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_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 EXPORT_SYMBOL(scsi_release_buffers);
794 * Bidi commands Must be complete as a whole, both sides at once.
795 * If part of the bytes were written and lld returned
796 * scsi_in()->resid and/or scsi_out()->resid this information will be left
797 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
798 * decide what to do with this information.
800 static void scsi_end_bidi_request(struct scsi_cmnd *cmd)
802 struct request *req = cmd->request;
803 unsigned int dlen = req->data_len;
804 unsigned int next_dlen = req->next_rq->data_len;
806 req->data_len = scsi_out(cmd)->resid;
807 req->next_rq->data_len = scsi_in(cmd)->resid;
809 /* The req and req->next_rq have not been completed */
810 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen));
812 scsi_release_buffers(cmd);
815 * This will goose the queue request function at the end, so we don't
816 * need to worry about launching another command.
818 scsi_next_command(cmd);
822 * Function: scsi_io_completion()
824 * Purpose: Completion processing for block device I/O requests.
826 * Arguments: cmd - command that is finished.
828 * Lock status: Assumed that no lock is held upon entry.
832 * Notes: This function is matched in terms of capabilities to
833 * the function that created the scatter-gather list.
834 * In other words, if there are no bounce buffers
835 * (the normal case for most drivers), we don't need
836 * the logic to deal with cleaning up afterwards.
838 * We must do one of several things here:
840 * a) Call scsi_end_request. This will finish off the
841 * specified number of sectors. If we are done, the
842 * command block will be released, and the queue
843 * function will be goosed. If we are not done, then
844 * scsi_end_request will directly goose the queue.
846 * b) We can just use scsi_requeue_command() here. This would
847 * be used if we just wanted to retry, for example.
849 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
851 int result = cmd->result;
852 int this_count = scsi_bufflen(cmd);
853 struct request_queue *q = cmd->device->request_queue;
854 struct request *req = cmd->request;
856 struct scsi_sense_hdr sshdr;
858 int sense_deferred = 0;
861 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
863 sense_deferred = scsi_sense_is_deferred(&sshdr);
866 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
867 req->errors = result;
869 if (sense_valid && req->sense) {
871 * SG_IO wants current and deferred errors
873 int len = 8 + cmd->sense_buffer[7];
875 if (len > SCSI_SENSE_BUFFERSIZE)
876 len = SCSI_SENSE_BUFFERSIZE;
877 memcpy(req->sense, cmd->sense_buffer, len);
878 req->sense_len = len;
883 if (scsi_bidi_cmnd(cmd)) {
884 /* will also release_buffers */
885 scsi_end_bidi_request(cmd);
888 req->data_len = scsi_get_resid(cmd);
891 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
892 scsi_release_buffers(cmd);
895 * Next deal with any sectors which we were able to correctly
898 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
900 req->nr_sectors, good_bytes));
902 /* A number of bytes were successfully read. If there
903 * are leftovers and there is some kind of error
904 * (result != 0), retry the rest.
906 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
909 /* good_bytes = 0, or (inclusive) there were leftovers and
910 * result = 0, so scsi_end_request couldn't retry.
912 if (sense_valid && !sense_deferred) {
913 switch (sshdr.sense_key) {
915 if (cmd->device->removable) {
916 /* Detected disc change. Set a bit
917 * and quietly refuse further access.
919 cmd->device->changed = 1;
920 scsi_end_request(cmd, -EIO, this_count, 1);
923 /* Must have been a power glitch, or a
924 * bus reset. Could not have been a
925 * media change, so we just retry the
926 * request and see what happens.
928 scsi_requeue_command(q, cmd);
932 case ILLEGAL_REQUEST:
933 /* If we had an ILLEGAL REQUEST returned, then
934 * we may have performed an unsupported
935 * command. The only thing this should be
936 * would be a ten byte read where only a six
937 * byte read was supported. Also, on a system
938 * where READ CAPACITY failed, we may have
939 * read past the end of the disk.
941 if ((cmd->device->use_10_for_rw &&
942 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
943 (cmd->cmnd[0] == READ_10 ||
944 cmd->cmnd[0] == WRITE_10)) {
945 cmd->device->use_10_for_rw = 0;
946 /* This will cause a retry with a
949 scsi_requeue_command(q, cmd);
952 scsi_end_request(cmd, -EIO, this_count, 1);
957 /* If the device is in the process of becoming
958 * ready, or has a temporary blockage, retry.
960 if (sshdr.asc == 0x04) {
961 switch (sshdr.ascq) {
962 case 0x01: /* becoming ready */
963 case 0x04: /* format in progress */
964 case 0x05: /* rebuild in progress */
965 case 0x06: /* recalculation in progress */
966 case 0x07: /* operation in progress */
967 case 0x08: /* Long write in progress */
968 case 0x09: /* self test in progress */
969 scsi_requeue_command(q, cmd);
975 if (!(req->cmd_flags & REQ_QUIET))
976 scsi_cmd_print_sense_hdr(cmd,
980 scsi_end_request(cmd, -EIO, this_count, 1);
982 case VOLUME_OVERFLOW:
983 if (!(req->cmd_flags & REQ_QUIET)) {
984 scmd_printk(KERN_INFO, cmd,
985 "Volume overflow, CDB: ");
986 __scsi_print_command(cmd->cmnd);
987 scsi_print_sense("", cmd);
989 /* See SSC3rXX or current. */
990 scsi_end_request(cmd, -EIO, this_count, 1);
996 if (host_byte(result) == DID_RESET) {
997 /* Third party bus reset or reset for error recovery
998 * reasons. Just retry the request and see what
1001 scsi_requeue_command(q, cmd);
1005 if (!(req->cmd_flags & REQ_QUIET)) {
1006 scsi_print_result(cmd);
1007 if (driver_byte(result) & DRIVER_SENSE)
1008 scsi_print_sense("", cmd);
1011 scsi_end_request(cmd, -EIO, this_count, !result);
1014 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1020 * If sg table allocation fails, requeue request later.
1022 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1024 return BLKPREP_DEFER;
1030 * Next, walk the list, and fill in the addresses and sizes of
1033 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1034 BUG_ON(count > sdb->table.nents);
1035 sdb->table.nents = count;
1036 if (blk_pc_request(req))
1037 sdb->length = req->data_len;
1039 sdb->length = req->nr_sectors << 9;
1044 * Function: scsi_init_io()
1046 * Purpose: SCSI I/O initialize function.
1048 * Arguments: cmd - Command descriptor we wish to initialize
1050 * Returns: 0 on success
1051 * BLKPREP_DEFER if the failure is retryable
1052 * BLKPREP_KILL if the failure is fatal
1054 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1056 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
1060 if (blk_bidi_rq(cmd->request)) {
1061 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1062 scsi_sdb_cache, GFP_ATOMIC);
1064 error = BLKPREP_DEFER;
1068 cmd->request->next_rq->special = bidi_sdb;
1069 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1078 scsi_release_buffers(cmd);
1079 if (error == BLKPREP_KILL)
1080 scsi_put_command(cmd);
1081 else /* BLKPREP_DEFER */
1082 scsi_unprep_request(cmd->request);
1086 EXPORT_SYMBOL(scsi_init_io);
1088 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1089 struct request *req)
1091 struct scsi_cmnd *cmd;
1093 if (!req->special) {
1094 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1102 /* pull a tag out of the request if we have one */
1103 cmd->tag = req->tag;
1106 cmd->cmnd = req->cmd;
1111 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1113 struct scsi_cmnd *cmd;
1114 int ret = scsi_prep_state_check(sdev, req);
1116 if (ret != BLKPREP_OK)
1119 cmd = scsi_get_cmd_from_req(sdev, req);
1121 return BLKPREP_DEFER;
1124 * BLOCK_PC requests may transfer data, in which case they must
1125 * a bio attached to them. Or they might contain a SCSI command
1126 * that does not transfer data, in which case they may optionally
1127 * submit a request without an attached bio.
1132 BUG_ON(!req->nr_phys_segments);
1134 ret = scsi_init_io(cmd, GFP_ATOMIC);
1138 BUG_ON(req->data_len);
1141 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1145 cmd->cmd_len = req->cmd_len;
1147 cmd->sc_data_direction = DMA_NONE;
1148 else if (rq_data_dir(req) == WRITE)
1149 cmd->sc_data_direction = DMA_TO_DEVICE;
1151 cmd->sc_data_direction = DMA_FROM_DEVICE;
1153 cmd->transfersize = req->data_len;
1154 cmd->allowed = req->retries;
1155 cmd->timeout_per_command = req->timeout;
1158 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1161 * Setup a REQ_TYPE_FS command. These are simple read/write request
1162 * from filesystems that still need to be translated to SCSI CDBs from
1165 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1167 struct scsi_cmnd *cmd;
1168 int ret = scsi_prep_state_check(sdev, req);
1170 if (ret != BLKPREP_OK)
1173 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1174 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1175 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1176 if (ret != BLKPREP_OK)
1181 * Filesystem requests must transfer data.
1183 BUG_ON(!req->nr_phys_segments);
1185 cmd = scsi_get_cmd_from_req(sdev, req);
1187 return BLKPREP_DEFER;
1189 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1190 return scsi_init_io(cmd, GFP_ATOMIC);
1192 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1194 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1196 int ret = BLKPREP_OK;
1199 * If the device is not in running state we will reject some
1202 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1203 switch (sdev->sdev_state) {
1206 * If the device is offline we refuse to process any
1207 * commands. The device must be brought online
1208 * before trying any recovery commands.
1210 sdev_printk(KERN_ERR, sdev,
1211 "rejecting I/O to offline device\n");
1216 * If the device is fully deleted, we refuse to
1217 * process any commands as well.
1219 sdev_printk(KERN_ERR, sdev,
1220 "rejecting I/O to dead device\n");
1226 * If the devices is blocked we defer normal commands.
1228 if (!(req->cmd_flags & REQ_PREEMPT))
1229 ret = BLKPREP_DEFER;
1233 * For any other not fully online state we only allow
1234 * special commands. In particular any user initiated
1235 * command is not allowed.
1237 if (!(req->cmd_flags & REQ_PREEMPT))
1244 EXPORT_SYMBOL(scsi_prep_state_check);
1246 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1248 struct scsi_device *sdev = q->queuedata;
1252 req->errors = DID_NO_CONNECT << 16;
1253 /* release the command and kill it */
1255 struct scsi_cmnd *cmd = req->special;
1256 scsi_release_buffers(cmd);
1257 scsi_put_command(cmd);
1258 req->special = NULL;
1263 * If we defer, the elv_next_request() returns NULL, but the
1264 * queue must be restarted, so we plug here if no returning
1265 * command will automatically do that.
1267 if (sdev->device_busy == 0)
1271 req->cmd_flags |= REQ_DONTPREP;
1276 EXPORT_SYMBOL(scsi_prep_return);
1278 int scsi_prep_fn(struct request_queue *q, struct request *req)
1280 struct scsi_device *sdev = q->queuedata;
1281 int ret = BLKPREP_KILL;
1283 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1284 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1285 return scsi_prep_return(q, req, ret);
1289 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1292 * Called with the queue_lock held.
1294 static inline int scsi_dev_queue_ready(struct request_queue *q,
1295 struct scsi_device *sdev)
1297 if (sdev->device_busy >= sdev->queue_depth)
1299 if (sdev->device_busy == 0 && sdev->device_blocked) {
1301 * unblock after device_blocked iterates to zero
1303 if (--sdev->device_blocked == 0) {
1305 sdev_printk(KERN_INFO, sdev,
1306 "unblocking device at zero depth\n"));
1312 if (sdev->device_blocked)
1319 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1320 * return 0. We must end up running the queue again whenever 0 is
1321 * returned, else IO can hang.
1323 * Called with host_lock held.
1325 static inline int scsi_host_queue_ready(struct request_queue *q,
1326 struct Scsi_Host *shost,
1327 struct scsi_device *sdev)
1329 if (scsi_host_in_recovery(shost))
1331 if (shost->host_busy == 0 && shost->host_blocked) {
1333 * unblock after host_blocked iterates to zero
1335 if (--shost->host_blocked == 0) {
1337 printk("scsi%d unblocking host at zero depth\n",
1343 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1344 shost->host_blocked || shost->host_self_blocked) {
1345 if (list_empty(&sdev->starved_entry))
1346 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1350 /* We're OK to process the command, so we can't be starved */
1351 if (!list_empty(&sdev->starved_entry))
1352 list_del_init(&sdev->starved_entry);
1358 * Kill a request for a dead device
1360 static void scsi_kill_request(struct request *req, struct request_queue *q)
1362 struct scsi_cmnd *cmd = req->special;
1363 struct scsi_device *sdev = cmd->device;
1364 struct Scsi_Host *shost = sdev->host;
1366 blkdev_dequeue_request(req);
1368 if (unlikely(cmd == NULL)) {
1369 printk(KERN_CRIT "impossible request in %s.\n",
1374 scsi_init_cmd_errh(cmd);
1375 cmd->result = DID_NO_CONNECT << 16;
1376 atomic_inc(&cmd->device->iorequest_cnt);
1379 * SCSI request completion path will do scsi_device_unbusy(),
1380 * bump busy counts. To bump the counters, we need to dance
1381 * with the locks as normal issue path does.
1383 sdev->device_busy++;
1384 spin_unlock(sdev->request_queue->queue_lock);
1385 spin_lock(shost->host_lock);
1387 spin_unlock(shost->host_lock);
1388 spin_lock(sdev->request_queue->queue_lock);
1393 static void scsi_softirq_done(struct request *rq)
1395 struct scsi_cmnd *cmd = rq->completion_data;
1396 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
1399 INIT_LIST_HEAD(&cmd->eh_entry);
1401 disposition = scsi_decide_disposition(cmd);
1402 if (disposition != SUCCESS &&
1403 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1404 sdev_printk(KERN_ERR, cmd->device,
1405 "timing out command, waited %lus\n",
1407 disposition = SUCCESS;
1410 scsi_log_completion(cmd, disposition);
1412 switch (disposition) {
1414 scsi_finish_command(cmd);
1417 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1419 case ADD_TO_MLQUEUE:
1420 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1423 if (!scsi_eh_scmd_add(cmd, 0))
1424 scsi_finish_command(cmd);
1429 * Function: scsi_request_fn()
1431 * Purpose: Main strategy routine for SCSI.
1433 * Arguments: q - Pointer to actual queue.
1437 * Lock status: IO request lock assumed to be held when called.
1439 static void scsi_request_fn(struct request_queue *q)
1441 struct scsi_device *sdev = q->queuedata;
1442 struct Scsi_Host *shost;
1443 struct scsi_cmnd *cmd;
1444 struct request *req;
1447 printk("scsi: killing requests for dead queue\n");
1448 while ((req = elv_next_request(q)) != NULL)
1449 scsi_kill_request(req, q);
1453 if(!get_device(&sdev->sdev_gendev))
1454 /* We must be tearing the block queue down already */
1458 * To start with, we keep looping until the queue is empty, or until
1459 * the host is no longer able to accept any more requests.
1462 while (!blk_queue_plugged(q)) {
1465 * get next queueable request. We do this early to make sure
1466 * that the request is fully prepared even if we cannot
1469 req = elv_next_request(q);
1470 if (!req || !scsi_dev_queue_ready(q, sdev))
1473 if (unlikely(!scsi_device_online(sdev))) {
1474 sdev_printk(KERN_ERR, sdev,
1475 "rejecting I/O to offline device\n");
1476 scsi_kill_request(req, q);
1482 * Remove the request from the request list.
1484 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1485 blkdev_dequeue_request(req);
1486 sdev->device_busy++;
1488 spin_unlock(q->queue_lock);
1490 if (unlikely(cmd == NULL)) {
1491 printk(KERN_CRIT "impossible request in %s.\n"
1492 "please mail a stack trace to "
1493 "linux-scsi@vger.kernel.org\n",
1495 blk_dump_rq_flags(req, "foo");
1498 spin_lock(shost->host_lock);
1501 * We hit this when the driver is using a host wide
1502 * tag map. For device level tag maps the queue_depth check
1503 * in the device ready fn would prevent us from trying
1504 * to allocate a tag. Since the map is a shared host resource
1505 * we add the dev to the starved list so it eventually gets
1506 * a run when a tag is freed.
1508 if (blk_queue_tagged(q) && (req->tag == -1)) {
1509 if (list_empty(&sdev->starved_entry))
1510 list_add_tail(&sdev->starved_entry,
1511 &shost->starved_list);
1515 if (!scsi_host_queue_ready(q, shost, sdev))
1517 if (scsi_target(sdev)->single_lun) {
1518 if (scsi_target(sdev)->starget_sdev_user &&
1519 scsi_target(sdev)->starget_sdev_user != sdev)
1521 scsi_target(sdev)->starget_sdev_user = sdev;
1526 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1527 * take the lock again.
1529 spin_unlock_irq(shost->host_lock);
1532 * Finally, initialize any error handling parameters, and set up
1533 * the timers for timeouts.
1535 scsi_init_cmd_errh(cmd);
1538 * Dispatch the command to the low-level driver.
1540 rtn = scsi_dispatch_cmd(cmd);
1541 spin_lock_irq(q->queue_lock);
1543 /* we're refusing the command; because of
1544 * the way locks get dropped, we need to
1545 * check here if plugging is required */
1546 if(sdev->device_busy == 0)
1556 spin_unlock_irq(shost->host_lock);
1559 * lock q, handle tag, requeue req, and decrement device_busy. We
1560 * must return with queue_lock held.
1562 * Decrementing device_busy without checking it is OK, as all such
1563 * cases (host limits or settings) should run the queue at some
1566 spin_lock_irq(q->queue_lock);
1567 blk_requeue_request(q, req);
1568 sdev->device_busy--;
1569 if(sdev->device_busy == 0)
1572 /* must be careful here...if we trigger the ->remove() function
1573 * we cannot be holding the q lock */
1574 spin_unlock_irq(q->queue_lock);
1575 put_device(&sdev->sdev_gendev);
1576 spin_lock_irq(q->queue_lock);
1579 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1581 struct device *host_dev;
1582 u64 bounce_limit = 0xffffffff;
1584 if (shost->unchecked_isa_dma)
1585 return BLK_BOUNCE_ISA;
1587 * Platforms with virtual-DMA translation
1588 * hardware have no practical limit.
1590 if (!PCI_DMA_BUS_IS_PHYS)
1591 return BLK_BOUNCE_ANY;
1593 host_dev = scsi_get_device(shost);
1594 if (host_dev && host_dev->dma_mask)
1595 bounce_limit = *host_dev->dma_mask;
1597 return bounce_limit;
1599 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1601 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1602 request_fn_proc *request_fn)
1604 struct request_queue *q;
1605 struct device *dev = shost->shost_gendev.parent;
1607 q = blk_init_queue(request_fn, NULL);
1612 * this limit is imposed by hardware restrictions
1614 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1615 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1617 blk_queue_max_sectors(q, shost->max_sectors);
1618 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1619 blk_queue_segment_boundary(q, shost->dma_boundary);
1620 dma_set_seg_boundary(dev, shost->dma_boundary);
1622 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1624 /* New queue, no concurrency on queue_flags */
1625 if (!shost->use_clustering)
1626 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1629 * set a reasonable default alignment on word boundaries: the
1630 * host and device may alter it using
1631 * blk_queue_update_dma_alignment() later.
1633 blk_queue_dma_alignment(q, 0x03);
1637 EXPORT_SYMBOL(__scsi_alloc_queue);
1639 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1641 struct request_queue *q;
1643 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1647 blk_queue_prep_rq(q, scsi_prep_fn);
1648 blk_queue_softirq_done(q, scsi_softirq_done);
1652 void scsi_free_queue(struct request_queue *q)
1654 blk_cleanup_queue(q);
1658 * Function: scsi_block_requests()
1660 * Purpose: Utility function used by low-level drivers to prevent further
1661 * commands from being queued to the device.
1663 * Arguments: shost - Host in question
1667 * Lock status: No locks are assumed held.
1669 * Notes: There is no timer nor any other means by which the requests
1670 * get unblocked other than the low-level driver calling
1671 * scsi_unblock_requests().
1673 void scsi_block_requests(struct Scsi_Host *shost)
1675 shost->host_self_blocked = 1;
1677 EXPORT_SYMBOL(scsi_block_requests);
1680 * Function: scsi_unblock_requests()
1682 * Purpose: Utility function used by low-level drivers to allow further
1683 * commands from being queued to the device.
1685 * Arguments: shost - Host in question
1689 * Lock status: No locks are assumed held.
1691 * Notes: There is no timer nor any other means by which the requests
1692 * get unblocked other than the low-level driver calling
1693 * scsi_unblock_requests().
1695 * This is done as an API function so that changes to the
1696 * internals of the scsi mid-layer won't require wholesale
1697 * changes to drivers that use this feature.
1699 void scsi_unblock_requests(struct Scsi_Host *shost)
1701 shost->host_self_blocked = 0;
1702 scsi_run_host_queues(shost);
1704 EXPORT_SYMBOL(scsi_unblock_requests);
1706 int __init scsi_init_queue(void)
1710 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1711 sizeof(struct scsi_io_context),
1713 if (!scsi_io_context_cache) {
1714 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1718 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1719 sizeof(struct scsi_data_buffer),
1721 if (!scsi_sdb_cache) {
1722 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1723 goto cleanup_io_context;
1726 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1727 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1728 int size = sgp->size * sizeof(struct scatterlist);
1730 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1731 SLAB_HWCACHE_ALIGN, NULL);
1733 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1738 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1741 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1750 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1751 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1753 mempool_destroy(sgp->pool);
1755 kmem_cache_destroy(sgp->slab);
1757 kmem_cache_destroy(scsi_sdb_cache);
1759 kmem_cache_destroy(scsi_io_context_cache);
1764 void scsi_exit_queue(void)
1768 kmem_cache_destroy(scsi_io_context_cache);
1769 kmem_cache_destroy(scsi_sdb_cache);
1771 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1772 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1773 mempool_destroy(sgp->pool);
1774 kmem_cache_destroy(sgp->slab);
1779 * scsi_mode_select - issue a mode select
1780 * @sdev: SCSI device to be queried
1781 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1782 * @sp: Save page bit (0 == don't save, 1 == save)
1783 * @modepage: mode page being requested
1784 * @buffer: request buffer (may not be smaller than eight bytes)
1785 * @len: length of request buffer.
1786 * @timeout: command timeout
1787 * @retries: number of retries before failing
1788 * @data: returns a structure abstracting the mode header data
1789 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1790 * must be SCSI_SENSE_BUFFERSIZE big.
1792 * Returns zero if successful; negative error number or scsi
1797 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1798 unsigned char *buffer, int len, int timeout, int retries,
1799 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1801 unsigned char cmd[10];
1802 unsigned char *real_buffer;
1805 memset(cmd, 0, sizeof(cmd));
1806 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1808 if (sdev->use_10_for_ms) {
1811 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1814 memcpy(real_buffer + 8, buffer, len);
1818 real_buffer[2] = data->medium_type;
1819 real_buffer[3] = data->device_specific;
1820 real_buffer[4] = data->longlba ? 0x01 : 0;
1822 real_buffer[6] = data->block_descriptor_length >> 8;
1823 real_buffer[7] = data->block_descriptor_length;
1825 cmd[0] = MODE_SELECT_10;
1829 if (len > 255 || data->block_descriptor_length > 255 ||
1833 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1836 memcpy(real_buffer + 4, buffer, len);
1839 real_buffer[1] = data->medium_type;
1840 real_buffer[2] = data->device_specific;
1841 real_buffer[3] = data->block_descriptor_length;
1844 cmd[0] = MODE_SELECT;
1848 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1849 sshdr, timeout, retries);
1853 EXPORT_SYMBOL_GPL(scsi_mode_select);
1856 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1857 * @sdev: SCSI device to be queried
1858 * @dbd: set if mode sense will allow block descriptors to be returned
1859 * @modepage: mode page being requested
1860 * @buffer: request buffer (may not be smaller than eight bytes)
1861 * @len: length of request buffer.
1862 * @timeout: command timeout
1863 * @retries: number of retries before failing
1864 * @data: returns a structure abstracting the mode header data
1865 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1866 * must be SCSI_SENSE_BUFFERSIZE big.
1868 * Returns zero if unsuccessful, or the header offset (either 4
1869 * or 8 depending on whether a six or ten byte command was
1870 * issued) if successful.
1873 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1874 unsigned char *buffer, int len, int timeout, int retries,
1875 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1877 unsigned char cmd[12];
1881 struct scsi_sense_hdr my_sshdr;
1883 memset(data, 0, sizeof(*data));
1884 memset(&cmd[0], 0, 12);
1885 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1888 /* caller might not be interested in sense, but we need it */
1893 use_10_for_ms = sdev->use_10_for_ms;
1895 if (use_10_for_ms) {
1899 cmd[0] = MODE_SENSE_10;
1906 cmd[0] = MODE_SENSE;
1911 memset(buffer, 0, len);
1913 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1914 sshdr, timeout, retries);
1916 /* This code looks awful: what it's doing is making sure an
1917 * ILLEGAL REQUEST sense return identifies the actual command
1918 * byte as the problem. MODE_SENSE commands can return
1919 * ILLEGAL REQUEST if the code page isn't supported */
1921 if (use_10_for_ms && !scsi_status_is_good(result) &&
1922 (driver_byte(result) & DRIVER_SENSE)) {
1923 if (scsi_sense_valid(sshdr)) {
1924 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1925 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1927 * Invalid command operation code
1929 sdev->use_10_for_ms = 0;
1935 if(scsi_status_is_good(result)) {
1936 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1937 (modepage == 6 || modepage == 8))) {
1938 /* Initio breakage? */
1941 data->medium_type = 0;
1942 data->device_specific = 0;
1944 data->block_descriptor_length = 0;
1945 } else if(use_10_for_ms) {
1946 data->length = buffer[0]*256 + buffer[1] + 2;
1947 data->medium_type = buffer[2];
1948 data->device_specific = buffer[3];
1949 data->longlba = buffer[4] & 0x01;
1950 data->block_descriptor_length = buffer[6]*256
1953 data->length = buffer[0] + 1;
1954 data->medium_type = buffer[1];
1955 data->device_specific = buffer[2];
1956 data->block_descriptor_length = buffer[3];
1958 data->header_length = header_length;
1963 EXPORT_SYMBOL(scsi_mode_sense);
1966 * scsi_test_unit_ready - test if unit is ready
1967 * @sdev: scsi device to change the state of.
1968 * @timeout: command timeout
1969 * @retries: number of retries before failing
1970 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1971 * returning sense. Make sure that this is cleared before passing
1974 * Returns zero if unsuccessful or an error if TUR failed. For
1975 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1976 * translated to success, with the ->changed flag updated.
1979 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1980 struct scsi_sense_hdr *sshdr_external)
1983 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1985 struct scsi_sense_hdr *sshdr;
1988 if (!sshdr_external)
1989 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1991 sshdr = sshdr_external;
1993 /* try to eat the UNIT_ATTENTION if there are enough retries */
1995 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
1997 } while ((driver_byte(result) & DRIVER_SENSE) &&
1998 sshdr && sshdr->sense_key == UNIT_ATTENTION &&
2002 /* could not allocate sense buffer, so can't process it */
2005 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
2007 if ((scsi_sense_valid(sshdr)) &&
2008 ((sshdr->sense_key == UNIT_ATTENTION) ||
2009 (sshdr->sense_key == NOT_READY))) {
2014 if (!sshdr_external)
2018 EXPORT_SYMBOL(scsi_test_unit_ready);
2021 * scsi_device_set_state - Take the given device through the device state model.
2022 * @sdev: scsi device to change the state of.
2023 * @state: state to change to.
2025 * Returns zero if unsuccessful or an error if the requested
2026 * transition is illegal.
2029 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2031 enum scsi_device_state oldstate = sdev->sdev_state;
2033 if (state == oldstate)
2038 /* There are no legal states that come back to
2039 * created. This is the manually initialised start
2113 sdev->sdev_state = state;
2117 SCSI_LOG_ERROR_RECOVERY(1,
2118 sdev_printk(KERN_ERR, sdev,
2119 "Illegal state transition %s->%s\n",
2120 scsi_device_state_name(oldstate),
2121 scsi_device_state_name(state))
2125 EXPORT_SYMBOL(scsi_device_set_state);
2128 * sdev_evt_emit - emit a single SCSI device uevent
2129 * @sdev: associated SCSI device
2130 * @evt: event to emit
2132 * Send a single uevent (scsi_event) to the associated scsi_device.
2134 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2139 switch (evt->evt_type) {
2140 case SDEV_EVT_MEDIA_CHANGE:
2141 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2151 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2155 * sdev_evt_thread - send a uevent for each scsi event
2156 * @work: work struct for scsi_device
2158 * Dispatch queued events to their associated scsi_device kobjects
2161 void scsi_evt_thread(struct work_struct *work)
2163 struct scsi_device *sdev;
2164 LIST_HEAD(event_list);
2166 sdev = container_of(work, struct scsi_device, event_work);
2169 struct scsi_event *evt;
2170 struct list_head *this, *tmp;
2171 unsigned long flags;
2173 spin_lock_irqsave(&sdev->list_lock, flags);
2174 list_splice_init(&sdev->event_list, &event_list);
2175 spin_unlock_irqrestore(&sdev->list_lock, flags);
2177 if (list_empty(&event_list))
2180 list_for_each_safe(this, tmp, &event_list) {
2181 evt = list_entry(this, struct scsi_event, node);
2182 list_del(&evt->node);
2183 scsi_evt_emit(sdev, evt);
2190 * sdev_evt_send - send asserted event to uevent thread
2191 * @sdev: scsi_device event occurred on
2192 * @evt: event to send
2194 * Assert scsi device event asynchronously.
2196 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2198 unsigned long flags;
2201 /* FIXME: currently this check eliminates all media change events
2202 * for polled devices. Need to update to discriminate between AN
2203 * and polled events */
2204 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2210 spin_lock_irqsave(&sdev->list_lock, flags);
2211 list_add_tail(&evt->node, &sdev->event_list);
2212 schedule_work(&sdev->event_work);
2213 spin_unlock_irqrestore(&sdev->list_lock, flags);
2215 EXPORT_SYMBOL_GPL(sdev_evt_send);
2218 * sdev_evt_alloc - allocate a new scsi event
2219 * @evt_type: type of event to allocate
2220 * @gfpflags: GFP flags for allocation
2222 * Allocates and returns a new scsi_event.
2224 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2227 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2231 evt->evt_type = evt_type;
2232 INIT_LIST_HEAD(&evt->node);
2234 /* evt_type-specific initialization, if any */
2236 case SDEV_EVT_MEDIA_CHANGE:
2244 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2247 * sdev_evt_send_simple - send asserted event to uevent thread
2248 * @sdev: scsi_device event occurred on
2249 * @evt_type: type of event to send
2250 * @gfpflags: GFP flags for allocation
2252 * Assert scsi device event asynchronously, given an event type.
2254 void sdev_evt_send_simple(struct scsi_device *sdev,
2255 enum scsi_device_event evt_type, gfp_t gfpflags)
2257 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2259 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2264 sdev_evt_send(sdev, evt);
2266 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2269 * scsi_device_quiesce - Block user issued commands.
2270 * @sdev: scsi device to quiesce.
2272 * This works by trying to transition to the SDEV_QUIESCE state
2273 * (which must be a legal transition). When the device is in this
2274 * state, only special requests will be accepted, all others will
2275 * be deferred. Since special requests may also be requeued requests,
2276 * a successful return doesn't guarantee the device will be
2277 * totally quiescent.
2279 * Must be called with user context, may sleep.
2281 * Returns zero if unsuccessful or an error if not.
2284 scsi_device_quiesce(struct scsi_device *sdev)
2286 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2290 scsi_run_queue(sdev->request_queue);
2291 while (sdev->device_busy) {
2292 msleep_interruptible(200);
2293 scsi_run_queue(sdev->request_queue);
2297 EXPORT_SYMBOL(scsi_device_quiesce);
2300 * scsi_device_resume - Restart user issued commands to a quiesced device.
2301 * @sdev: scsi device to resume.
2303 * Moves the device from quiesced back to running and restarts the
2306 * Must be called with user context, may sleep.
2309 scsi_device_resume(struct scsi_device *sdev)
2311 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2313 scsi_run_queue(sdev->request_queue);
2315 EXPORT_SYMBOL(scsi_device_resume);
2318 device_quiesce_fn(struct scsi_device *sdev, void *data)
2320 scsi_device_quiesce(sdev);
2324 scsi_target_quiesce(struct scsi_target *starget)
2326 starget_for_each_device(starget, NULL, device_quiesce_fn);
2328 EXPORT_SYMBOL(scsi_target_quiesce);
2331 device_resume_fn(struct scsi_device *sdev, void *data)
2333 scsi_device_resume(sdev);
2337 scsi_target_resume(struct scsi_target *starget)
2339 starget_for_each_device(starget, NULL, device_resume_fn);
2341 EXPORT_SYMBOL(scsi_target_resume);
2344 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2345 * @sdev: device to block
2347 * Block request made by scsi lld's to temporarily stop all
2348 * scsi commands on the specified device. Called from interrupt
2349 * or normal process context.
2351 * Returns zero if successful or error if not
2354 * This routine transitions the device to the SDEV_BLOCK state
2355 * (which must be a legal transition). When the device is in this
2356 * state, all commands are deferred until the scsi lld reenables
2357 * the device with scsi_device_unblock or device_block_tmo fires.
2358 * This routine assumes the host_lock is held on entry.
2361 scsi_internal_device_block(struct scsi_device *sdev)
2363 struct request_queue *q = sdev->request_queue;
2364 unsigned long flags;
2367 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2372 * The device has transitioned to SDEV_BLOCK. Stop the
2373 * block layer from calling the midlayer with this device's
2376 spin_lock_irqsave(q->queue_lock, flags);
2378 spin_unlock_irqrestore(q->queue_lock, flags);
2382 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2385 * scsi_internal_device_unblock - resume a device after a block request
2386 * @sdev: device to resume
2388 * Called by scsi lld's or the midlayer to restart the device queue
2389 * for the previously suspended scsi device. Called from interrupt or
2390 * normal process context.
2392 * Returns zero if successful or error if not.
2395 * This routine transitions the device to the SDEV_RUNNING state
2396 * (which must be a legal transition) allowing the midlayer to
2397 * goose the queue for this device. This routine assumes the
2398 * host_lock is held upon entry.
2401 scsi_internal_device_unblock(struct scsi_device *sdev)
2403 struct request_queue *q = sdev->request_queue;
2405 unsigned long flags;
2408 * Try to transition the scsi device to SDEV_RUNNING
2409 * and goose the device queue if successful.
2411 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2415 spin_lock_irqsave(q->queue_lock, flags);
2417 spin_unlock_irqrestore(q->queue_lock, flags);
2421 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2424 device_block(struct scsi_device *sdev, void *data)
2426 scsi_internal_device_block(sdev);
2430 target_block(struct device *dev, void *data)
2432 if (scsi_is_target_device(dev))
2433 starget_for_each_device(to_scsi_target(dev), NULL,
2439 scsi_target_block(struct device *dev)
2441 if (scsi_is_target_device(dev))
2442 starget_for_each_device(to_scsi_target(dev), NULL,
2445 device_for_each_child(dev, NULL, target_block);
2447 EXPORT_SYMBOL_GPL(scsi_target_block);
2450 device_unblock(struct scsi_device *sdev, void *data)
2452 scsi_internal_device_unblock(sdev);
2456 target_unblock(struct device *dev, void *data)
2458 if (scsi_is_target_device(dev))
2459 starget_for_each_device(to_scsi_target(dev), NULL,
2465 scsi_target_unblock(struct device *dev)
2467 if (scsi_is_target_device(dev))
2468 starget_for_each_device(to_scsi_target(dev), NULL,
2471 device_for_each_child(dev, NULL, target_unblock);
2473 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2476 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2477 * @sgl: scatter-gather list
2478 * @sg_count: number of segments in sg
2479 * @offset: offset in bytes into sg, on return offset into the mapped area
2480 * @len: bytes to map, on return number of bytes mapped
2482 * Returns virtual address of the start of the mapped page
2484 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2485 size_t *offset, size_t *len)
2488 size_t sg_len = 0, len_complete = 0;
2489 struct scatterlist *sg;
2492 WARN_ON(!irqs_disabled());
2494 for_each_sg(sgl, sg, sg_count, i) {
2495 len_complete = sg_len; /* Complete sg-entries */
2496 sg_len += sg->length;
2497 if (sg_len > *offset)
2501 if (unlikely(i == sg_count)) {
2502 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2504 __FUNCTION__, sg_len, *offset, sg_count);
2509 /* Offset starting from the beginning of first page in this sg-entry */
2510 *offset = *offset - len_complete + sg->offset;
2512 /* Assumption: contiguous pages can be accessed as "page + i" */
2513 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2514 *offset &= ~PAGE_MASK;
2516 /* Bytes in this sg-entry from *offset to the end of the page */
2517 sg_len = PAGE_SIZE - *offset;
2521 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2523 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2526 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2527 * @virt: virtual address to be unmapped
2529 void scsi_kunmap_atomic_sg(void *virt)
2531 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2533 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);