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/export.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_dbg.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_driver.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_host.h>
32 #include <trace/events/scsi.h>
34 #include "scsi_priv.h"
35 #include "scsi_logging.h"
38 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
39 #define SG_MEMPOOL_SIZE 2
41 struct scsi_host_sg_pool {
44 struct kmem_cache *slab;
48 #define SP(x) { x, "sgpool-" __stringify(x) }
49 #if (SCSI_MAX_SG_SEGMENTS < 32)
50 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
52 static struct scsi_host_sg_pool scsi_sg_pools[] = {
55 #if (SCSI_MAX_SG_SEGMENTS > 32)
57 #if (SCSI_MAX_SG_SEGMENTS > 64)
59 #if (SCSI_MAX_SG_SEGMENTS > 128)
61 #if (SCSI_MAX_SG_SEGMENTS > 256)
62 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
67 SP(SCSI_MAX_SG_SEGMENTS)
71 struct kmem_cache *scsi_sdb_cache;
74 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
75 * not change behaviour from the previous unplug mechanism, experimentation
76 * may prove this needs changing.
78 #define SCSI_QUEUE_DELAY 3
81 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
83 struct Scsi_Host *host = cmd->device->host;
84 struct scsi_device *device = cmd->device;
85 struct scsi_target *starget = scsi_target(device);
88 * Set the appropriate busy bit for the device/host.
90 * If the host/device isn't busy, assume that something actually
91 * completed, and that we should be able to queue a command now.
93 * Note that the prior mid-layer assumption that any host could
94 * always queue at least one command is now broken. The mid-layer
95 * will implement a user specifiable stall (see
96 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
97 * if a command is requeued with no other commands outstanding
98 * either for the device or for the host.
101 case SCSI_MLQUEUE_HOST_BUSY:
102 host->host_blocked = host->max_host_blocked;
104 case SCSI_MLQUEUE_DEVICE_BUSY:
105 case SCSI_MLQUEUE_EH_RETRY:
106 device->device_blocked = device->max_device_blocked;
108 case SCSI_MLQUEUE_TARGET_BUSY:
109 starget->target_blocked = starget->max_target_blocked;
115 * __scsi_queue_insert - private queue insertion
116 * @cmd: The SCSI command being requeued
117 * @reason: The reason for the requeue
118 * @unbusy: Whether the queue should be unbusied
120 * This is a private queue insertion. The public interface
121 * scsi_queue_insert() always assumes the queue should be unbusied
122 * because it's always called before the completion. This function is
123 * for a requeue after completion, which should only occur in this
126 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
128 struct scsi_device *device = cmd->device;
129 struct request_queue *q = device->request_queue;
132 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
133 "Inserting command %p into mlqueue\n", cmd));
135 scsi_set_blocked(cmd, reason);
138 * Decrement the counters, since these commands are no longer
139 * active on the host/device.
142 scsi_device_unbusy(device);
145 * Requeue this command. It will go before all other commands
146 * that are already in the queue. Schedule requeue work under
147 * lock such that the kblockd_schedule_work() call happens
148 * before blk_cleanup_queue() finishes.
151 spin_lock_irqsave(q->queue_lock, flags);
152 blk_requeue_request(q, cmd->request);
153 kblockd_schedule_work(&device->requeue_work);
154 spin_unlock_irqrestore(q->queue_lock, flags);
158 * Function: scsi_queue_insert()
160 * Purpose: Insert a command in the midlevel queue.
162 * Arguments: cmd - command that we are adding to queue.
163 * reason - why we are inserting command to queue.
165 * Lock status: Assumed that lock is not held upon entry.
169 * Notes: We do this for one of two cases. Either the host is busy
170 * and it cannot accept any more commands for the time being,
171 * or the device returned QUEUE_FULL and can accept no more
173 * Notes: This could be called either from an interrupt context or a
174 * normal process context.
176 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
178 __scsi_queue_insert(cmd, reason, 1);
181 * scsi_execute - insert request and wait for the result
184 * @data_direction: data direction
185 * @buffer: data buffer
186 * @bufflen: len of buffer
187 * @sense: optional sense buffer
188 * @timeout: request timeout in seconds
189 * @retries: number of times to retry request
190 * @flags: or into request flags;
191 * @resid: optional residual length
193 * returns the req->errors value which is the scsi_cmnd result
196 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
197 int data_direction, void *buffer, unsigned bufflen,
198 unsigned char *sense, int timeout, int retries, u64 flags,
202 int write = (data_direction == DMA_TO_DEVICE);
203 int ret = DRIVER_ERROR << 24;
205 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
208 blk_rq_set_block_pc(req);
210 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
211 buffer, bufflen, __GFP_WAIT))
214 req->cmd_len = COMMAND_SIZE(cmd[0]);
215 memcpy(req->cmd, cmd, req->cmd_len);
218 req->retries = retries;
219 req->timeout = timeout;
220 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
223 * head injection *required* here otherwise quiesce won't work
225 blk_execute_rq(req->q, NULL, req, 1);
228 * Some devices (USB mass-storage in particular) may transfer
229 * garbage data together with a residue indicating that the data
230 * is invalid. Prevent the garbage from being misinterpreted
231 * and prevent security leaks by zeroing out the excess data.
233 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
234 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
237 *resid = req->resid_len;
240 blk_put_request(req);
244 EXPORT_SYMBOL(scsi_execute);
246 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
247 int data_direction, void *buffer, unsigned bufflen,
248 struct scsi_sense_hdr *sshdr, int timeout, int retries,
249 int *resid, u64 flags)
255 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
257 return DRIVER_ERROR << 24;
259 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
260 sense, timeout, retries, flags, resid);
262 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
267 EXPORT_SYMBOL(scsi_execute_req_flags);
270 * Function: scsi_init_cmd_errh()
272 * Purpose: Initialize cmd fields related to error handling.
274 * Arguments: cmd - command that is ready to be queued.
276 * Notes: This function has the job of initializing a number of
277 * fields related to error handling. Typically this will
278 * be called once for each command, as required.
280 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
282 cmd->serial_number = 0;
283 scsi_set_resid(cmd, 0);
284 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
285 if (cmd->cmd_len == 0)
286 cmd->cmd_len = scsi_command_size(cmd->cmnd);
289 void scsi_device_unbusy(struct scsi_device *sdev)
291 struct Scsi_Host *shost = sdev->host;
292 struct scsi_target *starget = scsi_target(sdev);
295 spin_lock_irqsave(shost->host_lock, flags);
297 atomic_dec(&starget->target_busy);
298 if (unlikely(scsi_host_in_recovery(shost) &&
299 (shost->host_failed || shost->host_eh_scheduled)))
300 scsi_eh_wakeup(shost);
301 spin_unlock(shost->host_lock);
302 spin_lock(sdev->request_queue->queue_lock);
304 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
308 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
309 * and call blk_run_queue for all the scsi_devices on the target -
310 * including current_sdev first.
312 * Called with *no* scsi locks held.
314 static void scsi_single_lun_run(struct scsi_device *current_sdev)
316 struct Scsi_Host *shost = current_sdev->host;
317 struct scsi_device *sdev, *tmp;
318 struct scsi_target *starget = scsi_target(current_sdev);
321 spin_lock_irqsave(shost->host_lock, flags);
322 starget->starget_sdev_user = NULL;
323 spin_unlock_irqrestore(shost->host_lock, flags);
326 * Call blk_run_queue for all LUNs on the target, starting with
327 * current_sdev. We race with others (to set starget_sdev_user),
328 * but in most cases, we will be first. Ideally, each LU on the
329 * target would get some limited time or requests on the target.
331 blk_run_queue(current_sdev->request_queue);
333 spin_lock_irqsave(shost->host_lock, flags);
334 if (starget->starget_sdev_user)
336 list_for_each_entry_safe(sdev, tmp, &starget->devices,
337 same_target_siblings) {
338 if (sdev == current_sdev)
340 if (scsi_device_get(sdev))
343 spin_unlock_irqrestore(shost->host_lock, flags);
344 blk_run_queue(sdev->request_queue);
345 spin_lock_irqsave(shost->host_lock, flags);
347 scsi_device_put(sdev);
350 spin_unlock_irqrestore(shost->host_lock, flags);
353 static inline int scsi_device_is_busy(struct scsi_device *sdev)
355 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
361 static inline int scsi_target_is_busy(struct scsi_target *starget)
363 return ((starget->can_queue > 0 &&
364 atomic_read(&starget->target_busy) >= starget->can_queue) ||
365 starget->target_blocked);
368 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
370 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
371 shost->host_blocked || shost->host_self_blocked)
377 static void scsi_starved_list_run(struct Scsi_Host *shost)
379 LIST_HEAD(starved_list);
380 struct scsi_device *sdev;
383 spin_lock_irqsave(shost->host_lock, flags);
384 list_splice_init(&shost->starved_list, &starved_list);
386 while (!list_empty(&starved_list)) {
387 struct request_queue *slq;
390 * As long as shost is accepting commands and we have
391 * starved queues, call blk_run_queue. scsi_request_fn
392 * drops the queue_lock and can add us back to the
395 * host_lock protects the starved_list and starved_entry.
396 * scsi_request_fn must get the host_lock before checking
397 * or modifying starved_list or starved_entry.
399 if (scsi_host_is_busy(shost))
402 sdev = list_entry(starved_list.next,
403 struct scsi_device, starved_entry);
404 list_del_init(&sdev->starved_entry);
405 if (scsi_target_is_busy(scsi_target(sdev))) {
406 list_move_tail(&sdev->starved_entry,
407 &shost->starved_list);
412 * Once we drop the host lock, a racing scsi_remove_device()
413 * call may remove the sdev from the starved list and destroy
414 * it and the queue. Mitigate by taking a reference to the
415 * queue and never touching the sdev again after we drop the
416 * host lock. Note: if __scsi_remove_device() invokes
417 * blk_cleanup_queue() before the queue is run from this
418 * function then blk_run_queue() will return immediately since
419 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
421 slq = sdev->request_queue;
422 if (!blk_get_queue(slq))
424 spin_unlock_irqrestore(shost->host_lock, flags);
429 spin_lock_irqsave(shost->host_lock, flags);
431 /* put any unprocessed entries back */
432 list_splice(&starved_list, &shost->starved_list);
433 spin_unlock_irqrestore(shost->host_lock, flags);
437 * Function: scsi_run_queue()
439 * Purpose: Select a proper request queue to serve next
441 * Arguments: q - last request's queue
445 * Notes: The previous command was completely finished, start
446 * a new one if possible.
448 static void scsi_run_queue(struct request_queue *q)
450 struct scsi_device *sdev = q->queuedata;
452 if (scsi_target(sdev)->single_lun)
453 scsi_single_lun_run(sdev);
454 if (!list_empty(&sdev->host->starved_list))
455 scsi_starved_list_run(sdev->host);
460 void scsi_requeue_run_queue(struct work_struct *work)
462 struct scsi_device *sdev;
463 struct request_queue *q;
465 sdev = container_of(work, struct scsi_device, requeue_work);
466 q = sdev->request_queue;
471 * Function: scsi_requeue_command()
473 * Purpose: Handle post-processing of completed commands.
475 * Arguments: q - queue to operate on
476 * cmd - command that may need to be requeued.
480 * Notes: After command completion, there may be blocks left
481 * over which weren't finished by the previous command
482 * this can be for a number of reasons - the main one is
483 * I/O errors in the middle of the request, in which case
484 * we need to request the blocks that come after the bad
486 * Notes: Upon return, cmd is a stale pointer.
488 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
490 struct scsi_device *sdev = cmd->device;
491 struct request *req = cmd->request;
494 spin_lock_irqsave(q->queue_lock, flags);
495 blk_unprep_request(req);
497 scsi_put_command(cmd);
498 blk_requeue_request(q, req);
499 spin_unlock_irqrestore(q->queue_lock, flags);
503 put_device(&sdev->sdev_gendev);
506 void scsi_next_command(struct scsi_cmnd *cmd)
508 struct scsi_device *sdev = cmd->device;
509 struct request_queue *q = sdev->request_queue;
511 scsi_put_command(cmd);
514 put_device(&sdev->sdev_gendev);
517 void scsi_run_host_queues(struct Scsi_Host *shost)
519 struct scsi_device *sdev;
521 shost_for_each_device(sdev, shost)
522 scsi_run_queue(sdev->request_queue);
525 static inline unsigned int scsi_sgtable_index(unsigned short nents)
529 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
534 index = get_count_order(nents) - 3;
539 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
541 struct scsi_host_sg_pool *sgp;
543 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
544 mempool_free(sgl, sgp->pool);
547 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
549 struct scsi_host_sg_pool *sgp;
551 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
552 return mempool_alloc(sgp->pool, gfp_mask);
555 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
562 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
563 gfp_mask, scsi_sg_alloc);
565 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
571 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
573 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
577 * Function: scsi_release_buffers()
579 * Purpose: Free resources allocate for a scsi_command.
581 * Arguments: cmd - command that we are bailing.
583 * Lock status: Assumed that no lock is held upon entry.
587 * Notes: In the event that an upper level driver rejects a
588 * command, we must release resources allocated during
589 * the __init_io() function. Primarily this would involve
590 * the scatter-gather table.
592 static void scsi_release_buffers(struct scsi_cmnd *cmd)
594 if (cmd->sdb.table.nents)
595 scsi_free_sgtable(&cmd->sdb);
597 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
599 if (scsi_prot_sg_count(cmd))
600 scsi_free_sgtable(cmd->prot_sdb);
603 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
605 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
607 scsi_free_sgtable(bidi_sdb);
608 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
609 cmd->request->next_rq->special = NULL;
613 * __scsi_error_from_host_byte - translate SCSI error code into errno
614 * @cmd: SCSI command (unused)
615 * @result: scsi error code
617 * Translate SCSI error code into standard UNIX errno.
619 * -ENOLINK temporary transport failure
620 * -EREMOTEIO permanent target failure, do not retry
621 * -EBADE permanent nexus failure, retry on other path
622 * -ENOSPC No write space available
623 * -ENODATA Medium error
624 * -EIO unspecified I/O error
626 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
630 switch(host_byte(result)) {
631 case DID_TRANSPORT_FAILFAST:
634 case DID_TARGET_FAILURE:
635 set_host_byte(cmd, DID_OK);
638 case DID_NEXUS_FAILURE:
639 set_host_byte(cmd, DID_OK);
642 case DID_ALLOC_FAILURE:
643 set_host_byte(cmd, DID_OK);
646 case DID_MEDIUM_ERROR:
647 set_host_byte(cmd, DID_OK);
659 * Function: scsi_io_completion()
661 * Purpose: Completion processing for block device I/O requests.
663 * Arguments: cmd - command that is finished.
665 * Lock status: Assumed that no lock is held upon entry.
669 * Notes: We will finish off the specified number of sectors. If we
670 * are done, the command block will be released and the queue
671 * function will be goosed. If we are not done then we have to
672 * figure out what to do next:
674 * a) We can call scsi_requeue_command(). The request
675 * will be unprepared and put back on the queue. Then
676 * a new command will be created for it. This should
677 * be used if we made forward progress, or if we want
678 * to switch from READ(10) to READ(6) for example.
680 * b) We can call __scsi_queue_insert(). The request will
681 * be put back on the queue and retried using the same
682 * command as before, possibly after a delay.
684 * c) We can call blk_end_request() with -EIO to fail
685 * the remainder of the request.
687 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
689 int result = cmd->result;
690 struct request_queue *q = cmd->device->request_queue;
691 struct request *req = cmd->request;
693 struct scsi_sense_hdr sshdr;
695 int sense_deferred = 0;
696 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
697 ACTION_DELAYED_RETRY} action;
698 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
701 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
703 sense_deferred = scsi_sense_is_deferred(&sshdr);
706 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
708 if (sense_valid && req->sense) {
710 * SG_IO wants current and deferred errors
712 int len = 8 + cmd->sense_buffer[7];
714 if (len > SCSI_SENSE_BUFFERSIZE)
715 len = SCSI_SENSE_BUFFERSIZE;
716 memcpy(req->sense, cmd->sense_buffer, len);
717 req->sense_len = len;
720 error = __scsi_error_from_host_byte(cmd, result);
723 * __scsi_error_from_host_byte may have reset the host_byte
725 req->errors = cmd->result;
727 req->resid_len = scsi_get_resid(cmd);
729 if (scsi_bidi_cmnd(cmd)) {
731 * Bidi commands Must be complete as a whole,
732 * both sides at once.
734 req->next_rq->resid_len = scsi_in(cmd)->resid;
736 scsi_release_buffers(cmd);
737 scsi_release_bidi_buffers(cmd);
739 blk_end_request_all(req, 0);
741 scsi_next_command(cmd);
744 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
746 * Certain non BLOCK_PC requests are commands that don't
747 * actually transfer anything (FLUSH), so cannot use
748 * good_bytes != blk_rq_bytes(req) as the signal for an error.
749 * This sets the error explicitly for the problem case.
751 error = __scsi_error_from_host_byte(cmd, result);
754 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
755 BUG_ON(blk_bidi_rq(req));
758 * Next deal with any sectors which we were able to correctly
761 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
762 "%u sectors total, %d bytes done.\n",
763 blk_rq_sectors(req), good_bytes));
766 * Recovered errors need reporting, but they're always treated
767 * as success, so fiddle the result code here. For BLOCK_PC
768 * we already took a copy of the original into rq->errors which
769 * is what gets returned to the user
771 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
772 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
773 * print since caller wants ATA registers. Only occurs on
774 * SCSI ATA PASS_THROUGH commands when CK_COND=1
776 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
778 else if (!(req->cmd_flags & REQ_QUIET))
779 scsi_print_sense("", cmd);
781 /* BLOCK_PC may have set error */
786 * If we finished all bytes in the request we are done now.
788 if (!blk_end_request(req, error, good_bytes))
792 * Kill remainder if no retrys.
794 if (error && scsi_noretry_cmd(cmd)) {
795 blk_end_request_all(req, error);
800 * If there had been no error, but we have leftover bytes in the
801 * requeues just queue the command up again.
806 error = __scsi_error_from_host_byte(cmd, result);
808 if (host_byte(result) == DID_RESET) {
809 /* Third party bus reset or reset for error recovery
810 * reasons. Just retry the command and see what
813 action = ACTION_RETRY;
814 } else if (sense_valid && !sense_deferred) {
815 switch (sshdr.sense_key) {
817 if (cmd->device->removable) {
818 /* Detected disc change. Set a bit
819 * and quietly refuse further access.
821 cmd->device->changed = 1;
822 action = ACTION_FAIL;
824 /* Must have been a power glitch, or a
825 * bus reset. Could not have been a
826 * media change, so we just retry the
827 * command and see what happens.
829 action = ACTION_RETRY;
832 case ILLEGAL_REQUEST:
833 /* If we had an ILLEGAL REQUEST returned, then
834 * we may have performed an unsupported
835 * command. The only thing this should be
836 * would be a ten byte read where only a six
837 * byte read was supported. Also, on a system
838 * where READ CAPACITY failed, we may have
839 * read past the end of the disk.
841 if ((cmd->device->use_10_for_rw &&
842 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
843 (cmd->cmnd[0] == READ_10 ||
844 cmd->cmnd[0] == WRITE_10)) {
845 /* This will issue a new 6-byte command. */
846 cmd->device->use_10_for_rw = 0;
847 action = ACTION_REPREP;
848 } else if (sshdr.asc == 0x10) /* DIX */ {
849 action = ACTION_FAIL;
851 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
852 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
853 action = ACTION_FAIL;
856 action = ACTION_FAIL;
858 case ABORTED_COMMAND:
859 action = ACTION_FAIL;
860 if (sshdr.asc == 0x10) /* DIF */
864 /* If the device is in the process of becoming
865 * ready, or has a temporary blockage, retry.
867 if (sshdr.asc == 0x04) {
868 switch (sshdr.ascq) {
869 case 0x01: /* becoming ready */
870 case 0x04: /* format in progress */
871 case 0x05: /* rebuild in progress */
872 case 0x06: /* recalculation in progress */
873 case 0x07: /* operation in progress */
874 case 0x08: /* Long write in progress */
875 case 0x09: /* self test in progress */
876 case 0x14: /* space allocation in progress */
877 action = ACTION_DELAYED_RETRY;
880 action = ACTION_FAIL;
884 action = ACTION_FAIL;
886 case VOLUME_OVERFLOW:
887 /* See SSC3rXX or current. */
888 action = ACTION_FAIL;
891 action = ACTION_FAIL;
895 action = ACTION_FAIL;
897 if (action != ACTION_FAIL &&
898 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
899 action = ACTION_FAIL;
903 /* Give up and fail the remainder of the request */
904 if (!(req->cmd_flags & REQ_QUIET)) {
905 scsi_print_result(cmd);
906 if (driver_byte(result) & DRIVER_SENSE)
907 scsi_print_sense("", cmd);
908 scsi_print_command(cmd);
910 if (!blk_end_request_err(req, error))
915 /* Unprep the request and put it back at the head of the queue.
916 * A new command will be prepared and issued.
918 scsi_release_buffers(cmd);
919 scsi_requeue_command(q, cmd);
922 /* Retry the same command immediately */
923 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
925 case ACTION_DELAYED_RETRY:
926 /* Retry the same command after a delay */
927 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
933 scsi_release_buffers(cmd);
934 scsi_next_command(cmd);
937 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
943 * If sg table allocation fails, requeue request later.
945 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
947 return BLKPREP_DEFER;
951 * Next, walk the list, and fill in the addresses and sizes of
954 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
955 BUG_ON(count > sdb->table.nents);
956 sdb->table.nents = count;
957 sdb->length = blk_rq_bytes(req);
962 * Function: scsi_init_io()
964 * Purpose: SCSI I/O initialize function.
966 * Arguments: cmd - Command descriptor we wish to initialize
968 * Returns: 0 on success
969 * BLKPREP_DEFER if the failure is retryable
970 * BLKPREP_KILL if the failure is fatal
972 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
974 struct scsi_device *sdev = cmd->device;
975 struct request *rq = cmd->request;
978 BUG_ON(!rq->nr_phys_segments);
980 error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
984 if (blk_bidi_rq(rq)) {
985 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
986 scsi_sdb_cache, GFP_ATOMIC);
988 error = BLKPREP_DEFER;
992 rq->next_rq->special = bidi_sdb;
993 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
998 if (blk_integrity_rq(rq)) {
999 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1002 BUG_ON(prot_sdb == NULL);
1003 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1005 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1006 error = BLKPREP_DEFER;
1010 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1011 prot_sdb->table.sgl);
1012 BUG_ON(unlikely(count > ivecs));
1013 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1015 cmd->prot_sdb = prot_sdb;
1016 cmd->prot_sdb->table.nents = count;
1022 scsi_release_buffers(cmd);
1023 cmd->request->special = NULL;
1024 scsi_put_command(cmd);
1025 put_device(&sdev->sdev_gendev);
1028 EXPORT_SYMBOL(scsi_init_io);
1030 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1031 struct request *req)
1033 struct scsi_cmnd *cmd;
1035 if (!req->special) {
1036 /* Bail if we can't get a reference to the device */
1037 if (!get_device(&sdev->sdev_gendev))
1040 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1041 if (unlikely(!cmd)) {
1042 put_device(&sdev->sdev_gendev);
1050 /* pull a tag out of the request if we have one */
1051 cmd->tag = req->tag;
1054 cmd->cmnd = req->cmd;
1055 cmd->prot_op = SCSI_PROT_NORMAL;
1060 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1062 struct scsi_cmnd *cmd = req->special;
1065 * BLOCK_PC requests may transfer data, in which case they must
1066 * a bio attached to them. Or they might contain a SCSI command
1067 * that does not transfer data, in which case they may optionally
1068 * submit a request without an attached bio.
1071 int ret = scsi_init_io(cmd, GFP_ATOMIC);
1075 BUG_ON(blk_rq_bytes(req));
1077 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1080 cmd->cmd_len = req->cmd_len;
1081 cmd->transfersize = blk_rq_bytes(req);
1082 cmd->allowed = req->retries;
1087 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1088 * that still need to be translated to SCSI CDBs from the ULD.
1090 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1092 struct scsi_cmnd *cmd = req->special;
1094 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1095 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1096 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1097 if (ret != BLKPREP_OK)
1101 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1102 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1105 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1107 struct scsi_cmnd *cmd = req->special;
1109 if (!blk_rq_bytes(req))
1110 cmd->sc_data_direction = DMA_NONE;
1111 else if (rq_data_dir(req) == WRITE)
1112 cmd->sc_data_direction = DMA_TO_DEVICE;
1114 cmd->sc_data_direction = DMA_FROM_DEVICE;
1116 switch (req->cmd_type) {
1118 return scsi_setup_fs_cmnd(sdev, req);
1119 case REQ_TYPE_BLOCK_PC:
1120 return scsi_setup_blk_pc_cmnd(sdev, req);
1122 return BLKPREP_KILL;
1127 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1129 int ret = BLKPREP_OK;
1132 * If the device is not in running state we will reject some
1135 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1136 switch (sdev->sdev_state) {
1138 case SDEV_TRANSPORT_OFFLINE:
1140 * If the device is offline we refuse to process any
1141 * commands. The device must be brought online
1142 * before trying any recovery commands.
1144 sdev_printk(KERN_ERR, sdev,
1145 "rejecting I/O to offline device\n");
1150 * If the device is fully deleted, we refuse to
1151 * process any commands as well.
1153 sdev_printk(KERN_ERR, sdev,
1154 "rejecting I/O to dead device\n");
1159 case SDEV_CREATED_BLOCK:
1161 * If the devices is blocked we defer normal commands.
1163 if (!(req->cmd_flags & REQ_PREEMPT))
1164 ret = BLKPREP_DEFER;
1168 * For any other not fully online state we only allow
1169 * special commands. In particular any user initiated
1170 * command is not allowed.
1172 if (!(req->cmd_flags & REQ_PREEMPT))
1181 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1183 struct scsi_device *sdev = q->queuedata;
1187 req->errors = DID_NO_CONNECT << 16;
1188 /* release the command and kill it */
1190 struct scsi_cmnd *cmd = req->special;
1191 scsi_release_buffers(cmd);
1192 scsi_put_command(cmd);
1193 put_device(&sdev->sdev_gendev);
1194 req->special = NULL;
1199 * If we defer, the blk_peek_request() returns NULL, but the
1200 * queue must be restarted, so we schedule a callback to happen
1203 if (sdev->device_busy == 0)
1204 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1207 req->cmd_flags |= REQ_DONTPREP;
1213 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1215 struct scsi_device *sdev = q->queuedata;
1216 struct scsi_cmnd *cmd;
1219 ret = scsi_prep_state_check(sdev, req);
1220 if (ret != BLKPREP_OK)
1223 cmd = scsi_get_cmd_from_req(sdev, req);
1224 if (unlikely(!cmd)) {
1225 ret = BLKPREP_DEFER;
1229 ret = scsi_setup_cmnd(sdev, req);
1231 return scsi_prep_return(q, req, ret);
1234 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1236 if (req->cmd_type == REQ_TYPE_FS) {
1237 struct scsi_cmnd *cmd = req->special;
1238 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
1240 if (drv->uninit_command)
1241 drv->uninit_command(cmd);
1246 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1249 * Called with the queue_lock held.
1251 static inline int scsi_dev_queue_ready(struct request_queue *q,
1252 struct scsi_device *sdev)
1254 if (sdev->device_busy == 0 && sdev->device_blocked) {
1256 * unblock after device_blocked iterates to zero
1258 if (--sdev->device_blocked == 0) {
1260 sdev_printk(KERN_INFO, sdev,
1261 "unblocking device at zero depth\n"));
1263 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1267 if (scsi_device_is_busy(sdev))
1275 * scsi_target_queue_ready: checks if there we can send commands to target
1276 * @sdev: scsi device on starget to check.
1278 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1279 struct scsi_device *sdev)
1281 struct scsi_target *starget = scsi_target(sdev);
1284 if (starget->single_lun) {
1285 spin_lock_irq(shost->host_lock);
1286 if (starget->starget_sdev_user &&
1287 starget->starget_sdev_user != sdev) {
1288 spin_unlock_irq(shost->host_lock);
1291 starget->starget_sdev_user = sdev;
1292 spin_unlock_irq(shost->host_lock);
1295 busy = atomic_inc_return(&starget->target_busy) - 1;
1296 if (starget->target_blocked) {
1301 * unblock after target_blocked iterates to zero
1303 spin_lock_irq(shost->host_lock);
1304 if (--starget->target_blocked != 0) {
1305 spin_unlock_irq(shost->host_lock);
1308 spin_unlock_irq(shost->host_lock);
1310 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1311 "unblocking target at zero depth\n"));
1314 if (starget->can_queue > 0 && busy >= starget->can_queue)
1320 spin_lock_irq(shost->host_lock);
1321 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1322 spin_unlock_irq(shost->host_lock);
1324 atomic_dec(&starget->target_busy);
1329 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1330 * return 0. We must end up running the queue again whenever 0 is
1331 * returned, else IO can hang.
1333 static inline int scsi_host_queue_ready(struct request_queue *q,
1334 struct Scsi_Host *shost,
1335 struct scsi_device *sdev)
1339 spin_lock_irq(shost->host_lock);
1341 if (scsi_host_in_recovery(shost))
1343 if (shost->host_busy == 0 && shost->host_blocked) {
1345 * unblock after host_blocked iterates to zero
1347 if (--shost->host_blocked != 0)
1351 shost_printk(KERN_INFO, shost,
1352 "unblocking host at zero depth\n"));
1354 if (scsi_host_is_busy(shost)) {
1355 if (list_empty(&sdev->starved_entry))
1356 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1360 /* We're OK to process the command, so we can't be starved */
1361 if (!list_empty(&sdev->starved_entry))
1362 list_del_init(&sdev->starved_entry);
1367 spin_unlock_irq(shost->host_lock);
1372 * Busy state exporting function for request stacking drivers.
1374 * For efficiency, no lock is taken to check the busy state of
1375 * shost/starget/sdev, since the returned value is not guaranteed and
1376 * may be changed after request stacking drivers call the function,
1377 * regardless of taking lock or not.
1379 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1380 * needs to return 'not busy'. Otherwise, request stacking drivers
1381 * may hold requests forever.
1383 static int scsi_lld_busy(struct request_queue *q)
1385 struct scsi_device *sdev = q->queuedata;
1386 struct Scsi_Host *shost;
1388 if (blk_queue_dying(q))
1394 * Ignore host/starget busy state.
1395 * Since block layer does not have a concept of fairness across
1396 * multiple queues, congestion of host/starget needs to be handled
1399 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1406 * Kill a request for a dead device
1408 static void scsi_kill_request(struct request *req, struct request_queue *q)
1410 struct scsi_cmnd *cmd = req->special;
1411 struct scsi_device *sdev;
1412 struct scsi_target *starget;
1413 struct Scsi_Host *shost;
1415 blk_start_request(req);
1417 scmd_printk(KERN_INFO, cmd, "killing request\n");
1420 starget = scsi_target(sdev);
1422 scsi_init_cmd_errh(cmd);
1423 cmd->result = DID_NO_CONNECT << 16;
1424 atomic_inc(&cmd->device->iorequest_cnt);
1427 * SCSI request completion path will do scsi_device_unbusy(),
1428 * bump busy counts. To bump the counters, we need to dance
1429 * with the locks as normal issue path does.
1431 sdev->device_busy++;
1432 spin_unlock(sdev->request_queue->queue_lock);
1433 spin_lock(shost->host_lock);
1435 atomic_inc(&starget->target_busy);
1436 spin_unlock(shost->host_lock);
1437 spin_lock(sdev->request_queue->queue_lock);
1439 blk_complete_request(req);
1442 static void scsi_softirq_done(struct request *rq)
1444 struct scsi_cmnd *cmd = rq->special;
1445 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1448 INIT_LIST_HEAD(&cmd->eh_entry);
1450 atomic_inc(&cmd->device->iodone_cnt);
1452 atomic_inc(&cmd->device->ioerr_cnt);
1454 disposition = scsi_decide_disposition(cmd);
1455 if (disposition != SUCCESS &&
1456 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1457 sdev_printk(KERN_ERR, cmd->device,
1458 "timing out command, waited %lus\n",
1460 disposition = SUCCESS;
1463 scsi_log_completion(cmd, disposition);
1465 switch (disposition) {
1467 scsi_finish_command(cmd);
1470 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1472 case ADD_TO_MLQUEUE:
1473 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1476 if (!scsi_eh_scmd_add(cmd, 0))
1477 scsi_finish_command(cmd);
1482 * scsi_done - Invoke completion on finished SCSI command.
1483 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1484 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1486 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1487 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1488 * calls blk_complete_request() for further processing.
1490 * This function is interrupt context safe.
1492 static void scsi_done(struct scsi_cmnd *cmd)
1494 trace_scsi_dispatch_cmd_done(cmd);
1495 blk_complete_request(cmd->request);
1499 * Function: scsi_request_fn()
1501 * Purpose: Main strategy routine for SCSI.
1503 * Arguments: q - Pointer to actual queue.
1507 * Lock status: IO request lock assumed to be held when called.
1509 static void scsi_request_fn(struct request_queue *q)
1510 __releases(q->queue_lock)
1511 __acquires(q->queue_lock)
1513 struct scsi_device *sdev = q->queuedata;
1514 struct Scsi_Host *shost;
1515 struct scsi_cmnd *cmd;
1516 struct request *req;
1519 * To start with, we keep looping until the queue is empty, or until
1520 * the host is no longer able to accept any more requests.
1526 * get next queueable request. We do this early to make sure
1527 * that the request is fully prepared even if we cannot
1530 req = blk_peek_request(q);
1531 if (!req || !scsi_dev_queue_ready(q, sdev))
1534 if (unlikely(!scsi_device_online(sdev))) {
1535 sdev_printk(KERN_ERR, sdev,
1536 "rejecting I/O to offline device\n");
1537 scsi_kill_request(req, q);
1543 * Remove the request from the request list.
1545 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1546 blk_start_request(req);
1547 sdev->device_busy++;
1549 spin_unlock_irq(q->queue_lock);
1551 if (unlikely(cmd == NULL)) {
1552 printk(KERN_CRIT "impossible request in %s.\n"
1553 "please mail a stack trace to "
1554 "linux-scsi@vger.kernel.org\n",
1556 blk_dump_rq_flags(req, "foo");
1561 * We hit this when the driver is using a host wide
1562 * tag map. For device level tag maps the queue_depth check
1563 * in the device ready fn would prevent us from trying
1564 * to allocate a tag. Since the map is a shared host resource
1565 * we add the dev to the starved list so it eventually gets
1566 * a run when a tag is freed.
1568 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1569 spin_lock_irq(shost->host_lock);
1570 if (list_empty(&sdev->starved_entry))
1571 list_add_tail(&sdev->starved_entry,
1572 &shost->starved_list);
1573 spin_unlock_irq(shost->host_lock);
1577 if (!scsi_target_queue_ready(shost, sdev))
1580 if (!scsi_host_queue_ready(q, shost, sdev))
1581 goto host_not_ready;
1584 * Finally, initialize any error handling parameters, and set up
1585 * the timers for timeouts.
1587 scsi_init_cmd_errh(cmd);
1590 * Dispatch the command to the low-level driver.
1592 cmd->scsi_done = scsi_done;
1593 rtn = scsi_dispatch_cmd(cmd);
1595 scsi_queue_insert(cmd, rtn);
1596 spin_lock_irq(q->queue_lock);
1599 spin_lock_irq(q->queue_lock);
1605 atomic_dec(&scsi_target(sdev)->target_busy);
1608 * lock q, handle tag, requeue req, and decrement device_busy. We
1609 * must return with queue_lock held.
1611 * Decrementing device_busy without checking it is OK, as all such
1612 * cases (host limits or settings) should run the queue at some
1615 spin_lock_irq(q->queue_lock);
1616 blk_requeue_request(q, req);
1617 sdev->device_busy--;
1619 if (sdev->device_busy == 0 && !scsi_device_blocked(sdev))
1620 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1623 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1625 struct device *host_dev;
1626 u64 bounce_limit = 0xffffffff;
1628 if (shost->unchecked_isa_dma)
1629 return BLK_BOUNCE_ISA;
1631 * Platforms with virtual-DMA translation
1632 * hardware have no practical limit.
1634 if (!PCI_DMA_BUS_IS_PHYS)
1635 return BLK_BOUNCE_ANY;
1637 host_dev = scsi_get_device(shost);
1638 if (host_dev && host_dev->dma_mask)
1639 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1641 return bounce_limit;
1644 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1645 request_fn_proc *request_fn)
1647 struct request_queue *q;
1648 struct device *dev = shost->dma_dev;
1650 q = blk_init_queue(request_fn, NULL);
1655 * this limit is imposed by hardware restrictions
1657 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1658 SCSI_MAX_SG_CHAIN_SEGMENTS));
1660 if (scsi_host_prot_dma(shost)) {
1661 shost->sg_prot_tablesize =
1662 min_not_zero(shost->sg_prot_tablesize,
1663 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1664 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1665 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1668 blk_queue_max_hw_sectors(q, shost->max_sectors);
1669 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1670 blk_queue_segment_boundary(q, shost->dma_boundary);
1671 dma_set_seg_boundary(dev, shost->dma_boundary);
1673 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1675 if (!shost->use_clustering)
1676 q->limits.cluster = 0;
1679 * set a reasonable default alignment on word boundaries: the
1680 * host and device may alter it using
1681 * blk_queue_update_dma_alignment() later.
1683 blk_queue_dma_alignment(q, 0x03);
1687 EXPORT_SYMBOL(__scsi_alloc_queue);
1689 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1691 struct request_queue *q;
1693 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1697 blk_queue_prep_rq(q, scsi_prep_fn);
1698 blk_queue_unprep_rq(q, scsi_unprep_fn);
1699 blk_queue_softirq_done(q, scsi_softirq_done);
1700 blk_queue_rq_timed_out(q, scsi_times_out);
1701 blk_queue_lld_busy(q, scsi_lld_busy);
1706 * Function: scsi_block_requests()
1708 * Purpose: Utility function used by low-level drivers to prevent further
1709 * commands from being queued to the device.
1711 * Arguments: shost - Host in question
1715 * Lock status: No locks are assumed held.
1717 * Notes: There is no timer nor any other means by which the requests
1718 * get unblocked other than the low-level driver calling
1719 * scsi_unblock_requests().
1721 void scsi_block_requests(struct Scsi_Host *shost)
1723 shost->host_self_blocked = 1;
1725 EXPORT_SYMBOL(scsi_block_requests);
1728 * Function: scsi_unblock_requests()
1730 * Purpose: Utility function used by low-level drivers to allow further
1731 * commands from being queued to the device.
1733 * Arguments: shost - Host in question
1737 * Lock status: No locks are assumed held.
1739 * Notes: There is no timer nor any other means by which the requests
1740 * get unblocked other than the low-level driver calling
1741 * scsi_unblock_requests().
1743 * This is done as an API function so that changes to the
1744 * internals of the scsi mid-layer won't require wholesale
1745 * changes to drivers that use this feature.
1747 void scsi_unblock_requests(struct Scsi_Host *shost)
1749 shost->host_self_blocked = 0;
1750 scsi_run_host_queues(shost);
1752 EXPORT_SYMBOL(scsi_unblock_requests);
1754 int __init scsi_init_queue(void)
1758 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1759 sizeof(struct scsi_data_buffer),
1761 if (!scsi_sdb_cache) {
1762 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1766 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1767 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1768 int size = sgp->size * sizeof(struct scatterlist);
1770 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1771 SLAB_HWCACHE_ALIGN, NULL);
1773 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1778 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1781 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1790 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1791 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1793 mempool_destroy(sgp->pool);
1795 kmem_cache_destroy(sgp->slab);
1797 kmem_cache_destroy(scsi_sdb_cache);
1802 void scsi_exit_queue(void)
1806 kmem_cache_destroy(scsi_sdb_cache);
1808 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1809 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1810 mempool_destroy(sgp->pool);
1811 kmem_cache_destroy(sgp->slab);
1816 * scsi_mode_select - issue a mode select
1817 * @sdev: SCSI device to be queried
1818 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1819 * @sp: Save page bit (0 == don't save, 1 == save)
1820 * @modepage: mode page being requested
1821 * @buffer: request buffer (may not be smaller than eight bytes)
1822 * @len: length of request buffer.
1823 * @timeout: command timeout
1824 * @retries: number of retries before failing
1825 * @data: returns a structure abstracting the mode header data
1826 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1827 * must be SCSI_SENSE_BUFFERSIZE big.
1829 * Returns zero if successful; negative error number or scsi
1834 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1835 unsigned char *buffer, int len, int timeout, int retries,
1836 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1838 unsigned char cmd[10];
1839 unsigned char *real_buffer;
1842 memset(cmd, 0, sizeof(cmd));
1843 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1845 if (sdev->use_10_for_ms) {
1848 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1851 memcpy(real_buffer + 8, buffer, len);
1855 real_buffer[2] = data->medium_type;
1856 real_buffer[3] = data->device_specific;
1857 real_buffer[4] = data->longlba ? 0x01 : 0;
1859 real_buffer[6] = data->block_descriptor_length >> 8;
1860 real_buffer[7] = data->block_descriptor_length;
1862 cmd[0] = MODE_SELECT_10;
1866 if (len > 255 || data->block_descriptor_length > 255 ||
1870 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1873 memcpy(real_buffer + 4, buffer, len);
1876 real_buffer[1] = data->medium_type;
1877 real_buffer[2] = data->device_specific;
1878 real_buffer[3] = data->block_descriptor_length;
1881 cmd[0] = MODE_SELECT;
1885 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1886 sshdr, timeout, retries, NULL);
1890 EXPORT_SYMBOL_GPL(scsi_mode_select);
1893 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1894 * @sdev: SCSI device to be queried
1895 * @dbd: set if mode sense will allow block descriptors to be returned
1896 * @modepage: mode page being requested
1897 * @buffer: request buffer (may not be smaller than eight bytes)
1898 * @len: length of request buffer.
1899 * @timeout: command timeout
1900 * @retries: number of retries before failing
1901 * @data: returns a structure abstracting the mode header data
1902 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1903 * must be SCSI_SENSE_BUFFERSIZE big.
1905 * Returns zero if unsuccessful, or the header offset (either 4
1906 * or 8 depending on whether a six or ten byte command was
1907 * issued) if successful.
1910 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1911 unsigned char *buffer, int len, int timeout, int retries,
1912 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1914 unsigned char cmd[12];
1918 struct scsi_sense_hdr my_sshdr;
1920 memset(data, 0, sizeof(*data));
1921 memset(&cmd[0], 0, 12);
1922 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1925 /* caller might not be interested in sense, but we need it */
1930 use_10_for_ms = sdev->use_10_for_ms;
1932 if (use_10_for_ms) {
1936 cmd[0] = MODE_SENSE_10;
1943 cmd[0] = MODE_SENSE;
1948 memset(buffer, 0, len);
1950 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1951 sshdr, timeout, retries, NULL);
1953 /* This code looks awful: what it's doing is making sure an
1954 * ILLEGAL REQUEST sense return identifies the actual command
1955 * byte as the problem. MODE_SENSE commands can return
1956 * ILLEGAL REQUEST if the code page isn't supported */
1958 if (use_10_for_ms && !scsi_status_is_good(result) &&
1959 (driver_byte(result) & DRIVER_SENSE)) {
1960 if (scsi_sense_valid(sshdr)) {
1961 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1962 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1964 * Invalid command operation code
1966 sdev->use_10_for_ms = 0;
1972 if(scsi_status_is_good(result)) {
1973 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1974 (modepage == 6 || modepage == 8))) {
1975 /* Initio breakage? */
1978 data->medium_type = 0;
1979 data->device_specific = 0;
1981 data->block_descriptor_length = 0;
1982 } else if(use_10_for_ms) {
1983 data->length = buffer[0]*256 + buffer[1] + 2;
1984 data->medium_type = buffer[2];
1985 data->device_specific = buffer[3];
1986 data->longlba = buffer[4] & 0x01;
1987 data->block_descriptor_length = buffer[6]*256
1990 data->length = buffer[0] + 1;
1991 data->medium_type = buffer[1];
1992 data->device_specific = buffer[2];
1993 data->block_descriptor_length = buffer[3];
1995 data->header_length = header_length;
2000 EXPORT_SYMBOL(scsi_mode_sense);
2003 * scsi_test_unit_ready - test if unit is ready
2004 * @sdev: scsi device to change the state of.
2005 * @timeout: command timeout
2006 * @retries: number of retries before failing
2007 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2008 * returning sense. Make sure that this is cleared before passing
2011 * Returns zero if unsuccessful or an error if TUR failed. For
2012 * removable media, UNIT_ATTENTION sets ->changed flag.
2015 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2016 struct scsi_sense_hdr *sshdr_external)
2019 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2021 struct scsi_sense_hdr *sshdr;
2024 if (!sshdr_external)
2025 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2027 sshdr = sshdr_external;
2029 /* try to eat the UNIT_ATTENTION if there are enough retries */
2031 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2032 timeout, retries, NULL);
2033 if (sdev->removable && scsi_sense_valid(sshdr) &&
2034 sshdr->sense_key == UNIT_ATTENTION)
2036 } while (scsi_sense_valid(sshdr) &&
2037 sshdr->sense_key == UNIT_ATTENTION && --retries);
2039 if (!sshdr_external)
2043 EXPORT_SYMBOL(scsi_test_unit_ready);
2046 * scsi_device_set_state - Take the given device through the device state model.
2047 * @sdev: scsi device to change the state of.
2048 * @state: state to change to.
2050 * Returns zero if unsuccessful or an error if the requested
2051 * transition is illegal.
2054 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2056 enum scsi_device_state oldstate = sdev->sdev_state;
2058 if (state == oldstate)
2064 case SDEV_CREATED_BLOCK:
2075 case SDEV_TRANSPORT_OFFLINE:
2088 case SDEV_TRANSPORT_OFFLINE:
2096 case SDEV_TRANSPORT_OFFLINE:
2111 case SDEV_CREATED_BLOCK:
2118 case SDEV_CREATED_BLOCK:
2133 case SDEV_TRANSPORT_OFFLINE:
2146 case SDEV_TRANSPORT_OFFLINE:
2148 case SDEV_CREATED_BLOCK:
2156 sdev->sdev_state = state;
2160 SCSI_LOG_ERROR_RECOVERY(1,
2161 sdev_printk(KERN_ERR, sdev,
2162 "Illegal state transition %s->%s",
2163 scsi_device_state_name(oldstate),
2164 scsi_device_state_name(state))
2168 EXPORT_SYMBOL(scsi_device_set_state);
2171 * sdev_evt_emit - emit a single SCSI device uevent
2172 * @sdev: associated SCSI device
2173 * @evt: event to emit
2175 * Send a single uevent (scsi_event) to the associated scsi_device.
2177 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2182 switch (evt->evt_type) {
2183 case SDEV_EVT_MEDIA_CHANGE:
2184 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2186 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2187 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2189 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2190 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2192 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2193 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2195 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2196 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2198 case SDEV_EVT_LUN_CHANGE_REPORTED:
2199 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2208 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2212 * sdev_evt_thread - send a uevent for each scsi event
2213 * @work: work struct for scsi_device
2215 * Dispatch queued events to their associated scsi_device kobjects
2218 void scsi_evt_thread(struct work_struct *work)
2220 struct scsi_device *sdev;
2221 enum scsi_device_event evt_type;
2222 LIST_HEAD(event_list);
2224 sdev = container_of(work, struct scsi_device, event_work);
2226 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2227 if (test_and_clear_bit(evt_type, sdev->pending_events))
2228 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2231 struct scsi_event *evt;
2232 struct list_head *this, *tmp;
2233 unsigned long flags;
2235 spin_lock_irqsave(&sdev->list_lock, flags);
2236 list_splice_init(&sdev->event_list, &event_list);
2237 spin_unlock_irqrestore(&sdev->list_lock, flags);
2239 if (list_empty(&event_list))
2242 list_for_each_safe(this, tmp, &event_list) {
2243 evt = list_entry(this, struct scsi_event, node);
2244 list_del(&evt->node);
2245 scsi_evt_emit(sdev, evt);
2252 * sdev_evt_send - send asserted event to uevent thread
2253 * @sdev: scsi_device event occurred on
2254 * @evt: event to send
2256 * Assert scsi device event asynchronously.
2258 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2260 unsigned long flags;
2263 /* FIXME: currently this check eliminates all media change events
2264 * for polled devices. Need to update to discriminate between AN
2265 * and polled events */
2266 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2272 spin_lock_irqsave(&sdev->list_lock, flags);
2273 list_add_tail(&evt->node, &sdev->event_list);
2274 schedule_work(&sdev->event_work);
2275 spin_unlock_irqrestore(&sdev->list_lock, flags);
2277 EXPORT_SYMBOL_GPL(sdev_evt_send);
2280 * sdev_evt_alloc - allocate a new scsi event
2281 * @evt_type: type of event to allocate
2282 * @gfpflags: GFP flags for allocation
2284 * Allocates and returns a new scsi_event.
2286 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2289 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2293 evt->evt_type = evt_type;
2294 INIT_LIST_HEAD(&evt->node);
2296 /* evt_type-specific initialization, if any */
2298 case SDEV_EVT_MEDIA_CHANGE:
2299 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2300 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2301 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2302 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2303 case SDEV_EVT_LUN_CHANGE_REPORTED:
2311 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2314 * sdev_evt_send_simple - send asserted event to uevent thread
2315 * @sdev: scsi_device event occurred on
2316 * @evt_type: type of event to send
2317 * @gfpflags: GFP flags for allocation
2319 * Assert scsi device event asynchronously, given an event type.
2321 void sdev_evt_send_simple(struct scsi_device *sdev,
2322 enum scsi_device_event evt_type, gfp_t gfpflags)
2324 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2326 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2331 sdev_evt_send(sdev, evt);
2333 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2336 * scsi_device_quiesce - Block user issued commands.
2337 * @sdev: scsi device to quiesce.
2339 * This works by trying to transition to the SDEV_QUIESCE state
2340 * (which must be a legal transition). When the device is in this
2341 * state, only special requests will be accepted, all others will
2342 * be deferred. Since special requests may also be requeued requests,
2343 * a successful return doesn't guarantee the device will be
2344 * totally quiescent.
2346 * Must be called with user context, may sleep.
2348 * Returns zero if unsuccessful or an error if not.
2351 scsi_device_quiesce(struct scsi_device *sdev)
2353 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2357 scsi_run_queue(sdev->request_queue);
2358 while (sdev->device_busy) {
2359 msleep_interruptible(200);
2360 scsi_run_queue(sdev->request_queue);
2364 EXPORT_SYMBOL(scsi_device_quiesce);
2367 * scsi_device_resume - Restart user issued commands to a quiesced device.
2368 * @sdev: scsi device to resume.
2370 * Moves the device from quiesced back to running and restarts the
2373 * Must be called with user context, may sleep.
2375 void scsi_device_resume(struct scsi_device *sdev)
2377 /* check if the device state was mutated prior to resume, and if
2378 * so assume the state is being managed elsewhere (for example
2379 * device deleted during suspend)
2381 if (sdev->sdev_state != SDEV_QUIESCE ||
2382 scsi_device_set_state(sdev, SDEV_RUNNING))
2384 scsi_run_queue(sdev->request_queue);
2386 EXPORT_SYMBOL(scsi_device_resume);
2389 device_quiesce_fn(struct scsi_device *sdev, void *data)
2391 scsi_device_quiesce(sdev);
2395 scsi_target_quiesce(struct scsi_target *starget)
2397 starget_for_each_device(starget, NULL, device_quiesce_fn);
2399 EXPORT_SYMBOL(scsi_target_quiesce);
2402 device_resume_fn(struct scsi_device *sdev, void *data)
2404 scsi_device_resume(sdev);
2408 scsi_target_resume(struct scsi_target *starget)
2410 starget_for_each_device(starget, NULL, device_resume_fn);
2412 EXPORT_SYMBOL(scsi_target_resume);
2415 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2416 * @sdev: device to block
2418 * Block request made by scsi lld's to temporarily stop all
2419 * scsi commands on the specified device. Called from interrupt
2420 * or normal process context.
2422 * Returns zero if successful or error if not
2425 * This routine transitions the device to the SDEV_BLOCK state
2426 * (which must be a legal transition). When the device is in this
2427 * state, all commands are deferred until the scsi lld reenables
2428 * the device with scsi_device_unblock or device_block_tmo fires.
2431 scsi_internal_device_block(struct scsi_device *sdev)
2433 struct request_queue *q = sdev->request_queue;
2434 unsigned long flags;
2437 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2439 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2446 * The device has transitioned to SDEV_BLOCK. Stop the
2447 * block layer from calling the midlayer with this device's
2450 spin_lock_irqsave(q->queue_lock, flags);
2452 spin_unlock_irqrestore(q->queue_lock, flags);
2456 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2459 * scsi_internal_device_unblock - resume a device after a block request
2460 * @sdev: device to resume
2461 * @new_state: state to set devices to after unblocking
2463 * Called by scsi lld's or the midlayer to restart the device queue
2464 * for the previously suspended scsi device. Called from interrupt or
2465 * normal process context.
2467 * Returns zero if successful or error if not.
2470 * This routine transitions the device to the SDEV_RUNNING state
2471 * or to one of the offline states (which must be a legal transition)
2472 * allowing the midlayer to goose the queue for this device.
2475 scsi_internal_device_unblock(struct scsi_device *sdev,
2476 enum scsi_device_state new_state)
2478 struct request_queue *q = sdev->request_queue;
2479 unsigned long flags;
2482 * Try to transition the scsi device to SDEV_RUNNING or one of the
2483 * offlined states and goose the device queue if successful.
2485 if ((sdev->sdev_state == SDEV_BLOCK) ||
2486 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2487 sdev->sdev_state = new_state;
2488 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2489 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2490 new_state == SDEV_OFFLINE)
2491 sdev->sdev_state = new_state;
2493 sdev->sdev_state = SDEV_CREATED;
2494 } else if (sdev->sdev_state != SDEV_CANCEL &&
2495 sdev->sdev_state != SDEV_OFFLINE)
2498 spin_lock_irqsave(q->queue_lock, flags);
2500 spin_unlock_irqrestore(q->queue_lock, flags);
2504 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2507 device_block(struct scsi_device *sdev, void *data)
2509 scsi_internal_device_block(sdev);
2513 target_block(struct device *dev, void *data)
2515 if (scsi_is_target_device(dev))
2516 starget_for_each_device(to_scsi_target(dev), NULL,
2522 scsi_target_block(struct device *dev)
2524 if (scsi_is_target_device(dev))
2525 starget_for_each_device(to_scsi_target(dev), NULL,
2528 device_for_each_child(dev, NULL, target_block);
2530 EXPORT_SYMBOL_GPL(scsi_target_block);
2533 device_unblock(struct scsi_device *sdev, void *data)
2535 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2539 target_unblock(struct device *dev, void *data)
2541 if (scsi_is_target_device(dev))
2542 starget_for_each_device(to_scsi_target(dev), data,
2548 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2550 if (scsi_is_target_device(dev))
2551 starget_for_each_device(to_scsi_target(dev), &new_state,
2554 device_for_each_child(dev, &new_state, target_unblock);
2556 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2559 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2560 * @sgl: scatter-gather list
2561 * @sg_count: number of segments in sg
2562 * @offset: offset in bytes into sg, on return offset into the mapped area
2563 * @len: bytes to map, on return number of bytes mapped
2565 * Returns virtual address of the start of the mapped page
2567 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2568 size_t *offset, size_t *len)
2571 size_t sg_len = 0, len_complete = 0;
2572 struct scatterlist *sg;
2575 WARN_ON(!irqs_disabled());
2577 for_each_sg(sgl, sg, sg_count, i) {
2578 len_complete = sg_len; /* Complete sg-entries */
2579 sg_len += sg->length;
2580 if (sg_len > *offset)
2584 if (unlikely(i == sg_count)) {
2585 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2587 __func__, sg_len, *offset, sg_count);
2592 /* Offset starting from the beginning of first page in this sg-entry */
2593 *offset = *offset - len_complete + sg->offset;
2595 /* Assumption: contiguous pages can be accessed as "page + i" */
2596 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2597 *offset &= ~PAGE_MASK;
2599 /* Bytes in this sg-entry from *offset to the end of the page */
2600 sg_len = PAGE_SIZE - *offset;
2604 return kmap_atomic(page);
2606 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2609 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2610 * @virt: virtual address to be unmapped
2612 void scsi_kunmap_atomic_sg(void *virt)
2614 kunmap_atomic(virt);
2616 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2618 void sdev_disable_disk_events(struct scsi_device *sdev)
2620 atomic_inc(&sdev->disk_events_disable_depth);
2622 EXPORT_SYMBOL(sdev_disable_disk_events);
2624 void sdev_enable_disk_events(struct scsi_device *sdev)
2626 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2628 atomic_dec(&sdev->disk_events_disable_depth);
2630 EXPORT_SYMBOL(sdev_enable_disk_events);