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;
71 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
72 * not change behaviour from the previous unplug mechanism, experimentation
73 * may prove this needs changing.
75 #define SCSI_QUEUE_DELAY 3
77 static void scsi_run_queue(struct request_queue *q);
80 * Function: scsi_unprep_request()
82 * Purpose: Remove all preparation done for a request, including its
83 * associated scsi_cmnd, so that it can be requeued.
85 * Arguments: req - request to unprepare
87 * Lock status: Assumed that no locks are held upon entry.
91 static void scsi_unprep_request(struct request *req)
93 struct scsi_cmnd *cmd = req->special;
95 blk_unprep_request(req);
98 scsi_put_command(cmd);
102 * __scsi_queue_insert - private queue insertion
103 * @cmd: The SCSI command being requeued
104 * @reason: The reason for the requeue
105 * @unbusy: Whether the queue should be unbusied
107 * This is a private queue insertion. The public interface
108 * scsi_queue_insert() always assumes the queue should be unbusied
109 * because it's always called before the completion. This function is
110 * for a requeue after completion, which should only occur in this
113 static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
115 struct Scsi_Host *host = cmd->device->host;
116 struct scsi_device *device = cmd->device;
117 struct scsi_target *starget = scsi_target(device);
118 struct request_queue *q = device->request_queue;
122 printk("Inserting command %p into mlqueue\n", cmd));
125 * Set the appropriate busy bit for the device/host.
127 * If the host/device isn't busy, assume that something actually
128 * completed, and that we should be able to queue a command now.
130 * Note that the prior mid-layer assumption that any host could
131 * always queue at least one command is now broken. The mid-layer
132 * will implement a user specifiable stall (see
133 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
134 * if a command is requeued with no other commands outstanding
135 * either for the device or for the host.
138 case SCSI_MLQUEUE_HOST_BUSY:
139 host->host_blocked = host->max_host_blocked;
141 case SCSI_MLQUEUE_DEVICE_BUSY:
142 device->device_blocked = device->max_device_blocked;
144 case SCSI_MLQUEUE_TARGET_BUSY:
145 starget->target_blocked = starget->max_target_blocked;
150 * Decrement the counters, since these commands are no longer
151 * active on the host/device.
154 scsi_device_unbusy(device);
157 * Requeue this command. It will go before all other commands
158 * that are already in the queue.
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)) {
415 * As long as shost is accepting commands and we have
416 * starved queues, call blk_run_queue. scsi_request_fn
417 * drops the queue_lock and can add us back to the
420 * host_lock protects the starved_list and starved_entry.
421 * scsi_request_fn must get the host_lock before checking
422 * or modifying starved_list or starved_entry.
424 if (scsi_host_is_busy(shost))
427 sdev = list_entry(starved_list.next,
428 struct scsi_device, starved_entry);
429 list_del_init(&sdev->starved_entry);
430 if (scsi_target_is_busy(scsi_target(sdev))) {
431 list_move_tail(&sdev->starved_entry,
432 &shost->starved_list);
436 blk_run_queue_async(sdev->request_queue);
438 /* put any unprocessed entries back */
439 list_splice(&starved_list, &shost->starved_list);
440 spin_unlock_irqrestore(shost->host_lock, flags);
446 * Function: scsi_requeue_command()
448 * Purpose: Handle post-processing of completed commands.
450 * Arguments: q - queue to operate on
451 * cmd - command that may need to be requeued.
455 * Notes: After command completion, there may be blocks left
456 * over which weren't finished by the previous command
457 * this can be for a number of reasons - the main one is
458 * I/O errors in the middle of the request, in which case
459 * we need to request the blocks that come after the bad
461 * Notes: Upon return, cmd is a stale pointer.
463 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
465 struct request *req = cmd->request;
468 spin_lock_irqsave(q->queue_lock, flags);
469 scsi_unprep_request(req);
470 blk_requeue_request(q, req);
471 spin_unlock_irqrestore(q->queue_lock, flags);
476 void scsi_next_command(struct scsi_cmnd *cmd)
478 struct scsi_device *sdev = cmd->device;
479 struct request_queue *q = sdev->request_queue;
481 /* need to hold a reference on the device before we let go of the cmd */
482 get_device(&sdev->sdev_gendev);
484 scsi_put_command(cmd);
487 /* ok to remove device now */
488 put_device(&sdev->sdev_gendev);
491 void scsi_run_host_queues(struct Scsi_Host *shost)
493 struct scsi_device *sdev;
495 shost_for_each_device(sdev, shost)
496 scsi_run_queue(sdev->request_queue);
499 static void __scsi_release_buffers(struct scsi_cmnd *, int);
502 * Function: scsi_end_request()
504 * Purpose: Post-processing of completed commands (usually invoked at end
505 * of upper level post-processing and scsi_io_completion).
507 * Arguments: cmd - command that is complete.
508 * error - 0 if I/O indicates success, < 0 for I/O error.
509 * bytes - number of bytes of completed I/O
510 * requeue - indicates whether we should requeue leftovers.
512 * Lock status: Assumed that lock is not held upon entry.
514 * Returns: cmd if requeue required, NULL otherwise.
516 * Notes: This is called for block device requests in order to
517 * mark some number of sectors as complete.
519 * We are guaranteeing that the request queue will be goosed
520 * at some point during this call.
521 * Notes: If cmd was requeued, upon return it will be a stale pointer.
523 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
524 int bytes, int requeue)
526 struct request_queue *q = cmd->device->request_queue;
527 struct request *req = cmd->request;
530 * If there are blocks left over at the end, set up the command
531 * to queue the remainder of them.
533 if (blk_end_request(req, error, bytes)) {
534 /* kill remainder if no retrys */
535 if (error && scsi_noretry_cmd(cmd))
536 blk_end_request_all(req, error);
540 * Bleah. Leftovers again. Stick the
541 * leftovers in the front of the
542 * queue, and goose the queue again.
544 scsi_release_buffers(cmd);
545 scsi_requeue_command(q, cmd);
553 * This will goose the queue request function at the end, so we don't
554 * need to worry about launching another command.
556 __scsi_release_buffers(cmd, 0);
557 scsi_next_command(cmd);
561 static inline unsigned int scsi_sgtable_index(unsigned short nents)
565 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
570 index = get_count_order(nents) - 3;
575 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
577 struct scsi_host_sg_pool *sgp;
579 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
580 mempool_free(sgl, sgp->pool);
583 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
585 struct scsi_host_sg_pool *sgp;
587 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
588 return mempool_alloc(sgp->pool, gfp_mask);
591 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
598 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
599 gfp_mask, scsi_sg_alloc);
601 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
607 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
609 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
612 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
615 if (cmd->sdb.table.nents)
616 scsi_free_sgtable(&cmd->sdb);
618 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
620 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
621 struct scsi_data_buffer *bidi_sdb =
622 cmd->request->next_rq->special;
623 scsi_free_sgtable(bidi_sdb);
624 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
625 cmd->request->next_rq->special = NULL;
628 if (scsi_prot_sg_count(cmd))
629 scsi_free_sgtable(cmd->prot_sdb);
633 * Function: scsi_release_buffers()
635 * Purpose: Completion processing for block device I/O requests.
637 * Arguments: cmd - command that we are bailing.
639 * Lock status: Assumed that no lock is held upon entry.
643 * Notes: In the event that an upper level driver rejects a
644 * command, we must release resources allocated during
645 * the __init_io() function. Primarily this would involve
646 * the scatter-gather table, and potentially any bounce
649 void scsi_release_buffers(struct scsi_cmnd *cmd)
651 __scsi_release_buffers(cmd, 1);
653 EXPORT_SYMBOL(scsi_release_buffers);
655 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
659 switch(host_byte(result)) {
660 case DID_TRANSPORT_FAILFAST:
663 case DID_TARGET_FAILURE:
664 cmd->result |= (DID_OK << 16);
667 case DID_NEXUS_FAILURE:
668 cmd->result |= (DID_OK << 16);
680 * Function: scsi_io_completion()
682 * Purpose: Completion processing for block device I/O requests.
684 * Arguments: cmd - command that is finished.
686 * Lock status: Assumed that no lock is held upon entry.
690 * Notes: This function is matched in terms of capabilities to
691 * the function that created the scatter-gather list.
692 * In other words, if there are no bounce buffers
693 * (the normal case for most drivers), we don't need
694 * the logic to deal with cleaning up afterwards.
696 * We must call scsi_end_request(). This will finish off
697 * the specified number of sectors. If we are done, the
698 * command block will be released and the queue function
699 * will be goosed. If we are not done then we have to
700 * figure out what to do next:
702 * a) We can call scsi_requeue_command(). The request
703 * will be unprepared and put back on the queue. Then
704 * a new command will be created for it. This should
705 * be used if we made forward progress, or if we want
706 * to switch from READ(10) to READ(6) for example.
708 * b) We can call scsi_queue_insert(). The request will
709 * be put back on the queue and retried using the same
710 * command as before, possibly after a delay.
712 * c) We can call blk_end_request() with -EIO to fail
713 * the remainder of the request.
715 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
717 int result = cmd->result;
718 struct request_queue *q = cmd->device->request_queue;
719 struct request *req = cmd->request;
721 struct scsi_sense_hdr sshdr;
723 int sense_deferred = 0;
724 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
725 ACTION_DELAYED_RETRY} action;
726 char *description = NULL;
729 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
731 sense_deferred = scsi_sense_is_deferred(&sshdr);
734 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
735 req->errors = result;
737 if (sense_valid && req->sense) {
739 * SG_IO wants current and deferred errors
741 int len = 8 + cmd->sense_buffer[7];
743 if (len > SCSI_SENSE_BUFFERSIZE)
744 len = SCSI_SENSE_BUFFERSIZE;
745 memcpy(req->sense, cmd->sense_buffer, len);
746 req->sense_len = len;
749 error = __scsi_error_from_host_byte(cmd, result);
752 req->resid_len = scsi_get_resid(cmd);
754 if (scsi_bidi_cmnd(cmd)) {
756 * Bidi commands Must be complete as a whole,
757 * both sides at once.
759 req->next_rq->resid_len = scsi_in(cmd)->resid;
761 scsi_release_buffers(cmd);
762 blk_end_request_all(req, 0);
764 scsi_next_command(cmd);
769 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
770 BUG_ON(blk_bidi_rq(req));
773 * Next deal with any sectors which we were able to correctly
776 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
778 blk_rq_sectors(req), good_bytes));
781 * Recovered errors need reporting, but they're always treated
782 * as success, so fiddle the result code here. For BLOCK_PC
783 * we already took a copy of the original into rq->errors which
784 * is what gets returned to the user
786 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
787 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
788 * print since caller wants ATA registers. Only occurs on
789 * SCSI ATA PASS_THROUGH commands when CK_COND=1
791 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
793 else if (!(req->cmd_flags & REQ_QUIET))
794 scsi_print_sense("", cmd);
796 /* BLOCK_PC may have set error */
801 * A number of bytes were successfully read. If there
802 * are leftovers and there is some kind of error
803 * (result != 0), retry the rest.
805 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
808 error = __scsi_error_from_host_byte(cmd, result);
810 if (host_byte(result) == DID_RESET) {
811 /* Third party bus reset or reset for error recovery
812 * reasons. Just retry the command and see what
815 action = ACTION_RETRY;
816 } else if (sense_valid && !sense_deferred) {
817 switch (sshdr.sense_key) {
819 if (cmd->device->removable) {
820 /* Detected disc change. Set a bit
821 * and quietly refuse further access.
823 cmd->device->changed = 1;
824 description = "Media Changed";
825 action = ACTION_FAIL;
827 /* Must have been a power glitch, or a
828 * bus reset. Could not have been a
829 * media change, so we just retry the
830 * command and see what happens.
832 action = ACTION_RETRY;
835 case ILLEGAL_REQUEST:
836 /* If we had an ILLEGAL REQUEST returned, then
837 * we may have performed an unsupported
838 * command. The only thing this should be
839 * would be a ten byte read where only a six
840 * byte read was supported. Also, on a system
841 * where READ CAPACITY failed, we may have
842 * read past the end of the disk.
844 if ((cmd->device->use_10_for_rw &&
845 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
846 (cmd->cmnd[0] == READ_10 ||
847 cmd->cmnd[0] == WRITE_10)) {
848 /* This will issue a new 6-byte command. */
849 cmd->device->use_10_for_rw = 0;
850 action = ACTION_REPREP;
851 } else if (sshdr.asc == 0x10) /* DIX */ {
852 description = "Host Data Integrity Failure";
853 action = ACTION_FAIL;
855 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
856 } else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
857 (cmd->cmnd[0] == UNMAP ||
858 cmd->cmnd[0] == WRITE_SAME_16 ||
859 cmd->cmnd[0] == WRITE_SAME)) {
860 description = "Discard failure";
861 action = ACTION_FAIL;
863 action = ACTION_FAIL;
865 case ABORTED_COMMAND:
866 action = ACTION_FAIL;
867 if (sshdr.asc == 0x10) { /* DIF */
868 description = "Target Data Integrity Failure";
873 /* If the device is in the process of becoming
874 * ready, or has a temporary blockage, retry.
876 if (sshdr.asc == 0x04) {
877 switch (sshdr.ascq) {
878 case 0x01: /* becoming ready */
879 case 0x04: /* format in progress */
880 case 0x05: /* rebuild in progress */
881 case 0x06: /* recalculation in progress */
882 case 0x07: /* operation in progress */
883 case 0x08: /* Long write in progress */
884 case 0x09: /* self test in progress */
885 case 0x14: /* space allocation 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);
923 scsi_print_command(cmd);
925 if (blk_end_request_err(req, error))
926 scsi_requeue_command(q, cmd);
928 scsi_next_command(cmd);
931 /* Unprep the request and put it back at the head of the queue.
932 * A new command will be prepared and issued.
934 scsi_release_buffers(cmd);
935 scsi_requeue_command(q, cmd);
938 /* Retry the same command immediately */
939 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
941 case ACTION_DELAYED_RETRY:
942 /* Retry the same command after a delay */
943 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
948 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
954 * If sg table allocation fails, requeue request later.
956 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
958 return BLKPREP_DEFER;
964 * Next, walk the list, and fill in the addresses and sizes of
967 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
968 BUG_ON(count > sdb->table.nents);
969 sdb->table.nents = count;
970 sdb->length = blk_rq_bytes(req);
975 * Function: scsi_init_io()
977 * Purpose: SCSI I/O initialize function.
979 * Arguments: cmd - Command descriptor we wish to initialize
981 * Returns: 0 on success
982 * BLKPREP_DEFER if the failure is retryable
983 * BLKPREP_KILL if the failure is fatal
985 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
987 struct request *rq = cmd->request;
989 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
993 if (blk_bidi_rq(rq)) {
994 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
995 scsi_sdb_cache, GFP_ATOMIC);
997 error = BLKPREP_DEFER;
1001 rq->next_rq->special = bidi_sdb;
1002 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1007 if (blk_integrity_rq(rq)) {
1008 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1011 BUG_ON(prot_sdb == NULL);
1012 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1014 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1015 error = BLKPREP_DEFER;
1019 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1020 prot_sdb->table.sgl);
1021 BUG_ON(unlikely(count > ivecs));
1022 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1024 cmd->prot_sdb = prot_sdb;
1025 cmd->prot_sdb->table.nents = count;
1031 scsi_release_buffers(cmd);
1032 cmd->request->special = NULL;
1033 scsi_put_command(cmd);
1036 EXPORT_SYMBOL(scsi_init_io);
1038 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1039 struct request *req)
1041 struct scsi_cmnd *cmd;
1043 if (!req->special) {
1044 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1052 /* pull a tag out of the request if we have one */
1053 cmd->tag = req->tag;
1056 cmd->cmnd = req->cmd;
1057 cmd->prot_op = SCSI_PROT_NORMAL;
1062 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1064 struct scsi_cmnd *cmd;
1065 int ret = scsi_prep_state_check(sdev, req);
1067 if (ret != BLKPREP_OK)
1070 cmd = scsi_get_cmd_from_req(sdev, req);
1072 return BLKPREP_DEFER;
1075 * BLOCK_PC requests may transfer data, in which case they must
1076 * a bio attached to them. Or they might contain a SCSI command
1077 * that does not transfer data, in which case they may optionally
1078 * submit a request without an attached bio.
1083 BUG_ON(!req->nr_phys_segments);
1085 ret = scsi_init_io(cmd, GFP_ATOMIC);
1089 BUG_ON(blk_rq_bytes(req));
1091 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1095 cmd->cmd_len = req->cmd_len;
1096 if (!blk_rq_bytes(req))
1097 cmd->sc_data_direction = DMA_NONE;
1098 else if (rq_data_dir(req) == WRITE)
1099 cmd->sc_data_direction = DMA_TO_DEVICE;
1101 cmd->sc_data_direction = DMA_FROM_DEVICE;
1103 cmd->transfersize = blk_rq_bytes(req);
1104 cmd->allowed = req->retries;
1107 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1110 * Setup a REQ_TYPE_FS command. These are simple read/write request
1111 * from filesystems that still need to be translated to SCSI CDBs from
1114 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1116 struct scsi_cmnd *cmd;
1117 int ret = scsi_prep_state_check(sdev, req);
1119 if (ret != BLKPREP_OK)
1122 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1123 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1124 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1125 if (ret != BLKPREP_OK)
1130 * Filesystem requests must transfer data.
1132 BUG_ON(!req->nr_phys_segments);
1134 cmd = scsi_get_cmd_from_req(sdev, req);
1136 return BLKPREP_DEFER;
1138 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1139 return scsi_init_io(cmd, GFP_ATOMIC);
1141 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1143 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1145 int ret = BLKPREP_OK;
1148 * If the device is not in running state we will reject some
1151 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1152 switch (sdev->sdev_state) {
1155 * If the device is offline we refuse to process any
1156 * commands. The device must be brought online
1157 * before trying any recovery commands.
1159 sdev_printk(KERN_ERR, sdev,
1160 "rejecting I/O to offline device\n");
1165 * If the device is fully deleted, we refuse to
1166 * process any commands as well.
1168 sdev_printk(KERN_ERR, sdev,
1169 "rejecting I/O to dead device\n");
1174 case SDEV_CREATED_BLOCK:
1176 * If the devices is blocked we defer normal commands.
1178 if (!(req->cmd_flags & REQ_PREEMPT))
1179 ret = BLKPREP_DEFER;
1183 * For any other not fully online state we only allow
1184 * special commands. In particular any user initiated
1185 * command is not allowed.
1187 if (!(req->cmd_flags & REQ_PREEMPT))
1194 EXPORT_SYMBOL(scsi_prep_state_check);
1196 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1198 struct scsi_device *sdev = q->queuedata;
1202 req->errors = DID_NO_CONNECT << 16;
1203 /* release the command and kill it */
1205 struct scsi_cmnd *cmd = req->special;
1206 scsi_release_buffers(cmd);
1207 scsi_put_command(cmd);
1208 req->special = NULL;
1213 * If we defer, the blk_peek_request() returns NULL, but the
1214 * queue must be restarted, so we schedule a callback to happen
1217 if (sdev->device_busy == 0)
1218 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1221 req->cmd_flags |= REQ_DONTPREP;
1226 EXPORT_SYMBOL(scsi_prep_return);
1228 int scsi_prep_fn(struct request_queue *q, struct request *req)
1230 struct scsi_device *sdev = q->queuedata;
1231 int ret = BLKPREP_KILL;
1233 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1234 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1235 return scsi_prep_return(q, req, ret);
1237 EXPORT_SYMBOL(scsi_prep_fn);
1240 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1243 * Called with the queue_lock held.
1245 static inline int scsi_dev_queue_ready(struct request_queue *q,
1246 struct scsi_device *sdev)
1248 if (sdev->device_busy == 0 && sdev->device_blocked) {
1250 * unblock after device_blocked iterates to zero
1252 if (--sdev->device_blocked == 0) {
1254 sdev_printk(KERN_INFO, sdev,
1255 "unblocking device at zero depth\n"));
1257 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1261 if (scsi_device_is_busy(sdev))
1269 * scsi_target_queue_ready: checks if there we can send commands to target
1270 * @sdev: scsi device on starget to check.
1272 * Called with the host lock held.
1274 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1275 struct scsi_device *sdev)
1277 struct scsi_target *starget = scsi_target(sdev);
1279 if (starget->single_lun) {
1280 if (starget->starget_sdev_user &&
1281 starget->starget_sdev_user != sdev)
1283 starget->starget_sdev_user = sdev;
1286 if (starget->target_busy == 0 && starget->target_blocked) {
1288 * unblock after target_blocked iterates to zero
1290 if (--starget->target_blocked == 0) {
1291 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1292 "unblocking target at zero depth\n"));
1297 if (scsi_target_is_busy(starget)) {
1298 if (list_empty(&sdev->starved_entry))
1299 list_add_tail(&sdev->starved_entry,
1300 &shost->starved_list);
1304 /* We're OK to process the command, so we can't be starved */
1305 if (!list_empty(&sdev->starved_entry))
1306 list_del_init(&sdev->starved_entry);
1311 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1312 * return 0. We must end up running the queue again whenever 0 is
1313 * returned, else IO can hang.
1315 * Called with host_lock held.
1317 static inline int scsi_host_queue_ready(struct request_queue *q,
1318 struct Scsi_Host *shost,
1319 struct scsi_device *sdev)
1321 if (scsi_host_in_recovery(shost))
1323 if (shost->host_busy == 0 && shost->host_blocked) {
1325 * unblock after host_blocked iterates to zero
1327 if (--shost->host_blocked == 0) {
1329 printk("scsi%d unblocking host at zero depth\n",
1335 if (scsi_host_is_busy(shost)) {
1336 if (list_empty(&sdev->starved_entry))
1337 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1341 /* We're OK to process the command, so we can't be starved */
1342 if (!list_empty(&sdev->starved_entry))
1343 list_del_init(&sdev->starved_entry);
1349 * Busy state exporting function for request stacking drivers.
1351 * For efficiency, no lock is taken to check the busy state of
1352 * shost/starget/sdev, since the returned value is not guaranteed and
1353 * may be changed after request stacking drivers call the function,
1354 * regardless of taking lock or not.
1356 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1357 * (e.g. !sdev), scsi needs to return 'not busy'.
1358 * Otherwise, request stacking drivers may hold requests forever.
1360 static int scsi_lld_busy(struct request_queue *q)
1362 struct scsi_device *sdev = q->queuedata;
1363 struct Scsi_Host *shost;
1364 struct scsi_target *starget;
1370 starget = scsi_target(sdev);
1372 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1373 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1380 * Kill a request for a dead device
1382 static void scsi_kill_request(struct request *req, struct request_queue *q)
1384 struct scsi_cmnd *cmd = req->special;
1385 struct scsi_device *sdev;
1386 struct scsi_target *starget;
1387 struct Scsi_Host *shost;
1389 blk_start_request(req);
1392 starget = scsi_target(sdev);
1394 scsi_init_cmd_errh(cmd);
1395 cmd->result = DID_NO_CONNECT << 16;
1396 atomic_inc(&cmd->device->iorequest_cnt);
1399 * SCSI request completion path will do scsi_device_unbusy(),
1400 * bump busy counts. To bump the counters, we need to dance
1401 * with the locks as normal issue path does.
1403 sdev->device_busy++;
1404 spin_unlock(sdev->request_queue->queue_lock);
1405 spin_lock(shost->host_lock);
1407 starget->target_busy++;
1408 spin_unlock(shost->host_lock);
1409 spin_lock(sdev->request_queue->queue_lock);
1411 blk_complete_request(req);
1414 static void scsi_softirq_done(struct request *rq)
1416 struct scsi_cmnd *cmd = rq->special;
1417 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1420 INIT_LIST_HEAD(&cmd->eh_entry);
1422 atomic_inc(&cmd->device->iodone_cnt);
1424 atomic_inc(&cmd->device->ioerr_cnt);
1426 disposition = scsi_decide_disposition(cmd);
1427 if (disposition != SUCCESS &&
1428 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1429 sdev_printk(KERN_ERR, cmd->device,
1430 "timing out command, waited %lus\n",
1432 disposition = SUCCESS;
1435 scsi_log_completion(cmd, disposition);
1437 switch (disposition) {
1439 scsi_finish_command(cmd);
1442 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1444 case ADD_TO_MLQUEUE:
1445 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1448 if (!scsi_eh_scmd_add(cmd, 0))
1449 scsi_finish_command(cmd);
1454 * Function: scsi_request_fn()
1456 * Purpose: Main strategy routine for SCSI.
1458 * Arguments: q - Pointer to actual queue.
1462 * Lock status: IO request lock assumed to be held when called.
1464 static void scsi_request_fn(struct request_queue *q)
1466 struct scsi_device *sdev = q->queuedata;
1467 struct Scsi_Host *shost;
1468 struct scsi_cmnd *cmd;
1469 struct request *req;
1472 printk("scsi: killing requests for dead queue\n");
1473 while ((req = blk_peek_request(q)) != NULL)
1474 scsi_kill_request(req, q);
1478 if(!get_device(&sdev->sdev_gendev))
1479 /* We must be tearing the block queue down already */
1483 * To start with, we keep looping until the queue is empty, or until
1484 * the host is no longer able to accept any more requests.
1490 * get next queueable request. We do this early to make sure
1491 * that the request is fully prepared even if we cannot
1494 req = blk_peek_request(q);
1495 if (!req || !scsi_dev_queue_ready(q, sdev))
1498 if (unlikely(!scsi_device_online(sdev))) {
1499 sdev_printk(KERN_ERR, sdev,
1500 "rejecting I/O to offline device\n");
1501 scsi_kill_request(req, q);
1507 * Remove the request from the request list.
1509 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1510 blk_start_request(req);
1511 sdev->device_busy++;
1513 spin_unlock(q->queue_lock);
1515 if (unlikely(cmd == NULL)) {
1516 printk(KERN_CRIT "impossible request in %s.\n"
1517 "please mail a stack trace to "
1518 "linux-scsi@vger.kernel.org\n",
1520 blk_dump_rq_flags(req, "foo");
1523 spin_lock(shost->host_lock);
1526 * We hit this when the driver is using a host wide
1527 * tag map. For device level tag maps the queue_depth check
1528 * in the device ready fn would prevent us from trying
1529 * to allocate a tag. Since the map is a shared host resource
1530 * we add the dev to the starved list so it eventually gets
1531 * a run when a tag is freed.
1533 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1534 if (list_empty(&sdev->starved_entry))
1535 list_add_tail(&sdev->starved_entry,
1536 &shost->starved_list);
1540 if (!scsi_target_queue_ready(shost, sdev))
1543 if (!scsi_host_queue_ready(q, shost, sdev))
1546 scsi_target(sdev)->target_busy++;
1550 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1551 * take the lock again.
1553 spin_unlock_irq(shost->host_lock);
1556 * Finally, initialize any error handling parameters, and set up
1557 * the timers for timeouts.
1559 scsi_init_cmd_errh(cmd);
1562 * Dispatch the command to the low-level driver.
1564 rtn = scsi_dispatch_cmd(cmd);
1565 spin_lock_irq(q->queue_lock);
1573 spin_unlock_irq(shost->host_lock);
1576 * lock q, handle tag, requeue req, and decrement device_busy. We
1577 * must return with queue_lock held.
1579 * Decrementing device_busy without checking it is OK, as all such
1580 * cases (host limits or settings) should run the queue at some
1583 spin_lock_irq(q->queue_lock);
1584 blk_requeue_request(q, req);
1585 sdev->device_busy--;
1587 if (sdev->device_busy == 0)
1588 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1590 /* must be careful here...if we trigger the ->remove() function
1591 * we cannot be holding the q lock */
1592 spin_unlock_irq(q->queue_lock);
1593 put_device(&sdev->sdev_gendev);
1594 spin_lock_irq(q->queue_lock);
1597 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1599 struct device *host_dev;
1600 u64 bounce_limit = 0xffffffff;
1602 if (shost->unchecked_isa_dma)
1603 return BLK_BOUNCE_ISA;
1605 * Platforms with virtual-DMA translation
1606 * hardware have no practical limit.
1608 if (!PCI_DMA_BUS_IS_PHYS)
1609 return BLK_BOUNCE_ANY;
1611 host_dev = scsi_get_device(shost);
1612 if (host_dev && host_dev->dma_mask)
1613 bounce_limit = *host_dev->dma_mask;
1615 return bounce_limit;
1617 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1619 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1620 request_fn_proc *request_fn)
1622 struct request_queue *q;
1623 struct device *dev = shost->shost_gendev.parent;
1625 q = blk_init_queue(request_fn, NULL);
1630 * this limit is imposed by hardware restrictions
1632 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1633 SCSI_MAX_SG_CHAIN_SEGMENTS));
1635 if (scsi_host_prot_dma(shost)) {
1636 shost->sg_prot_tablesize =
1637 min_not_zero(shost->sg_prot_tablesize,
1638 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1639 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1640 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1643 blk_queue_max_hw_sectors(q, shost->max_sectors);
1644 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1645 blk_queue_segment_boundary(q, shost->dma_boundary);
1646 dma_set_seg_boundary(dev, shost->dma_boundary);
1648 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1650 if (!shost->use_clustering)
1651 q->limits.cluster = 0;
1654 * set a reasonable default alignment on word boundaries: the
1655 * host and device may alter it using
1656 * blk_queue_update_dma_alignment() later.
1658 blk_queue_dma_alignment(q, 0x03);
1662 EXPORT_SYMBOL(__scsi_alloc_queue);
1664 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1666 struct request_queue *q;
1668 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1672 blk_queue_prep_rq(q, scsi_prep_fn);
1673 blk_queue_softirq_done(q, scsi_softirq_done);
1674 blk_queue_rq_timed_out(q, scsi_times_out);
1675 blk_queue_lld_busy(q, scsi_lld_busy);
1679 void scsi_free_queue(struct request_queue *q)
1681 blk_cleanup_queue(q);
1685 * Function: scsi_block_requests()
1687 * Purpose: Utility function used by low-level drivers to prevent further
1688 * commands from being queued to the device.
1690 * Arguments: shost - Host in question
1694 * Lock status: No locks are assumed held.
1696 * Notes: There is no timer nor any other means by which the requests
1697 * get unblocked other than the low-level driver calling
1698 * scsi_unblock_requests().
1700 void scsi_block_requests(struct Scsi_Host *shost)
1702 shost->host_self_blocked = 1;
1704 EXPORT_SYMBOL(scsi_block_requests);
1707 * Function: scsi_unblock_requests()
1709 * Purpose: Utility function used by low-level drivers to allow further
1710 * commands from being queued to the device.
1712 * Arguments: shost - Host in question
1716 * Lock status: No locks are assumed held.
1718 * Notes: There is no timer nor any other means by which the requests
1719 * get unblocked other than the low-level driver calling
1720 * scsi_unblock_requests().
1722 * This is done as an API function so that changes to the
1723 * internals of the scsi mid-layer won't require wholesale
1724 * changes to drivers that use this feature.
1726 void scsi_unblock_requests(struct Scsi_Host *shost)
1728 shost->host_self_blocked = 0;
1729 scsi_run_host_queues(shost);
1731 EXPORT_SYMBOL(scsi_unblock_requests);
1733 int __init scsi_init_queue(void)
1737 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1738 sizeof(struct scsi_data_buffer),
1740 if (!scsi_sdb_cache) {
1741 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1745 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1746 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1747 int size = sgp->size * sizeof(struct scatterlist);
1749 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1750 SLAB_HWCACHE_ALIGN, NULL);
1752 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1757 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1760 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1769 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1770 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1772 mempool_destroy(sgp->pool);
1774 kmem_cache_destroy(sgp->slab);
1776 kmem_cache_destroy(scsi_sdb_cache);
1781 void scsi_exit_queue(void)
1785 kmem_cache_destroy(scsi_sdb_cache);
1787 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1788 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1789 mempool_destroy(sgp->pool);
1790 kmem_cache_destroy(sgp->slab);
1795 * scsi_mode_select - issue a mode select
1796 * @sdev: SCSI device to be queried
1797 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1798 * @sp: Save page bit (0 == don't save, 1 == save)
1799 * @modepage: mode page being requested
1800 * @buffer: request buffer (may not be smaller than eight bytes)
1801 * @len: length of request buffer.
1802 * @timeout: command timeout
1803 * @retries: number of retries before failing
1804 * @data: returns a structure abstracting the mode header data
1805 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1806 * must be SCSI_SENSE_BUFFERSIZE big.
1808 * Returns zero if successful; negative error number or scsi
1813 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1814 unsigned char *buffer, int len, int timeout, int retries,
1815 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1817 unsigned char cmd[10];
1818 unsigned char *real_buffer;
1821 memset(cmd, 0, sizeof(cmd));
1822 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1824 if (sdev->use_10_for_ms) {
1827 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1830 memcpy(real_buffer + 8, buffer, len);
1834 real_buffer[2] = data->medium_type;
1835 real_buffer[3] = data->device_specific;
1836 real_buffer[4] = data->longlba ? 0x01 : 0;
1838 real_buffer[6] = data->block_descriptor_length >> 8;
1839 real_buffer[7] = data->block_descriptor_length;
1841 cmd[0] = MODE_SELECT_10;
1845 if (len > 255 || data->block_descriptor_length > 255 ||
1849 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1852 memcpy(real_buffer + 4, buffer, len);
1855 real_buffer[1] = data->medium_type;
1856 real_buffer[2] = data->device_specific;
1857 real_buffer[3] = data->block_descriptor_length;
1860 cmd[0] = MODE_SELECT;
1864 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1865 sshdr, timeout, retries, NULL);
1869 EXPORT_SYMBOL_GPL(scsi_mode_select);
1872 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1873 * @sdev: SCSI device to be queried
1874 * @dbd: set if mode sense will allow block descriptors to be returned
1875 * @modepage: mode page being requested
1876 * @buffer: request buffer (may not be smaller than eight bytes)
1877 * @len: length of request buffer.
1878 * @timeout: command timeout
1879 * @retries: number of retries before failing
1880 * @data: returns a structure abstracting the mode header data
1881 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1882 * must be SCSI_SENSE_BUFFERSIZE big.
1884 * Returns zero if unsuccessful, or the header offset (either 4
1885 * or 8 depending on whether a six or ten byte command was
1886 * issued) if successful.
1889 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1890 unsigned char *buffer, int len, int timeout, int retries,
1891 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1893 unsigned char cmd[12];
1897 struct scsi_sense_hdr my_sshdr;
1899 memset(data, 0, sizeof(*data));
1900 memset(&cmd[0], 0, 12);
1901 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1904 /* caller might not be interested in sense, but we need it */
1909 use_10_for_ms = sdev->use_10_for_ms;
1911 if (use_10_for_ms) {
1915 cmd[0] = MODE_SENSE_10;
1922 cmd[0] = MODE_SENSE;
1927 memset(buffer, 0, len);
1929 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1930 sshdr, timeout, retries, NULL);
1932 /* This code looks awful: what it's doing is making sure an
1933 * ILLEGAL REQUEST sense return identifies the actual command
1934 * byte as the problem. MODE_SENSE commands can return
1935 * ILLEGAL REQUEST if the code page isn't supported */
1937 if (use_10_for_ms && !scsi_status_is_good(result) &&
1938 (driver_byte(result) & DRIVER_SENSE)) {
1939 if (scsi_sense_valid(sshdr)) {
1940 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1941 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1943 * Invalid command operation code
1945 sdev->use_10_for_ms = 0;
1951 if(scsi_status_is_good(result)) {
1952 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1953 (modepage == 6 || modepage == 8))) {
1954 /* Initio breakage? */
1957 data->medium_type = 0;
1958 data->device_specific = 0;
1960 data->block_descriptor_length = 0;
1961 } else if(use_10_for_ms) {
1962 data->length = buffer[0]*256 + buffer[1] + 2;
1963 data->medium_type = buffer[2];
1964 data->device_specific = buffer[3];
1965 data->longlba = buffer[4] & 0x01;
1966 data->block_descriptor_length = buffer[6]*256
1969 data->length = buffer[0] + 1;
1970 data->medium_type = buffer[1];
1971 data->device_specific = buffer[2];
1972 data->block_descriptor_length = buffer[3];
1974 data->header_length = header_length;
1979 EXPORT_SYMBOL(scsi_mode_sense);
1982 * scsi_test_unit_ready - test if unit is ready
1983 * @sdev: scsi device to change the state of.
1984 * @timeout: command timeout
1985 * @retries: number of retries before failing
1986 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1987 * returning sense. Make sure that this is cleared before passing
1990 * Returns zero if unsuccessful or an error if TUR failed. For
1991 * removable media, UNIT_ATTENTION sets ->changed flag.
1994 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1995 struct scsi_sense_hdr *sshdr_external)
1998 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2000 struct scsi_sense_hdr *sshdr;
2003 if (!sshdr_external)
2004 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2006 sshdr = sshdr_external;
2008 /* try to eat the UNIT_ATTENTION if there are enough retries */
2010 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2011 timeout, retries, NULL);
2012 if (sdev->removable && scsi_sense_valid(sshdr) &&
2013 sshdr->sense_key == UNIT_ATTENTION)
2015 } while (scsi_sense_valid(sshdr) &&
2016 sshdr->sense_key == UNIT_ATTENTION && --retries);
2018 if (!sshdr_external)
2022 EXPORT_SYMBOL(scsi_test_unit_ready);
2025 * scsi_device_set_state - Take the given device through the device state model.
2026 * @sdev: scsi device to change the state of.
2027 * @state: state to change to.
2029 * Returns zero if unsuccessful or an error if the requested
2030 * transition is illegal.
2033 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2035 enum scsi_device_state oldstate = sdev->sdev_state;
2037 if (state == oldstate)
2043 case SDEV_CREATED_BLOCK:
2087 case SDEV_CREATED_BLOCK:
2094 case SDEV_CREATED_BLOCK:
2129 sdev->sdev_state = state;
2133 SCSI_LOG_ERROR_RECOVERY(1,
2134 sdev_printk(KERN_ERR, sdev,
2135 "Illegal state transition %s->%s\n",
2136 scsi_device_state_name(oldstate),
2137 scsi_device_state_name(state))
2141 EXPORT_SYMBOL(scsi_device_set_state);
2144 * sdev_evt_emit - emit a single SCSI device uevent
2145 * @sdev: associated SCSI device
2146 * @evt: event to emit
2148 * Send a single uevent (scsi_event) to the associated scsi_device.
2150 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2155 switch (evt->evt_type) {
2156 case SDEV_EVT_MEDIA_CHANGE:
2157 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2167 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2171 * sdev_evt_thread - send a uevent for each scsi event
2172 * @work: work struct for scsi_device
2174 * Dispatch queued events to their associated scsi_device kobjects
2177 void scsi_evt_thread(struct work_struct *work)
2179 struct scsi_device *sdev;
2180 LIST_HEAD(event_list);
2182 sdev = container_of(work, struct scsi_device, event_work);
2185 struct scsi_event *evt;
2186 struct list_head *this, *tmp;
2187 unsigned long flags;
2189 spin_lock_irqsave(&sdev->list_lock, flags);
2190 list_splice_init(&sdev->event_list, &event_list);
2191 spin_unlock_irqrestore(&sdev->list_lock, flags);
2193 if (list_empty(&event_list))
2196 list_for_each_safe(this, tmp, &event_list) {
2197 evt = list_entry(this, struct scsi_event, node);
2198 list_del(&evt->node);
2199 scsi_evt_emit(sdev, evt);
2206 * sdev_evt_send - send asserted event to uevent thread
2207 * @sdev: scsi_device event occurred on
2208 * @evt: event to send
2210 * Assert scsi device event asynchronously.
2212 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2214 unsigned long flags;
2217 /* FIXME: currently this check eliminates all media change events
2218 * for polled devices. Need to update to discriminate between AN
2219 * and polled events */
2220 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2226 spin_lock_irqsave(&sdev->list_lock, flags);
2227 list_add_tail(&evt->node, &sdev->event_list);
2228 schedule_work(&sdev->event_work);
2229 spin_unlock_irqrestore(&sdev->list_lock, flags);
2231 EXPORT_SYMBOL_GPL(sdev_evt_send);
2234 * sdev_evt_alloc - allocate a new scsi event
2235 * @evt_type: type of event to allocate
2236 * @gfpflags: GFP flags for allocation
2238 * Allocates and returns a new scsi_event.
2240 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2243 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2247 evt->evt_type = evt_type;
2248 INIT_LIST_HEAD(&evt->node);
2250 /* evt_type-specific initialization, if any */
2252 case SDEV_EVT_MEDIA_CHANGE:
2260 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2263 * sdev_evt_send_simple - send asserted event to uevent thread
2264 * @sdev: scsi_device event occurred on
2265 * @evt_type: type of event to send
2266 * @gfpflags: GFP flags for allocation
2268 * Assert scsi device event asynchronously, given an event type.
2270 void sdev_evt_send_simple(struct scsi_device *sdev,
2271 enum scsi_device_event evt_type, gfp_t gfpflags)
2273 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2275 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2280 sdev_evt_send(sdev, evt);
2282 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2285 * scsi_device_quiesce - Block user issued commands.
2286 * @sdev: scsi device to quiesce.
2288 * This works by trying to transition to the SDEV_QUIESCE state
2289 * (which must be a legal transition). When the device is in this
2290 * state, only special requests will be accepted, all others will
2291 * be deferred. Since special requests may also be requeued requests,
2292 * a successful return doesn't guarantee the device will be
2293 * totally quiescent.
2295 * Must be called with user context, may sleep.
2297 * Returns zero if unsuccessful or an error if not.
2300 scsi_device_quiesce(struct scsi_device *sdev)
2302 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2306 scsi_run_queue(sdev->request_queue);
2307 while (sdev->device_busy) {
2308 msleep_interruptible(200);
2309 scsi_run_queue(sdev->request_queue);
2313 EXPORT_SYMBOL(scsi_device_quiesce);
2316 * scsi_device_resume - Restart user issued commands to a quiesced device.
2317 * @sdev: scsi device to resume.
2319 * Moves the device from quiesced back to running and restarts the
2322 * Must be called with user context, may sleep.
2325 scsi_device_resume(struct scsi_device *sdev)
2327 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2329 scsi_run_queue(sdev->request_queue);
2331 EXPORT_SYMBOL(scsi_device_resume);
2334 device_quiesce_fn(struct scsi_device *sdev, void *data)
2336 scsi_device_quiesce(sdev);
2340 scsi_target_quiesce(struct scsi_target *starget)
2342 starget_for_each_device(starget, NULL, device_quiesce_fn);
2344 EXPORT_SYMBOL(scsi_target_quiesce);
2347 device_resume_fn(struct scsi_device *sdev, void *data)
2349 scsi_device_resume(sdev);
2353 scsi_target_resume(struct scsi_target *starget)
2355 starget_for_each_device(starget, NULL, device_resume_fn);
2357 EXPORT_SYMBOL(scsi_target_resume);
2360 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2361 * @sdev: device to block
2363 * Block request made by scsi lld's to temporarily stop all
2364 * scsi commands on the specified device. Called from interrupt
2365 * or normal process context.
2367 * Returns zero if successful or error if not
2370 * This routine transitions the device to the SDEV_BLOCK state
2371 * (which must be a legal transition). When the device is in this
2372 * state, all commands are deferred until the scsi lld reenables
2373 * the device with scsi_device_unblock or device_block_tmo fires.
2374 * This routine assumes the host_lock is held on entry.
2377 scsi_internal_device_block(struct scsi_device *sdev)
2379 struct request_queue *q = sdev->request_queue;
2380 unsigned long flags;
2383 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2385 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2392 * The device has transitioned to SDEV_BLOCK. Stop the
2393 * block layer from calling the midlayer with this device's
2396 spin_lock_irqsave(q->queue_lock, flags);
2398 spin_unlock_irqrestore(q->queue_lock, flags);
2402 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2405 * scsi_internal_device_unblock - resume a device after a block request
2406 * @sdev: device to resume
2408 * Called by scsi lld's or the midlayer to restart the device queue
2409 * for the previously suspended scsi device. Called from interrupt or
2410 * normal process context.
2412 * Returns zero if successful or error if not.
2415 * This routine transitions the device to the SDEV_RUNNING state
2416 * (which must be a legal transition) allowing the midlayer to
2417 * goose the queue for this device. This routine assumes the
2418 * host_lock is held upon entry.
2421 scsi_internal_device_unblock(struct scsi_device *sdev)
2423 struct request_queue *q = sdev->request_queue;
2424 unsigned long flags;
2427 * Try to transition the scsi device to SDEV_RUNNING
2428 * and goose the device queue if successful.
2430 if (sdev->sdev_state == SDEV_BLOCK)
2431 sdev->sdev_state = SDEV_RUNNING;
2432 else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2433 sdev->sdev_state = SDEV_CREATED;
2434 else if (sdev->sdev_state != SDEV_CANCEL &&
2435 sdev->sdev_state != SDEV_OFFLINE)
2438 spin_lock_irqsave(q->queue_lock, flags);
2440 spin_unlock_irqrestore(q->queue_lock, flags);
2444 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2447 device_block(struct scsi_device *sdev, void *data)
2449 scsi_internal_device_block(sdev);
2453 target_block(struct device *dev, void *data)
2455 if (scsi_is_target_device(dev))
2456 starget_for_each_device(to_scsi_target(dev), NULL,
2462 scsi_target_block(struct device *dev)
2464 if (scsi_is_target_device(dev))
2465 starget_for_each_device(to_scsi_target(dev), NULL,
2468 device_for_each_child(dev, NULL, target_block);
2470 EXPORT_SYMBOL_GPL(scsi_target_block);
2473 device_unblock(struct scsi_device *sdev, void *data)
2475 scsi_internal_device_unblock(sdev);
2479 target_unblock(struct device *dev, void *data)
2481 if (scsi_is_target_device(dev))
2482 starget_for_each_device(to_scsi_target(dev), NULL,
2488 scsi_target_unblock(struct device *dev)
2490 if (scsi_is_target_device(dev))
2491 starget_for_each_device(to_scsi_target(dev), NULL,
2494 device_for_each_child(dev, NULL, target_unblock);
2496 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2499 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2500 * @sgl: scatter-gather list
2501 * @sg_count: number of segments in sg
2502 * @offset: offset in bytes into sg, on return offset into the mapped area
2503 * @len: bytes to map, on return number of bytes mapped
2505 * Returns virtual address of the start of the mapped page
2507 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2508 size_t *offset, size_t *len)
2511 size_t sg_len = 0, len_complete = 0;
2512 struct scatterlist *sg;
2515 WARN_ON(!irqs_disabled());
2517 for_each_sg(sgl, sg, sg_count, i) {
2518 len_complete = sg_len; /* Complete sg-entries */
2519 sg_len += sg->length;
2520 if (sg_len > *offset)
2524 if (unlikely(i == sg_count)) {
2525 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2527 __func__, sg_len, *offset, sg_count);
2532 /* Offset starting from the beginning of first page in this sg-entry */
2533 *offset = *offset - len_complete + sg->offset;
2535 /* Assumption: contiguous pages can be accessed as "page + i" */
2536 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2537 *offset &= ~PAGE_MASK;
2539 /* Bytes in this sg-entry from *offset to the end of the page */
2540 sg_len = PAGE_SIZE - *offset;
2544 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2546 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2549 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2550 * @virt: virtual address to be unmapped
2552 void scsi_kunmap_atomic_sg(void *virt)
2554 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2556 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);