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);
738 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
739 BUG_ON(blk_bidi_rq(req));
742 * Next deal with any sectors which we were able to correctly
745 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
747 blk_rq_sectors(req), good_bytes));
750 * Recovered errors need reporting, but they're always treated
751 * as success, so fiddle the result code here. For BLOCK_PC
752 * we already took a copy of the original into rq->errors which
753 * is what gets returned to the user
755 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
756 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
757 * print since caller wants ATA registers. Only occurs on
758 * SCSI ATA PASS_THROUGH commands when CK_COND=1
760 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
762 else if (!(req->cmd_flags & REQ_QUIET))
763 scsi_print_sense("", cmd);
765 /* BLOCK_PC may have set error */
770 * If we finished all bytes in the request we are done now.
772 if (!blk_end_request(req, error, good_bytes))
776 * Kill remainder if no retrys.
778 if (error && scsi_noretry_cmd(cmd)) {
779 blk_end_request_all(req, error);
784 * If there had been no error, but we have leftover bytes in the
785 * requeues just queue the command up again.
790 error = __scsi_error_from_host_byte(cmd, result);
792 if (host_byte(result) == DID_RESET) {
793 /* Third party bus reset or reset for error recovery
794 * reasons. Just retry the command and see what
797 action = ACTION_RETRY;
798 } else if (sense_valid && !sense_deferred) {
799 switch (sshdr.sense_key) {
801 if (cmd->device->removable) {
802 /* Detected disc change. Set a bit
803 * and quietly refuse further access.
805 cmd->device->changed = 1;
806 description = "Media Changed";
807 action = ACTION_FAIL;
809 /* Must have been a power glitch, or a
810 * bus reset. Could not have been a
811 * media change, so we just retry the
812 * command and see what happens.
814 action = ACTION_RETRY;
817 case ILLEGAL_REQUEST:
818 /* If we had an ILLEGAL REQUEST returned, then
819 * we may have performed an unsupported
820 * command. The only thing this should be
821 * would be a ten byte read where only a six
822 * byte read was supported. Also, on a system
823 * where READ CAPACITY failed, we may have
824 * read past the end of the disk.
826 if ((cmd->device->use_10_for_rw &&
827 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
828 (cmd->cmnd[0] == READ_10 ||
829 cmd->cmnd[0] == WRITE_10)) {
830 /* This will issue a new 6-byte command. */
831 cmd->device->use_10_for_rw = 0;
832 action = ACTION_REPREP;
833 } else if (sshdr.asc == 0x10) /* DIX */ {
834 description = "Host Data Integrity Failure";
835 action = ACTION_FAIL;
837 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
838 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
839 switch (cmd->cmnd[0]) {
841 description = "Discard failure";
845 if (cmd->cmnd[1] & 0x8)
846 description = "Discard failure";
849 "Write same failure";
852 description = "Invalid command failure";
855 action = ACTION_FAIL;
858 action = ACTION_FAIL;
860 case ABORTED_COMMAND:
861 action = ACTION_FAIL;
862 if (sshdr.asc == 0x10) { /* DIF */
863 description = "Target Data Integrity Failure";
868 /* If the device is in the process of becoming
869 * ready, or has a temporary blockage, retry.
871 if (sshdr.asc == 0x04) {
872 switch (sshdr.ascq) {
873 case 0x01: /* becoming ready */
874 case 0x04: /* format in progress */
875 case 0x05: /* rebuild in progress */
876 case 0x06: /* recalculation in progress */
877 case 0x07: /* operation in progress */
878 case 0x08: /* Long write in progress */
879 case 0x09: /* self test in progress */
880 case 0x14: /* space allocation in progress */
881 action = ACTION_DELAYED_RETRY;
884 description = "Device not ready";
885 action = ACTION_FAIL;
889 description = "Device not ready";
890 action = ACTION_FAIL;
893 case VOLUME_OVERFLOW:
894 /* See SSC3rXX or current. */
895 action = ACTION_FAIL;
898 description = "Unhandled sense code";
899 action = ACTION_FAIL;
903 description = "Unhandled error code";
904 action = ACTION_FAIL;
907 if (action != ACTION_FAIL &&
908 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
909 action = ACTION_FAIL;
910 description = "Command timed out";
915 /* Give up and fail the remainder of the request */
916 if (!(req->cmd_flags & REQ_QUIET)) {
918 scmd_printk(KERN_INFO, cmd, "%s\n",
920 scsi_print_result(cmd);
921 if (driver_byte(result) & DRIVER_SENSE)
922 scsi_print_sense("", cmd);
923 scsi_print_command(cmd);
925 if (!blk_end_request_err(req, error))
930 /* Unprep the request and put it back at the head of the queue.
931 * A new command will be prepared and issued.
933 scsi_release_buffers(cmd);
934 scsi_requeue_command(q, cmd);
937 /* Retry the same command immediately */
938 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
940 case ACTION_DELAYED_RETRY:
941 /* Retry the same command after a delay */
942 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
948 scsi_release_buffers(cmd);
949 scsi_next_command(cmd);
952 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
958 * If sg table allocation fails, requeue request later.
960 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
962 return BLKPREP_DEFER;
966 * Next, walk the list, and fill in the addresses and sizes of
969 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
970 BUG_ON(count > sdb->table.nents);
971 sdb->table.nents = count;
972 sdb->length = blk_rq_bytes(req);
977 * Function: scsi_init_io()
979 * Purpose: SCSI I/O initialize function.
981 * Arguments: cmd - Command descriptor we wish to initialize
983 * Returns: 0 on success
984 * BLKPREP_DEFER if the failure is retryable
985 * BLKPREP_KILL if the failure is fatal
987 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
989 struct scsi_device *sdev = cmd->device;
990 struct request *rq = cmd->request;
992 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
996 if (blk_bidi_rq(rq)) {
997 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
998 scsi_sdb_cache, GFP_ATOMIC);
1000 error = BLKPREP_DEFER;
1004 rq->next_rq->special = bidi_sdb;
1005 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1010 if (blk_integrity_rq(rq)) {
1011 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1014 BUG_ON(prot_sdb == NULL);
1015 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1017 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1018 error = BLKPREP_DEFER;
1022 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1023 prot_sdb->table.sgl);
1024 BUG_ON(unlikely(count > ivecs));
1025 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1027 cmd->prot_sdb = prot_sdb;
1028 cmd->prot_sdb->table.nents = count;
1034 scsi_release_buffers(cmd);
1035 cmd->request->special = NULL;
1036 scsi_put_command(cmd);
1037 put_device(&sdev->sdev_gendev);
1040 EXPORT_SYMBOL(scsi_init_io);
1042 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1043 struct request *req)
1045 struct scsi_cmnd *cmd;
1047 if (!req->special) {
1048 /* Bail if we can't get a reference to the device */
1049 if (!get_device(&sdev->sdev_gendev))
1052 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1053 if (unlikely(!cmd)) {
1054 put_device(&sdev->sdev_gendev);
1062 /* pull a tag out of the request if we have one */
1063 cmd->tag = req->tag;
1066 cmd->cmnd = req->cmd;
1067 cmd->prot_op = SCSI_PROT_NORMAL;
1072 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1074 struct scsi_cmnd *cmd = req->special;
1077 * BLOCK_PC requests may transfer data, in which case they must
1078 * a bio attached to them. Or they might contain a SCSI command
1079 * that does not transfer data, in which case they may optionally
1080 * submit a request without an attached bio.
1085 BUG_ON(!req->nr_phys_segments);
1087 ret = scsi_init_io(cmd, GFP_ATOMIC);
1091 BUG_ON(blk_rq_bytes(req));
1093 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1096 cmd->cmd_len = req->cmd_len;
1097 if (!blk_rq_bytes(req))
1098 cmd->sc_data_direction = DMA_NONE;
1099 else if (rq_data_dir(req) == WRITE)
1100 cmd->sc_data_direction = DMA_TO_DEVICE;
1102 cmd->sc_data_direction = DMA_FROM_DEVICE;
1104 cmd->transfersize = blk_rq_bytes(req);
1105 cmd->allowed = req->retries;
1108 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1111 * Setup a REQ_TYPE_FS command. These are simple read/write request
1112 * from filesystems that still need to be translated to SCSI CDBs from
1115 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1117 struct scsi_cmnd *cmd = req->special;
1119 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1120 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1121 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1122 if (ret != BLKPREP_OK)
1127 * Filesystem requests must transfer data.
1129 BUG_ON(!req->nr_phys_segments);
1131 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1132 return scsi_init_io(cmd, GFP_ATOMIC);
1134 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1137 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1139 int ret = BLKPREP_OK;
1142 * If the device is not in running state we will reject some
1145 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1146 switch (sdev->sdev_state) {
1148 case SDEV_TRANSPORT_OFFLINE:
1150 * If the device is offline we refuse to process any
1151 * commands. The device must be brought online
1152 * before trying any recovery commands.
1154 sdev_printk(KERN_ERR, sdev,
1155 "rejecting I/O to offline device\n");
1160 * If the device is fully deleted, we refuse to
1161 * process any commands as well.
1163 sdev_printk(KERN_ERR, sdev,
1164 "rejecting I/O to dead device\n");
1169 case SDEV_CREATED_BLOCK:
1171 * If the devices is blocked we defer normal commands.
1173 if (!(req->cmd_flags & REQ_PREEMPT))
1174 ret = BLKPREP_DEFER;
1178 * For any other not fully online state we only allow
1179 * special commands. In particular any user initiated
1180 * command is not allowed.
1182 if (!(req->cmd_flags & REQ_PREEMPT))
1191 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1193 struct scsi_device *sdev = q->queuedata;
1197 req->errors = DID_NO_CONNECT << 16;
1198 /* release the command and kill it */
1200 struct scsi_cmnd *cmd = req->special;
1201 scsi_release_buffers(cmd);
1202 scsi_put_command(cmd);
1203 put_device(&sdev->sdev_gendev);
1204 req->special = NULL;
1209 * If we defer, the blk_peek_request() returns NULL, but the
1210 * queue must be restarted, so we schedule a callback to happen
1213 if (sdev->device_busy == 0)
1214 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1217 req->cmd_flags |= REQ_DONTPREP;
1223 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1225 struct scsi_device *sdev = q->queuedata;
1226 struct scsi_cmnd *cmd;
1229 ret = scsi_prep_state_check(sdev, req);
1230 if (ret != BLKPREP_OK)
1233 cmd = scsi_get_cmd_from_req(sdev, req);
1234 if (unlikely(!cmd)) {
1235 ret = BLKPREP_DEFER;
1239 if (req->cmd_type == REQ_TYPE_FS)
1240 ret = scsi_cmd_to_driver(cmd)->init_command(cmd);
1241 else if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1242 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1247 return scsi_prep_return(q, req, ret);
1250 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1252 if (req->cmd_type == REQ_TYPE_FS) {
1253 struct scsi_cmnd *cmd = req->special;
1254 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
1256 if (drv->uninit_command)
1257 drv->uninit_command(cmd);
1262 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1265 * Called with the queue_lock held.
1267 static inline int scsi_dev_queue_ready(struct request_queue *q,
1268 struct scsi_device *sdev)
1270 if (sdev->device_busy == 0 && sdev->device_blocked) {
1272 * unblock after device_blocked iterates to zero
1274 if (--sdev->device_blocked == 0) {
1276 sdev_printk(KERN_INFO, sdev,
1277 "unblocking device at zero depth\n"));
1279 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1283 if (scsi_device_is_busy(sdev))
1291 * scsi_target_queue_ready: checks if there we can send commands to target
1292 * @sdev: scsi device on starget to check.
1294 * Called with the host lock held.
1296 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1297 struct scsi_device *sdev)
1299 struct scsi_target *starget = scsi_target(sdev);
1301 if (starget->single_lun) {
1302 if (starget->starget_sdev_user &&
1303 starget->starget_sdev_user != sdev)
1305 starget->starget_sdev_user = sdev;
1308 if (starget->target_busy == 0 && starget->target_blocked) {
1310 * unblock after target_blocked iterates to zero
1312 if (--starget->target_blocked == 0) {
1313 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1314 "unblocking target at zero depth\n"));
1319 if (scsi_target_is_busy(starget)) {
1320 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1328 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1329 * return 0. We must end up running the queue again whenever 0 is
1330 * returned, else IO can hang.
1332 * Called with host_lock held.
1334 static inline int scsi_host_queue_ready(struct request_queue *q,
1335 struct Scsi_Host *shost,
1336 struct scsi_device *sdev)
1338 if (scsi_host_in_recovery(shost))
1340 if (shost->host_busy == 0 && shost->host_blocked) {
1342 * unblock after host_blocked iterates to zero
1344 if (--shost->host_blocked == 0) {
1346 printk("scsi%d unblocking host at zero depth\n",
1352 if (scsi_host_is_busy(shost)) {
1353 if (list_empty(&sdev->starved_entry))
1354 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1358 /* We're OK to process the command, so we can't be starved */
1359 if (!list_empty(&sdev->starved_entry))
1360 list_del_init(&sdev->starved_entry);
1366 * Busy state exporting function for request stacking drivers.
1368 * For efficiency, no lock is taken to check the busy state of
1369 * shost/starget/sdev, since the returned value is not guaranteed and
1370 * may be changed after request stacking drivers call the function,
1371 * regardless of taking lock or not.
1373 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1374 * needs to return 'not busy'. Otherwise, request stacking drivers
1375 * may hold requests forever.
1377 static int scsi_lld_busy(struct request_queue *q)
1379 struct scsi_device *sdev = q->queuedata;
1380 struct Scsi_Host *shost;
1382 if (blk_queue_dying(q))
1388 * Ignore host/starget busy state.
1389 * Since block layer does not have a concept of fairness across
1390 * multiple queues, congestion of host/starget needs to be handled
1393 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1400 * Kill a request for a dead device
1402 static void scsi_kill_request(struct request *req, struct request_queue *q)
1404 struct scsi_cmnd *cmd = req->special;
1405 struct scsi_device *sdev;
1406 struct scsi_target *starget;
1407 struct Scsi_Host *shost;
1409 blk_start_request(req);
1411 scmd_printk(KERN_INFO, cmd, "killing request\n");
1414 starget = scsi_target(sdev);
1416 scsi_init_cmd_errh(cmd);
1417 cmd->result = DID_NO_CONNECT << 16;
1418 atomic_inc(&cmd->device->iorequest_cnt);
1421 * SCSI request completion path will do scsi_device_unbusy(),
1422 * bump busy counts. To bump the counters, we need to dance
1423 * with the locks as normal issue path does.
1425 sdev->device_busy++;
1426 spin_unlock(sdev->request_queue->queue_lock);
1427 spin_lock(shost->host_lock);
1429 starget->target_busy++;
1430 spin_unlock(shost->host_lock);
1431 spin_lock(sdev->request_queue->queue_lock);
1433 blk_complete_request(req);
1436 static void scsi_softirq_done(struct request *rq)
1438 struct scsi_cmnd *cmd = rq->special;
1439 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1442 INIT_LIST_HEAD(&cmd->eh_entry);
1444 atomic_inc(&cmd->device->iodone_cnt);
1446 atomic_inc(&cmd->device->ioerr_cnt);
1448 disposition = scsi_decide_disposition(cmd);
1449 if (disposition != SUCCESS &&
1450 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1451 sdev_printk(KERN_ERR, cmd->device,
1452 "timing out command, waited %lus\n",
1454 disposition = SUCCESS;
1457 scsi_log_completion(cmd, disposition);
1459 switch (disposition) {
1461 scsi_finish_command(cmd);
1464 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1466 case ADD_TO_MLQUEUE:
1467 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1470 if (!scsi_eh_scmd_add(cmd, 0))
1471 scsi_finish_command(cmd);
1476 * Function: scsi_request_fn()
1478 * Purpose: Main strategy routine for SCSI.
1480 * Arguments: q - Pointer to actual queue.
1484 * Lock status: IO request lock assumed to be held when called.
1486 static void scsi_request_fn(struct request_queue *q)
1487 __releases(q->queue_lock)
1488 __acquires(q->queue_lock)
1490 struct scsi_device *sdev = q->queuedata;
1491 struct Scsi_Host *shost;
1492 struct scsi_cmnd *cmd;
1493 struct request *req;
1496 * To start with, we keep looping until the queue is empty, or until
1497 * the host is no longer able to accept any more requests.
1503 * get next queueable request. We do this early to make sure
1504 * that the request is fully prepared even if we cannot
1507 req = blk_peek_request(q);
1508 if (!req || !scsi_dev_queue_ready(q, sdev))
1511 if (unlikely(!scsi_device_online(sdev))) {
1512 sdev_printk(KERN_ERR, sdev,
1513 "rejecting I/O to offline device\n");
1514 scsi_kill_request(req, q);
1520 * Remove the request from the request list.
1522 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1523 blk_start_request(req);
1524 sdev->device_busy++;
1526 spin_unlock(q->queue_lock);
1528 if (unlikely(cmd == NULL)) {
1529 printk(KERN_CRIT "impossible request in %s.\n"
1530 "please mail a stack trace to "
1531 "linux-scsi@vger.kernel.org\n",
1533 blk_dump_rq_flags(req, "foo");
1536 spin_lock(shost->host_lock);
1539 * We hit this when the driver is using a host wide
1540 * tag map. For device level tag maps the queue_depth check
1541 * in the device ready fn would prevent us from trying
1542 * to allocate a tag. Since the map is a shared host resource
1543 * we add the dev to the starved list so it eventually gets
1544 * a run when a tag is freed.
1546 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1547 if (list_empty(&sdev->starved_entry))
1548 list_add_tail(&sdev->starved_entry,
1549 &shost->starved_list);
1553 if (!scsi_target_queue_ready(shost, sdev))
1556 if (!scsi_host_queue_ready(q, shost, sdev))
1559 scsi_target(sdev)->target_busy++;
1563 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1564 * take the lock again.
1566 spin_unlock_irq(shost->host_lock);
1569 * Finally, initialize any error handling parameters, and set up
1570 * the timers for timeouts.
1572 scsi_init_cmd_errh(cmd);
1575 * Dispatch the command to the low-level driver.
1577 rtn = scsi_dispatch_cmd(cmd);
1578 spin_lock_irq(q->queue_lock);
1586 spin_unlock_irq(shost->host_lock);
1589 * lock q, handle tag, requeue req, and decrement device_busy. We
1590 * must return with queue_lock held.
1592 * Decrementing device_busy without checking it is OK, as all such
1593 * cases (host limits or settings) should run the queue at some
1596 spin_lock_irq(q->queue_lock);
1597 blk_requeue_request(q, req);
1598 sdev->device_busy--;
1600 if (sdev->device_busy == 0)
1601 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1604 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1606 struct device *host_dev;
1607 u64 bounce_limit = 0xffffffff;
1609 if (shost->unchecked_isa_dma)
1610 return BLK_BOUNCE_ISA;
1612 * Platforms with virtual-DMA translation
1613 * hardware have no practical limit.
1615 if (!PCI_DMA_BUS_IS_PHYS)
1616 return BLK_BOUNCE_ANY;
1618 host_dev = scsi_get_device(shost);
1619 if (host_dev && host_dev->dma_mask)
1620 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1622 return bounce_limit;
1624 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1626 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1627 request_fn_proc *request_fn)
1629 struct request_queue *q;
1630 struct device *dev = shost->dma_dev;
1632 q = blk_init_queue(request_fn, NULL);
1637 * this limit is imposed by hardware restrictions
1639 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1640 SCSI_MAX_SG_CHAIN_SEGMENTS));
1642 if (scsi_host_prot_dma(shost)) {
1643 shost->sg_prot_tablesize =
1644 min_not_zero(shost->sg_prot_tablesize,
1645 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1646 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1647 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1650 blk_queue_max_hw_sectors(q, shost->max_sectors);
1651 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1652 blk_queue_segment_boundary(q, shost->dma_boundary);
1653 dma_set_seg_boundary(dev, shost->dma_boundary);
1655 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1657 if (!shost->use_clustering)
1658 q->limits.cluster = 0;
1661 * set a reasonable default alignment on word boundaries: the
1662 * host and device may alter it using
1663 * blk_queue_update_dma_alignment() later.
1665 blk_queue_dma_alignment(q, 0x03);
1669 EXPORT_SYMBOL(__scsi_alloc_queue);
1671 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1673 struct request_queue *q;
1675 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1679 blk_queue_prep_rq(q, scsi_prep_fn);
1680 blk_queue_unprep_rq(q, scsi_unprep_fn);
1681 blk_queue_softirq_done(q, scsi_softirq_done);
1682 blk_queue_rq_timed_out(q, scsi_times_out);
1683 blk_queue_lld_busy(q, scsi_lld_busy);
1688 * Function: scsi_block_requests()
1690 * Purpose: Utility function used by low-level drivers to prevent further
1691 * commands from being queued to the device.
1693 * Arguments: shost - Host in question
1697 * Lock status: No locks are assumed held.
1699 * Notes: There is no timer nor any other means by which the requests
1700 * get unblocked other than the low-level driver calling
1701 * scsi_unblock_requests().
1703 void scsi_block_requests(struct Scsi_Host *shost)
1705 shost->host_self_blocked = 1;
1707 EXPORT_SYMBOL(scsi_block_requests);
1710 * Function: scsi_unblock_requests()
1712 * Purpose: Utility function used by low-level drivers to allow further
1713 * commands from being queued to the device.
1715 * Arguments: shost - Host in question
1719 * Lock status: No locks are assumed held.
1721 * Notes: There is no timer nor any other means by which the requests
1722 * get unblocked other than the low-level driver calling
1723 * scsi_unblock_requests().
1725 * This is done as an API function so that changes to the
1726 * internals of the scsi mid-layer won't require wholesale
1727 * changes to drivers that use this feature.
1729 void scsi_unblock_requests(struct Scsi_Host *shost)
1731 shost->host_self_blocked = 0;
1732 scsi_run_host_queues(shost);
1734 EXPORT_SYMBOL(scsi_unblock_requests);
1736 int __init scsi_init_queue(void)
1740 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1741 sizeof(struct scsi_data_buffer),
1743 if (!scsi_sdb_cache) {
1744 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1748 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1749 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1750 int size = sgp->size * sizeof(struct scatterlist);
1752 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1753 SLAB_HWCACHE_ALIGN, NULL);
1755 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1760 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1763 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1772 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1773 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1775 mempool_destroy(sgp->pool);
1777 kmem_cache_destroy(sgp->slab);
1779 kmem_cache_destroy(scsi_sdb_cache);
1784 void scsi_exit_queue(void)
1788 kmem_cache_destroy(scsi_sdb_cache);
1790 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1791 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1792 mempool_destroy(sgp->pool);
1793 kmem_cache_destroy(sgp->slab);
1798 * scsi_mode_select - issue a mode select
1799 * @sdev: SCSI device to be queried
1800 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1801 * @sp: Save page bit (0 == don't save, 1 == save)
1802 * @modepage: mode page being requested
1803 * @buffer: request buffer (may not be smaller than eight bytes)
1804 * @len: length of request buffer.
1805 * @timeout: command timeout
1806 * @retries: number of retries before failing
1807 * @data: returns a structure abstracting the mode header data
1808 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1809 * must be SCSI_SENSE_BUFFERSIZE big.
1811 * Returns zero if successful; negative error number or scsi
1816 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1817 unsigned char *buffer, int len, int timeout, int retries,
1818 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1820 unsigned char cmd[10];
1821 unsigned char *real_buffer;
1824 memset(cmd, 0, sizeof(cmd));
1825 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1827 if (sdev->use_10_for_ms) {
1830 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1833 memcpy(real_buffer + 8, buffer, len);
1837 real_buffer[2] = data->medium_type;
1838 real_buffer[3] = data->device_specific;
1839 real_buffer[4] = data->longlba ? 0x01 : 0;
1841 real_buffer[6] = data->block_descriptor_length >> 8;
1842 real_buffer[7] = data->block_descriptor_length;
1844 cmd[0] = MODE_SELECT_10;
1848 if (len > 255 || data->block_descriptor_length > 255 ||
1852 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1855 memcpy(real_buffer + 4, buffer, len);
1858 real_buffer[1] = data->medium_type;
1859 real_buffer[2] = data->device_specific;
1860 real_buffer[3] = data->block_descriptor_length;
1863 cmd[0] = MODE_SELECT;
1867 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1868 sshdr, timeout, retries, NULL);
1872 EXPORT_SYMBOL_GPL(scsi_mode_select);
1875 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1876 * @sdev: SCSI device to be queried
1877 * @dbd: set if mode sense will allow block descriptors to be returned
1878 * @modepage: mode page being requested
1879 * @buffer: request buffer (may not be smaller than eight bytes)
1880 * @len: length of request buffer.
1881 * @timeout: command timeout
1882 * @retries: number of retries before failing
1883 * @data: returns a structure abstracting the mode header data
1884 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1885 * must be SCSI_SENSE_BUFFERSIZE big.
1887 * Returns zero if unsuccessful, or the header offset (either 4
1888 * or 8 depending on whether a six or ten byte command was
1889 * issued) if successful.
1892 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1893 unsigned char *buffer, int len, int timeout, int retries,
1894 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1896 unsigned char cmd[12];
1900 struct scsi_sense_hdr my_sshdr;
1902 memset(data, 0, sizeof(*data));
1903 memset(&cmd[0], 0, 12);
1904 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1907 /* caller might not be interested in sense, but we need it */
1912 use_10_for_ms = sdev->use_10_for_ms;
1914 if (use_10_for_ms) {
1918 cmd[0] = MODE_SENSE_10;
1925 cmd[0] = MODE_SENSE;
1930 memset(buffer, 0, len);
1932 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1933 sshdr, timeout, retries, NULL);
1935 /* This code looks awful: what it's doing is making sure an
1936 * ILLEGAL REQUEST sense return identifies the actual command
1937 * byte as the problem. MODE_SENSE commands can return
1938 * ILLEGAL REQUEST if the code page isn't supported */
1940 if (use_10_for_ms && !scsi_status_is_good(result) &&
1941 (driver_byte(result) & DRIVER_SENSE)) {
1942 if (scsi_sense_valid(sshdr)) {
1943 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1944 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1946 * Invalid command operation code
1948 sdev->use_10_for_ms = 0;
1954 if(scsi_status_is_good(result)) {
1955 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1956 (modepage == 6 || modepage == 8))) {
1957 /* Initio breakage? */
1960 data->medium_type = 0;
1961 data->device_specific = 0;
1963 data->block_descriptor_length = 0;
1964 } else if(use_10_for_ms) {
1965 data->length = buffer[0]*256 + buffer[1] + 2;
1966 data->medium_type = buffer[2];
1967 data->device_specific = buffer[3];
1968 data->longlba = buffer[4] & 0x01;
1969 data->block_descriptor_length = buffer[6]*256
1972 data->length = buffer[0] + 1;
1973 data->medium_type = buffer[1];
1974 data->device_specific = buffer[2];
1975 data->block_descriptor_length = buffer[3];
1977 data->header_length = header_length;
1982 EXPORT_SYMBOL(scsi_mode_sense);
1985 * scsi_test_unit_ready - test if unit is ready
1986 * @sdev: scsi device to change the state of.
1987 * @timeout: command timeout
1988 * @retries: number of retries before failing
1989 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1990 * returning sense. Make sure that this is cleared before passing
1993 * Returns zero if unsuccessful or an error if TUR failed. For
1994 * removable media, UNIT_ATTENTION sets ->changed flag.
1997 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1998 struct scsi_sense_hdr *sshdr_external)
2001 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2003 struct scsi_sense_hdr *sshdr;
2006 if (!sshdr_external)
2007 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2009 sshdr = sshdr_external;
2011 /* try to eat the UNIT_ATTENTION if there are enough retries */
2013 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2014 timeout, retries, NULL);
2015 if (sdev->removable && scsi_sense_valid(sshdr) &&
2016 sshdr->sense_key == UNIT_ATTENTION)
2018 } while (scsi_sense_valid(sshdr) &&
2019 sshdr->sense_key == UNIT_ATTENTION && --retries);
2021 if (!sshdr_external)
2025 EXPORT_SYMBOL(scsi_test_unit_ready);
2028 * scsi_device_set_state - Take the given device through the device state model.
2029 * @sdev: scsi device to change the state of.
2030 * @state: state to change to.
2032 * Returns zero if unsuccessful or an error if the requested
2033 * transition is illegal.
2036 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2038 enum scsi_device_state oldstate = sdev->sdev_state;
2040 if (state == oldstate)
2046 case SDEV_CREATED_BLOCK:
2057 case SDEV_TRANSPORT_OFFLINE:
2070 case SDEV_TRANSPORT_OFFLINE:
2078 case SDEV_TRANSPORT_OFFLINE:
2093 case SDEV_CREATED_BLOCK:
2100 case SDEV_CREATED_BLOCK:
2115 case SDEV_TRANSPORT_OFFLINE:
2128 case SDEV_TRANSPORT_OFFLINE:
2130 case SDEV_CREATED_BLOCK:
2138 sdev->sdev_state = state;
2142 SCSI_LOG_ERROR_RECOVERY(1,
2143 sdev_printk(KERN_ERR, sdev,
2144 "Illegal state transition %s->%s\n",
2145 scsi_device_state_name(oldstate),
2146 scsi_device_state_name(state))
2150 EXPORT_SYMBOL(scsi_device_set_state);
2153 * sdev_evt_emit - emit a single SCSI device uevent
2154 * @sdev: associated SCSI device
2155 * @evt: event to emit
2157 * Send a single uevent (scsi_event) to the associated scsi_device.
2159 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2164 switch (evt->evt_type) {
2165 case SDEV_EVT_MEDIA_CHANGE:
2166 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2168 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2169 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2171 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2172 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2174 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2175 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2177 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2178 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2180 case SDEV_EVT_LUN_CHANGE_REPORTED:
2181 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2190 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2194 * sdev_evt_thread - send a uevent for each scsi event
2195 * @work: work struct for scsi_device
2197 * Dispatch queued events to their associated scsi_device kobjects
2200 void scsi_evt_thread(struct work_struct *work)
2202 struct scsi_device *sdev;
2203 enum scsi_device_event evt_type;
2204 LIST_HEAD(event_list);
2206 sdev = container_of(work, struct scsi_device, event_work);
2208 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2209 if (test_and_clear_bit(evt_type, sdev->pending_events))
2210 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2213 struct scsi_event *evt;
2214 struct list_head *this, *tmp;
2215 unsigned long flags;
2217 spin_lock_irqsave(&sdev->list_lock, flags);
2218 list_splice_init(&sdev->event_list, &event_list);
2219 spin_unlock_irqrestore(&sdev->list_lock, flags);
2221 if (list_empty(&event_list))
2224 list_for_each_safe(this, tmp, &event_list) {
2225 evt = list_entry(this, struct scsi_event, node);
2226 list_del(&evt->node);
2227 scsi_evt_emit(sdev, evt);
2234 * sdev_evt_send - send asserted event to uevent thread
2235 * @sdev: scsi_device event occurred on
2236 * @evt: event to send
2238 * Assert scsi device event asynchronously.
2240 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2242 unsigned long flags;
2245 /* FIXME: currently this check eliminates all media change events
2246 * for polled devices. Need to update to discriminate between AN
2247 * and polled events */
2248 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2254 spin_lock_irqsave(&sdev->list_lock, flags);
2255 list_add_tail(&evt->node, &sdev->event_list);
2256 schedule_work(&sdev->event_work);
2257 spin_unlock_irqrestore(&sdev->list_lock, flags);
2259 EXPORT_SYMBOL_GPL(sdev_evt_send);
2262 * sdev_evt_alloc - allocate a new scsi event
2263 * @evt_type: type of event to allocate
2264 * @gfpflags: GFP flags for allocation
2266 * Allocates and returns a new scsi_event.
2268 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2271 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2275 evt->evt_type = evt_type;
2276 INIT_LIST_HEAD(&evt->node);
2278 /* evt_type-specific initialization, if any */
2280 case SDEV_EVT_MEDIA_CHANGE:
2281 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2282 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2283 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2284 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2285 case SDEV_EVT_LUN_CHANGE_REPORTED:
2293 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2296 * sdev_evt_send_simple - send asserted event to uevent thread
2297 * @sdev: scsi_device event occurred on
2298 * @evt_type: type of event to send
2299 * @gfpflags: GFP flags for allocation
2301 * Assert scsi device event asynchronously, given an event type.
2303 void sdev_evt_send_simple(struct scsi_device *sdev,
2304 enum scsi_device_event evt_type, gfp_t gfpflags)
2306 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2308 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2313 sdev_evt_send(sdev, evt);
2315 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2318 * scsi_device_quiesce - Block user issued commands.
2319 * @sdev: scsi device to quiesce.
2321 * This works by trying to transition to the SDEV_QUIESCE state
2322 * (which must be a legal transition). When the device is in this
2323 * state, only special requests will be accepted, all others will
2324 * be deferred. Since special requests may also be requeued requests,
2325 * a successful return doesn't guarantee the device will be
2326 * totally quiescent.
2328 * Must be called with user context, may sleep.
2330 * Returns zero if unsuccessful or an error if not.
2333 scsi_device_quiesce(struct scsi_device *sdev)
2335 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2339 scsi_run_queue(sdev->request_queue);
2340 while (sdev->device_busy) {
2341 msleep_interruptible(200);
2342 scsi_run_queue(sdev->request_queue);
2346 EXPORT_SYMBOL(scsi_device_quiesce);
2349 * scsi_device_resume - Restart user issued commands to a quiesced device.
2350 * @sdev: scsi device to resume.
2352 * Moves the device from quiesced back to running and restarts the
2355 * Must be called with user context, may sleep.
2357 void scsi_device_resume(struct scsi_device *sdev)
2359 /* check if the device state was mutated prior to resume, and if
2360 * so assume the state is being managed elsewhere (for example
2361 * device deleted during suspend)
2363 if (sdev->sdev_state != SDEV_QUIESCE ||
2364 scsi_device_set_state(sdev, SDEV_RUNNING))
2366 scsi_run_queue(sdev->request_queue);
2368 EXPORT_SYMBOL(scsi_device_resume);
2371 device_quiesce_fn(struct scsi_device *sdev, void *data)
2373 scsi_device_quiesce(sdev);
2377 scsi_target_quiesce(struct scsi_target *starget)
2379 starget_for_each_device(starget, NULL, device_quiesce_fn);
2381 EXPORT_SYMBOL(scsi_target_quiesce);
2384 device_resume_fn(struct scsi_device *sdev, void *data)
2386 scsi_device_resume(sdev);
2390 scsi_target_resume(struct scsi_target *starget)
2392 starget_for_each_device(starget, NULL, device_resume_fn);
2394 EXPORT_SYMBOL(scsi_target_resume);
2397 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2398 * @sdev: device to block
2400 * Block request made by scsi lld's to temporarily stop all
2401 * scsi commands on the specified device. Called from interrupt
2402 * or normal process context.
2404 * Returns zero if successful or error if not
2407 * This routine transitions the device to the SDEV_BLOCK state
2408 * (which must be a legal transition). When the device is in this
2409 * state, all commands are deferred until the scsi lld reenables
2410 * the device with scsi_device_unblock or device_block_tmo fires.
2413 scsi_internal_device_block(struct scsi_device *sdev)
2415 struct request_queue *q = sdev->request_queue;
2416 unsigned long flags;
2419 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2421 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2428 * The device has transitioned to SDEV_BLOCK. Stop the
2429 * block layer from calling the midlayer with this device's
2432 spin_lock_irqsave(q->queue_lock, flags);
2434 spin_unlock_irqrestore(q->queue_lock, flags);
2438 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2441 * scsi_internal_device_unblock - resume a device after a block request
2442 * @sdev: device to resume
2443 * @new_state: state to set devices to after unblocking
2445 * Called by scsi lld's or the midlayer to restart the device queue
2446 * for the previously suspended scsi device. Called from interrupt or
2447 * normal process context.
2449 * Returns zero if successful or error if not.
2452 * This routine transitions the device to the SDEV_RUNNING state
2453 * or to one of the offline states (which must be a legal transition)
2454 * allowing the midlayer to goose the queue for this device.
2457 scsi_internal_device_unblock(struct scsi_device *sdev,
2458 enum scsi_device_state new_state)
2460 struct request_queue *q = sdev->request_queue;
2461 unsigned long flags;
2464 * Try to transition the scsi device to SDEV_RUNNING or one of the
2465 * offlined states and goose the device queue if successful.
2467 if ((sdev->sdev_state == SDEV_BLOCK) ||
2468 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2469 sdev->sdev_state = new_state;
2470 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2471 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2472 new_state == SDEV_OFFLINE)
2473 sdev->sdev_state = new_state;
2475 sdev->sdev_state = SDEV_CREATED;
2476 } else if (sdev->sdev_state != SDEV_CANCEL &&
2477 sdev->sdev_state != SDEV_OFFLINE)
2480 spin_lock_irqsave(q->queue_lock, flags);
2482 spin_unlock_irqrestore(q->queue_lock, flags);
2486 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2489 device_block(struct scsi_device *sdev, void *data)
2491 scsi_internal_device_block(sdev);
2495 target_block(struct device *dev, void *data)
2497 if (scsi_is_target_device(dev))
2498 starget_for_each_device(to_scsi_target(dev), NULL,
2504 scsi_target_block(struct device *dev)
2506 if (scsi_is_target_device(dev))
2507 starget_for_each_device(to_scsi_target(dev), NULL,
2510 device_for_each_child(dev, NULL, target_block);
2512 EXPORT_SYMBOL_GPL(scsi_target_block);
2515 device_unblock(struct scsi_device *sdev, void *data)
2517 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2521 target_unblock(struct device *dev, void *data)
2523 if (scsi_is_target_device(dev))
2524 starget_for_each_device(to_scsi_target(dev), data,
2530 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2532 if (scsi_is_target_device(dev))
2533 starget_for_each_device(to_scsi_target(dev), &new_state,
2536 device_for_each_child(dev, &new_state, target_unblock);
2538 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2541 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2542 * @sgl: scatter-gather list
2543 * @sg_count: number of segments in sg
2544 * @offset: offset in bytes into sg, on return offset into the mapped area
2545 * @len: bytes to map, on return number of bytes mapped
2547 * Returns virtual address of the start of the mapped page
2549 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2550 size_t *offset, size_t *len)
2553 size_t sg_len = 0, len_complete = 0;
2554 struct scatterlist *sg;
2557 WARN_ON(!irqs_disabled());
2559 for_each_sg(sgl, sg, sg_count, i) {
2560 len_complete = sg_len; /* Complete sg-entries */
2561 sg_len += sg->length;
2562 if (sg_len > *offset)
2566 if (unlikely(i == sg_count)) {
2567 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2569 __func__, sg_len, *offset, sg_count);
2574 /* Offset starting from the beginning of first page in this sg-entry */
2575 *offset = *offset - len_complete + sg->offset;
2577 /* Assumption: contiguous pages can be accessed as "page + i" */
2578 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2579 *offset &= ~PAGE_MASK;
2581 /* Bytes in this sg-entry from *offset to the end of the page */
2582 sg_len = PAGE_SIZE - *offset;
2586 return kmap_atomic(page);
2588 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2591 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2592 * @virt: virtual address to be unmapped
2594 void scsi_kunmap_atomic_sg(void *virt)
2596 kunmap_atomic(virt);
2598 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2600 void sdev_disable_disk_events(struct scsi_device *sdev)
2602 atomic_inc(&sdev->disk_events_disable_depth);
2604 EXPORT_SYMBOL(sdev_disable_disk_events);
2606 void sdev_enable_disk_events(struct scsi_device *sdev)
2608 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2610 atomic_dec(&sdev->disk_events_disable_depth);
2612 EXPORT_SYMBOL(sdev_enable_disk_events);