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 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
106 static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
108 struct Scsi_Host *host = cmd->device->host;
109 struct scsi_device *device = cmd->device;
110 struct scsi_target *starget = scsi_target(device);
111 struct request_queue *q = device->request_queue;
115 printk("Inserting command %p into mlqueue\n", cmd));
118 * Set the appropriate busy bit for the device/host.
120 * If the host/device isn't busy, assume that something actually
121 * completed, and that we should be able to queue a command now.
123 * Note that the prior mid-layer assumption that any host could
124 * always queue at least one command is now broken. The mid-layer
125 * will implement a user specifiable stall (see
126 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
127 * if a command is requeued with no other commands outstanding
128 * either for the device or for the host.
131 case SCSI_MLQUEUE_HOST_BUSY:
132 host->host_blocked = host->max_host_blocked;
134 case SCSI_MLQUEUE_DEVICE_BUSY:
135 device->device_blocked = device->max_device_blocked;
137 case SCSI_MLQUEUE_TARGET_BUSY:
138 starget->target_blocked = starget->max_target_blocked;
143 * Decrement the counters, since these commands are no longer
144 * active on the host/device.
147 scsi_device_unbusy(device);
150 * Requeue this command. It will go before all other commands
151 * that are already in the queue.
153 * NOTE: there is magic here about the way the queue is plugged if
154 * we have no outstanding commands.
156 * Although we *don't* plug the queue, we call the request
157 * function. The SCSI request function detects the blocked condition
158 * and plugs the queue appropriately.
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
170 * Function: scsi_queue_insert()
172 * Purpose: Insert a command in the midlevel queue.
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
177 * Lock status: Assumed that lock is not held upon entry.
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
188 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
190 return __scsi_queue_insert(cmd, reason, 1);
193 * scsi_execute - insert request and wait for the result
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
205 * returns the req->errors value which is the scsi_cmnd result
208 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209 int data_direction, void *buffer, unsigned bufflen,
210 unsigned char *sense, int timeout, int retries, int flags,
214 int write = (data_direction == DMA_TO_DEVICE);
215 int ret = DRIVER_ERROR << 24;
217 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
219 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
220 buffer, bufflen, __GFP_WAIT))
223 req->cmd_len = COMMAND_SIZE(cmd[0]);
224 memcpy(req->cmd, cmd, req->cmd_len);
227 req->retries = retries;
228 req->timeout = timeout;
229 req->cmd_type = REQ_TYPE_BLOCK_PC;
230 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
233 * head injection *required* here otherwise quiesce won't work
235 blk_execute_rq(req->q, NULL, req, 1);
238 * Some devices (USB mass-storage in particular) may transfer
239 * garbage data together with a residue indicating that the data
240 * is invalid. Prevent the garbage from being misinterpreted
241 * and prevent security leaks by zeroing out the excess data.
243 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
244 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
247 *resid = req->resid_len;
250 blk_put_request(req);
254 EXPORT_SYMBOL(scsi_execute);
257 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
258 int data_direction, void *buffer, unsigned bufflen,
259 struct scsi_sense_hdr *sshdr, int timeout, int retries,
266 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
268 return DRIVER_ERROR << 24;
270 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
271 sense, timeout, retries, 0, resid);
273 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
278 EXPORT_SYMBOL(scsi_execute_req);
281 * Function: scsi_init_cmd_errh()
283 * Purpose: Initialize cmd fields related to error handling.
285 * Arguments: cmd - command that is ready to be queued.
287 * Notes: This function has the job of initializing a number of
288 * fields related to error handling. Typically this will
289 * be called once for each command, as required.
291 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
293 cmd->serial_number = 0;
294 scsi_set_resid(cmd, 0);
295 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
296 if (cmd->cmd_len == 0)
297 cmd->cmd_len = scsi_command_size(cmd->cmnd);
300 void scsi_device_unbusy(struct scsi_device *sdev)
302 struct Scsi_Host *shost = sdev->host;
303 struct scsi_target *starget = scsi_target(sdev);
306 spin_lock_irqsave(shost->host_lock, flags);
308 starget->target_busy--;
309 if (unlikely(scsi_host_in_recovery(shost) &&
310 (shost->host_failed || shost->host_eh_scheduled)))
311 scsi_eh_wakeup(shost);
312 spin_unlock(shost->host_lock);
313 spin_lock(sdev->request_queue->queue_lock);
315 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
319 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
320 * and call blk_run_queue for all the scsi_devices on the target -
321 * including current_sdev first.
323 * Called with *no* scsi locks held.
325 static void scsi_single_lun_run(struct scsi_device *current_sdev)
327 struct Scsi_Host *shost = current_sdev->host;
328 struct scsi_device *sdev, *tmp;
329 struct scsi_target *starget = scsi_target(current_sdev);
332 spin_lock_irqsave(shost->host_lock, flags);
333 starget->starget_sdev_user = NULL;
334 spin_unlock_irqrestore(shost->host_lock, flags);
337 * Call blk_run_queue for all LUNs on the target, starting with
338 * current_sdev. We race with others (to set starget_sdev_user),
339 * but in most cases, we will be first. Ideally, each LU on the
340 * target would get some limited time or requests on the target.
342 blk_run_queue(current_sdev->request_queue);
344 spin_lock_irqsave(shost->host_lock, flags);
345 if (starget->starget_sdev_user)
347 list_for_each_entry_safe(sdev, tmp, &starget->devices,
348 same_target_siblings) {
349 if (sdev == current_sdev)
351 if (scsi_device_get(sdev))
354 spin_unlock_irqrestore(shost->host_lock, flags);
355 blk_run_queue(sdev->request_queue);
356 spin_lock_irqsave(shost->host_lock, flags);
358 scsi_device_put(sdev);
361 spin_unlock_irqrestore(shost->host_lock, flags);
364 static inline int scsi_device_is_busy(struct scsi_device *sdev)
366 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
372 static inline int scsi_target_is_busy(struct scsi_target *starget)
374 return ((starget->can_queue > 0 &&
375 starget->target_busy >= starget->can_queue) ||
376 starget->target_blocked);
379 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
381 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
382 shost->host_blocked || shost->host_self_blocked)
389 * Function: scsi_run_queue()
391 * Purpose: Select a proper request queue to serve next
393 * Arguments: q - last request's queue
397 * Notes: The previous command was completely finished, start
398 * a new one if possible.
400 static void scsi_run_queue(struct request_queue *q)
402 struct scsi_device *sdev = q->queuedata;
403 struct Scsi_Host *shost = sdev->host;
404 LIST_HEAD(starved_list);
407 if (scsi_target(sdev)->single_lun)
408 scsi_single_lun_run(sdev);
410 spin_lock_irqsave(shost->host_lock, flags);
411 list_splice_init(&shost->starved_list, &starved_list);
413 while (!list_empty(&starved_list)) {
417 * As long as shost is accepting commands and we have
418 * starved queues, call blk_run_queue. scsi_request_fn
419 * drops the queue_lock and can add us back to the
422 * host_lock protects the starved_list and starved_entry.
423 * scsi_request_fn must get the host_lock before checking
424 * or modifying starved_list or starved_entry.
426 if (scsi_host_is_busy(shost))
429 sdev = list_entry(starved_list.next,
430 struct scsi_device, starved_entry);
431 list_del_init(&sdev->starved_entry);
432 if (scsi_target_is_busy(scsi_target(sdev))) {
433 list_move_tail(&sdev->starved_entry,
434 &shost->starved_list);
438 spin_unlock(shost->host_lock);
440 spin_lock(sdev->request_queue->queue_lock);
441 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
442 !test_bit(QUEUE_FLAG_REENTER,
443 &sdev->request_queue->queue_flags);
445 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
446 __blk_run_queue(sdev->request_queue);
448 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
449 spin_unlock(sdev->request_queue->queue_lock);
451 spin_lock(shost->host_lock);
453 /* put any unprocessed entries back */
454 list_splice(&starved_list, &shost->starved_list);
455 spin_unlock_irqrestore(shost->host_lock, flags);
461 * Function: scsi_requeue_command()
463 * Purpose: Handle post-processing of completed commands.
465 * Arguments: q - queue to operate on
466 * cmd - command that may need to be requeued.
470 * Notes: After command completion, there may be blocks left
471 * over which weren't finished by the previous command
472 * this can be for a number of reasons - the main one is
473 * I/O errors in the middle of the request, in which case
474 * we need to request the blocks that come after the bad
476 * Notes: Upon return, cmd is a stale pointer.
478 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
480 struct request *req = cmd->request;
483 spin_lock_irqsave(q->queue_lock, flags);
484 scsi_unprep_request(req);
485 blk_requeue_request(q, req);
486 spin_unlock_irqrestore(q->queue_lock, flags);
491 void scsi_next_command(struct scsi_cmnd *cmd)
493 struct scsi_device *sdev = cmd->device;
494 struct request_queue *q = sdev->request_queue;
496 /* need to hold a reference on the device before we let go of the cmd */
497 get_device(&sdev->sdev_gendev);
499 scsi_put_command(cmd);
502 /* ok to remove device now */
503 put_device(&sdev->sdev_gendev);
506 void scsi_run_host_queues(struct Scsi_Host *shost)
508 struct scsi_device *sdev;
510 shost_for_each_device(sdev, shost)
511 scsi_run_queue(sdev->request_queue);
514 static void __scsi_release_buffers(struct scsi_cmnd *, int);
517 * Function: scsi_end_request()
519 * Purpose: Post-processing of completed commands (usually invoked at end
520 * of upper level post-processing and scsi_io_completion).
522 * Arguments: cmd - command that is complete.
523 * error - 0 if I/O indicates success, < 0 for I/O error.
524 * bytes - number of bytes of completed I/O
525 * requeue - indicates whether we should requeue leftovers.
527 * Lock status: Assumed that lock is not held upon entry.
529 * Returns: cmd if requeue required, NULL otherwise.
531 * Notes: This is called for block device requests in order to
532 * mark some number of sectors as complete.
534 * We are guaranteeing that the request queue will be goosed
535 * at some point during this call.
536 * Notes: If cmd was requeued, upon return it will be a stale pointer.
538 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
539 int bytes, int requeue)
541 struct request_queue *q = cmd->device->request_queue;
542 struct request *req = cmd->request;
545 * If there are blocks left over at the end, set up the command
546 * to queue the remainder of them.
548 if (blk_end_request(req, error, bytes)) {
549 int leftover = blk_rq_bytes(req);
551 if (blk_pc_request(req))
552 leftover = req->resid_len;
554 /* kill remainder if no retrys */
555 if (error && scsi_noretry_cmd(cmd))
556 blk_end_request(req, error, leftover);
560 * Bleah. Leftovers again. Stick the
561 * leftovers in the front of the
562 * queue, and goose the queue again.
564 scsi_release_buffers(cmd);
565 scsi_requeue_command(q, cmd);
573 * This will goose the queue request function at the end, so we don't
574 * need to worry about launching another command.
576 __scsi_release_buffers(cmd, 0);
577 scsi_next_command(cmd);
581 static inline unsigned int scsi_sgtable_index(unsigned short nents)
585 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
590 index = get_count_order(nents) - 3;
595 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
597 struct scsi_host_sg_pool *sgp;
599 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
600 mempool_free(sgl, sgp->pool);
603 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
605 struct scsi_host_sg_pool *sgp;
607 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
608 return mempool_alloc(sgp->pool, gfp_mask);
611 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
618 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
619 gfp_mask, scsi_sg_alloc);
621 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
627 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
629 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
632 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
635 if (cmd->sdb.table.nents)
636 scsi_free_sgtable(&cmd->sdb);
638 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
640 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
641 struct scsi_data_buffer *bidi_sdb =
642 cmd->request->next_rq->special;
643 scsi_free_sgtable(bidi_sdb);
644 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
645 cmd->request->next_rq->special = NULL;
648 if (scsi_prot_sg_count(cmd))
649 scsi_free_sgtable(cmd->prot_sdb);
653 * Function: scsi_release_buffers()
655 * Purpose: Completion processing for block device I/O requests.
657 * Arguments: cmd - command that we are bailing.
659 * Lock status: Assumed that no lock is held upon entry.
663 * Notes: In the event that an upper level driver rejects a
664 * command, we must release resources allocated during
665 * the __init_io() function. Primarily this would involve
666 * the scatter-gather table, and potentially any bounce
669 void scsi_release_buffers(struct scsi_cmnd *cmd)
671 __scsi_release_buffers(cmd, 1);
673 EXPORT_SYMBOL(scsi_release_buffers);
676 * Bidi commands Must be complete as a whole, both sides at once. If
677 * part of the bytes were written and lld returned scsi_in()->resid
678 * and/or scsi_out()->resid this information will be left in
679 * req->resid_len and req->next_rq->resid_len. The upper-layer driver
680 * can decide what to do with this information.
682 static void scsi_end_bidi_request(struct scsi_cmnd *cmd)
684 struct request *req = cmd->request;
686 req->resid_len = scsi_out(cmd)->resid;
687 req->next_rq->resid_len = scsi_in(cmd)->resid;
689 /* The req and req->next_rq have not been completed */
690 BUG_ON(blk_end_bidi_request(req, 0, blk_rq_bytes(req),
691 blk_rq_bytes(req->next_rq)));
693 scsi_release_buffers(cmd);
696 * This will goose the queue request function at the end, so we don't
697 * need to worry about launching another command.
699 scsi_next_command(cmd);
703 * Function: scsi_io_completion()
705 * Purpose: Completion processing for block device I/O requests.
707 * Arguments: cmd - command that is finished.
709 * Lock status: Assumed that no lock is held upon entry.
713 * Notes: This function is matched in terms of capabilities to
714 * the function that created the scatter-gather list.
715 * In other words, if there are no bounce buffers
716 * (the normal case for most drivers), we don't need
717 * the logic to deal with cleaning up afterwards.
719 * We must call scsi_end_request(). This will finish off
720 * the specified number of sectors. If we are done, the
721 * command block will be released and the queue function
722 * will be goosed. If we are not done then we have to
723 * figure out what to do next:
725 * a) We can call scsi_requeue_command(). The request
726 * will be unprepared and put back on the queue. Then
727 * a new command will be created for it. This should
728 * be used if we made forward progress, or if we want
729 * to switch from READ(10) to READ(6) for example.
731 * b) We can call scsi_queue_insert(). The request will
732 * be put back on the queue and retried using the same
733 * command as before, possibly after a delay.
735 * c) We can call blk_end_request() with -EIO to fail
736 * the remainder of the request.
738 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
740 int result = cmd->result;
742 struct request_queue *q = cmd->device->request_queue;
743 struct request *req = cmd->request;
745 struct scsi_sense_hdr sshdr;
747 int sense_deferred = 0;
748 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
749 ACTION_DELAYED_RETRY} action;
750 char *description = NULL;
753 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
755 sense_deferred = scsi_sense_is_deferred(&sshdr);
758 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
759 req->errors = result;
761 if (sense_valid && req->sense) {
763 * SG_IO wants current and deferred errors
765 int len = 8 + cmd->sense_buffer[7];
767 if (len > SCSI_SENSE_BUFFERSIZE)
768 len = SCSI_SENSE_BUFFERSIZE;
769 memcpy(req->sense, cmd->sense_buffer, len);
770 req->sense_len = len;
775 if (scsi_bidi_cmnd(cmd)) {
776 /* will also release_buffers */
777 scsi_end_bidi_request(cmd);
780 req->resid_len = scsi_get_resid(cmd);
783 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
786 * Next deal with any sectors which we were able to correctly
789 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
791 blk_rq_sectors(req), good_bytes));
794 * Recovered errors need reporting, but they're always treated
795 * as success, so fiddle the result code here. For BLOCK_PC
796 * we already took a copy of the original into rq->errors which
797 * is what gets returned to the user
799 if (sense_valid && sshdr.sense_key == RECOVERED_ERROR) {
800 if (!(req->cmd_flags & REQ_QUIET))
801 scsi_print_sense("", cmd);
803 /* BLOCK_PC may have set error */
808 * A number of bytes were successfully read. If there
809 * are leftovers and there is some kind of error
810 * (result != 0), retry the rest.
812 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
814 this_count = blk_rq_bytes(req);
818 if (host_byte(result) == DID_RESET) {
819 /* Third party bus reset or reset for error recovery
820 * reasons. Just retry the command and see what
823 action = ACTION_RETRY;
824 } else if (sense_valid && !sense_deferred) {
825 switch (sshdr.sense_key) {
827 if (cmd->device->removable) {
828 /* Detected disc change. Set a bit
829 * and quietly refuse further access.
831 cmd->device->changed = 1;
832 description = "Media Changed";
833 action = ACTION_FAIL;
835 /* Must have been a power glitch, or a
836 * bus reset. Could not have been a
837 * media change, so we just retry the
838 * command and see what happens.
840 action = ACTION_RETRY;
843 case ILLEGAL_REQUEST:
844 /* If we had an ILLEGAL REQUEST returned, then
845 * we may have performed an unsupported
846 * command. The only thing this should be
847 * would be a ten byte read where only a six
848 * byte read was supported. Also, on a system
849 * where READ CAPACITY failed, we may have
850 * read past the end of the disk.
852 if ((cmd->device->use_10_for_rw &&
853 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
854 (cmd->cmnd[0] == READ_10 ||
855 cmd->cmnd[0] == WRITE_10)) {
856 /* This will issue a new 6-byte command. */
857 cmd->device->use_10_for_rw = 0;
858 action = ACTION_REPREP;
859 } else if (sshdr.asc == 0x10) /* DIX */ {
860 description = "Host Data Integrity Failure";
861 action = ACTION_FAIL;
864 action = ACTION_FAIL;
866 case ABORTED_COMMAND:
867 action = ACTION_FAIL;
868 if (sshdr.asc == 0x10) { /* DIF */
869 description = "Target Data Integrity Failure";
874 /* If the device is in the process of becoming
875 * ready, or has a temporary blockage, retry.
877 if (sshdr.asc == 0x04) {
878 switch (sshdr.ascq) {
879 case 0x01: /* becoming ready */
880 case 0x04: /* format in progress */
881 case 0x05: /* rebuild in progress */
882 case 0x06: /* recalculation in progress */
883 case 0x07: /* operation in progress */
884 case 0x08: /* Long write in progress */
885 case 0x09: /* self test in progress */
886 action = ACTION_DELAYED_RETRY;
889 description = "Device not ready";
890 action = ACTION_FAIL;
894 description = "Device not ready";
895 action = ACTION_FAIL;
898 case VOLUME_OVERFLOW:
899 /* See SSC3rXX or current. */
900 action = ACTION_FAIL;
903 description = "Unhandled sense code";
904 action = ACTION_FAIL;
908 description = "Unhandled error code";
909 action = ACTION_FAIL;
914 /* Give up and fail the remainder of the request */
915 scsi_release_buffers(cmd);
916 if (!(req->cmd_flags & REQ_QUIET)) {
918 scmd_printk(KERN_INFO, cmd, "%s\n",
920 scsi_print_result(cmd);
921 if (driver_byte(result) & DRIVER_SENSE)
922 scsi_print_sense("", cmd);
924 blk_end_request_all(req, -EIO);
925 scsi_next_command(cmd);
928 /* Unprep the request and put it back at the head of the queue.
929 * A new command will be prepared and issued.
931 scsi_release_buffers(cmd);
932 scsi_requeue_command(q, cmd);
935 /* Retry the same command immediately */
936 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
938 case ACTION_DELAYED_RETRY:
939 /* Retry the same command after a delay */
940 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
945 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
951 * If sg table allocation fails, requeue request later.
953 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
955 return BLKPREP_DEFER;
961 * Next, walk the list, and fill in the addresses and sizes of
964 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
965 BUG_ON(count > sdb->table.nents);
966 sdb->table.nents = count;
967 sdb->length = blk_rq_bytes(req);
972 * Function: scsi_init_io()
974 * Purpose: SCSI I/O initialize function.
976 * Arguments: cmd - Command descriptor we wish to initialize
978 * Returns: 0 on success
979 * BLKPREP_DEFER if the failure is retryable
980 * BLKPREP_KILL if the failure is fatal
982 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
984 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
988 if (blk_bidi_rq(cmd->request)) {
989 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
990 scsi_sdb_cache, GFP_ATOMIC);
992 error = BLKPREP_DEFER;
996 cmd->request->next_rq->special = bidi_sdb;
997 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1003 if (blk_integrity_rq(cmd->request)) {
1004 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1007 BUG_ON(prot_sdb == NULL);
1008 ivecs = blk_rq_count_integrity_sg(cmd->request);
1010 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1011 error = BLKPREP_DEFER;
1015 count = blk_rq_map_integrity_sg(cmd->request,
1016 prot_sdb->table.sgl);
1017 BUG_ON(unlikely(count > ivecs));
1019 cmd->prot_sdb = prot_sdb;
1020 cmd->prot_sdb->table.nents = count;
1026 scsi_release_buffers(cmd);
1027 if (error == BLKPREP_KILL)
1028 scsi_put_command(cmd);
1029 else /* BLKPREP_DEFER */
1030 scsi_unprep_request(cmd->request);
1034 EXPORT_SYMBOL(scsi_init_io);
1036 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1037 struct request *req)
1039 struct scsi_cmnd *cmd;
1041 if (!req->special) {
1042 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1050 /* pull a tag out of the request if we have one */
1051 cmd->tag = req->tag;
1054 cmd->cmnd = req->cmd;
1059 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1061 struct scsi_cmnd *cmd;
1062 int ret = scsi_prep_state_check(sdev, req);
1064 if (ret != BLKPREP_OK)
1067 cmd = scsi_get_cmd_from_req(sdev, req);
1069 return BLKPREP_DEFER;
1072 * BLOCK_PC requests may transfer data, in which case they must
1073 * a bio attached to them. Or they might contain a SCSI command
1074 * that does not transfer data, in which case they may optionally
1075 * submit a request without an attached bio.
1080 BUG_ON(!req->nr_phys_segments);
1082 ret = scsi_init_io(cmd, GFP_ATOMIC);
1086 BUG_ON(blk_rq_bytes(req));
1088 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1092 cmd->cmd_len = req->cmd_len;
1093 if (!blk_rq_bytes(req))
1094 cmd->sc_data_direction = DMA_NONE;
1095 else if (rq_data_dir(req) == WRITE)
1096 cmd->sc_data_direction = DMA_TO_DEVICE;
1098 cmd->sc_data_direction = DMA_FROM_DEVICE;
1100 cmd->transfersize = blk_rq_bytes(req);
1101 cmd->allowed = req->retries;
1104 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1107 * Setup a REQ_TYPE_FS command. These are simple read/write request
1108 * from filesystems that still need to be translated to SCSI CDBs from
1111 int scsi_setup_fs_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 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1120 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1121 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1122 if (ret != BLKPREP_OK)
1127 * Filesystem requests must transfer data.
1129 BUG_ON(!req->nr_phys_segments);
1131 cmd = scsi_get_cmd_from_req(sdev, req);
1133 return BLKPREP_DEFER;
1135 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1136 return scsi_init_io(cmd, GFP_ATOMIC);
1138 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1140 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1142 int ret = BLKPREP_OK;
1145 * If the device is not in running state we will reject some
1148 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1149 switch (sdev->sdev_state) {
1152 * If the device is offline we refuse to process any
1153 * commands. The device must be brought online
1154 * before trying any recovery commands.
1156 sdev_printk(KERN_ERR, sdev,
1157 "rejecting I/O to offline device\n");
1162 * If the device is fully deleted, we refuse to
1163 * process any commands as well.
1165 sdev_printk(KERN_ERR, sdev,
1166 "rejecting I/O to dead device\n");
1171 case SDEV_CREATED_BLOCK:
1173 * If the devices is blocked we defer normal commands.
1175 if (!(req->cmd_flags & REQ_PREEMPT))
1176 ret = BLKPREP_DEFER;
1180 * For any other not fully online state we only allow
1181 * special commands. In particular any user initiated
1182 * command is not allowed.
1184 if (!(req->cmd_flags & REQ_PREEMPT))
1191 EXPORT_SYMBOL(scsi_prep_state_check);
1193 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1195 struct scsi_device *sdev = q->queuedata;
1199 req->errors = DID_NO_CONNECT << 16;
1200 /* release the command and kill it */
1202 struct scsi_cmnd *cmd = req->special;
1203 scsi_release_buffers(cmd);
1204 scsi_put_command(cmd);
1205 req->special = NULL;
1210 * If we defer, the elv_next_request() returns NULL, but the
1211 * queue must be restarted, so we plug here if no returning
1212 * command will automatically do that.
1214 if (sdev->device_busy == 0)
1218 req->cmd_flags |= REQ_DONTPREP;
1223 EXPORT_SYMBOL(scsi_prep_return);
1225 int scsi_prep_fn(struct request_queue *q, struct request *req)
1227 struct scsi_device *sdev = q->queuedata;
1228 int ret = BLKPREP_KILL;
1230 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1231 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1232 return scsi_prep_return(q, req, ret);
1236 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1239 * Called with the queue_lock held.
1241 static inline int scsi_dev_queue_ready(struct request_queue *q,
1242 struct scsi_device *sdev)
1244 if (sdev->device_busy == 0 && sdev->device_blocked) {
1246 * unblock after device_blocked iterates to zero
1248 if (--sdev->device_blocked == 0) {
1250 sdev_printk(KERN_INFO, sdev,
1251 "unblocking device at zero depth\n"));
1257 if (scsi_device_is_busy(sdev))
1265 * scsi_target_queue_ready: checks if there we can send commands to target
1266 * @sdev: scsi device on starget to check.
1268 * Called with the host lock held.
1270 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1271 struct scsi_device *sdev)
1273 struct scsi_target *starget = scsi_target(sdev);
1275 if (starget->single_lun) {
1276 if (starget->starget_sdev_user &&
1277 starget->starget_sdev_user != sdev)
1279 starget->starget_sdev_user = sdev;
1282 if (starget->target_busy == 0 && starget->target_blocked) {
1284 * unblock after target_blocked iterates to zero
1286 if (--starget->target_blocked == 0) {
1287 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1288 "unblocking target at zero depth\n"));
1290 blk_plug_device(sdev->request_queue);
1295 if (scsi_target_is_busy(starget)) {
1296 if (list_empty(&sdev->starved_entry)) {
1297 list_add_tail(&sdev->starved_entry,
1298 &shost->starved_list);
1303 /* We're OK to process the command, so we can't be starved */
1304 if (!list_empty(&sdev->starved_entry))
1305 list_del_init(&sdev->starved_entry);
1310 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1311 * return 0. We must end up running the queue again whenever 0 is
1312 * returned, else IO can hang.
1314 * Called with host_lock held.
1316 static inline int scsi_host_queue_ready(struct request_queue *q,
1317 struct Scsi_Host *shost,
1318 struct scsi_device *sdev)
1320 if (scsi_host_in_recovery(shost))
1322 if (shost->host_busy == 0 && shost->host_blocked) {
1324 * unblock after host_blocked iterates to zero
1326 if (--shost->host_blocked == 0) {
1328 printk("scsi%d unblocking host at zero depth\n",
1334 if (scsi_host_is_busy(shost)) {
1335 if (list_empty(&sdev->starved_entry))
1336 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1340 /* We're OK to process the command, so we can't be starved */
1341 if (!list_empty(&sdev->starved_entry))
1342 list_del_init(&sdev->starved_entry);
1348 * Busy state exporting function for request stacking drivers.
1350 * For efficiency, no lock is taken to check the busy state of
1351 * shost/starget/sdev, since the returned value is not guaranteed and
1352 * may be changed after request stacking drivers call the function,
1353 * regardless of taking lock or not.
1355 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1356 * (e.g. !sdev), scsi needs to return 'not busy'.
1357 * Otherwise, request stacking drivers may hold requests forever.
1359 static int scsi_lld_busy(struct request_queue *q)
1361 struct scsi_device *sdev = q->queuedata;
1362 struct Scsi_Host *shost;
1363 struct scsi_target *starget;
1369 starget = scsi_target(sdev);
1371 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1372 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1379 * Kill a request for a dead device
1381 static void scsi_kill_request(struct request *req, struct request_queue *q)
1383 struct scsi_cmnd *cmd = req->special;
1384 struct scsi_device *sdev = cmd->device;
1385 struct scsi_target *starget = scsi_target(sdev);
1386 struct Scsi_Host *shost = sdev->host;
1388 blkdev_dequeue_request(req);
1390 if (unlikely(cmd == NULL)) {
1391 printk(KERN_CRIT "impossible request in %s.\n",
1396 scsi_init_cmd_errh(cmd);
1397 cmd->result = DID_NO_CONNECT << 16;
1398 atomic_inc(&cmd->device->iorequest_cnt);
1401 * SCSI request completion path will do scsi_device_unbusy(),
1402 * bump busy counts. To bump the counters, we need to dance
1403 * with the locks as normal issue path does.
1405 sdev->device_busy++;
1406 spin_unlock(sdev->request_queue->queue_lock);
1407 spin_lock(shost->host_lock);
1409 starget->target_busy++;
1410 spin_unlock(shost->host_lock);
1411 spin_lock(sdev->request_queue->queue_lock);
1413 blk_complete_request(req);
1416 static void scsi_softirq_done(struct request *rq)
1418 struct scsi_cmnd *cmd = rq->special;
1419 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1422 INIT_LIST_HEAD(&cmd->eh_entry);
1425 * Set the serial numbers back to zero
1427 cmd->serial_number = 0;
1429 atomic_inc(&cmd->device->iodone_cnt);
1431 atomic_inc(&cmd->device->ioerr_cnt);
1433 disposition = scsi_decide_disposition(cmd);
1434 if (disposition != SUCCESS &&
1435 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1436 sdev_printk(KERN_ERR, cmd->device,
1437 "timing out command, waited %lus\n",
1439 disposition = SUCCESS;
1442 scsi_log_completion(cmd, disposition);
1444 switch (disposition) {
1446 scsi_finish_command(cmd);
1449 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1451 case ADD_TO_MLQUEUE:
1452 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1455 if (!scsi_eh_scmd_add(cmd, 0))
1456 scsi_finish_command(cmd);
1461 * Function: scsi_request_fn()
1463 * Purpose: Main strategy routine for SCSI.
1465 * Arguments: q - Pointer to actual queue.
1469 * Lock status: IO request lock assumed to be held when called.
1471 static void scsi_request_fn(struct request_queue *q)
1473 struct scsi_device *sdev = q->queuedata;
1474 struct Scsi_Host *shost;
1475 struct scsi_cmnd *cmd;
1476 struct request *req;
1479 printk("scsi: killing requests for dead queue\n");
1480 while ((req = elv_next_request(q)) != NULL)
1481 scsi_kill_request(req, q);
1485 if(!get_device(&sdev->sdev_gendev))
1486 /* We must be tearing the block queue down already */
1490 * To start with, we keep looping until the queue is empty, or until
1491 * the host is no longer able to accept any more requests.
1494 while (!blk_queue_plugged(q)) {
1497 * get next queueable request. We do this early to make sure
1498 * that the request is fully prepared even if we cannot
1501 req = elv_next_request(q);
1502 if (!req || !scsi_dev_queue_ready(q, sdev))
1505 if (unlikely(!scsi_device_online(sdev))) {
1506 sdev_printk(KERN_ERR, sdev,
1507 "rejecting I/O to offline device\n");
1508 scsi_kill_request(req, q);
1514 * Remove the request from the request list.
1516 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1517 blkdev_dequeue_request(req);
1518 sdev->device_busy++;
1520 spin_unlock(q->queue_lock);
1522 if (unlikely(cmd == NULL)) {
1523 printk(KERN_CRIT "impossible request in %s.\n"
1524 "please mail a stack trace to "
1525 "linux-scsi@vger.kernel.org\n",
1527 blk_dump_rq_flags(req, "foo");
1530 spin_lock(shost->host_lock);
1533 * We hit this when the driver is using a host wide
1534 * tag map. For device level tag maps the queue_depth check
1535 * in the device ready fn would prevent us from trying
1536 * to allocate a tag. Since the map is a shared host resource
1537 * we add the dev to the starved list so it eventually gets
1538 * a run when a tag is freed.
1540 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1541 if (list_empty(&sdev->starved_entry))
1542 list_add_tail(&sdev->starved_entry,
1543 &shost->starved_list);
1547 if (!scsi_target_queue_ready(shost, sdev))
1550 if (!scsi_host_queue_ready(q, shost, sdev))
1553 scsi_target(sdev)->target_busy++;
1557 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1558 * take the lock again.
1560 spin_unlock_irq(shost->host_lock);
1563 * Finally, initialize any error handling parameters, and set up
1564 * the timers for timeouts.
1566 scsi_init_cmd_errh(cmd);
1569 * Dispatch the command to the low-level driver.
1571 rtn = scsi_dispatch_cmd(cmd);
1572 spin_lock_irq(q->queue_lock);
1574 /* we're refusing the command; because of
1575 * the way locks get dropped, we need to
1576 * check here if plugging is required */
1577 if(sdev->device_busy == 0)
1587 spin_unlock_irq(shost->host_lock);
1590 * lock q, handle tag, requeue req, and decrement device_busy. We
1591 * must return with queue_lock held.
1593 * Decrementing device_busy without checking it is OK, as all such
1594 * cases (host limits or settings) should run the queue at some
1597 spin_lock_irq(q->queue_lock);
1598 blk_requeue_request(q, req);
1599 sdev->device_busy--;
1600 if(sdev->device_busy == 0)
1603 /* must be careful here...if we trigger the ->remove() function
1604 * we cannot be holding the q lock */
1605 spin_unlock_irq(q->queue_lock);
1606 put_device(&sdev->sdev_gendev);
1607 spin_lock_irq(q->queue_lock);
1610 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1612 struct device *host_dev;
1613 u64 bounce_limit = 0xffffffff;
1615 if (shost->unchecked_isa_dma)
1616 return BLK_BOUNCE_ISA;
1618 * Platforms with virtual-DMA translation
1619 * hardware have no practical limit.
1621 if (!PCI_DMA_BUS_IS_PHYS)
1622 return BLK_BOUNCE_ANY;
1624 host_dev = scsi_get_device(shost);
1625 if (host_dev && host_dev->dma_mask)
1626 bounce_limit = *host_dev->dma_mask;
1628 return bounce_limit;
1630 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1632 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1633 request_fn_proc *request_fn)
1635 struct request_queue *q;
1636 struct device *dev = shost->shost_gendev.parent;
1638 q = blk_init_queue(request_fn, NULL);
1643 * this limit is imposed by hardware restrictions
1645 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1646 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1648 blk_queue_max_sectors(q, shost->max_sectors);
1649 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1650 blk_queue_segment_boundary(q, shost->dma_boundary);
1651 dma_set_seg_boundary(dev, shost->dma_boundary);
1653 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1655 /* New queue, no concurrency on queue_flags */
1656 if (!shost->use_clustering)
1657 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1660 * set a reasonable default alignment on word boundaries: the
1661 * host and device may alter it using
1662 * blk_queue_update_dma_alignment() later.
1664 blk_queue_dma_alignment(q, 0x03);
1668 EXPORT_SYMBOL(__scsi_alloc_queue);
1670 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1672 struct request_queue *q;
1674 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1678 blk_queue_prep_rq(q, scsi_prep_fn);
1679 blk_queue_softirq_done(q, scsi_softirq_done);
1680 blk_queue_rq_timed_out(q, scsi_times_out);
1681 blk_queue_lld_busy(q, scsi_lld_busy);
1685 void scsi_free_queue(struct request_queue *q)
1687 blk_cleanup_queue(q);
1691 * Function: scsi_block_requests()
1693 * Purpose: Utility function used by low-level drivers to prevent further
1694 * commands from being queued to the device.
1696 * Arguments: shost - Host in question
1700 * Lock status: No locks are assumed held.
1702 * Notes: There is no timer nor any other means by which the requests
1703 * get unblocked other than the low-level driver calling
1704 * scsi_unblock_requests().
1706 void scsi_block_requests(struct Scsi_Host *shost)
1708 shost->host_self_blocked = 1;
1710 EXPORT_SYMBOL(scsi_block_requests);
1713 * Function: scsi_unblock_requests()
1715 * Purpose: Utility function used by low-level drivers to allow further
1716 * commands from being queued to the device.
1718 * Arguments: shost - Host in question
1722 * Lock status: No locks are assumed held.
1724 * Notes: There is no timer nor any other means by which the requests
1725 * get unblocked other than the low-level driver calling
1726 * scsi_unblock_requests().
1728 * This is done as an API function so that changes to the
1729 * internals of the scsi mid-layer won't require wholesale
1730 * changes to drivers that use this feature.
1732 void scsi_unblock_requests(struct Scsi_Host *shost)
1734 shost->host_self_blocked = 0;
1735 scsi_run_host_queues(shost);
1737 EXPORT_SYMBOL(scsi_unblock_requests);
1739 int __init scsi_init_queue(void)
1743 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1744 sizeof(struct scsi_data_buffer),
1746 if (!scsi_sdb_cache) {
1747 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1751 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1752 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1753 int size = sgp->size * sizeof(struct scatterlist);
1755 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1756 SLAB_HWCACHE_ALIGN, NULL);
1758 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1763 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1766 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1775 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1776 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1778 mempool_destroy(sgp->pool);
1780 kmem_cache_destroy(sgp->slab);
1782 kmem_cache_destroy(scsi_sdb_cache);
1787 void scsi_exit_queue(void)
1791 kmem_cache_destroy(scsi_sdb_cache);
1793 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1794 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1795 mempool_destroy(sgp->pool);
1796 kmem_cache_destroy(sgp->slab);
1801 * scsi_mode_select - issue a mode select
1802 * @sdev: SCSI device to be queried
1803 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1804 * @sp: Save page bit (0 == don't save, 1 == save)
1805 * @modepage: mode page being requested
1806 * @buffer: request buffer (may not be smaller than eight bytes)
1807 * @len: length of request buffer.
1808 * @timeout: command timeout
1809 * @retries: number of retries before failing
1810 * @data: returns a structure abstracting the mode header data
1811 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1812 * must be SCSI_SENSE_BUFFERSIZE big.
1814 * Returns zero if successful; negative error number or scsi
1819 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1820 unsigned char *buffer, int len, int timeout, int retries,
1821 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1823 unsigned char cmd[10];
1824 unsigned char *real_buffer;
1827 memset(cmd, 0, sizeof(cmd));
1828 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1830 if (sdev->use_10_for_ms) {
1833 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1836 memcpy(real_buffer + 8, buffer, len);
1840 real_buffer[2] = data->medium_type;
1841 real_buffer[3] = data->device_specific;
1842 real_buffer[4] = data->longlba ? 0x01 : 0;
1844 real_buffer[6] = data->block_descriptor_length >> 8;
1845 real_buffer[7] = data->block_descriptor_length;
1847 cmd[0] = MODE_SELECT_10;
1851 if (len > 255 || data->block_descriptor_length > 255 ||
1855 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1858 memcpy(real_buffer + 4, buffer, len);
1861 real_buffer[1] = data->medium_type;
1862 real_buffer[2] = data->device_specific;
1863 real_buffer[3] = data->block_descriptor_length;
1866 cmd[0] = MODE_SELECT;
1870 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1871 sshdr, timeout, retries, NULL);
1875 EXPORT_SYMBOL_GPL(scsi_mode_select);
1878 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1879 * @sdev: SCSI device to be queried
1880 * @dbd: set if mode sense will allow block descriptors to be returned
1881 * @modepage: mode page being requested
1882 * @buffer: request buffer (may not be smaller than eight bytes)
1883 * @len: length of request buffer.
1884 * @timeout: command timeout
1885 * @retries: number of retries before failing
1886 * @data: returns a structure abstracting the mode header data
1887 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1888 * must be SCSI_SENSE_BUFFERSIZE big.
1890 * Returns zero if unsuccessful, or the header offset (either 4
1891 * or 8 depending on whether a six or ten byte command was
1892 * issued) if successful.
1895 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1896 unsigned char *buffer, int len, int timeout, int retries,
1897 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1899 unsigned char cmd[12];
1903 struct scsi_sense_hdr my_sshdr;
1905 memset(data, 0, sizeof(*data));
1906 memset(&cmd[0], 0, 12);
1907 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1910 /* caller might not be interested in sense, but we need it */
1915 use_10_for_ms = sdev->use_10_for_ms;
1917 if (use_10_for_ms) {
1921 cmd[0] = MODE_SENSE_10;
1928 cmd[0] = MODE_SENSE;
1933 memset(buffer, 0, len);
1935 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1936 sshdr, timeout, retries, NULL);
1938 /* This code looks awful: what it's doing is making sure an
1939 * ILLEGAL REQUEST sense return identifies the actual command
1940 * byte as the problem. MODE_SENSE commands can return
1941 * ILLEGAL REQUEST if the code page isn't supported */
1943 if (use_10_for_ms && !scsi_status_is_good(result) &&
1944 (driver_byte(result) & DRIVER_SENSE)) {
1945 if (scsi_sense_valid(sshdr)) {
1946 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1947 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1949 * Invalid command operation code
1951 sdev->use_10_for_ms = 0;
1957 if(scsi_status_is_good(result)) {
1958 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1959 (modepage == 6 || modepage == 8))) {
1960 /* Initio breakage? */
1963 data->medium_type = 0;
1964 data->device_specific = 0;
1966 data->block_descriptor_length = 0;
1967 } else if(use_10_for_ms) {
1968 data->length = buffer[0]*256 + buffer[1] + 2;
1969 data->medium_type = buffer[2];
1970 data->device_specific = buffer[3];
1971 data->longlba = buffer[4] & 0x01;
1972 data->block_descriptor_length = buffer[6]*256
1975 data->length = buffer[0] + 1;
1976 data->medium_type = buffer[1];
1977 data->device_specific = buffer[2];
1978 data->block_descriptor_length = buffer[3];
1980 data->header_length = header_length;
1985 EXPORT_SYMBOL(scsi_mode_sense);
1988 * scsi_test_unit_ready - test if unit is ready
1989 * @sdev: scsi device to change the state of.
1990 * @timeout: command timeout
1991 * @retries: number of retries before failing
1992 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1993 * returning sense. Make sure that this is cleared before passing
1996 * Returns zero if unsuccessful or an error if TUR failed. For
1997 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1998 * translated to success, with the ->changed flag updated.
2001 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2002 struct scsi_sense_hdr *sshdr_external)
2005 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2007 struct scsi_sense_hdr *sshdr;
2010 if (!sshdr_external)
2011 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2013 sshdr = sshdr_external;
2015 /* try to eat the UNIT_ATTENTION if there are enough retries */
2017 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2018 timeout, retries, NULL);
2019 if (sdev->removable && scsi_sense_valid(sshdr) &&
2020 sshdr->sense_key == UNIT_ATTENTION)
2022 } while (scsi_sense_valid(sshdr) &&
2023 sshdr->sense_key == UNIT_ATTENTION && --retries);
2026 /* could not allocate sense buffer, so can't process it */
2029 if (sdev->removable && scsi_sense_valid(sshdr) &&
2030 (sshdr->sense_key == UNIT_ATTENTION ||
2031 sshdr->sense_key == NOT_READY)) {
2035 if (!sshdr_external)
2039 EXPORT_SYMBOL(scsi_test_unit_ready);
2042 * scsi_device_set_state - Take the given device through the device state model.
2043 * @sdev: scsi device to change the state of.
2044 * @state: state to change to.
2046 * Returns zero if unsuccessful or an error if the requested
2047 * transition is illegal.
2050 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2052 enum scsi_device_state oldstate = sdev->sdev_state;
2054 if (state == oldstate)
2060 case SDEV_CREATED_BLOCK:
2104 case SDEV_CREATED_BLOCK:
2111 case SDEV_CREATED_BLOCK:
2146 sdev->sdev_state = state;
2150 SCSI_LOG_ERROR_RECOVERY(1,
2151 sdev_printk(KERN_ERR, sdev,
2152 "Illegal state transition %s->%s\n",
2153 scsi_device_state_name(oldstate),
2154 scsi_device_state_name(state))
2158 EXPORT_SYMBOL(scsi_device_set_state);
2161 * sdev_evt_emit - emit a single SCSI device uevent
2162 * @sdev: associated SCSI device
2163 * @evt: event to emit
2165 * Send a single uevent (scsi_event) to the associated scsi_device.
2167 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2172 switch (evt->evt_type) {
2173 case SDEV_EVT_MEDIA_CHANGE:
2174 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2184 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2188 * sdev_evt_thread - send a uevent for each scsi event
2189 * @work: work struct for scsi_device
2191 * Dispatch queued events to their associated scsi_device kobjects
2194 void scsi_evt_thread(struct work_struct *work)
2196 struct scsi_device *sdev;
2197 LIST_HEAD(event_list);
2199 sdev = container_of(work, struct scsi_device, event_work);
2202 struct scsi_event *evt;
2203 struct list_head *this, *tmp;
2204 unsigned long flags;
2206 spin_lock_irqsave(&sdev->list_lock, flags);
2207 list_splice_init(&sdev->event_list, &event_list);
2208 spin_unlock_irqrestore(&sdev->list_lock, flags);
2210 if (list_empty(&event_list))
2213 list_for_each_safe(this, tmp, &event_list) {
2214 evt = list_entry(this, struct scsi_event, node);
2215 list_del(&evt->node);
2216 scsi_evt_emit(sdev, evt);
2223 * sdev_evt_send - send asserted event to uevent thread
2224 * @sdev: scsi_device event occurred on
2225 * @evt: event to send
2227 * Assert scsi device event asynchronously.
2229 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2231 unsigned long flags;
2234 /* FIXME: currently this check eliminates all media change events
2235 * for polled devices. Need to update to discriminate between AN
2236 * and polled events */
2237 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2243 spin_lock_irqsave(&sdev->list_lock, flags);
2244 list_add_tail(&evt->node, &sdev->event_list);
2245 schedule_work(&sdev->event_work);
2246 spin_unlock_irqrestore(&sdev->list_lock, flags);
2248 EXPORT_SYMBOL_GPL(sdev_evt_send);
2251 * sdev_evt_alloc - allocate a new scsi event
2252 * @evt_type: type of event to allocate
2253 * @gfpflags: GFP flags for allocation
2255 * Allocates and returns a new scsi_event.
2257 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2260 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2264 evt->evt_type = evt_type;
2265 INIT_LIST_HEAD(&evt->node);
2267 /* evt_type-specific initialization, if any */
2269 case SDEV_EVT_MEDIA_CHANGE:
2277 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2280 * sdev_evt_send_simple - send asserted event to uevent thread
2281 * @sdev: scsi_device event occurred on
2282 * @evt_type: type of event to send
2283 * @gfpflags: GFP flags for allocation
2285 * Assert scsi device event asynchronously, given an event type.
2287 void sdev_evt_send_simple(struct scsi_device *sdev,
2288 enum scsi_device_event evt_type, gfp_t gfpflags)
2290 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2292 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2297 sdev_evt_send(sdev, evt);
2299 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2302 * scsi_device_quiesce - Block user issued commands.
2303 * @sdev: scsi device to quiesce.
2305 * This works by trying to transition to the SDEV_QUIESCE state
2306 * (which must be a legal transition). When the device is in this
2307 * state, only special requests will be accepted, all others will
2308 * be deferred. Since special requests may also be requeued requests,
2309 * a successful return doesn't guarantee the device will be
2310 * totally quiescent.
2312 * Must be called with user context, may sleep.
2314 * Returns zero if unsuccessful or an error if not.
2317 scsi_device_quiesce(struct scsi_device *sdev)
2319 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2323 scsi_run_queue(sdev->request_queue);
2324 while (sdev->device_busy) {
2325 msleep_interruptible(200);
2326 scsi_run_queue(sdev->request_queue);
2330 EXPORT_SYMBOL(scsi_device_quiesce);
2333 * scsi_device_resume - Restart user issued commands to a quiesced device.
2334 * @sdev: scsi device to resume.
2336 * Moves the device from quiesced back to running and restarts the
2339 * Must be called with user context, may sleep.
2342 scsi_device_resume(struct scsi_device *sdev)
2344 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2346 scsi_run_queue(sdev->request_queue);
2348 EXPORT_SYMBOL(scsi_device_resume);
2351 device_quiesce_fn(struct scsi_device *sdev, void *data)
2353 scsi_device_quiesce(sdev);
2357 scsi_target_quiesce(struct scsi_target *starget)
2359 starget_for_each_device(starget, NULL, device_quiesce_fn);
2361 EXPORT_SYMBOL(scsi_target_quiesce);
2364 device_resume_fn(struct scsi_device *sdev, void *data)
2366 scsi_device_resume(sdev);
2370 scsi_target_resume(struct scsi_target *starget)
2372 starget_for_each_device(starget, NULL, device_resume_fn);
2374 EXPORT_SYMBOL(scsi_target_resume);
2377 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2378 * @sdev: device to block
2380 * Block request made by scsi lld's to temporarily stop all
2381 * scsi commands on the specified device. Called from interrupt
2382 * or normal process context.
2384 * Returns zero if successful or error if not
2387 * This routine transitions the device to the SDEV_BLOCK state
2388 * (which must be a legal transition). When the device is in this
2389 * state, all commands are deferred until the scsi lld reenables
2390 * the device with scsi_device_unblock or device_block_tmo fires.
2391 * This routine assumes the host_lock is held on entry.
2394 scsi_internal_device_block(struct scsi_device *sdev)
2396 struct request_queue *q = sdev->request_queue;
2397 unsigned long flags;
2400 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2402 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2409 * The device has transitioned to SDEV_BLOCK. Stop the
2410 * block layer from calling the midlayer with this device's
2413 spin_lock_irqsave(q->queue_lock, flags);
2415 spin_unlock_irqrestore(q->queue_lock, flags);
2419 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2422 * scsi_internal_device_unblock - resume a device after a block request
2423 * @sdev: device to resume
2425 * Called by scsi lld's or the midlayer to restart the device queue
2426 * for the previously suspended scsi device. Called from interrupt or
2427 * normal process context.
2429 * Returns zero if successful or error if not.
2432 * This routine transitions the device to the SDEV_RUNNING state
2433 * (which must be a legal transition) allowing the midlayer to
2434 * goose the queue for this device. This routine assumes the
2435 * host_lock is held upon entry.
2438 scsi_internal_device_unblock(struct scsi_device *sdev)
2440 struct request_queue *q = sdev->request_queue;
2442 unsigned long flags;
2445 * Try to transition the scsi device to SDEV_RUNNING
2446 * and goose the device queue if successful.
2448 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2450 err = scsi_device_set_state(sdev, SDEV_CREATED);
2456 spin_lock_irqsave(q->queue_lock, flags);
2458 spin_unlock_irqrestore(q->queue_lock, flags);
2462 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2465 device_block(struct scsi_device *sdev, void *data)
2467 scsi_internal_device_block(sdev);
2471 target_block(struct device *dev, void *data)
2473 if (scsi_is_target_device(dev))
2474 starget_for_each_device(to_scsi_target(dev), NULL,
2480 scsi_target_block(struct device *dev)
2482 if (scsi_is_target_device(dev))
2483 starget_for_each_device(to_scsi_target(dev), NULL,
2486 device_for_each_child(dev, NULL, target_block);
2488 EXPORT_SYMBOL_GPL(scsi_target_block);
2491 device_unblock(struct scsi_device *sdev, void *data)
2493 scsi_internal_device_unblock(sdev);
2497 target_unblock(struct device *dev, void *data)
2499 if (scsi_is_target_device(dev))
2500 starget_for_each_device(to_scsi_target(dev), NULL,
2506 scsi_target_unblock(struct device *dev)
2508 if (scsi_is_target_device(dev))
2509 starget_for_each_device(to_scsi_target(dev), NULL,
2512 device_for_each_child(dev, NULL, target_unblock);
2514 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2517 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2518 * @sgl: scatter-gather list
2519 * @sg_count: number of segments in sg
2520 * @offset: offset in bytes into sg, on return offset into the mapped area
2521 * @len: bytes to map, on return number of bytes mapped
2523 * Returns virtual address of the start of the mapped page
2525 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2526 size_t *offset, size_t *len)
2529 size_t sg_len = 0, len_complete = 0;
2530 struct scatterlist *sg;
2533 WARN_ON(!irqs_disabled());
2535 for_each_sg(sgl, sg, sg_count, i) {
2536 len_complete = sg_len; /* Complete sg-entries */
2537 sg_len += sg->length;
2538 if (sg_len > *offset)
2542 if (unlikely(i == sg_count)) {
2543 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2545 __func__, sg_len, *offset, sg_count);
2550 /* Offset starting from the beginning of first page in this sg-entry */
2551 *offset = *offset - len_complete + sg->offset;
2553 /* Assumption: contiguous pages can be accessed as "page + i" */
2554 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2555 *offset &= ~PAGE_MASK;
2557 /* Bytes in this sg-entry from *offset to the end of the page */
2558 sg_len = PAGE_SIZE - *offset;
2562 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2564 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2567 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2568 * @virt: virtual address to be unmapped
2570 void scsi_kunmap_atomic_sg(void *virt)
2572 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2574 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);