failure is fatal
1063 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1065 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
1069 if (blk_bidi_rq(cmd->request)) {
1070 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1071 scsi_sdb_cache, GFP_ATOMIC);
1073 error = BLKPREP_DEFER;
1077 cmd->request->next_rq->special = bidi_sdb;
1078 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1084 if (blk_integrity_rq(cmd->request)) {
1085 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1088 BUG_ON(prot_sdb == NULL);
1089 ivecs = blk_rq_count_integrity_sg(cmd->request);
1091 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1092 error = BLKPREP_DEFER;
1096 count = blk_rq_map_integrity_sg(cmd->request,
1097 prot_sdb->table.sgl);
1098 BUG_ON(unlikely(count > ivecs));
1100 cmd->prot_sdb = prot_sdb;
1101 cmd->prot_sdb->table.nents = count;
1107 scsi_release_buffers(cmd);
1108 if (error == BLKPREP_KILL)
1109 scsi_put_command(cmd);
1110 else /* BLKPREP_DEFER */
1111 scsi_unprep_request(cmd->request);
1115 EXPORT_SYMBOL(scsi_init_io);
1117 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1118 struct request *req)
1120 struct scsi_cmnd *cmd;
1122 if (!req->special) {
1123 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1131 /* pull a tag out of the request if we have one */
1132 cmd->tag = req->tag;
1135 cmd->cmnd = req->cmd;
1140 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1142 struct scsi_cmnd *cmd;
1143 int ret = scsi_prep_state_check(sdev, req);
1145 if (ret != BLKPREP_OK)
1148 cmd = scsi_get_cmd_from_req(sdev, req);
1150 return BLKPREP_DEFER;
1153 * BLOCK_PC requests may transfer data, in which case they must
1154 * a bio attached to them. Or they might contain a SCSI command
1155 * that does not transfer data, in which case they may optionally
1156 * submit a request without an attached bio.
1161 BUG_ON(!req->nr_phys_segments);
1163 ret = scsi_init_io(cmd, GFP_ATOMIC);
1167 BUG_ON(req->data_len);
1170 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1174 cmd->cmd_len = req->cmd_len;
1176 cmd->sc_data_direction = DMA_NONE;
1177 else if (rq_data_dir(req) == WRITE)
1178 cmd->sc_data_direction = DMA_TO_DEVICE;
1180 cmd->sc_data_direction = DMA_FROM_DEVICE;
1182 cmd->transfersize = req->data_len;
1183 cmd->allowed = req->retries;
1184 cmd->timeout_per_command = req->timeout;
1187 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1190 * Setup a REQ_TYPE_FS command. These are simple read/write request
1191 * from filesystems that still need to be translated to SCSI CDBs from
1194 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1196 struct scsi_cmnd *cmd;
1197 int ret = scsi_prep_state_check(sdev, req);
1199 if (ret != BLKPREP_OK)
1202 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1203 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1204 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1205 if (ret != BLKPREP_OK)
1210 * Filesystem requests must transfer data.
1212 BUG_ON(!req->nr_phys_segments);
1214 cmd = scsi_get_cmd_from_req(sdev, req);
1216 return BLKPREP_DEFER;
1218 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1219 return scsi_init_io(cmd, GFP_ATOMIC);
1221 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1223 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1225 int ret = BLKPREP_OK;
1228 * If the device is not in running state we will reject some
1231 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1232 switch (sdev->sdev_state) {
1235 * If the device is offline we refuse to process any
1236 * commands. The device must be brought online
1237 * before trying any recovery commands.
1239 sdev_printk(KERN_ERR, sdev,
1240 "rejecting I/O to offline device\n");
1245 * If the device is fully deleted, we refuse to
1246 * process any commands as well.
1248 sdev_printk(KERN_ERR, sdev,
1249 "rejecting I/O to dead device\n");
1254 case SDEV_CREATED_BLOCK:
1256 * If the devices is blocked we defer normal commands.
1258 if (!(req->cmd_flags & REQ_PREEMPT))
1259 ret = BLKPREP_DEFER;
1263 * For any other not fully online state we only allow
1264 * special commands. In particular any user initiated
1265 * command is not allowed.
1267 if (!(req->cmd_flags & REQ_PREEMPT))
1274 EXPORT_SYMBOL(scsi_prep_state_check);
1276 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1278 struct scsi_device *sdev = q->queuedata;
1282 req->errors = DID_NO_CONNECT << 16;
1283 /* release the command and kill it */
1285 struct scsi_cmnd *cmd = req->special;
1286 scsi_release_buffers(cmd);
1287 scsi_put_command(cmd);
1288 req->special = NULL;
1293 * If we defer, the elv_next_request() returns NULL, but the
1294 * queue must be restarted, so we plug here if no returning
1295 * command will automatically do that.
1297 if (sdev->device_busy == 0)
1301 req->cmd_flags |= REQ_DONTPREP;
1306 EXPORT_SYMBOL(scsi_prep_return);
1308 int scsi_prep_fn(struct request_queue *q, struct request *req)
1310 struct scsi_device *sdev = q->queuedata;
1311 int ret = BLKPREP_KILL;
1313 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1314 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1315 return scsi_prep_return(q, req, ret);
1319 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1322 * Called with the queue_lock held.
1324 static inline int scsi_dev_queue_ready(struct request_queue *q,
1325 struct scsi_device *sdev)
1327 if (sdev->device_busy >= sdev->queue_depth)
1329 if (sdev->device_busy == 0 && sdev->device_blocked) {
1331 * unblock after device_blocked iterates to zero
1333 if (--sdev->device_blocked == 0) {
1335 sdev_printk(KERN_INFO, sdev,
1336 "unblocking device at zero depth\n"));
1342 if (sdev->device_blocked)
1349 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1350 * return 0. We must end up running the queue again whenever 0 is
1351 * returned, else IO can hang.
1353 * Called with host_lock held.
1355 static inline int scsi_host_queue_ready(struct request_queue *q,
1356 struct Scsi_Host *shost,
1357 struct scsi_device *sdev)
1359 if (scsi_host_in_recovery(shost))
1361 if (shost->host_busy == 0 && shost->host_blocked) {
1363 * unblock after host_blocked iterates to zero
1365 if (--shost->host_blocked == 0) {
1367 printk("scsi%d unblocking host at zero depth\n",
1373 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1374 shost->host_blocked || shost->host_self_blocked) {
1375 if (list_empty(&sdev->starved_entry))
1376 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1380 /* We're OK to process the command, so we can't be starved */
1381 if (!list_empty(&sdev->starved_entry))
1382 list_del_init(&sdev->starved_entry);
1388 * Kill a request for a dead device
1390 static void scsi_kill_request(struct request *req, struct request_queue *q)
1392 struct scsi_cmnd *cmd = req->special;
1393 struct scsi_device *sdev = cmd->device;
1394 struct Scsi_Host *shost = sdev->host;
1396 blkdev_dequeue_request(req);
1398 if (unlikely(cmd == NULL)) {
1399 printk(KERN_CRIT "impossible request in %s.\n",
1404 scsi_init_cmd_errh(cmd);
1405 cmd->result = DID_NO_CONNECT << 16;
1406 atomic_inc(&cmd->device->iorequest_cnt);
1409 * SCSI request completion path will do scsi_device_unbusy(),
1410 * bump busy counts. To bump the counters, we need to dance
1411 * with the locks as normal issue path does.
1413 sdev->device_busy++;
1414 spin_unlock(sdev->request_queue->queue_lock);
1415 spin_lock(shost->host_lock);
1417 spin_unlock(shost->host_lock);
1418 spin_lock(sdev->request_queue->queue_lock);
1423 static void scsi_softirq_done(struct request *rq)
1425 struct scsi_cmnd *cmd = rq->completion_data;
1426 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
1429 INIT_LIST_HEAD(&cmd->eh_entry);
1431 disposition = scsi_decide_disposition(cmd);
1432 if (disposition != SUCCESS &&
1433 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1434 sdev_printk(KERN_ERR, cmd->device,
1435 "timing out command, waited %lus\n",
1437 disposition = SUCCESS;
1440 scsi_log_completion(cmd, disposition);
1442 switch (disposition) {
1444 scsi_finish_command(cmd);
1447 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1449 case ADD_TO_MLQUEUE:
1450 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1453 if (!scsi_eh_scmd_add(cmd, 0))
1454 scsi_finish_command(cmd);
1459 * Function: scsi_request_fn()
1461 * Purpose: Main strategy routine for SCSI.
1463 * Arguments: q - Pointer to actual queue.
1467 * Lock status: IO request lock assumed to be held when called.
1469 static void scsi_request_fn(struct request_queue *q)
1471 struct scsi_device *sdev = q->queuedata;
1472 struct Scsi_Host *shost;
1473 struct scsi_cmnd *cmd;
1474 struct request *req;
1477 printk("scsi: killing requests for dead queue\n");
1478 while ((req = elv_next_request(q)) != NULL)
1479 scsi_kill_request(req, q);
1483 if(!get_device(&sdev->sdev_gendev))
1484 /* We must be tearing the block queue down already */
1488 * To start with, we keep looping until the queue is empty, or until
1489 * the host is no longer able to accept any more requests.
1492 while (!blk_queue_plugged(q)) {
1495 * get next queueable request. We do this early to make sure
1496 * that the request is fully prepared even if we cannot
1499 req = elv_next_request(q);
1500 if (!req || !scsi_dev_queue_ready(q, sdev))
1503 if (unlikely(!scsi_device_online(sdev))) {
1504 sdev_printk(KERN_ERR, sdev,
1505 "rejecting I/O to offline device\n");
1506 scsi_kill_request(req, q);
1512 * Remove the request from the request list.
1514 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1515 blkdev_dequeue_request(req);
1516 sdev->device_busy++;
1518 spin_unlock(q->queue_lock);
1520 if (unlikely(cmd == NULL)) {
1521 printk(KERN_CRIT "impossible request in %s.\n"
1522 "please mail a stack trace to "
1523 "linux-scsi@vger.kernel.org\n",
1525 blk_dump_rq_flags(req, "foo");
1528 spin_lock(shost->host_lock);
1531 * We hit this when the driver is using a host wide
1532 * tag map. For device level tag maps the queue_depth check
1533 * in the device ready fn would prevent us from trying
1534 * to allocate a tag. Since the map is a shared host resource
1535 * we add the dev to the starved list so it eventually gets
1536 * a run when a tag is freed.
1538 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1539 if (list_empty(&sdev->starved_entry))
1540 list_add_tail(&sdev->starved_entry,
1541 &shost->starved_list);
1545 if (!scsi_host_queue_ready(q, shost, sdev))
1547 if (scsi_target(sdev)->single_lun) {
1548 if (scsi_target(sdev)->starget_sdev_user &&
1549 scsi_target(sdev)->starget_sdev_user != sdev)
1551 scsi_target(sdev)->starget_sdev_user = sdev;
1556 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1557 * take the lock again.
1559 spin_unlock_irq(shost->host_lock);
1562 * Finally, initialize any error handling parameters, and set up
1563 * the timers for timeouts.
1565 scsi_init_cmd_errh(cmd);
1568 * Dispatch the command to the low-level driver.
1570 rtn = scsi_dispatch_cmd(cmd);
1571 spin_lock_irq(q->queue_lock);
1573 /* we're refusing the command; because of
1574 * the way locks get dropped, we need to
1575 * check here if plugging is required */
1576 if(sdev->device_busy == 0)
1586 spin_unlock_irq(shost->host_lock);
1589 * lock q, handle tag, requeue req, and decrement device_busy. We
1590 * must return with queue_lock held.
1592 * Decrementing device_busy without checking it is OK, as all such
1593 * cases (host limits or settings) should run the queue at some
1596 spin_lock_irq(q->queue_lock);
1597 blk_requeue_request(q, req);
1598 sdev->device_busy--;
1599 if(sdev->device_busy == 0)
1602 /* must be careful here...if we trigger the ->remove() function
1603 * we cannot be holding the q lock */
1604 spin_unlock_irq(q->queue_lock);
1605 put_device(&sdev->sdev_gendev);
1606 spin_lock_irq(q->queue_lock);
1609 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1611 struct device *host_dev;
1612 u64 bounce_limit = 0xffffffff;
1614 if (shost->unchecked_isa_dma)
1615 return BLK_BOUNCE_ISA;
1617 * Platforms with virtual-DMA translation
1618 * hardware have no practical limit.
1620 if (!PCI_DMA_BUS_IS_PHYS)
1621 return BLK_BOUNCE_ANY;
1623 host_dev = scsi_get_device(shost);
1624 if (host_dev && host_dev->dma_mask)
1625 bounce_limit = *host_dev->dma_mask;
1627 return bounce_limit;
1629 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1631 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1632 request_fn_proc *request_fn)
1634 struct request_queue *q;
1635 struct device *dev = shost->shost_gendev.parent;
1637 q = blk_init_queue(request_fn, NULL);
1642 * this limit is imposed by hardware restrictions
1644 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1645 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1647 blk_queue_max_sectors(q, shost->max_sectors);
1648 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1649 blk_queue_segment_boundary(q, shost->dma_boundary);
1650 dma_set_seg_boundary(dev, shost->dma_boundary);
1652 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1654 /* New queue, no concurrency on queue_flags */
1655 if (!shost->use_clustering)
1656 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1659 * set a reasonable default alignment on word boundaries: the
1660 * host and device may alter it using
1661 * blk_queue_update_dma_alignment() later.
1663 blk_queue_dma_alignment(q, 0x03);
1667 EXPORT_SYMBOL(__scsi_alloc_queue);
1669 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1671 struct request_queue *q;
1673 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1677 blk_queue_prep_rq(q, scsi_prep_fn);
1678 blk_queue_softirq_done(q, scsi_softirq_done);
1682 void scsi_free_queue(struct request_queue *q)
1684 blk_cleanup_queue(q);
1688 * Function: scsi_block_requests()
1690 * Purpose: Utility function used by low-level drivers to prevent further
1691 * commands from being queued to the device.
1693 * Arguments: shost - Host in question
1697 * Lock status: No locks are assumed held.
1699 * Notes: There is no timer nor any other means by which the requests
1700 * get unblocked other than the low-level driver calling
1701 * scsi_unblock_requests().
1703 void scsi_block_requests(struct Scsi_Host *shost)
1705 shost->host_self_blocked = 1;
1707 EXPORT_SYMBOL(scsi_block_requests);
1710 * Function: scsi_unblock_requests()
1712 * Purpose: Utility function used by low-level drivers to allow further
1713 * commands from being queued to the device.
1715 * Arguments: shost - Host in question
1719 * Lock status: No locks are assumed held.
1721 * Notes: There is no timer nor any other means by which the requests
1722 * get unblocked other than the low-level driver calling
1723 * scsi_unblock_requests().
1725 * This is done as an API function so that changes to the
1726 * internals of the scsi mid-layer won't require wholesale
1727 * changes to drivers that use this feature.
1729 void scsi_unblock_requests(struct Scsi_Host *shost)
1731 shost->host_self_blocked = 0;
1732 scsi_run_host_queues(shost);
1734 EXPORT_SYMBOL(scsi_unblock_requests);
1736 int __init scsi_init_queue(void)
1740 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1741 sizeof(struct scsi_io_context),
1743 if (!scsi_io_context_cache) {
1744 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1748 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1749 sizeof(struct scsi_data_buffer),
1751 if (!scsi_sdb_cache) {
1752 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1753 goto cleanup_io_context;
1756 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1757 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1758 int size = sgp->size * sizeof(struct scatterlist);
1760 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1761 SLAB_HWCACHE_ALIGN, NULL);
1763 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1768 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1771 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1780 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1781 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1783 mempool_destroy(sgp->pool);
1785 kmem_cache_destroy(sgp->slab);
1787 kmem_cache_destroy(scsi_sdb_cache);
1789 kmem_cache_destroy(scsi_io_context_cache);
1794 void scsi_exit_queue(void)
1798 kmem_cache_destroy(scsi_io_context_cache);
1799 kmem_cache_destroy(scsi_sdb_cache);
1801 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1802 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1803 mempool_destroy(sgp->pool);
1804 kmem_cache_destroy(sgp->slab);
1809 * scsi_mode_select - issue a mode select
1810 * @sdev: SCSI device to be queried
1811 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1812 * @sp: Save page bit (0 == don't save, 1 == save)
1813 * @modepage: mode page being requested
1814 * @buffer: request buffer (may not be smaller than eight bytes)
1815 * @len: length of request buffer.
1816 * @timeout: command timeout
1817 * @retries: number of retries before failing
1818 * @data: returns a structure abstracting the mode header data
1819 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1820 * must be SCSI_SENSE_BUFFERSIZE big.
1822 * Returns zero if successful; negative error number or scsi
1827 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1828 unsigned char *buffer, int len, int timeout, int retries,
1829 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1831 unsigned char cmd[10];
1832 unsigned char *real_buffer;
1835 memset(cmd, 0, sizeof(cmd));
1836 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1838 if (sdev->use_10_for_ms) {
1841 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1844 memcpy(real_buffer + 8, buffer, len);
1848 real_buffer[2] = data->medium_type;
1849 real_buffer[3] = data->device_specific;
1850 real_buffer[4] = data->longlba ? 0x01 : 0;
1852 real_buffer[6] = data->block_descriptor_length >> 8;
1853 real_buffer[7] = data->block_descriptor_length;
1855 cmd[0] = MODE_SELECT_10;
1859 if (len > 255 || data->block_descriptor_length > 255 ||
1863 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1866 memcpy(real_buffer + 4, buffer, len);
1869 real_buffer[1] = data->medium_type;
1870 real_buffer[2] = data->device_specific;
1871 real_buffer[3] = data->block_descriptor_length;
1874 cmd[0] = MODE_SELECT;
1878 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1879 sshdr, timeout, retries);
1883 EXPORT_SYMBOL_GPL(scsi_mode_select);
1886 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1887 * @sdev: SCSI device to be queried
1888 * @dbd: set if mode sense will allow block descriptors to be returned
1889 * @modepage: mode page being requested
1890 * @buffer: request buffer (may not be smaller than eight bytes)
1891 * @len: length of request buffer.
1892 * @timeout: command timeout
1893 * @retries: number of retries before failing
1894 * @data: returns a structure abstracting the mode header data
1895 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1896 * must be SCSI_SENSE_BUFFERSIZE big.
1898 * Returns zero if unsuccessful, or the header offset (either 4
1899 * or 8 depending on whether a six or ten byte command was
1900 * issued) if successful.
1903 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1904 unsigned char *buffer, int len, int timeout, int retries,
1905 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1907 unsigned char cmd[12];
1911 struct scsi_sense_hdr my_sshdr;
1913 memset(data, 0, sizeof(*data));
1914 memset(&cmd[0], 0, 12);
1915 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1918 /* caller might not be interested in sense, but we need it */
1923 use_10_for_ms = sdev->use_10_for_ms;
1925 if (use_10_for_ms) {
1929 cmd[0] = MODE_SENSE_10;
1936 cmd[0] = MODE_SENSE;
1941 memset(buffer, 0, len);
1943 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1944 sshdr, timeout, retries);
1946 /* This code looks awful: what it's doing is making sure an
1947 * ILLEGAL REQUEST sense return identifies the actual command
1948 * byte as the problem. MODE_SENSE commands can return
1949 * ILLEGAL REQUEST if the code page isn't supported */
1951 if (use_10_for_ms && !scsi_status_is_good(result) &&
1952 (driver_byte(result) & DRIVER_SENSE)) {
1953 if (scsi_sense_valid(sshdr)) {
1954 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1955 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1957 * Invalid command operation code
1959 sdev->use_10_for_ms = 0;
1965 if(scsi_status_is_good(result)) {
1966 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1967 (modepage == 6 || modepage == 8))) {
1968 /* Initio breakage? */
1971 data->medium_type = 0;
1972 data->device_specific = 0;
1974 data->block_descriptor_length = 0;
1975 } else if(use_10_for_ms) {
1976 data->length = buffer[0]*256 + buffer[1] + 2;
1977 data->medium_type = buffer[2];
1978 data->device_specific = buffer[3];
1979 data->longlba = buffer[4] & 0x01;
1980 data->block_descriptor_length = buffer[6]*256
1983 data->length = buffer[0] + 1;
1984 data->medium_type = buffer[1];
1985 data->device_specific = buffer[2];
1986 data->block_descriptor_length = buffer[3];
1988 data->header_length = header_length;
1993 EXPORT_SYMBOL(scsi_mode_sense);
1996 * scsi_test_unit_ready - test if unit is ready
1997 * @sdev: scsi device to change the state of.
1998 * @timeout: command timeout
1999 * @retries: number of retries before failing
2000 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2001 * returning sense. Make sure that this is cleared before passing
2004 * Returns zero if unsuccessful or an error if TUR failed. For
2005 * removable media, a return of NOT_READY or UNIT_ATTENTION is
2006 * translated to success, with the ->changed flag updated.
2009 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2010 struct scsi_sense_hdr *sshdr_external)
2013 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2015 struct scsi_sense_hdr *sshdr;
2018 if (!sshdr_external)
2019 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2021 sshdr = sshdr_external;
2023 /* try to eat the UNIT_ATTENTION if there are enough retries */
2025 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2027 } while ((driver_byte(result) & DRIVER_SENSE) &&
2028 sshdr && sshdr->sense_key == UNIT_ATTENTION &&
2032 /* could not allocate sense buffer, so can't process it */
2035 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
2037 if ((scsi_sense_valid(sshdr)) &&
2038 ((sshdr->sense_key == UNIT_ATTENTION) ||
2039 (sshdr->sense_key == NOT_READY))) {
2044 if (!sshdr_external)
2048 EXPORT_SYMBOL(scsi_test_unit_ready);
2051 * scsi_device_set_state - Take the given device through the device state model.
2052 * @sdev: scsi device to change the state of.
2053 * @state: state to change to.
2055 * Returns zero if unsuccessful or an error if the requested
2056 * transition is illegal.
2059 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2061 enum scsi_device_state oldstate = sdev->sdev_state;
2063 if (state == oldstate)
2069 case SDEV_CREATED_BLOCK:
2113 case SDEV_CREATED_BLOCK:
2120 case SDEV_CREATED_BLOCK:
2155 sdev->sdev_state = state;
2159 SCSI_LOG_ERROR_RECOVERY(1,
2160 sdev_printk(KERN_ERR, sdev,
2161 "Illegal state transition %s->%s\n",
2162 scsi_device_state_name(oldstate),
2163 scsi_device_state_name(state))
2167 EXPORT_SYMBOL(scsi_device_set_state);
2170 * sdev_evt_emit - emit a single SCSI device uevent
2171 * @sdev: associated SCSI device
2172 * @evt: event to emit
2174 * Send a single uevent (scsi_event) to the associated scsi_device.
2176 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2181 switch (evt->evt_type) {
2182 case SDEV_EVT_MEDIA_CHANGE:
2183 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2193 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2197 * sdev_evt_thread - send a uevent for each scsi event
2198 * @work: work struct for scsi_device
2200 * Dispatch queued events to their associated scsi_device kobjects
2203 void scsi_evt_thread(struct work_struct *work)
2205 struct scsi_device *sdev;
2206 LIST_HEAD(event_list);
2208 sdev = container_of(work, struct scsi_device, event_work);
2211 struct scsi_event *evt;
2212 struct list_head *this, *tmp;
2213 unsigned long flags;
2215 spin_lock_irqsave(&sdev->list_lock, flags);
2216 list_splice_init(&sdev->event_list, &event_list);
2217 spin_unlock_irqrestore(&sdev->list_lock, flags);
2219 if (list_empty(&event_list))
2222 list_for_each_safe(this, tmp, &event_list) {
2223 evt = list_entry(this, struct scsi_event, node);
2224 list_del(&evt->node);
2225 scsi_evt_emit(sdev, evt);
2232 * sdev_evt_send - send asserted event to uevent thread
2233 * @sdev: scsi_device event occurred on
2234 * @evt: event to send
2236 * Assert scsi device event asynchronously.
2238 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2240 unsigned long flags;
2243 /* FIXME: currently this check eliminates all media change events
2244 * for polled devices. Need to update to discriminate between AN
2245 * and polled events */
2246 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2252 spin_lock_irqsave(&sdev->list_lock, flags);
2253 list_add_tail(&evt->node, &sdev->event_list);
2254 schedule_work(&sdev->event_work);
2255 spin_unlock_irqrestore(&sdev->list_lock, flags);
2257 EXPORT_SYMBOL_GPL(sdev_evt_send);
2260 * sdev_evt_alloc - allocate a new scsi event
2261 * @evt_type: type of event to allocate
2262 * @gfpflags: GFP flags for allocation
2264 * Allocates and returns a new scsi_event.
2266 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2269 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2273 evt->evt_type = evt_type;
2274 INIT_LIST_HEAD(&evt->node);
2276 /* evt_type-specific initialization, if any */
2278 case SDEV_EVT_MEDIA_CHANGE:
2286 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2289 * sdev_evt_send_simple - send asserted event to uevent thread
2290 * @sdev: scsi_device event occurred on
2291 * @evt_type: type of event to send
2292 * @gfpflags: GFP flags for allocation
2294 * Assert scsi device event asynchronously, given an event type.
2296 void sdev_evt_send_simple(struct scsi_device *sdev,
2297 enum scsi_device_event evt_type, gfp_t gfpflags)
2299 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2301 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2306 sdev_evt_send(sdev, evt);
2308 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2311 * scsi_device_quiesce - Block user issued commands.
2312 * @sdev: scsi device to quiesce.
2314 * This works by trying to transition to the SDEV_QUIESCE state
2315 * (which must be a legal transition). When the device is in this
2316 * state, only special requests will be accepted, all others will
2317 * be deferred. Since special requests may also be requeued requests,
2318 * a successful return doesn't guarantee the device will be
2319 * totally quiescent.
2321 * Must be called with user context, may sleep.
2323 * Returns zero if unsuccessful or an error if not.
2326 scsi_device_quiesce(struct scsi_device *sdev)
2328 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2332 scsi_run_queue(sdev->request_queue);
2333 while (sdev->device_busy) {
2334 msleep_interruptible(200);
2335 scsi_run_queue(sdev->request_queue);
2339 EXPORT_SYMBOL(scsi_device_quiesce);
2342 * scsi_device_resume - Restart user issued commands to a quiesced device.
2343 * @sdev: scsi device to resume.
2345 * Moves the device from quiesced back to running and restarts the
2348 * Must be called with user context, may sleep.
2351 scsi_device_resume(struct scsi_device *sdev)
2353 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2355 scsi_run_queue(sdev->request_queue);
2357 EXPORT_SYMBOL(scsi_device_resume);
2360 device_quiesce_fn(struct scsi_device *sdev, void *data)
2362 scsi_device_quiesce(sdev);
2366 scsi_target_quiesce(struct scsi_target *starget)
2368 starget_for_each_device(starget, NULL, device_quiesce_fn);
2370 EXPORT_SYMBOL(scsi_target_quiesce);
2373 device_resume_fn(struct scsi_device *sdev, void *data)
2375 scsi_device_resume(sdev);
2379 scsi_target_resume(struct scsi_target *starget)
2381 starget_for_each_device(starget, NULL, device_resume_fn);
2383 EXPORT_SYMBOL(scsi_target_resume);
2386 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2387 * @sdev: device to block
2389 * Block request made by scsi lld's to temporarily stop all
2390 * scsi commands on the specified device. Called from interrupt
2391 * or normal process context.
2393 * Returns zero if successful or error if not
2396 * This routine transitions the device to the SDEV_BLOCK state
2397 * (which must be a legal transition). When the device is in this
2398 * state, all commands are deferred until the scsi lld reenables
2399 * the device with scsi_device_unblock or device_block_tmo fires.
2400 * This routine assumes the host_lock is held on entry.
2403 scsi_internal_device_block(struct scsi_device *sdev)
2405 struct request_queue *q = sdev->request_queue;
2406 unsigned long flags;
2409 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2411 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2418 * The device has transitioned to SDEV_BLOCK. Stop the
2419 * block layer from calling the midlayer with this device's
2422 spin_lock_irqsave(q->queue_lock, flags);
2424 spin_unlock_irqrestore(q->queue_lock, flags);
2428 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2431 * scsi_internal_device_unblock - resume a device after a block request
2432 * @sdev: device to resume
2434 * Called by scsi lld's or the midlayer to restart the device queue
2435 * for the previously suspended scsi device. Called from interrupt or
2436 * normal process context.
2438 * Returns zero if successful or error if not.
2441 * This routine transitions the device to the SDEV_RUNNING state
2442 * (which must be a legal transition) allowing the midlayer to
2443 * goose the queue for this device. This routine assumes the
2444 * host_lock is held upon entry.
2447 scsi_internal_device_unblock(struct scsi_device *sdev)
2449 struct request_queue *q = sdev->request_queue;
2451 unsigned long flags;
2454 * Try to transition the scsi device to SDEV_RUNNING
2455 * and goose the device queue if successful.
2457 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2459 err = scsi_device_set_state(sdev, SDEV_CREATED);
2465 spin_lock_irqsave(q->queue_lock, flags);
2467 spin_unlock_irqrestore(q->queue_lock, flags);
2471 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2474 device_block(struct scsi_device *sdev, void *data)
2476 scsi_internal_device_block(sdev);
2480 target_block(struct device *dev, void *data)
2482 if (scsi_is_target_device(dev))
2483 starget_for_each_device(to_scsi_target(dev), NULL,
2489 scsi_target_block(struct device *dev)
2491 if (scsi_is_target_device(dev))
2492 starget_for_each_device(to_scsi_target(dev), NULL,
2495 device_for_each_child(dev, NULL, target_block);
2497 EXPORT_SYMBOL_GPL(scsi_target_block);
2500 device_unblock(struct scsi_device *sdev, void *data)
2502 scsi_internal_device_unblock(sdev);
2506 target_unblock(struct device *dev, void *data)
2508 if (scsi_is_target_device(dev))
2509 starget_for_each_device(to_scsi_target(dev), NULL,
2515 scsi_target_unblock(struct device *dev)
2517 if (scsi_is_target_device(dev))
2518 starget_for_each_device(to_scsi_target(dev), NULL,
2521 device_for_each_child(dev, NULL, target_unblock);
2523 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2526 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2527 * @sgl: scatter-gather list
2528 * @sg_count: number of segments in sg
2529 * @offset: offset in bytes into sg, on return offset into the mapped area
2530 * @len: bytes to map, on return number of bytes mapped
2532 * Returns virtual address of the start of the mapped page
2534 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2535 size_t *offset, size_t *len)
2538 size_t sg_len = 0, len_complete = 0;
2539 struct scatterlist *sg;
2542 WARN_ON(!irqs_disabled());
2544 for_each_sg(sgl, sg, sg_count, i) {
2545 len_complete = sg_len; /* Complete sg-entries */
2546 sg_len += sg->length;
2547 if (sg_len > *offset)
2551 if (unlikely(i == sg_count)) {
2552 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2554 __func__, sg_len, *offset, sg_count);
2559 /* Offset starting from the beginning of first page in this sg-entry */
2560 *offset = *offset - len_complete + sg->offset;
2562 /* Assumption: contiguous pages can be accessed as "page + i" */
2563 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2564 *offset &= ~PAGE_MASK;
2566 /* Bytes in this sg-entry from *offset to the end of the page */
2567 sg_len = PAGE_SIZE - *offset;
2571 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2573 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2576 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2577 * @virt: virtual address to be unmapped
2579 void scsi_kunmap_atomic_sg(void *virt)
2581 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2583 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);