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 "scsi_priv.h"
33 #include "scsi_logging.h"
36 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
37 #define SG_MEMPOOL_SIZE 2
39 struct scsi_host_sg_pool {
42 struct kmem_cache *slab;
46 #define SP(x) { x, "sgpool-" __stringify(x) }
47 #if (SCSI_MAX_SG_SEGMENTS < 32)
48 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
50 static struct scsi_host_sg_pool scsi_sg_pools[] = {
53 #if (SCSI_MAX_SG_SEGMENTS > 32)
55 #if (SCSI_MAX_SG_SEGMENTS > 64)
57 #if (SCSI_MAX_SG_SEGMENTS > 128)
59 #if (SCSI_MAX_SG_SEGMENTS > 256)
60 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
65 SP(SCSI_MAX_SG_SEGMENTS)
69 struct kmem_cache *scsi_sdb_cache;
72 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
73 * not change behaviour from the previous unplug mechanism, experimentation
74 * may prove this needs changing.
76 #define SCSI_QUEUE_DELAY 3
79 * __scsi_queue_insert - private queue insertion
80 * @cmd: The SCSI command being requeued
81 * @reason: The reason for the requeue
82 * @unbusy: Whether the queue should be unbusied
84 * This is a private queue insertion. The public interface
85 * scsi_queue_insert() always assumes the queue should be unbusied
86 * because it's always called before the completion. This function is
87 * for a requeue after completion, which should only occur in this
90 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
92 struct Scsi_Host *host = cmd->device->host;
93 struct scsi_device *device = cmd->device;
94 struct scsi_target *starget = scsi_target(device);
95 struct request_queue *q = device->request_queue;
99 printk("Inserting command %p into mlqueue\n", cmd));
102 * Set the appropriate busy bit for the device/host.
104 * If the host/device isn't busy, assume that something actually
105 * completed, and that we should be able to queue a command now.
107 * Note that the prior mid-layer assumption that any host could
108 * always queue at least one command is now broken. The mid-layer
109 * will implement a user specifiable stall (see
110 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
111 * if a command is requeued with no other commands outstanding
112 * either for the device or for the host.
115 case SCSI_MLQUEUE_HOST_BUSY:
116 host->host_blocked = host->max_host_blocked;
118 case SCSI_MLQUEUE_DEVICE_BUSY:
119 case SCSI_MLQUEUE_EH_RETRY:
120 device->device_blocked = device->max_device_blocked;
122 case SCSI_MLQUEUE_TARGET_BUSY:
123 starget->target_blocked = starget->max_target_blocked;
128 * Decrement the counters, since these commands are no longer
129 * active on the host/device.
132 scsi_device_unbusy(device);
135 * Requeue this command. It will go before all other commands
136 * that are already in the queue. Schedule requeue work under
137 * lock such that the kblockd_schedule_work() call happens
138 * before blk_cleanup_queue() finishes.
141 spin_lock_irqsave(q->queue_lock, flags);
142 blk_requeue_request(q, cmd->request);
143 kblockd_schedule_work(&device->requeue_work);
144 spin_unlock_irqrestore(q->queue_lock, flags);
148 * Function: scsi_queue_insert()
150 * Purpose: Insert a command in the midlevel queue.
152 * Arguments: cmd - command that we are adding to queue.
153 * reason - why we are inserting command to queue.
155 * Lock status: Assumed that lock is not held upon entry.
159 * Notes: We do this for one of two cases. Either the host is busy
160 * and it cannot accept any more commands for the time being,
161 * or the device returned QUEUE_FULL and can accept no more
163 * Notes: This could be called either from an interrupt context or a
164 * normal process context.
166 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
168 __scsi_queue_insert(cmd, reason, 1);
171 * scsi_execute - insert request and wait for the result
174 * @data_direction: data direction
175 * @buffer: data buffer
176 * @bufflen: len of buffer
177 * @sense: optional sense buffer
178 * @timeout: request timeout in seconds
179 * @retries: number of times to retry request
180 * @flags: or into request flags;
181 * @resid: optional residual length
183 * returns the req->errors value which is the scsi_cmnd result
186 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
187 int data_direction, void *buffer, unsigned bufflen,
188 unsigned char *sense, int timeout, int retries, u64 flags,
192 int write = (data_direction == DMA_TO_DEVICE);
193 int ret = DRIVER_ERROR << 24;
195 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
198 blk_rq_set_block_pc(req);
200 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
201 buffer, bufflen, __GFP_WAIT))
204 req->cmd_len = COMMAND_SIZE(cmd[0]);
205 memcpy(req->cmd, cmd, req->cmd_len);
208 req->retries = retries;
209 req->timeout = timeout;
210 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
213 * head injection *required* here otherwise quiesce won't work
215 blk_execute_rq(req->q, NULL, req, 1);
218 * Some devices (USB mass-storage in particular) may transfer
219 * garbage data together with a residue indicating that the data
220 * is invalid. Prevent the garbage from being misinterpreted
221 * and prevent security leaks by zeroing out the excess data.
223 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
224 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
227 *resid = req->resid_len;
230 blk_put_request(req);
234 EXPORT_SYMBOL(scsi_execute);
236 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
237 int data_direction, void *buffer, unsigned bufflen,
238 struct scsi_sense_hdr *sshdr, int timeout, int retries,
239 int *resid, u64 flags)
245 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
247 return DRIVER_ERROR << 24;
249 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
250 sense, timeout, retries, flags, resid);
252 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
257 EXPORT_SYMBOL(scsi_execute_req_flags);
260 * Function: scsi_init_cmd_errh()
262 * Purpose: Initialize cmd fields related to error handling.
264 * Arguments: cmd - command that is ready to be queued.
266 * Notes: This function has the job of initializing a number of
267 * fields related to error handling. Typically this will
268 * be called once for each command, as required.
270 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
272 cmd->serial_number = 0;
273 scsi_set_resid(cmd, 0);
274 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
275 if (cmd->cmd_len == 0)
276 cmd->cmd_len = scsi_command_size(cmd->cmnd);
279 void scsi_device_unbusy(struct scsi_device *sdev)
281 struct Scsi_Host *shost = sdev->host;
282 struct scsi_target *starget = scsi_target(sdev);
285 spin_lock_irqsave(shost->host_lock, flags);
287 starget->target_busy--;
288 if (unlikely(scsi_host_in_recovery(shost) &&
289 (shost->host_failed || shost->host_eh_scheduled)))
290 scsi_eh_wakeup(shost);
291 spin_unlock(shost->host_lock);
292 spin_lock(sdev->request_queue->queue_lock);
294 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
298 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
299 * and call blk_run_queue for all the scsi_devices on the target -
300 * including current_sdev first.
302 * Called with *no* scsi locks held.
304 static void scsi_single_lun_run(struct scsi_device *current_sdev)
306 struct Scsi_Host *shost = current_sdev->host;
307 struct scsi_device *sdev, *tmp;
308 struct scsi_target *starget = scsi_target(current_sdev);
311 spin_lock_irqsave(shost->host_lock, flags);
312 starget->starget_sdev_user = NULL;
313 spin_unlock_irqrestore(shost->host_lock, flags);
316 * Call blk_run_queue for all LUNs on the target, starting with
317 * current_sdev. We race with others (to set starget_sdev_user),
318 * but in most cases, we will be first. Ideally, each LU on the
319 * target would get some limited time or requests on the target.
321 blk_run_queue(current_sdev->request_queue);
323 spin_lock_irqsave(shost->host_lock, flags);
324 if (starget->starget_sdev_user)
326 list_for_each_entry_safe(sdev, tmp, &starget->devices,
327 same_target_siblings) {
328 if (sdev == current_sdev)
330 if (scsi_device_get(sdev))
333 spin_unlock_irqrestore(shost->host_lock, flags);
334 blk_run_queue(sdev->request_queue);
335 spin_lock_irqsave(shost->host_lock, flags);
337 scsi_device_put(sdev);
340 spin_unlock_irqrestore(shost->host_lock, flags);
343 static inline int scsi_device_is_busy(struct scsi_device *sdev)
345 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
351 static inline int scsi_target_is_busy(struct scsi_target *starget)
353 return ((starget->can_queue > 0 &&
354 starget->target_busy >= starget->can_queue) ||
355 starget->target_blocked);
358 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
360 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
361 shost->host_blocked || shost->host_self_blocked)
367 static void scsi_starved_list_run(struct Scsi_Host *shost)
369 LIST_HEAD(starved_list);
370 struct scsi_device *sdev;
373 spin_lock_irqsave(shost->host_lock, flags);
374 list_splice_init(&shost->starved_list, &starved_list);
376 while (!list_empty(&starved_list)) {
377 struct request_queue *slq;
380 * As long as shost is accepting commands and we have
381 * starved queues, call blk_run_queue. scsi_request_fn
382 * drops the queue_lock and can add us back to the
385 * host_lock protects the starved_list and starved_entry.
386 * scsi_request_fn must get the host_lock before checking
387 * or modifying starved_list or starved_entry.
389 if (scsi_host_is_busy(shost))
392 sdev = list_entry(starved_list.next,
393 struct scsi_device, starved_entry);
394 list_del_init(&sdev->starved_entry);
395 if (scsi_target_is_busy(scsi_target(sdev))) {
396 list_move_tail(&sdev->starved_entry,
397 &shost->starved_list);
402 * Once we drop the host lock, a racing scsi_remove_device()
403 * call may remove the sdev from the starved list and destroy
404 * it and the queue. Mitigate by taking a reference to the
405 * queue and never touching the sdev again after we drop the
406 * host lock. Note: if __scsi_remove_device() invokes
407 * blk_cleanup_queue() before the queue is run from this
408 * function then blk_run_queue() will return immediately since
409 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
411 slq = sdev->request_queue;
412 if (!blk_get_queue(slq))
414 spin_unlock_irqrestore(shost->host_lock, flags);
419 spin_lock_irqsave(shost->host_lock, flags);
421 /* put any unprocessed entries back */
422 list_splice(&starved_list, &shost->starved_list);
423 spin_unlock_irqrestore(shost->host_lock, flags);
427 * Function: scsi_run_queue()
429 * Purpose: Select a proper request queue to serve next
431 * Arguments: q - last request's queue
435 * Notes: The previous command was completely finished, start
436 * a new one if possible.
438 static void scsi_run_queue(struct request_queue *q)
440 struct scsi_device *sdev = q->queuedata;
442 if (scsi_target(sdev)->single_lun)
443 scsi_single_lun_run(sdev);
444 if (!list_empty(&sdev->host->starved_list))
445 scsi_starved_list_run(sdev->host);
450 void scsi_requeue_run_queue(struct work_struct *work)
452 struct scsi_device *sdev;
453 struct request_queue *q;
455 sdev = container_of(work, struct scsi_device, requeue_work);
456 q = sdev->request_queue;
461 * Function: scsi_requeue_command()
463 * Purpose: Handle post-processing of completed commands.
465 * Arguments: q - queue to operate on
466 * cmd - command that may need to be requeued.
470 * Notes: After command completion, there may be blocks left
471 * over which weren't finished by the previous command
472 * this can be for a number of reasons - the main one is
473 * I/O errors in the middle of the request, in which case
474 * we need to request the blocks that come after the bad
476 * Notes: Upon return, cmd is a stale pointer.
478 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
480 struct scsi_device *sdev = cmd->device;
481 struct request *req = cmd->request;
484 spin_lock_irqsave(q->queue_lock, flags);
485 blk_unprep_request(req);
487 scsi_put_command(cmd);
488 blk_requeue_request(q, req);
489 spin_unlock_irqrestore(q->queue_lock, flags);
493 put_device(&sdev->sdev_gendev);
496 void scsi_next_command(struct scsi_cmnd *cmd)
498 struct scsi_device *sdev = cmd->device;
499 struct request_queue *q = sdev->request_queue;
501 scsi_put_command(cmd);
504 put_device(&sdev->sdev_gendev);
507 void scsi_run_host_queues(struct Scsi_Host *shost)
509 struct scsi_device *sdev;
511 shost_for_each_device(sdev, shost)
512 scsi_run_queue(sdev->request_queue);
515 static inline unsigned int scsi_sgtable_index(unsigned short nents)
519 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
524 index = get_count_order(nents) - 3;
529 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
531 struct scsi_host_sg_pool *sgp;
533 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
534 mempool_free(sgl, sgp->pool);
537 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
539 struct scsi_host_sg_pool *sgp;
541 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
542 return mempool_alloc(sgp->pool, gfp_mask);
545 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
552 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
553 gfp_mask, scsi_sg_alloc);
555 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
561 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
563 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
567 * Function: scsi_release_buffers()
569 * Purpose: Free resources allocate for a scsi_command.
571 * Arguments: cmd - command that we are bailing.
573 * Lock status: Assumed that no lock is held upon entry.
577 * Notes: In the event that an upper level driver rejects a
578 * command, we must release resources allocated during
579 * the __init_io() function. Primarily this would involve
580 * the scatter-gather table.
582 void scsi_release_buffers(struct scsi_cmnd *cmd)
584 if (cmd->sdb.table.nents)
585 scsi_free_sgtable(&cmd->sdb);
587 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
589 if (scsi_prot_sg_count(cmd))
590 scsi_free_sgtable(cmd->prot_sdb);
592 EXPORT_SYMBOL(scsi_release_buffers);
594 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
596 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
598 scsi_free_sgtable(bidi_sdb);
599 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
600 cmd->request->next_rq->special = NULL;
604 * __scsi_error_from_host_byte - translate SCSI error code into errno
605 * @cmd: SCSI command (unused)
606 * @result: scsi error code
608 * Translate SCSI error code into standard UNIX errno.
610 * -ENOLINK temporary transport failure
611 * -EREMOTEIO permanent target failure, do not retry
612 * -EBADE permanent nexus failure, retry on other path
613 * -ENOSPC No write space available
614 * -ENODATA Medium error
615 * -EIO unspecified I/O error
617 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
621 switch(host_byte(result)) {
622 case DID_TRANSPORT_FAILFAST:
625 case DID_TARGET_FAILURE:
626 set_host_byte(cmd, DID_OK);
629 case DID_NEXUS_FAILURE:
630 set_host_byte(cmd, DID_OK);
633 case DID_ALLOC_FAILURE:
634 set_host_byte(cmd, DID_OK);
637 case DID_MEDIUM_ERROR:
638 set_host_byte(cmd, DID_OK);
650 * Function: scsi_io_completion()
652 * Purpose: Completion processing for block device I/O requests.
654 * Arguments: cmd - command that is finished.
656 * Lock status: Assumed that no lock is held upon entry.
660 * Notes: We will finish off the specified number of sectors. If we
661 * are done, the command block will be released and the queue
662 * function will be goosed. If we are not done then we have to
663 * figure out what to do next:
665 * a) We can call scsi_requeue_command(). The request
666 * will be unprepared and put back on the queue. Then
667 * a new command will be created for it. This should
668 * be used if we made forward progress, or if we want
669 * to switch from READ(10) to READ(6) for example.
671 * b) We can call __scsi_queue_insert(). The request will
672 * be put back on the queue and retried using the same
673 * command as before, possibly after a delay.
675 * c) We can call blk_end_request() with -EIO to fail
676 * the remainder of the request.
678 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
680 int result = cmd->result;
681 struct request_queue *q = cmd->device->request_queue;
682 struct request *req = cmd->request;
684 struct scsi_sense_hdr sshdr;
686 int sense_deferred = 0;
687 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
688 ACTION_DELAYED_RETRY} action;
689 char *description = NULL;
690 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
693 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
695 sense_deferred = scsi_sense_is_deferred(&sshdr);
698 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
700 if (sense_valid && req->sense) {
702 * SG_IO wants current and deferred errors
704 int len = 8 + cmd->sense_buffer[7];
706 if (len > SCSI_SENSE_BUFFERSIZE)
707 len = SCSI_SENSE_BUFFERSIZE;
708 memcpy(req->sense, cmd->sense_buffer, len);
709 req->sense_len = len;
712 error = __scsi_error_from_host_byte(cmd, result);
715 * __scsi_error_from_host_byte may have reset the host_byte
717 req->errors = cmd->result;
719 req->resid_len = scsi_get_resid(cmd);
721 if (scsi_bidi_cmnd(cmd)) {
723 * Bidi commands Must be complete as a whole,
724 * both sides at once.
726 req->next_rq->resid_len = scsi_in(cmd)->resid;
728 scsi_release_buffers(cmd);
729 scsi_release_bidi_buffers(cmd);
731 blk_end_request_all(req, 0);
733 scsi_next_command(cmd);
736 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
738 * Certain non BLOCK_PC requests are commands that don't
739 * actually transfer anything (FLUSH), so cannot use
740 * good_bytes != blk_rq_bytes(req) as the signal for an error.
741 * This sets the error explicitly for the problem case.
743 error = __scsi_error_from_host_byte(cmd, result);
746 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
747 BUG_ON(blk_bidi_rq(req));
750 * Next deal with any sectors which we were able to correctly
753 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
755 blk_rq_sectors(req), good_bytes));
758 * Recovered errors need reporting, but they're always treated
759 * as success, so fiddle the result code here. For BLOCK_PC
760 * we already took a copy of the original into rq->errors which
761 * is what gets returned to the user
763 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
764 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
765 * print since caller wants ATA registers. Only occurs on
766 * SCSI ATA PASS_THROUGH commands when CK_COND=1
768 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
770 else if (!(req->cmd_flags & REQ_QUIET))
771 scsi_print_sense("", cmd);
773 /* BLOCK_PC may have set error */
778 * If we finished all bytes in the request we are done now.
780 if (!blk_end_request(req, error, good_bytes))
784 * Kill remainder if no retrys.
786 if (error && scsi_noretry_cmd(cmd)) {
787 blk_end_request_all(req, error);
792 * If there had been no error, but we have leftover bytes in the
793 * requeues just queue the command up again.
798 error = __scsi_error_from_host_byte(cmd, result);
800 if (host_byte(result) == DID_RESET) {
801 /* Third party bus reset or reset for error recovery
802 * reasons. Just retry the command and see what
805 action = ACTION_RETRY;
806 } else if (sense_valid && !sense_deferred) {
807 switch (sshdr.sense_key) {
809 if (cmd->device->removable) {
810 /* Detected disc change. Set a bit
811 * and quietly refuse further access.
813 cmd->device->changed = 1;
814 description = "Media Changed";
815 action = ACTION_FAIL;
817 /* Must have been a power glitch, or a
818 * bus reset. Could not have been a
819 * media change, so we just retry the
820 * command and see what happens.
822 action = ACTION_RETRY;
825 case ILLEGAL_REQUEST:
826 /* If we had an ILLEGAL REQUEST returned, then
827 * we may have performed an unsupported
828 * command. The only thing this should be
829 * would be a ten byte read where only a six
830 * byte read was supported. Also, on a system
831 * where READ CAPACITY failed, we may have
832 * read past the end of the disk.
834 if ((cmd->device->use_10_for_rw &&
835 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
836 (cmd->cmnd[0] == READ_10 ||
837 cmd->cmnd[0] == WRITE_10)) {
838 /* This will issue a new 6-byte command. */
839 cmd->device->use_10_for_rw = 0;
840 action = ACTION_REPREP;
841 } else if (sshdr.asc == 0x10) /* DIX */ {
842 description = "Host Data Integrity Failure";
843 action = ACTION_FAIL;
845 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
846 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
847 switch (cmd->cmnd[0]) {
849 description = "Discard failure";
853 if (cmd->cmnd[1] & 0x8)
854 description = "Discard failure";
857 "Write same failure";
860 description = "Invalid command failure";
863 action = ACTION_FAIL;
866 action = ACTION_FAIL;
868 case ABORTED_COMMAND:
869 action = ACTION_FAIL;
870 if (sshdr.asc == 0x10) { /* DIF */
871 description = "Target Data Integrity Failure";
876 /* If the device is in the process of becoming
877 * ready, or has a temporary blockage, retry.
879 if (sshdr.asc == 0x04) {
880 switch (sshdr.ascq) {
881 case 0x01: /* becoming ready */
882 case 0x04: /* format in progress */
883 case 0x05: /* rebuild in progress */
884 case 0x06: /* recalculation in progress */
885 case 0x07: /* operation in progress */
886 case 0x08: /* Long write in progress */
887 case 0x09: /* self test in progress */
888 case 0x14: /* space allocation in progress */
889 action = ACTION_DELAYED_RETRY;
892 description = "Device not ready";
893 action = ACTION_FAIL;
897 description = "Device not ready";
898 action = ACTION_FAIL;
901 case VOLUME_OVERFLOW:
902 /* See SSC3rXX or current. */
903 action = ACTION_FAIL;
906 description = "Unhandled sense code";
907 action = ACTION_FAIL;
911 description = "Unhandled error code";
912 action = ACTION_FAIL;
915 if (action != ACTION_FAIL &&
916 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
917 action = ACTION_FAIL;
918 description = "Command timed out";
923 /* Give up and fail the remainder of the request */
924 if (!(req->cmd_flags & REQ_QUIET)) {
926 scmd_printk(KERN_INFO, cmd, "%s\n",
928 scsi_print_result(cmd);
929 if (driver_byte(result) & DRIVER_SENSE)
930 scsi_print_sense("", cmd);
931 scsi_print_command(cmd);
933 if (!blk_end_request_err(req, error))
938 /* Unprep the request and put it back at the head of the queue.
939 * A new command will be prepared and issued.
941 scsi_release_buffers(cmd);
942 scsi_requeue_command(q, cmd);
945 /* Retry the same command immediately */
946 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
948 case ACTION_DELAYED_RETRY:
949 /* Retry the same command after a delay */
950 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
956 scsi_release_buffers(cmd);
957 scsi_next_command(cmd);
960 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
966 * If sg table allocation fails, requeue request later.
968 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
970 return BLKPREP_DEFER;
974 * Next, walk the list, and fill in the addresses and sizes of
977 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
978 BUG_ON(count > sdb->table.nents);
979 sdb->table.nents = count;
980 sdb->length = blk_rq_bytes(req);
985 * Function: scsi_init_io()
987 * Purpose: SCSI I/O initialize function.
989 * Arguments: cmd - Command descriptor we wish to initialize
991 * Returns: 0 on success
992 * BLKPREP_DEFER if the failure is retryable
993 * BLKPREP_KILL if the failure is fatal
995 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
997 struct scsi_device *sdev = cmd->device;
998 struct request *rq = cmd->request;
1000 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1004 if (blk_bidi_rq(rq)) {
1005 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1006 scsi_sdb_cache, GFP_ATOMIC);
1008 error = BLKPREP_DEFER;
1012 rq->next_rq->special = bidi_sdb;
1013 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1018 if (blk_integrity_rq(rq)) {
1019 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1022 BUG_ON(prot_sdb == NULL);
1023 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1025 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1026 error = BLKPREP_DEFER;
1030 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1031 prot_sdb->table.sgl);
1032 BUG_ON(unlikely(count > ivecs));
1033 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1035 cmd->prot_sdb = prot_sdb;
1036 cmd->prot_sdb->table.nents = count;
1042 scsi_release_buffers(cmd);
1043 cmd->request->special = NULL;
1044 scsi_put_command(cmd);
1045 put_device(&sdev->sdev_gendev);
1048 EXPORT_SYMBOL(scsi_init_io);
1050 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1051 struct request *req)
1053 struct scsi_cmnd *cmd;
1055 if (!req->special) {
1056 /* Bail if we can't get a reference to the device */
1057 if (!get_device(&sdev->sdev_gendev))
1060 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1061 if (unlikely(!cmd)) {
1062 put_device(&sdev->sdev_gendev);
1070 /* pull a tag out of the request if we have one */
1071 cmd->tag = req->tag;
1074 cmd->cmnd = req->cmd;
1075 cmd->prot_op = SCSI_PROT_NORMAL;
1080 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1082 struct scsi_cmnd *cmd = req->special;
1085 * BLOCK_PC requests may transfer data, in which case they must
1086 * a bio attached to them. Or they might contain a SCSI command
1087 * that does not transfer data, in which case they may optionally
1088 * submit a request without an attached bio.
1093 BUG_ON(!req->nr_phys_segments);
1095 ret = scsi_init_io(cmd, GFP_ATOMIC);
1099 BUG_ON(blk_rq_bytes(req));
1101 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1104 cmd->cmd_len = req->cmd_len;
1105 if (!blk_rq_bytes(req))
1106 cmd->sc_data_direction = DMA_NONE;
1107 else if (rq_data_dir(req) == WRITE)
1108 cmd->sc_data_direction = DMA_TO_DEVICE;
1110 cmd->sc_data_direction = DMA_FROM_DEVICE;
1112 cmd->transfersize = blk_rq_bytes(req);
1113 cmd->allowed = req->retries;
1116 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1119 * Setup a REQ_TYPE_FS command. These are simple read/write request
1120 * from filesystems that still need to be translated to SCSI CDBs from
1123 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1125 struct scsi_cmnd *cmd = req->special;
1127 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1128 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1129 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1130 if (ret != BLKPREP_OK)
1135 * Filesystem requests must transfer data.
1137 BUG_ON(!req->nr_phys_segments);
1139 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1140 return scsi_init_io(cmd, GFP_ATOMIC);
1142 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1145 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1147 int ret = BLKPREP_OK;
1150 * If the device is not in running state we will reject some
1153 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1154 switch (sdev->sdev_state) {
1156 case SDEV_TRANSPORT_OFFLINE:
1158 * If the device is offline we refuse to process any
1159 * commands. The device must be brought online
1160 * before trying any recovery commands.
1162 sdev_printk(KERN_ERR, sdev,
1163 "rejecting I/O to offline device\n");
1168 * If the device is fully deleted, we refuse to
1169 * process any commands as well.
1171 sdev_printk(KERN_ERR, sdev,
1172 "rejecting I/O to dead device\n");
1177 case SDEV_CREATED_BLOCK:
1179 * If the devices is blocked we defer normal commands.
1181 if (!(req->cmd_flags & REQ_PREEMPT))
1182 ret = BLKPREP_DEFER;
1186 * For any other not fully online state we only allow
1187 * special commands. In particular any user initiated
1188 * command is not allowed.
1190 if (!(req->cmd_flags & REQ_PREEMPT))
1199 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1201 struct scsi_device *sdev = q->queuedata;
1205 req->errors = DID_NO_CONNECT << 16;
1206 /* release the command and kill it */
1208 struct scsi_cmnd *cmd = req->special;
1209 scsi_release_buffers(cmd);
1210 scsi_put_command(cmd);
1211 put_device(&sdev->sdev_gendev);
1212 req->special = NULL;
1217 * If we defer, the blk_peek_request() returns NULL, but the
1218 * queue must be restarted, so we schedule a callback to happen
1221 if (sdev->device_busy == 0)
1222 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1225 req->cmd_flags |= REQ_DONTPREP;
1231 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1233 struct scsi_device *sdev = q->queuedata;
1234 struct scsi_cmnd *cmd;
1237 ret = scsi_prep_state_check(sdev, req);
1238 if (ret != BLKPREP_OK)
1241 cmd = scsi_get_cmd_from_req(sdev, req);
1242 if (unlikely(!cmd)) {
1243 ret = BLKPREP_DEFER;
1247 if (req->cmd_type == REQ_TYPE_FS)
1248 ret = scsi_cmd_to_driver(cmd)->init_command(cmd);
1249 else if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1250 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1255 return scsi_prep_return(q, req, ret);
1258 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1260 if (req->cmd_type == REQ_TYPE_FS) {
1261 struct scsi_cmnd *cmd = req->special;
1262 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
1264 if (drv->uninit_command)
1265 drv->uninit_command(cmd);
1270 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1273 * Called with the queue_lock held.
1275 static inline int scsi_dev_queue_ready(struct request_queue *q,
1276 struct scsi_device *sdev)
1278 if (sdev->device_busy == 0 && sdev->device_blocked) {
1280 * unblock after device_blocked iterates to zero
1282 if (--sdev->device_blocked == 0) {
1284 sdev_printk(KERN_INFO, sdev,
1285 "unblocking device at zero depth\n"));
1287 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1291 if (scsi_device_is_busy(sdev))
1299 * scsi_target_queue_ready: checks if there we can send commands to target
1300 * @sdev: scsi device on starget to check.
1302 * Called with the host lock held.
1304 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1305 struct scsi_device *sdev)
1307 struct scsi_target *starget = scsi_target(sdev);
1309 if (starget->single_lun) {
1310 if (starget->starget_sdev_user &&
1311 starget->starget_sdev_user != sdev)
1313 starget->starget_sdev_user = sdev;
1316 if (starget->target_busy == 0 && starget->target_blocked) {
1318 * unblock after target_blocked iterates to zero
1320 if (--starget->target_blocked == 0) {
1321 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1322 "unblocking target at zero depth\n"));
1327 if (scsi_target_is_busy(starget)) {
1328 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1336 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1337 * return 0. We must end up running the queue again whenever 0 is
1338 * returned, else IO can hang.
1340 * Called with host_lock held.
1342 static inline int scsi_host_queue_ready(struct request_queue *q,
1343 struct Scsi_Host *shost,
1344 struct scsi_device *sdev)
1346 if (scsi_host_in_recovery(shost))
1348 if (shost->host_busy == 0 && shost->host_blocked) {
1350 * unblock after host_blocked iterates to zero
1352 if (--shost->host_blocked == 0) {
1354 printk("scsi%d unblocking host at zero depth\n",
1360 if (scsi_host_is_busy(shost)) {
1361 if (list_empty(&sdev->starved_entry))
1362 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1366 /* We're OK to process the command, so we can't be starved */
1367 if (!list_empty(&sdev->starved_entry))
1368 list_del_init(&sdev->starved_entry);
1374 * Busy state exporting function for request stacking drivers.
1376 * For efficiency, no lock is taken to check the busy state of
1377 * shost/starget/sdev, since the returned value is not guaranteed and
1378 * may be changed after request stacking drivers call the function,
1379 * regardless of taking lock or not.
1381 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1382 * needs to return 'not busy'. Otherwise, request stacking drivers
1383 * may hold requests forever.
1385 static int scsi_lld_busy(struct request_queue *q)
1387 struct scsi_device *sdev = q->queuedata;
1388 struct Scsi_Host *shost;
1390 if (blk_queue_dying(q))
1396 * Ignore host/starget busy state.
1397 * Since block layer does not have a concept of fairness across
1398 * multiple queues, congestion of host/starget needs to be handled
1401 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1408 * Kill a request for a dead device
1410 static void scsi_kill_request(struct request *req, struct request_queue *q)
1412 struct scsi_cmnd *cmd = req->special;
1413 struct scsi_device *sdev;
1414 struct scsi_target *starget;
1415 struct Scsi_Host *shost;
1417 blk_start_request(req);
1419 scmd_printk(KERN_INFO, cmd, "killing request\n");
1422 starget = scsi_target(sdev);
1424 scsi_init_cmd_errh(cmd);
1425 cmd->result = DID_NO_CONNECT << 16;
1426 atomic_inc(&cmd->device->iorequest_cnt);
1429 * SCSI request completion path will do scsi_device_unbusy(),
1430 * bump busy counts. To bump the counters, we need to dance
1431 * with the locks as normal issue path does.
1433 sdev->device_busy++;
1434 spin_unlock(sdev->request_queue->queue_lock);
1435 spin_lock(shost->host_lock);
1437 starget->target_busy++;
1438 spin_unlock(shost->host_lock);
1439 spin_lock(sdev->request_queue->queue_lock);
1441 blk_complete_request(req);
1444 static void scsi_softirq_done(struct request *rq)
1446 struct scsi_cmnd *cmd = rq->special;
1447 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1450 INIT_LIST_HEAD(&cmd->eh_entry);
1452 atomic_inc(&cmd->device->iodone_cnt);
1454 atomic_inc(&cmd->device->ioerr_cnt);
1456 disposition = scsi_decide_disposition(cmd);
1457 if (disposition != SUCCESS &&
1458 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1459 sdev_printk(KERN_ERR, cmd->device,
1460 "timing out command, waited %lus\n",
1462 disposition = SUCCESS;
1465 scsi_log_completion(cmd, disposition);
1467 switch (disposition) {
1469 scsi_finish_command(cmd);
1472 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1474 case ADD_TO_MLQUEUE:
1475 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1478 if (!scsi_eh_scmd_add(cmd, 0))
1479 scsi_finish_command(cmd);
1484 * Function: scsi_request_fn()
1486 * Purpose: Main strategy routine for SCSI.
1488 * Arguments: q - Pointer to actual queue.
1492 * Lock status: IO request lock assumed to be held when called.
1494 static void scsi_request_fn(struct request_queue *q)
1495 __releases(q->queue_lock)
1496 __acquires(q->queue_lock)
1498 struct scsi_device *sdev = q->queuedata;
1499 struct Scsi_Host *shost;
1500 struct scsi_cmnd *cmd;
1501 struct request *req;
1504 * To start with, we keep looping until the queue is empty, or until
1505 * the host is no longer able to accept any more requests.
1511 * get next queueable request. We do this early to make sure
1512 * that the request is fully prepared even if we cannot
1515 req = blk_peek_request(q);
1516 if (!req || !scsi_dev_queue_ready(q, sdev))
1519 if (unlikely(!scsi_device_online(sdev))) {
1520 sdev_printk(KERN_ERR, sdev,
1521 "rejecting I/O to offline device\n");
1522 scsi_kill_request(req, q);
1528 * Remove the request from the request list.
1530 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1531 blk_start_request(req);
1532 sdev->device_busy++;
1534 spin_unlock(q->queue_lock);
1536 if (unlikely(cmd == NULL)) {
1537 printk(KERN_CRIT "impossible request in %s.\n"
1538 "please mail a stack trace to "
1539 "linux-scsi@vger.kernel.org\n",
1541 blk_dump_rq_flags(req, "foo");
1544 spin_lock(shost->host_lock);
1547 * We hit this when the driver is using a host wide
1548 * tag map. For device level tag maps the queue_depth check
1549 * in the device ready fn would prevent us from trying
1550 * to allocate a tag. Since the map is a shared host resource
1551 * we add the dev to the starved list so it eventually gets
1552 * a run when a tag is freed.
1554 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1555 if (list_empty(&sdev->starved_entry))
1556 list_add_tail(&sdev->starved_entry,
1557 &shost->starved_list);
1561 if (!scsi_target_queue_ready(shost, sdev))
1564 if (!scsi_host_queue_ready(q, shost, sdev))
1567 scsi_target(sdev)->target_busy++;
1571 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1572 * take the lock again.
1574 spin_unlock_irq(shost->host_lock);
1577 * Finally, initialize any error handling parameters, and set up
1578 * the timers for timeouts.
1580 scsi_init_cmd_errh(cmd);
1583 * Dispatch the command to the low-level driver.
1585 rtn = scsi_dispatch_cmd(cmd);
1586 spin_lock_irq(q->queue_lock);
1594 spin_unlock_irq(shost->host_lock);
1597 * lock q, handle tag, requeue req, and decrement device_busy. We
1598 * must return with queue_lock held.
1600 * Decrementing device_busy without checking it is OK, as all such
1601 * cases (host limits or settings) should run the queue at some
1604 spin_lock_irq(q->queue_lock);
1605 blk_requeue_request(q, req);
1606 sdev->device_busy--;
1608 if (sdev->device_busy == 0)
1609 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1612 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1614 struct device *host_dev;
1615 u64 bounce_limit = 0xffffffff;
1617 if (shost->unchecked_isa_dma)
1618 return BLK_BOUNCE_ISA;
1620 * Platforms with virtual-DMA translation
1621 * hardware have no practical limit.
1623 if (!PCI_DMA_BUS_IS_PHYS)
1624 return BLK_BOUNCE_ANY;
1626 host_dev = scsi_get_device(shost);
1627 if (host_dev && host_dev->dma_mask)
1628 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1630 return bounce_limit;
1632 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1634 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1635 request_fn_proc *request_fn)
1637 struct request_queue *q;
1638 struct device *dev = shost->dma_dev;
1640 q = blk_init_queue(request_fn, NULL);
1645 * this limit is imposed by hardware restrictions
1647 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1648 SCSI_MAX_SG_CHAIN_SEGMENTS));
1650 if (scsi_host_prot_dma(shost)) {
1651 shost->sg_prot_tablesize =
1652 min_not_zero(shost->sg_prot_tablesize,
1653 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1654 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1655 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1658 blk_queue_max_hw_sectors(q, shost->max_sectors);
1659 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1660 blk_queue_segment_boundary(q, shost->dma_boundary);
1661 dma_set_seg_boundary(dev, shost->dma_boundary);
1663 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1665 if (!shost->use_clustering)
1666 q->limits.cluster = 0;
1669 * set a reasonable default alignment on word boundaries: the
1670 * host and device may alter it using
1671 * blk_queue_update_dma_alignment() later.
1673 blk_queue_dma_alignment(q, 0x03);
1677 EXPORT_SYMBOL(__scsi_alloc_queue);
1679 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1681 struct request_queue *q;
1683 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1687 blk_queue_prep_rq(q, scsi_prep_fn);
1688 blk_queue_unprep_rq(q, scsi_unprep_fn);
1689 blk_queue_softirq_done(q, scsi_softirq_done);
1690 blk_queue_rq_timed_out(q, scsi_times_out);
1691 blk_queue_lld_busy(q, scsi_lld_busy);
1696 * Function: scsi_block_requests()
1698 * Purpose: Utility function used by low-level drivers to prevent further
1699 * commands from being queued to the device.
1701 * Arguments: shost - Host in question
1705 * Lock status: No locks are assumed held.
1707 * Notes: There is no timer nor any other means by which the requests
1708 * get unblocked other than the low-level driver calling
1709 * scsi_unblock_requests().
1711 void scsi_block_requests(struct Scsi_Host *shost)
1713 shost->host_self_blocked = 1;
1715 EXPORT_SYMBOL(scsi_block_requests);
1718 * Function: scsi_unblock_requests()
1720 * Purpose: Utility function used by low-level drivers to allow further
1721 * commands from being queued to the device.
1723 * Arguments: shost - Host in question
1727 * Lock status: No locks are assumed held.
1729 * Notes: There is no timer nor any other means by which the requests
1730 * get unblocked other than the low-level driver calling
1731 * scsi_unblock_requests().
1733 * This is done as an API function so that changes to the
1734 * internals of the scsi mid-layer won't require wholesale
1735 * changes to drivers that use this feature.
1737 void scsi_unblock_requests(struct Scsi_Host *shost)
1739 shost->host_self_blocked = 0;
1740 scsi_run_host_queues(shost);
1742 EXPORT_SYMBOL(scsi_unblock_requests);
1744 int __init scsi_init_queue(void)
1748 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1749 sizeof(struct scsi_data_buffer),
1751 if (!scsi_sdb_cache) {
1752 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1756 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1757 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1758 int size = sgp->size * sizeof(struct scatterlist);
1760 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1761 SLAB_HWCACHE_ALIGN, NULL);
1763 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1768 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1771 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1780 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1781 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1783 mempool_destroy(sgp->pool);
1785 kmem_cache_destroy(sgp->slab);
1787 kmem_cache_destroy(scsi_sdb_cache);
1792 void scsi_exit_queue(void)
1796 kmem_cache_destroy(scsi_sdb_cache);
1798 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1799 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1800 mempool_destroy(sgp->pool);
1801 kmem_cache_destroy(sgp->slab);
1806 * scsi_mode_select - issue a mode select
1807 * @sdev: SCSI device to be queried
1808 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1809 * @sp: Save page bit (0 == don't save, 1 == save)
1810 * @modepage: mode page being requested
1811 * @buffer: request buffer (may not be smaller than eight bytes)
1812 * @len: length of request buffer.
1813 * @timeout: command timeout
1814 * @retries: number of retries before failing
1815 * @data: returns a structure abstracting the mode header data
1816 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1817 * must be SCSI_SENSE_BUFFERSIZE big.
1819 * Returns zero if successful; negative error number or scsi
1824 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1825 unsigned char *buffer, int len, int timeout, int retries,
1826 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1828 unsigned char cmd[10];
1829 unsigned char *real_buffer;
1832 memset(cmd, 0, sizeof(cmd));
1833 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1835 if (sdev->use_10_for_ms) {
1838 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1841 memcpy(real_buffer + 8, buffer, len);
1845 real_buffer[2] = data->medium_type;
1846 real_buffer[3] = data->device_specific;
1847 real_buffer[4] = data->longlba ? 0x01 : 0;
1849 real_buffer[6] = data->block_descriptor_length >> 8;
1850 real_buffer[7] = data->block_descriptor_length;
1852 cmd[0] = MODE_SELECT_10;
1856 if (len > 255 || data->block_descriptor_length > 255 ||
1860 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1863 memcpy(real_buffer + 4, buffer, len);
1866 real_buffer[1] = data->medium_type;
1867 real_buffer[2] = data->device_specific;
1868 real_buffer[3] = data->block_descriptor_length;
1871 cmd[0] = MODE_SELECT;
1875 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1876 sshdr, timeout, retries, NULL);
1880 EXPORT_SYMBOL_GPL(scsi_mode_select);
1883 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1884 * @sdev: SCSI device to be queried
1885 * @dbd: set if mode sense will allow block descriptors to be returned
1886 * @modepage: mode page being requested
1887 * @buffer: request buffer (may not be smaller than eight bytes)
1888 * @len: length of request buffer.
1889 * @timeout: command timeout
1890 * @retries: number of retries before failing
1891 * @data: returns a structure abstracting the mode header data
1892 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1893 * must be SCSI_SENSE_BUFFERSIZE big.
1895 * Returns zero if unsuccessful, or the header offset (either 4
1896 * or 8 depending on whether a six or ten byte command was
1897 * issued) if successful.
1900 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1901 unsigned char *buffer, int len, int timeout, int retries,
1902 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1904 unsigned char cmd[12];
1908 struct scsi_sense_hdr my_sshdr;
1910 memset(data, 0, sizeof(*data));
1911 memset(&cmd[0], 0, 12);
1912 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1915 /* caller might not be interested in sense, but we need it */
1920 use_10_for_ms = sdev->use_10_for_ms;
1922 if (use_10_for_ms) {
1926 cmd[0] = MODE_SENSE_10;
1933 cmd[0] = MODE_SENSE;
1938 memset(buffer, 0, len);
1940 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1941 sshdr, timeout, retries, NULL);
1943 /* This code looks awful: what it's doing is making sure an
1944 * ILLEGAL REQUEST sense return identifies the actual command
1945 * byte as the problem. MODE_SENSE commands can return
1946 * ILLEGAL REQUEST if the code page isn't supported */
1948 if (use_10_for_ms && !scsi_status_is_good(result) &&
1949 (driver_byte(result) & DRIVER_SENSE)) {
1950 if (scsi_sense_valid(sshdr)) {
1951 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1952 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1954 * Invalid command operation code
1956 sdev->use_10_for_ms = 0;
1962 if(scsi_status_is_good(result)) {
1963 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1964 (modepage == 6 || modepage == 8))) {
1965 /* Initio breakage? */
1968 data->medium_type = 0;
1969 data->device_specific = 0;
1971 data->block_descriptor_length = 0;
1972 } else if(use_10_for_ms) {
1973 data->length = buffer[0]*256 + buffer[1] + 2;
1974 data->medium_type = buffer[2];
1975 data->device_specific = buffer[3];
1976 data->longlba = buffer[4] & 0x01;
1977 data->block_descriptor_length = buffer[6]*256
1980 data->length = buffer[0] + 1;
1981 data->medium_type = buffer[1];
1982 data->device_specific = buffer[2];
1983 data->block_descriptor_length = buffer[3];
1985 data->header_length = header_length;
1990 EXPORT_SYMBOL(scsi_mode_sense);
1993 * scsi_test_unit_ready - test if unit is ready
1994 * @sdev: scsi device to change the state of.
1995 * @timeout: command timeout
1996 * @retries: number of retries before failing
1997 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1998 * returning sense. Make sure that this is cleared before passing
2001 * Returns zero if unsuccessful or an error if TUR failed. For
2002 * removable media, UNIT_ATTENTION sets ->changed flag.
2005 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2006 struct scsi_sense_hdr *sshdr_external)
2009 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2011 struct scsi_sense_hdr *sshdr;
2014 if (!sshdr_external)
2015 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2017 sshdr = sshdr_external;
2019 /* try to eat the UNIT_ATTENTION if there are enough retries */
2021 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2022 timeout, retries, NULL);
2023 if (sdev->removable && scsi_sense_valid(sshdr) &&
2024 sshdr->sense_key == UNIT_ATTENTION)
2026 } while (scsi_sense_valid(sshdr) &&
2027 sshdr->sense_key == UNIT_ATTENTION && --retries);
2029 if (!sshdr_external)
2033 EXPORT_SYMBOL(scsi_test_unit_ready);
2036 * scsi_device_set_state - Take the given device through the device state model.
2037 * @sdev: scsi device to change the state of.
2038 * @state: state to change to.
2040 * Returns zero if unsuccessful or an error if the requested
2041 * transition is illegal.
2044 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2046 enum scsi_device_state oldstate = sdev->sdev_state;
2048 if (state == oldstate)
2054 case SDEV_CREATED_BLOCK:
2065 case SDEV_TRANSPORT_OFFLINE:
2078 case SDEV_TRANSPORT_OFFLINE:
2086 case SDEV_TRANSPORT_OFFLINE:
2101 case SDEV_CREATED_BLOCK:
2108 case SDEV_CREATED_BLOCK:
2123 case SDEV_TRANSPORT_OFFLINE:
2136 case SDEV_TRANSPORT_OFFLINE:
2138 case SDEV_CREATED_BLOCK:
2146 sdev->sdev_state = state;
2150 SCSI_LOG_ERROR_RECOVERY(1,
2151 sdev_printk(KERN_ERR, sdev,
2152 "Illegal state transition %s->%s\n",
2153 scsi_device_state_name(oldstate),
2154 scsi_device_state_name(state))
2158 EXPORT_SYMBOL(scsi_device_set_state);
2161 * sdev_evt_emit - emit a single SCSI device uevent
2162 * @sdev: associated SCSI device
2163 * @evt: event to emit
2165 * Send a single uevent (scsi_event) to the associated scsi_device.
2167 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2172 switch (evt->evt_type) {
2173 case SDEV_EVT_MEDIA_CHANGE:
2174 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2176 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2177 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2179 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2180 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2182 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2183 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2185 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2186 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2188 case SDEV_EVT_LUN_CHANGE_REPORTED:
2189 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2198 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2202 * sdev_evt_thread - send a uevent for each scsi event
2203 * @work: work struct for scsi_device
2205 * Dispatch queued events to their associated scsi_device kobjects
2208 void scsi_evt_thread(struct work_struct *work)
2210 struct scsi_device *sdev;
2211 enum scsi_device_event evt_type;
2212 LIST_HEAD(event_list);
2214 sdev = container_of(work, struct scsi_device, event_work);
2216 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2217 if (test_and_clear_bit(evt_type, sdev->pending_events))
2218 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2221 struct scsi_event *evt;
2222 struct list_head *this, *tmp;
2223 unsigned long flags;
2225 spin_lock_irqsave(&sdev->list_lock, flags);
2226 list_splice_init(&sdev->event_list, &event_list);
2227 spin_unlock_irqrestore(&sdev->list_lock, flags);
2229 if (list_empty(&event_list))
2232 list_for_each_safe(this, tmp, &event_list) {
2233 evt = list_entry(this, struct scsi_event, node);
2234 list_del(&evt->node);
2235 scsi_evt_emit(sdev, evt);
2242 * sdev_evt_send - send asserted event to uevent thread
2243 * @sdev: scsi_device event occurred on
2244 * @evt: event to send
2246 * Assert scsi device event asynchronously.
2248 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2250 unsigned long flags;
2253 /* FIXME: currently this check eliminates all media change events
2254 * for polled devices. Need to update to discriminate between AN
2255 * and polled events */
2256 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2262 spin_lock_irqsave(&sdev->list_lock, flags);
2263 list_add_tail(&evt->node, &sdev->event_list);
2264 schedule_work(&sdev->event_work);
2265 spin_unlock_irqrestore(&sdev->list_lock, flags);
2267 EXPORT_SYMBOL_GPL(sdev_evt_send);
2270 * sdev_evt_alloc - allocate a new scsi event
2271 * @evt_type: type of event to allocate
2272 * @gfpflags: GFP flags for allocation
2274 * Allocates and returns a new scsi_event.
2276 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2279 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2283 evt->evt_type = evt_type;
2284 INIT_LIST_HEAD(&evt->node);
2286 /* evt_type-specific initialization, if any */
2288 case SDEV_EVT_MEDIA_CHANGE:
2289 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2290 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2291 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2292 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2293 case SDEV_EVT_LUN_CHANGE_REPORTED:
2301 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2304 * sdev_evt_send_simple - send asserted event to uevent thread
2305 * @sdev: scsi_device event occurred on
2306 * @evt_type: type of event to send
2307 * @gfpflags: GFP flags for allocation
2309 * Assert scsi device event asynchronously, given an event type.
2311 void sdev_evt_send_simple(struct scsi_device *sdev,
2312 enum scsi_device_event evt_type, gfp_t gfpflags)
2314 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2316 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2321 sdev_evt_send(sdev, evt);
2323 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2326 * scsi_device_quiesce - Block user issued commands.
2327 * @sdev: scsi device to quiesce.
2329 * This works by trying to transition to the SDEV_QUIESCE state
2330 * (which must be a legal transition). When the device is in this
2331 * state, only special requests will be accepted, all others will
2332 * be deferred. Since special requests may also be requeued requests,
2333 * a successful return doesn't guarantee the device will be
2334 * totally quiescent.
2336 * Must be called with user context, may sleep.
2338 * Returns zero if unsuccessful or an error if not.
2341 scsi_device_quiesce(struct scsi_device *sdev)
2343 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2347 scsi_run_queue(sdev->request_queue);
2348 while (sdev->device_busy) {
2349 msleep_interruptible(200);
2350 scsi_run_queue(sdev->request_queue);
2354 EXPORT_SYMBOL(scsi_device_quiesce);
2357 * scsi_device_resume - Restart user issued commands to a quiesced device.
2358 * @sdev: scsi device to resume.
2360 * Moves the device from quiesced back to running and restarts the
2363 * Must be called with user context, may sleep.
2365 void scsi_device_resume(struct scsi_device *sdev)
2367 /* check if the device state was mutated prior to resume, and if
2368 * so assume the state is being managed elsewhere (for example
2369 * device deleted during suspend)
2371 if (sdev->sdev_state != SDEV_QUIESCE ||
2372 scsi_device_set_state(sdev, SDEV_RUNNING))
2374 scsi_run_queue(sdev->request_queue);
2376 EXPORT_SYMBOL(scsi_device_resume);
2379 device_quiesce_fn(struct scsi_device *sdev, void *data)
2381 scsi_device_quiesce(sdev);
2385 scsi_target_quiesce(struct scsi_target *starget)
2387 starget_for_each_device(starget, NULL, device_quiesce_fn);
2389 EXPORT_SYMBOL(scsi_target_quiesce);
2392 device_resume_fn(struct scsi_device *sdev, void *data)
2394 scsi_device_resume(sdev);
2398 scsi_target_resume(struct scsi_target *starget)
2400 starget_for_each_device(starget, NULL, device_resume_fn);
2402 EXPORT_SYMBOL(scsi_target_resume);
2405 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2406 * @sdev: device to block
2408 * Block request made by scsi lld's to temporarily stop all
2409 * scsi commands on the specified device. Called from interrupt
2410 * or normal process context.
2412 * Returns zero if successful or error if not
2415 * This routine transitions the device to the SDEV_BLOCK state
2416 * (which must be a legal transition). When the device is in this
2417 * state, all commands are deferred until the scsi lld reenables
2418 * the device with scsi_device_unblock or device_block_tmo fires.
2421 scsi_internal_device_block(struct scsi_device *sdev)
2423 struct request_queue *q = sdev->request_queue;
2424 unsigned long flags;
2427 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2429 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2436 * The device has transitioned to SDEV_BLOCK. Stop the
2437 * block layer from calling the midlayer with this device's
2440 spin_lock_irqsave(q->queue_lock, flags);
2442 spin_unlock_irqrestore(q->queue_lock, flags);
2446 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2449 * scsi_internal_device_unblock - resume a device after a block request
2450 * @sdev: device to resume
2451 * @new_state: state to set devices to after unblocking
2453 * Called by scsi lld's or the midlayer to restart the device queue
2454 * for the previously suspended scsi device. Called from interrupt or
2455 * normal process context.
2457 * Returns zero if successful or error if not.
2460 * This routine transitions the device to the SDEV_RUNNING state
2461 * or to one of the offline states (which must be a legal transition)
2462 * allowing the midlayer to goose the queue for this device.
2465 scsi_internal_device_unblock(struct scsi_device *sdev,
2466 enum scsi_device_state new_state)
2468 struct request_queue *q = sdev->request_queue;
2469 unsigned long flags;
2472 * Try to transition the scsi device to SDEV_RUNNING or one of the
2473 * offlined states and goose the device queue if successful.
2475 if ((sdev->sdev_state == SDEV_BLOCK) ||
2476 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2477 sdev->sdev_state = new_state;
2478 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2479 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2480 new_state == SDEV_OFFLINE)
2481 sdev->sdev_state = new_state;
2483 sdev->sdev_state = SDEV_CREATED;
2484 } else if (sdev->sdev_state != SDEV_CANCEL &&
2485 sdev->sdev_state != SDEV_OFFLINE)
2488 spin_lock_irqsave(q->queue_lock, flags);
2490 spin_unlock_irqrestore(q->queue_lock, flags);
2494 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2497 device_block(struct scsi_device *sdev, void *data)
2499 scsi_internal_device_block(sdev);
2503 target_block(struct device *dev, void *data)
2505 if (scsi_is_target_device(dev))
2506 starget_for_each_device(to_scsi_target(dev), NULL,
2512 scsi_target_block(struct device *dev)
2514 if (scsi_is_target_device(dev))
2515 starget_for_each_device(to_scsi_target(dev), NULL,
2518 device_for_each_child(dev, NULL, target_block);
2520 EXPORT_SYMBOL_GPL(scsi_target_block);
2523 device_unblock(struct scsi_device *sdev, void *data)
2525 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2529 target_unblock(struct device *dev, void *data)
2531 if (scsi_is_target_device(dev))
2532 starget_for_each_device(to_scsi_target(dev), data,
2538 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2540 if (scsi_is_target_device(dev))
2541 starget_for_each_device(to_scsi_target(dev), &new_state,
2544 device_for_each_child(dev, &new_state, target_unblock);
2546 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2549 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2550 * @sgl: scatter-gather list
2551 * @sg_count: number of segments in sg
2552 * @offset: offset in bytes into sg, on return offset into the mapped area
2553 * @len: bytes to map, on return number of bytes mapped
2555 * Returns virtual address of the start of the mapped page
2557 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2558 size_t *offset, size_t *len)
2561 size_t sg_len = 0, len_complete = 0;
2562 struct scatterlist *sg;
2565 WARN_ON(!irqs_disabled());
2567 for_each_sg(sgl, sg, sg_count, i) {
2568 len_complete = sg_len; /* Complete sg-entries */
2569 sg_len += sg->length;
2570 if (sg_len > *offset)
2574 if (unlikely(i == sg_count)) {
2575 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2577 __func__, sg_len, *offset, sg_count);
2582 /* Offset starting from the beginning of first page in this sg-entry */
2583 *offset = *offset - len_complete + sg->offset;
2585 /* Assumption: contiguous pages can be accessed as "page + i" */
2586 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2587 *offset &= ~PAGE_MASK;
2589 /* Bytes in this sg-entry from *offset to the end of the page */
2590 sg_len = PAGE_SIZE - *offset;
2594 return kmap_atomic(page);
2596 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2599 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2600 * @virt: virtual address to be unmapped
2602 void scsi_kunmap_atomic_sg(void *virt)
2604 kunmap_atomic(virt);
2606 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2608 void sdev_disable_disk_events(struct scsi_device *sdev)
2610 atomic_inc(&sdev->disk_events_disable_depth);
2612 EXPORT_SYMBOL(sdev_disable_disk_events);
2614 void sdev_enable_disk_events(struct scsi_device *sdev)
2616 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2618 atomic_dec(&sdev->disk_events_disable_depth);
2620 EXPORT_SYMBOL(sdev_enable_disk_events);