2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
10 #include <linux/bio.h>
11 #include <linux/bitops.h>
12 #include <linux/blkdev.h>
13 #include <linux/completion.h>
14 #include <linux/kernel.h>
15 #include <linux/export.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_dbg.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_driver.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_host.h>
32 #include <trace/events/scsi.h>
34 #include "scsi_priv.h"
35 #include "scsi_logging.h"
38 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
39 #define SG_MEMPOOL_SIZE 2
41 struct scsi_host_sg_pool {
44 struct kmem_cache *slab;
48 #define SP(x) { x, "sgpool-" __stringify(x) }
49 #if (SCSI_MAX_SG_SEGMENTS < 32)
50 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
52 static struct scsi_host_sg_pool scsi_sg_pools[] = {
55 #if (SCSI_MAX_SG_SEGMENTS > 32)
57 #if (SCSI_MAX_SG_SEGMENTS > 64)
59 #if (SCSI_MAX_SG_SEGMENTS > 128)
61 #if (SCSI_MAX_SG_SEGMENTS > 256)
62 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
67 SP(SCSI_MAX_SG_SEGMENTS)
71 struct kmem_cache *scsi_sdb_cache;
74 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
75 * not change behaviour from the previous unplug mechanism, experimentation
76 * may prove this needs changing.
78 #define SCSI_QUEUE_DELAY 3
81 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
83 struct Scsi_Host *host = cmd->device->host;
84 struct scsi_device *device = cmd->device;
85 struct scsi_target *starget = scsi_target(device);
88 * Set the appropriate busy bit for the device/host.
90 * If the host/device isn't busy, assume that something actually
91 * completed, and that we should be able to queue a command now.
93 * Note that the prior mid-layer assumption that any host could
94 * always queue at least one command is now broken. The mid-layer
95 * will implement a user specifiable stall (see
96 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
97 * if a command is requeued with no other commands outstanding
98 * either for the device or for the host.
101 case SCSI_MLQUEUE_HOST_BUSY:
102 host->host_blocked = host->max_host_blocked;
104 case SCSI_MLQUEUE_DEVICE_BUSY:
105 case SCSI_MLQUEUE_EH_RETRY:
106 device->device_blocked = device->max_device_blocked;
108 case SCSI_MLQUEUE_TARGET_BUSY:
109 starget->target_blocked = starget->max_target_blocked;
115 * __scsi_queue_insert - private queue insertion
116 * @cmd: The SCSI command being requeued
117 * @reason: The reason for the requeue
118 * @unbusy: Whether the queue should be unbusied
120 * This is a private queue insertion. The public interface
121 * scsi_queue_insert() always assumes the queue should be unbusied
122 * because it's always called before the completion. This function is
123 * for a requeue after completion, which should only occur in this
126 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
128 struct scsi_device *device = cmd->device;
129 struct request_queue *q = device->request_queue;
132 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
133 "Inserting command %p into mlqueue\n", cmd));
135 scsi_set_blocked(cmd, reason);
138 * Decrement the counters, since these commands are no longer
139 * active on the host/device.
142 scsi_device_unbusy(device);
145 * Requeue this command. It will go before all other commands
146 * that are already in the queue. Schedule requeue work under
147 * lock such that the kblockd_schedule_work() call happens
148 * before blk_cleanup_queue() finishes.
151 spin_lock_irqsave(q->queue_lock, flags);
152 blk_requeue_request(q, cmd->request);
153 kblockd_schedule_work(&device->requeue_work);
154 spin_unlock_irqrestore(q->queue_lock, flags);
158 * Function: scsi_queue_insert()
160 * Purpose: Insert a command in the midlevel queue.
162 * Arguments: cmd - command that we are adding to queue.
163 * reason - why we are inserting command to queue.
165 * Lock status: Assumed that lock is not held upon entry.
169 * Notes: We do this for one of two cases. Either the host is busy
170 * and it cannot accept any more commands for the time being,
171 * or the device returned QUEUE_FULL and can accept no more
173 * Notes: This could be called either from an interrupt context or a
174 * normal process context.
176 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
178 __scsi_queue_insert(cmd, reason, 1);
181 * scsi_execute - insert request and wait for the result
184 * @data_direction: data direction
185 * @buffer: data buffer
186 * @bufflen: len of buffer
187 * @sense: optional sense buffer
188 * @timeout: request timeout in seconds
189 * @retries: number of times to retry request
190 * @flags: or into request flags;
191 * @resid: optional residual length
193 * returns the req->errors value which is the scsi_cmnd result
196 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
197 int data_direction, void *buffer, unsigned bufflen,
198 unsigned char *sense, int timeout, int retries, u64 flags,
202 int write = (data_direction == DMA_TO_DEVICE);
203 int ret = DRIVER_ERROR << 24;
205 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
208 blk_rq_set_block_pc(req);
210 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
211 buffer, bufflen, __GFP_WAIT))
214 req->cmd_len = COMMAND_SIZE(cmd[0]);
215 memcpy(req->cmd, cmd, req->cmd_len);
218 req->retries = retries;
219 req->timeout = timeout;
220 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
223 * head injection *required* here otherwise quiesce won't work
225 blk_execute_rq(req->q, NULL, req, 1);
228 * Some devices (USB mass-storage in particular) may transfer
229 * garbage data together with a residue indicating that the data
230 * is invalid. Prevent the garbage from being misinterpreted
231 * and prevent security leaks by zeroing out the excess data.
233 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
234 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
237 *resid = req->resid_len;
240 blk_put_request(req);
244 EXPORT_SYMBOL(scsi_execute);
246 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
247 int data_direction, void *buffer, unsigned bufflen,
248 struct scsi_sense_hdr *sshdr, int timeout, int retries,
249 int *resid, u64 flags)
255 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
257 return DRIVER_ERROR << 24;
259 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
260 sense, timeout, retries, flags, resid);
262 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
267 EXPORT_SYMBOL(scsi_execute_req_flags);
270 * Function: scsi_init_cmd_errh()
272 * Purpose: Initialize cmd fields related to error handling.
274 * Arguments: cmd - command that is ready to be queued.
276 * Notes: This function has the job of initializing a number of
277 * fields related to error handling. Typically this will
278 * be called once for each command, as required.
280 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
282 cmd->serial_number = 0;
283 scsi_set_resid(cmd, 0);
284 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
285 if (cmd->cmd_len == 0)
286 cmd->cmd_len = scsi_command_size(cmd->cmnd);
289 void scsi_device_unbusy(struct scsi_device *sdev)
291 struct Scsi_Host *shost = sdev->host;
292 struct scsi_target *starget = scsi_target(sdev);
295 atomic_dec(&shost->host_busy);
296 atomic_dec(&starget->target_busy);
298 if (unlikely(scsi_host_in_recovery(shost) &&
299 (shost->host_failed || shost->host_eh_scheduled))) {
300 spin_lock_irqsave(shost->host_lock, flags);
301 scsi_eh_wakeup(shost);
302 spin_unlock_irqrestore(shost->host_lock, flags);
305 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
307 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
311 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
312 * and call blk_run_queue for all the scsi_devices on the target -
313 * including current_sdev first.
315 * Called with *no* scsi locks held.
317 static void scsi_single_lun_run(struct scsi_device *current_sdev)
319 struct Scsi_Host *shost = current_sdev->host;
320 struct scsi_device *sdev, *tmp;
321 struct scsi_target *starget = scsi_target(current_sdev);
324 spin_lock_irqsave(shost->host_lock, flags);
325 starget->starget_sdev_user = NULL;
326 spin_unlock_irqrestore(shost->host_lock, flags);
329 * Call blk_run_queue for all LUNs on the target, starting with
330 * current_sdev. We race with others (to set starget_sdev_user),
331 * but in most cases, we will be first. Ideally, each LU on the
332 * target would get some limited time or requests on the target.
334 blk_run_queue(current_sdev->request_queue);
336 spin_lock_irqsave(shost->host_lock, flags);
337 if (starget->starget_sdev_user)
339 list_for_each_entry_safe(sdev, tmp, &starget->devices,
340 same_target_siblings) {
341 if (sdev == current_sdev)
343 if (scsi_device_get(sdev))
346 spin_unlock_irqrestore(shost->host_lock, flags);
347 blk_run_queue(sdev->request_queue);
348 spin_lock_irqsave(shost->host_lock, flags);
350 scsi_device_put(sdev);
353 spin_unlock_irqrestore(shost->host_lock, flags);
356 static inline int scsi_device_is_busy(struct scsi_device *sdev)
358 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
364 static inline int scsi_target_is_busy(struct scsi_target *starget)
366 return ((starget->can_queue > 0 &&
367 atomic_read(&starget->target_busy) >= starget->can_queue) ||
368 starget->target_blocked);
371 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
373 if ((shost->can_queue > 0 &&
374 atomic_read(&shost->host_busy) >= shost->can_queue) ||
375 shost->host_blocked || shost->host_self_blocked)
381 static void scsi_starved_list_run(struct Scsi_Host *shost)
383 LIST_HEAD(starved_list);
384 struct scsi_device *sdev;
387 spin_lock_irqsave(shost->host_lock, flags);
388 list_splice_init(&shost->starved_list, &starved_list);
390 while (!list_empty(&starved_list)) {
391 struct request_queue *slq;
394 * As long as shost is accepting commands and we have
395 * starved queues, call blk_run_queue. scsi_request_fn
396 * drops the queue_lock and can add us back to the
399 * host_lock protects the starved_list and starved_entry.
400 * scsi_request_fn must get the host_lock before checking
401 * or modifying starved_list or starved_entry.
403 if (scsi_host_is_busy(shost))
406 sdev = list_entry(starved_list.next,
407 struct scsi_device, starved_entry);
408 list_del_init(&sdev->starved_entry);
409 if (scsi_target_is_busy(scsi_target(sdev))) {
410 list_move_tail(&sdev->starved_entry,
411 &shost->starved_list);
416 * Once we drop the host lock, a racing scsi_remove_device()
417 * call may remove the sdev from the starved list and destroy
418 * it and the queue. Mitigate by taking a reference to the
419 * queue and never touching the sdev again after we drop the
420 * host lock. Note: if __scsi_remove_device() invokes
421 * blk_cleanup_queue() before the queue is run from this
422 * function then blk_run_queue() will return immediately since
423 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
425 slq = sdev->request_queue;
426 if (!blk_get_queue(slq))
428 spin_unlock_irqrestore(shost->host_lock, flags);
433 spin_lock_irqsave(shost->host_lock, flags);
435 /* put any unprocessed entries back */
436 list_splice(&starved_list, &shost->starved_list);
437 spin_unlock_irqrestore(shost->host_lock, flags);
441 * Function: scsi_run_queue()
443 * Purpose: Select a proper request queue to serve next
445 * Arguments: q - last request's queue
449 * Notes: The previous command was completely finished, start
450 * a new one if possible.
452 static void scsi_run_queue(struct request_queue *q)
454 struct scsi_device *sdev = q->queuedata;
456 if (scsi_target(sdev)->single_lun)
457 scsi_single_lun_run(sdev);
458 if (!list_empty(&sdev->host->starved_list))
459 scsi_starved_list_run(sdev->host);
464 void scsi_requeue_run_queue(struct work_struct *work)
466 struct scsi_device *sdev;
467 struct request_queue *q;
469 sdev = container_of(work, struct scsi_device, requeue_work);
470 q = sdev->request_queue;
475 * Function: scsi_requeue_command()
477 * Purpose: Handle post-processing of completed commands.
479 * Arguments: q - queue to operate on
480 * cmd - command that may need to be requeued.
484 * Notes: After command completion, there may be blocks left
485 * over which weren't finished by the previous command
486 * this can be for a number of reasons - the main one is
487 * I/O errors in the middle of the request, in which case
488 * we need to request the blocks that come after the bad
490 * Notes: Upon return, cmd is a stale pointer.
492 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
494 struct scsi_device *sdev = cmd->device;
495 struct request *req = cmd->request;
498 spin_lock_irqsave(q->queue_lock, flags);
499 blk_unprep_request(req);
501 scsi_put_command(cmd);
502 blk_requeue_request(q, req);
503 spin_unlock_irqrestore(q->queue_lock, flags);
507 put_device(&sdev->sdev_gendev);
510 void scsi_next_command(struct scsi_cmnd *cmd)
512 struct scsi_device *sdev = cmd->device;
513 struct request_queue *q = sdev->request_queue;
515 scsi_put_command(cmd);
518 put_device(&sdev->sdev_gendev);
521 void scsi_run_host_queues(struct Scsi_Host *shost)
523 struct scsi_device *sdev;
525 shost_for_each_device(sdev, shost)
526 scsi_run_queue(sdev->request_queue);
529 static inline unsigned int scsi_sgtable_index(unsigned short nents)
533 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
538 index = get_count_order(nents) - 3;
543 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
545 struct scsi_host_sg_pool *sgp;
547 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
548 mempool_free(sgl, sgp->pool);
551 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
553 struct scsi_host_sg_pool *sgp;
555 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
556 return mempool_alloc(sgp->pool, gfp_mask);
559 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
566 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
567 gfp_mask, scsi_sg_alloc);
569 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
575 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
577 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
581 * Function: scsi_release_buffers()
583 * Purpose: Free resources allocate for a scsi_command.
585 * Arguments: cmd - command that we are bailing.
587 * Lock status: Assumed that no lock is held upon entry.
591 * Notes: In the event that an upper level driver rejects a
592 * command, we must release resources allocated during
593 * the __init_io() function. Primarily this would involve
594 * the scatter-gather table.
596 static void scsi_release_buffers(struct scsi_cmnd *cmd)
598 if (cmd->sdb.table.nents)
599 scsi_free_sgtable(&cmd->sdb);
601 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
603 if (scsi_prot_sg_count(cmd))
604 scsi_free_sgtable(cmd->prot_sdb);
607 static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
609 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
611 scsi_free_sgtable(bidi_sdb);
612 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
613 cmd->request->next_rq->special = NULL;
617 * __scsi_error_from_host_byte - translate SCSI error code into errno
618 * @cmd: SCSI command (unused)
619 * @result: scsi error code
621 * Translate SCSI error code into standard UNIX errno.
623 * -ENOLINK temporary transport failure
624 * -EREMOTEIO permanent target failure, do not retry
625 * -EBADE permanent nexus failure, retry on other path
626 * -ENOSPC No write space available
627 * -ENODATA Medium error
628 * -EIO unspecified I/O error
630 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
634 switch(host_byte(result)) {
635 case DID_TRANSPORT_FAILFAST:
638 case DID_TARGET_FAILURE:
639 set_host_byte(cmd, DID_OK);
642 case DID_NEXUS_FAILURE:
643 set_host_byte(cmd, DID_OK);
646 case DID_ALLOC_FAILURE:
647 set_host_byte(cmd, DID_OK);
650 case DID_MEDIUM_ERROR:
651 set_host_byte(cmd, DID_OK);
663 * Function: scsi_io_completion()
665 * Purpose: Completion processing for block device I/O requests.
667 * Arguments: cmd - command that is finished.
669 * Lock status: Assumed that no lock is held upon entry.
673 * Notes: We will finish off the specified number of sectors. If we
674 * are done, the command block will be released and the queue
675 * function will be goosed. If we are not done then we have to
676 * figure out what to do next:
678 * a) We can call scsi_requeue_command(). The request
679 * will be unprepared and put back on the queue. Then
680 * a new command will be created for it. This should
681 * be used if we made forward progress, or if we want
682 * to switch from READ(10) to READ(6) for example.
684 * b) We can call __scsi_queue_insert(). The request will
685 * be put back on the queue and retried using the same
686 * command as before, possibly after a delay.
688 * c) We can call blk_end_request() with -EIO to fail
689 * the remainder of the request.
691 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
693 int result = cmd->result;
694 struct request_queue *q = cmd->device->request_queue;
695 struct request *req = cmd->request;
697 struct scsi_sense_hdr sshdr;
699 int sense_deferred = 0;
700 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
701 ACTION_DELAYED_RETRY} action;
702 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
705 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
707 sense_deferred = scsi_sense_is_deferred(&sshdr);
710 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
712 if (sense_valid && req->sense) {
714 * SG_IO wants current and deferred errors
716 int len = 8 + cmd->sense_buffer[7];
718 if (len > SCSI_SENSE_BUFFERSIZE)
719 len = SCSI_SENSE_BUFFERSIZE;
720 memcpy(req->sense, cmd->sense_buffer, len);
721 req->sense_len = len;
724 error = __scsi_error_from_host_byte(cmd, result);
727 * __scsi_error_from_host_byte may have reset the host_byte
729 req->errors = cmd->result;
731 req->resid_len = scsi_get_resid(cmd);
733 if (scsi_bidi_cmnd(cmd)) {
735 * Bidi commands Must be complete as a whole,
736 * both sides at once.
738 req->next_rq->resid_len = scsi_in(cmd)->resid;
740 scsi_release_buffers(cmd);
741 scsi_release_bidi_buffers(cmd);
743 blk_end_request_all(req, 0);
745 scsi_next_command(cmd);
748 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
750 * Certain non BLOCK_PC requests are commands that don't
751 * actually transfer anything (FLUSH), so cannot use
752 * good_bytes != blk_rq_bytes(req) as the signal for an error.
753 * This sets the error explicitly for the problem case.
755 error = __scsi_error_from_host_byte(cmd, result);
758 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
759 BUG_ON(blk_bidi_rq(req));
762 * Next deal with any sectors which we were able to correctly
765 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
766 "%u sectors total, %d bytes done.\n",
767 blk_rq_sectors(req), good_bytes));
770 * Recovered errors need reporting, but they're always treated
771 * as success, so fiddle the result code here. For BLOCK_PC
772 * we already took a copy of the original into rq->errors which
773 * is what gets returned to the user
775 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
776 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
777 * print since caller wants ATA registers. Only occurs on
778 * SCSI ATA PASS_THROUGH commands when CK_COND=1
780 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
782 else if (!(req->cmd_flags & REQ_QUIET))
783 scsi_print_sense("", cmd);
785 /* BLOCK_PC may have set error */
790 * If we finished all bytes in the request we are done now.
792 if (!blk_end_request(req, error, good_bytes))
796 * Kill remainder if no retrys.
798 if (error && scsi_noretry_cmd(cmd)) {
799 blk_end_request_all(req, error);
804 * If there had been no error, but we have leftover bytes in the
805 * requeues just queue the command up again.
810 error = __scsi_error_from_host_byte(cmd, result);
812 if (host_byte(result) == DID_RESET) {
813 /* Third party bus reset or reset for error recovery
814 * reasons. Just retry the command and see what
817 action = ACTION_RETRY;
818 } else if (sense_valid && !sense_deferred) {
819 switch (sshdr.sense_key) {
821 if (cmd->device->removable) {
822 /* Detected disc change. Set a bit
823 * and quietly refuse further access.
825 cmd->device->changed = 1;
826 action = ACTION_FAIL;
828 /* Must have been a power glitch, or a
829 * bus reset. Could not have been a
830 * media change, so we just retry the
831 * command and see what happens.
833 action = ACTION_RETRY;
836 case ILLEGAL_REQUEST:
837 /* If we had an ILLEGAL REQUEST returned, then
838 * we may have performed an unsupported
839 * command. The only thing this should be
840 * would be a ten byte read where only a six
841 * byte read was supported. Also, on a system
842 * where READ CAPACITY failed, we may have
843 * read past the end of the disk.
845 if ((cmd->device->use_10_for_rw &&
846 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
847 (cmd->cmnd[0] == READ_10 ||
848 cmd->cmnd[0] == WRITE_10)) {
849 /* This will issue a new 6-byte command. */
850 cmd->device->use_10_for_rw = 0;
851 action = ACTION_REPREP;
852 } else if (sshdr.asc == 0x10) /* DIX */ {
853 action = ACTION_FAIL;
855 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
856 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
857 action = ACTION_FAIL;
860 action = ACTION_FAIL;
862 case ABORTED_COMMAND:
863 action = ACTION_FAIL;
864 if (sshdr.asc == 0x10) /* DIF */
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 action = ACTION_FAIL;
888 action = ACTION_FAIL;
890 case VOLUME_OVERFLOW:
891 /* See SSC3rXX or current. */
892 action = ACTION_FAIL;
895 action = ACTION_FAIL;
899 action = ACTION_FAIL;
901 if (action != ACTION_FAIL &&
902 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
903 action = ACTION_FAIL;
907 /* Give up and fail the remainder of the request */
908 if (!(req->cmd_flags & REQ_QUIET)) {
909 scsi_print_result(cmd);
910 if (driver_byte(result) & DRIVER_SENSE)
911 scsi_print_sense("", cmd);
912 scsi_print_command(cmd);
914 if (!blk_end_request_err(req, error))
919 /* Unprep the request and put it back at the head of the queue.
920 * A new command will be prepared and issued.
922 scsi_release_buffers(cmd);
923 scsi_requeue_command(q, cmd);
926 /* Retry the same command immediately */
927 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
929 case ACTION_DELAYED_RETRY:
930 /* Retry the same command after a delay */
931 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
937 scsi_release_buffers(cmd);
938 scsi_next_command(cmd);
941 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
947 * If sg table allocation fails, requeue request later.
949 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
951 return BLKPREP_DEFER;
955 * Next, walk the list, and fill in the addresses and sizes of
958 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
959 BUG_ON(count > sdb->table.nents);
960 sdb->table.nents = count;
961 sdb->length = blk_rq_bytes(req);
966 * Function: scsi_init_io()
968 * Purpose: SCSI I/O initialize function.
970 * Arguments: cmd - Command descriptor we wish to initialize
972 * Returns: 0 on success
973 * BLKPREP_DEFER if the failure is retryable
974 * BLKPREP_KILL if the failure is fatal
976 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
978 struct scsi_device *sdev = cmd->device;
979 struct request *rq = cmd->request;
982 BUG_ON(!rq->nr_phys_segments);
984 error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
988 if (blk_bidi_rq(rq)) {
989 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
990 scsi_sdb_cache, GFP_ATOMIC);
992 error = BLKPREP_DEFER;
996 rq->next_rq->special = bidi_sdb;
997 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1002 if (blk_integrity_rq(rq)) {
1003 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1006 BUG_ON(prot_sdb == NULL);
1007 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1009 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1010 error = BLKPREP_DEFER;
1014 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1015 prot_sdb->table.sgl);
1016 BUG_ON(unlikely(count > ivecs));
1017 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1019 cmd->prot_sdb = prot_sdb;
1020 cmd->prot_sdb->table.nents = count;
1026 scsi_release_buffers(cmd);
1027 cmd->request->special = NULL;
1028 scsi_put_command(cmd);
1029 put_device(&sdev->sdev_gendev);
1032 EXPORT_SYMBOL(scsi_init_io);
1034 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1035 struct request *req)
1037 struct scsi_cmnd *cmd;
1039 if (!req->special) {
1040 /* Bail if we can't get a reference to the device */
1041 if (!get_device(&sdev->sdev_gendev))
1044 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1045 if (unlikely(!cmd)) {
1046 put_device(&sdev->sdev_gendev);
1054 /* pull a tag out of the request if we have one */
1055 cmd->tag = req->tag;
1058 cmd->cmnd = req->cmd;
1059 cmd->prot_op = SCSI_PROT_NORMAL;
1064 static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1066 struct scsi_cmnd *cmd = req->special;
1069 * BLOCK_PC requests may transfer data, in which case they must
1070 * a bio attached to them. Or they might contain a SCSI command
1071 * that does not transfer data, in which case they may optionally
1072 * submit a request without an attached bio.
1075 int ret = scsi_init_io(cmd, GFP_ATOMIC);
1079 BUG_ON(blk_rq_bytes(req));
1081 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1084 cmd->cmd_len = req->cmd_len;
1085 cmd->transfersize = blk_rq_bytes(req);
1086 cmd->allowed = req->retries;
1091 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1092 * that still need to be translated to SCSI CDBs from the ULD.
1094 static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1096 struct scsi_cmnd *cmd = req->special;
1098 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1099 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1100 int ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1101 if (ret != BLKPREP_OK)
1105 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1106 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1109 static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1111 struct scsi_cmnd *cmd = req->special;
1113 if (!blk_rq_bytes(req))
1114 cmd->sc_data_direction = DMA_NONE;
1115 else if (rq_data_dir(req) == WRITE)
1116 cmd->sc_data_direction = DMA_TO_DEVICE;
1118 cmd->sc_data_direction = DMA_FROM_DEVICE;
1120 switch (req->cmd_type) {
1122 return scsi_setup_fs_cmnd(sdev, req);
1123 case REQ_TYPE_BLOCK_PC:
1124 return scsi_setup_blk_pc_cmnd(sdev, req);
1126 return BLKPREP_KILL;
1131 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1133 int ret = BLKPREP_OK;
1136 * If the device is not in running state we will reject some
1139 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1140 switch (sdev->sdev_state) {
1142 case SDEV_TRANSPORT_OFFLINE:
1144 * If the device is offline we refuse to process any
1145 * commands. The device must be brought online
1146 * before trying any recovery commands.
1148 sdev_printk(KERN_ERR, sdev,
1149 "rejecting I/O to offline device\n");
1154 * If the device is fully deleted, we refuse to
1155 * process any commands as well.
1157 sdev_printk(KERN_ERR, sdev,
1158 "rejecting I/O to dead device\n");
1163 case SDEV_CREATED_BLOCK:
1165 * If the devices is blocked we defer normal commands.
1167 if (!(req->cmd_flags & REQ_PREEMPT))
1168 ret = BLKPREP_DEFER;
1172 * For any other not fully online state we only allow
1173 * special commands. In particular any user initiated
1174 * command is not allowed.
1176 if (!(req->cmd_flags & REQ_PREEMPT))
1185 scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1187 struct scsi_device *sdev = q->queuedata;
1191 req->errors = DID_NO_CONNECT << 16;
1192 /* release the command and kill it */
1194 struct scsi_cmnd *cmd = req->special;
1195 scsi_release_buffers(cmd);
1196 scsi_put_command(cmd);
1197 put_device(&sdev->sdev_gendev);
1198 req->special = NULL;
1203 * If we defer, the blk_peek_request() returns NULL, but the
1204 * queue must be restarted, so we schedule a callback to happen
1207 if (sdev->device_busy == 0)
1208 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1211 req->cmd_flags |= REQ_DONTPREP;
1217 static int scsi_prep_fn(struct request_queue *q, struct request *req)
1219 struct scsi_device *sdev = q->queuedata;
1220 struct scsi_cmnd *cmd;
1223 ret = scsi_prep_state_check(sdev, req);
1224 if (ret != BLKPREP_OK)
1227 cmd = scsi_get_cmd_from_req(sdev, req);
1228 if (unlikely(!cmd)) {
1229 ret = BLKPREP_DEFER;
1233 ret = scsi_setup_cmnd(sdev, req);
1235 return scsi_prep_return(q, req, ret);
1238 static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1240 if (req->cmd_type == REQ_TYPE_FS) {
1241 struct scsi_cmnd *cmd = req->special;
1242 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
1244 if (drv->uninit_command)
1245 drv->uninit_command(cmd);
1250 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1253 * Called with the queue_lock held.
1255 static inline int scsi_dev_queue_ready(struct request_queue *q,
1256 struct scsi_device *sdev)
1258 if (sdev->device_busy == 0 && sdev->device_blocked) {
1260 * unblock after device_blocked iterates to zero
1262 if (--sdev->device_blocked == 0) {
1264 sdev_printk(KERN_INFO, sdev,
1265 "unblocking device at zero depth\n"));
1267 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1271 if (scsi_device_is_busy(sdev))
1279 * scsi_target_queue_ready: checks if there we can send commands to target
1280 * @sdev: scsi device on starget to check.
1282 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1283 struct scsi_device *sdev)
1285 struct scsi_target *starget = scsi_target(sdev);
1288 if (starget->single_lun) {
1289 spin_lock_irq(shost->host_lock);
1290 if (starget->starget_sdev_user &&
1291 starget->starget_sdev_user != sdev) {
1292 spin_unlock_irq(shost->host_lock);
1295 starget->starget_sdev_user = sdev;
1296 spin_unlock_irq(shost->host_lock);
1299 busy = atomic_inc_return(&starget->target_busy) - 1;
1300 if (starget->target_blocked) {
1305 * unblock after target_blocked iterates to zero
1307 spin_lock_irq(shost->host_lock);
1308 if (--starget->target_blocked != 0) {
1309 spin_unlock_irq(shost->host_lock);
1312 spin_unlock_irq(shost->host_lock);
1314 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1315 "unblocking target at zero depth\n"));
1318 if (starget->can_queue > 0 && busy >= starget->can_queue)
1324 spin_lock_irq(shost->host_lock);
1325 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1326 spin_unlock_irq(shost->host_lock);
1328 atomic_dec(&starget->target_busy);
1333 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1334 * return 0. We must end up running the queue again whenever 0 is
1335 * returned, else IO can hang.
1337 static inline int scsi_host_queue_ready(struct request_queue *q,
1338 struct Scsi_Host *shost,
1339 struct scsi_device *sdev)
1343 if (scsi_host_in_recovery(shost))
1346 busy = atomic_inc_return(&shost->host_busy) - 1;
1347 if (shost->host_blocked) {
1352 * unblock after host_blocked iterates to zero
1354 spin_lock_irq(shost->host_lock);
1355 if (--shost->host_blocked != 0) {
1356 spin_unlock_irq(shost->host_lock);
1359 spin_unlock_irq(shost->host_lock);
1362 shost_printk(KERN_INFO, shost,
1363 "unblocking host at zero depth\n"));
1366 if (shost->can_queue > 0 && busy >= shost->can_queue)
1368 if (shost->host_self_blocked)
1371 /* We're OK to process the command, so we can't be starved */
1372 if (!list_empty(&sdev->starved_entry)) {
1373 spin_lock_irq(shost->host_lock);
1374 if (!list_empty(&sdev->starved_entry))
1375 list_del_init(&sdev->starved_entry);
1376 spin_unlock_irq(shost->host_lock);
1382 spin_lock_irq(shost->host_lock);
1383 if (list_empty(&sdev->starved_entry))
1384 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1385 spin_unlock_irq(shost->host_lock);
1387 atomic_dec(&shost->host_busy);
1392 * Busy state exporting function for request stacking drivers.
1394 * For efficiency, no lock is taken to check the busy state of
1395 * shost/starget/sdev, since the returned value is not guaranteed and
1396 * may be changed after request stacking drivers call the function,
1397 * regardless of taking lock or not.
1399 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1400 * needs to return 'not busy'. Otherwise, request stacking drivers
1401 * may hold requests forever.
1403 static int scsi_lld_busy(struct request_queue *q)
1405 struct scsi_device *sdev = q->queuedata;
1406 struct Scsi_Host *shost;
1408 if (blk_queue_dying(q))
1414 * Ignore host/starget busy state.
1415 * Since block layer does not have a concept of fairness across
1416 * multiple queues, congestion of host/starget needs to be handled
1419 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1426 * Kill a request for a dead device
1428 static void scsi_kill_request(struct request *req, struct request_queue *q)
1430 struct scsi_cmnd *cmd = req->special;
1431 struct scsi_device *sdev;
1432 struct scsi_target *starget;
1433 struct Scsi_Host *shost;
1435 blk_start_request(req);
1437 scmd_printk(KERN_INFO, cmd, "killing request\n");
1440 starget = scsi_target(sdev);
1442 scsi_init_cmd_errh(cmd);
1443 cmd->result = DID_NO_CONNECT << 16;
1444 atomic_inc(&cmd->device->iorequest_cnt);
1447 * SCSI request completion path will do scsi_device_unbusy(),
1448 * bump busy counts. To bump the counters, we need to dance
1449 * with the locks as normal issue path does.
1451 sdev->device_busy++;
1452 atomic_inc(&shost->host_busy);
1453 atomic_inc(&starget->target_busy);
1455 blk_complete_request(req);
1458 static void scsi_softirq_done(struct request *rq)
1460 struct scsi_cmnd *cmd = rq->special;
1461 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1464 INIT_LIST_HEAD(&cmd->eh_entry);
1466 atomic_inc(&cmd->device->iodone_cnt);
1468 atomic_inc(&cmd->device->ioerr_cnt);
1470 disposition = scsi_decide_disposition(cmd);
1471 if (disposition != SUCCESS &&
1472 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1473 sdev_printk(KERN_ERR, cmd->device,
1474 "timing out command, waited %lus\n",
1476 disposition = SUCCESS;
1479 scsi_log_completion(cmd, disposition);
1481 switch (disposition) {
1483 scsi_finish_command(cmd);
1486 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1488 case ADD_TO_MLQUEUE:
1489 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1492 if (!scsi_eh_scmd_add(cmd, 0))
1493 scsi_finish_command(cmd);
1498 * scsi_done - Invoke completion on finished SCSI command.
1499 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1500 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1502 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1503 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1504 * calls blk_complete_request() for further processing.
1506 * This function is interrupt context safe.
1508 static void scsi_done(struct scsi_cmnd *cmd)
1510 trace_scsi_dispatch_cmd_done(cmd);
1511 blk_complete_request(cmd->request);
1515 * Function: scsi_request_fn()
1517 * Purpose: Main strategy routine for SCSI.
1519 * Arguments: q - Pointer to actual queue.
1523 * Lock status: IO request lock assumed to be held when called.
1525 static void scsi_request_fn(struct request_queue *q)
1526 __releases(q->queue_lock)
1527 __acquires(q->queue_lock)
1529 struct scsi_device *sdev = q->queuedata;
1530 struct Scsi_Host *shost;
1531 struct scsi_cmnd *cmd;
1532 struct request *req;
1535 * To start with, we keep looping until the queue is empty, or until
1536 * the host is no longer able to accept any more requests.
1542 * get next queueable request. We do this early to make sure
1543 * that the request is fully prepared even if we cannot
1546 req = blk_peek_request(q);
1547 if (!req || !scsi_dev_queue_ready(q, sdev))
1550 if (unlikely(!scsi_device_online(sdev))) {
1551 sdev_printk(KERN_ERR, sdev,
1552 "rejecting I/O to offline device\n");
1553 scsi_kill_request(req, q);
1559 * Remove the request from the request list.
1561 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1562 blk_start_request(req);
1563 sdev->device_busy++;
1565 spin_unlock_irq(q->queue_lock);
1567 if (unlikely(cmd == NULL)) {
1568 printk(KERN_CRIT "impossible request in %s.\n"
1569 "please mail a stack trace to "
1570 "linux-scsi@vger.kernel.org\n",
1572 blk_dump_rq_flags(req, "foo");
1577 * We hit this when the driver is using a host wide
1578 * tag map. For device level tag maps the queue_depth check
1579 * in the device ready fn would prevent us from trying
1580 * to allocate a tag. Since the map is a shared host resource
1581 * we add the dev to the starved list so it eventually gets
1582 * a run when a tag is freed.
1584 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1585 spin_lock_irq(shost->host_lock);
1586 if (list_empty(&sdev->starved_entry))
1587 list_add_tail(&sdev->starved_entry,
1588 &shost->starved_list);
1589 spin_unlock_irq(shost->host_lock);
1593 if (!scsi_target_queue_ready(shost, sdev))
1596 if (!scsi_host_queue_ready(q, shost, sdev))
1597 goto host_not_ready;
1600 * Finally, initialize any error handling parameters, and set up
1601 * the timers for timeouts.
1603 scsi_init_cmd_errh(cmd);
1606 * Dispatch the command to the low-level driver.
1608 cmd->scsi_done = scsi_done;
1609 rtn = scsi_dispatch_cmd(cmd);
1611 scsi_queue_insert(cmd, rtn);
1612 spin_lock_irq(q->queue_lock);
1615 spin_lock_irq(q->queue_lock);
1621 atomic_dec(&scsi_target(sdev)->target_busy);
1624 * lock q, handle tag, requeue req, and decrement device_busy. We
1625 * must return with queue_lock held.
1627 * Decrementing device_busy without checking it is OK, as all such
1628 * cases (host limits or settings) should run the queue at some
1631 spin_lock_irq(q->queue_lock);
1632 blk_requeue_request(q, req);
1633 sdev->device_busy--;
1635 if (sdev->device_busy == 0 && !scsi_device_blocked(sdev))
1636 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1639 static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1641 struct device *host_dev;
1642 u64 bounce_limit = 0xffffffff;
1644 if (shost->unchecked_isa_dma)
1645 return BLK_BOUNCE_ISA;
1647 * Platforms with virtual-DMA translation
1648 * hardware have no practical limit.
1650 if (!PCI_DMA_BUS_IS_PHYS)
1651 return BLK_BOUNCE_ANY;
1653 host_dev = scsi_get_device(shost);
1654 if (host_dev && host_dev->dma_mask)
1655 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1657 return bounce_limit;
1660 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1661 request_fn_proc *request_fn)
1663 struct request_queue *q;
1664 struct device *dev = shost->dma_dev;
1666 q = blk_init_queue(request_fn, NULL);
1671 * this limit is imposed by hardware restrictions
1673 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1674 SCSI_MAX_SG_CHAIN_SEGMENTS));
1676 if (scsi_host_prot_dma(shost)) {
1677 shost->sg_prot_tablesize =
1678 min_not_zero(shost->sg_prot_tablesize,
1679 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1680 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1681 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1684 blk_queue_max_hw_sectors(q, shost->max_sectors);
1685 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1686 blk_queue_segment_boundary(q, shost->dma_boundary);
1687 dma_set_seg_boundary(dev, shost->dma_boundary);
1689 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1691 if (!shost->use_clustering)
1692 q->limits.cluster = 0;
1695 * set a reasonable default alignment on word boundaries: the
1696 * host and device may alter it using
1697 * blk_queue_update_dma_alignment() later.
1699 blk_queue_dma_alignment(q, 0x03);
1703 EXPORT_SYMBOL(__scsi_alloc_queue);
1705 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1707 struct request_queue *q;
1709 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1713 blk_queue_prep_rq(q, scsi_prep_fn);
1714 blk_queue_unprep_rq(q, scsi_unprep_fn);
1715 blk_queue_softirq_done(q, scsi_softirq_done);
1716 blk_queue_rq_timed_out(q, scsi_times_out);
1717 blk_queue_lld_busy(q, scsi_lld_busy);
1722 * Function: scsi_block_requests()
1724 * Purpose: Utility function used by low-level drivers to prevent further
1725 * commands from being queued to the device.
1727 * Arguments: shost - Host in question
1731 * Lock status: No locks are assumed held.
1733 * Notes: There is no timer nor any other means by which the requests
1734 * get unblocked other than the low-level driver calling
1735 * scsi_unblock_requests().
1737 void scsi_block_requests(struct Scsi_Host *shost)
1739 shost->host_self_blocked = 1;
1741 EXPORT_SYMBOL(scsi_block_requests);
1744 * Function: scsi_unblock_requests()
1746 * Purpose: Utility function used by low-level drivers to allow further
1747 * commands from being queued to the device.
1749 * Arguments: shost - Host in question
1753 * Lock status: No locks are assumed held.
1755 * Notes: There is no timer nor any other means by which the requests
1756 * get unblocked other than the low-level driver calling
1757 * scsi_unblock_requests().
1759 * This is done as an API function so that changes to the
1760 * internals of the scsi mid-layer won't require wholesale
1761 * changes to drivers that use this feature.
1763 void scsi_unblock_requests(struct Scsi_Host *shost)
1765 shost->host_self_blocked = 0;
1766 scsi_run_host_queues(shost);
1768 EXPORT_SYMBOL(scsi_unblock_requests);
1770 int __init scsi_init_queue(void)
1774 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1775 sizeof(struct scsi_data_buffer),
1777 if (!scsi_sdb_cache) {
1778 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1782 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1783 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1784 int size = sgp->size * sizeof(struct scatterlist);
1786 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1787 SLAB_HWCACHE_ALIGN, NULL);
1789 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1794 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1797 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1806 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1807 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1809 mempool_destroy(sgp->pool);
1811 kmem_cache_destroy(sgp->slab);
1813 kmem_cache_destroy(scsi_sdb_cache);
1818 void scsi_exit_queue(void)
1822 kmem_cache_destroy(scsi_sdb_cache);
1824 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1825 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1826 mempool_destroy(sgp->pool);
1827 kmem_cache_destroy(sgp->slab);
1832 * scsi_mode_select - issue a mode select
1833 * @sdev: SCSI device to be queried
1834 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1835 * @sp: Save page bit (0 == don't save, 1 == save)
1836 * @modepage: mode page being requested
1837 * @buffer: request buffer (may not be smaller than eight bytes)
1838 * @len: length of request buffer.
1839 * @timeout: command timeout
1840 * @retries: number of retries before failing
1841 * @data: returns a structure abstracting the mode header data
1842 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1843 * must be SCSI_SENSE_BUFFERSIZE big.
1845 * Returns zero if successful; negative error number or scsi
1850 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1851 unsigned char *buffer, int len, int timeout, int retries,
1852 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1854 unsigned char cmd[10];
1855 unsigned char *real_buffer;
1858 memset(cmd, 0, sizeof(cmd));
1859 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1861 if (sdev->use_10_for_ms) {
1864 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1867 memcpy(real_buffer + 8, buffer, len);
1871 real_buffer[2] = data->medium_type;
1872 real_buffer[3] = data->device_specific;
1873 real_buffer[4] = data->longlba ? 0x01 : 0;
1875 real_buffer[6] = data->block_descriptor_length >> 8;
1876 real_buffer[7] = data->block_descriptor_length;
1878 cmd[0] = MODE_SELECT_10;
1882 if (len > 255 || data->block_descriptor_length > 255 ||
1886 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1889 memcpy(real_buffer + 4, buffer, len);
1892 real_buffer[1] = data->medium_type;
1893 real_buffer[2] = data->device_specific;
1894 real_buffer[3] = data->block_descriptor_length;
1897 cmd[0] = MODE_SELECT;
1901 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1902 sshdr, timeout, retries, NULL);
1906 EXPORT_SYMBOL_GPL(scsi_mode_select);
1909 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1910 * @sdev: SCSI device to be queried
1911 * @dbd: set if mode sense will allow block descriptors to be returned
1912 * @modepage: mode page being requested
1913 * @buffer: request buffer (may not be smaller than eight bytes)
1914 * @len: length of request buffer.
1915 * @timeout: command timeout
1916 * @retries: number of retries before failing
1917 * @data: returns a structure abstracting the mode header data
1918 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1919 * must be SCSI_SENSE_BUFFERSIZE big.
1921 * Returns zero if unsuccessful, or the header offset (either 4
1922 * or 8 depending on whether a six or ten byte command was
1923 * issued) if successful.
1926 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1927 unsigned char *buffer, int len, int timeout, int retries,
1928 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1930 unsigned char cmd[12];
1934 struct scsi_sense_hdr my_sshdr;
1936 memset(data, 0, sizeof(*data));
1937 memset(&cmd[0], 0, 12);
1938 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1941 /* caller might not be interested in sense, but we need it */
1946 use_10_for_ms = sdev->use_10_for_ms;
1948 if (use_10_for_ms) {
1952 cmd[0] = MODE_SENSE_10;
1959 cmd[0] = MODE_SENSE;
1964 memset(buffer, 0, len);
1966 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1967 sshdr, timeout, retries, NULL);
1969 /* This code looks awful: what it's doing is making sure an
1970 * ILLEGAL REQUEST sense return identifies the actual command
1971 * byte as the problem. MODE_SENSE commands can return
1972 * ILLEGAL REQUEST if the code page isn't supported */
1974 if (use_10_for_ms && !scsi_status_is_good(result) &&
1975 (driver_byte(result) & DRIVER_SENSE)) {
1976 if (scsi_sense_valid(sshdr)) {
1977 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1978 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1980 * Invalid command operation code
1982 sdev->use_10_for_ms = 0;
1988 if(scsi_status_is_good(result)) {
1989 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1990 (modepage == 6 || modepage == 8))) {
1991 /* Initio breakage? */
1994 data->medium_type = 0;
1995 data->device_specific = 0;
1997 data->block_descriptor_length = 0;
1998 } else if(use_10_for_ms) {
1999 data->length = buffer[0]*256 + buffer[1] + 2;
2000 data->medium_type = buffer[2];
2001 data->device_specific = buffer[3];
2002 data->longlba = buffer[4] & 0x01;
2003 data->block_descriptor_length = buffer[6]*256
2006 data->length = buffer[0] + 1;
2007 data->medium_type = buffer[1];
2008 data->device_specific = buffer[2];
2009 data->block_descriptor_length = buffer[3];
2011 data->header_length = header_length;
2016 EXPORT_SYMBOL(scsi_mode_sense);
2019 * scsi_test_unit_ready - test if unit is ready
2020 * @sdev: scsi device to change the state of.
2021 * @timeout: command timeout
2022 * @retries: number of retries before failing
2023 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2024 * returning sense. Make sure that this is cleared before passing
2027 * Returns zero if unsuccessful or an error if TUR failed. For
2028 * removable media, UNIT_ATTENTION sets ->changed flag.
2031 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2032 struct scsi_sense_hdr *sshdr_external)
2035 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2037 struct scsi_sense_hdr *sshdr;
2040 if (!sshdr_external)
2041 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2043 sshdr = sshdr_external;
2045 /* try to eat the UNIT_ATTENTION if there are enough retries */
2047 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2048 timeout, retries, NULL);
2049 if (sdev->removable && scsi_sense_valid(sshdr) &&
2050 sshdr->sense_key == UNIT_ATTENTION)
2052 } while (scsi_sense_valid(sshdr) &&
2053 sshdr->sense_key == UNIT_ATTENTION && --retries);
2055 if (!sshdr_external)
2059 EXPORT_SYMBOL(scsi_test_unit_ready);
2062 * scsi_device_set_state - Take the given device through the device state model.
2063 * @sdev: scsi device to change the state of.
2064 * @state: state to change to.
2066 * Returns zero if unsuccessful or an error if the requested
2067 * transition is illegal.
2070 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2072 enum scsi_device_state oldstate = sdev->sdev_state;
2074 if (state == oldstate)
2080 case SDEV_CREATED_BLOCK:
2091 case SDEV_TRANSPORT_OFFLINE:
2104 case SDEV_TRANSPORT_OFFLINE:
2112 case SDEV_TRANSPORT_OFFLINE:
2127 case SDEV_CREATED_BLOCK:
2134 case SDEV_CREATED_BLOCK:
2149 case SDEV_TRANSPORT_OFFLINE:
2162 case SDEV_TRANSPORT_OFFLINE:
2164 case SDEV_CREATED_BLOCK:
2172 sdev->sdev_state = state;
2176 SCSI_LOG_ERROR_RECOVERY(1,
2177 sdev_printk(KERN_ERR, sdev,
2178 "Illegal state transition %s->%s",
2179 scsi_device_state_name(oldstate),
2180 scsi_device_state_name(state))
2184 EXPORT_SYMBOL(scsi_device_set_state);
2187 * sdev_evt_emit - emit a single SCSI device uevent
2188 * @sdev: associated SCSI device
2189 * @evt: event to emit
2191 * Send a single uevent (scsi_event) to the associated scsi_device.
2193 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2198 switch (evt->evt_type) {
2199 case SDEV_EVT_MEDIA_CHANGE:
2200 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2202 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2203 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2205 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2206 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2208 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2209 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2211 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2212 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2214 case SDEV_EVT_LUN_CHANGE_REPORTED:
2215 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2224 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2228 * sdev_evt_thread - send a uevent for each scsi event
2229 * @work: work struct for scsi_device
2231 * Dispatch queued events to their associated scsi_device kobjects
2234 void scsi_evt_thread(struct work_struct *work)
2236 struct scsi_device *sdev;
2237 enum scsi_device_event evt_type;
2238 LIST_HEAD(event_list);
2240 sdev = container_of(work, struct scsi_device, event_work);
2242 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2243 if (test_and_clear_bit(evt_type, sdev->pending_events))
2244 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2247 struct scsi_event *evt;
2248 struct list_head *this, *tmp;
2249 unsigned long flags;
2251 spin_lock_irqsave(&sdev->list_lock, flags);
2252 list_splice_init(&sdev->event_list, &event_list);
2253 spin_unlock_irqrestore(&sdev->list_lock, flags);
2255 if (list_empty(&event_list))
2258 list_for_each_safe(this, tmp, &event_list) {
2259 evt = list_entry(this, struct scsi_event, node);
2260 list_del(&evt->node);
2261 scsi_evt_emit(sdev, evt);
2268 * sdev_evt_send - send asserted event to uevent thread
2269 * @sdev: scsi_device event occurred on
2270 * @evt: event to send
2272 * Assert scsi device event asynchronously.
2274 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2276 unsigned long flags;
2279 /* FIXME: currently this check eliminates all media change events
2280 * for polled devices. Need to update to discriminate between AN
2281 * and polled events */
2282 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2288 spin_lock_irqsave(&sdev->list_lock, flags);
2289 list_add_tail(&evt->node, &sdev->event_list);
2290 schedule_work(&sdev->event_work);
2291 spin_unlock_irqrestore(&sdev->list_lock, flags);
2293 EXPORT_SYMBOL_GPL(sdev_evt_send);
2296 * sdev_evt_alloc - allocate a new scsi event
2297 * @evt_type: type of event to allocate
2298 * @gfpflags: GFP flags for allocation
2300 * Allocates and returns a new scsi_event.
2302 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2305 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2309 evt->evt_type = evt_type;
2310 INIT_LIST_HEAD(&evt->node);
2312 /* evt_type-specific initialization, if any */
2314 case SDEV_EVT_MEDIA_CHANGE:
2315 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2316 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2317 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2318 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2319 case SDEV_EVT_LUN_CHANGE_REPORTED:
2327 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2330 * sdev_evt_send_simple - send asserted event to uevent thread
2331 * @sdev: scsi_device event occurred on
2332 * @evt_type: type of event to send
2333 * @gfpflags: GFP flags for allocation
2335 * Assert scsi device event asynchronously, given an event type.
2337 void sdev_evt_send_simple(struct scsi_device *sdev,
2338 enum scsi_device_event evt_type, gfp_t gfpflags)
2340 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2342 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2347 sdev_evt_send(sdev, evt);
2349 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2352 * scsi_device_quiesce - Block user issued commands.
2353 * @sdev: scsi device to quiesce.
2355 * This works by trying to transition to the SDEV_QUIESCE state
2356 * (which must be a legal transition). When the device is in this
2357 * state, only special requests will be accepted, all others will
2358 * be deferred. Since special requests may also be requeued requests,
2359 * a successful return doesn't guarantee the device will be
2360 * totally quiescent.
2362 * Must be called with user context, may sleep.
2364 * Returns zero if unsuccessful or an error if not.
2367 scsi_device_quiesce(struct scsi_device *sdev)
2369 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2373 scsi_run_queue(sdev->request_queue);
2374 while (sdev->device_busy) {
2375 msleep_interruptible(200);
2376 scsi_run_queue(sdev->request_queue);
2380 EXPORT_SYMBOL(scsi_device_quiesce);
2383 * scsi_device_resume - Restart user issued commands to a quiesced device.
2384 * @sdev: scsi device to resume.
2386 * Moves the device from quiesced back to running and restarts the
2389 * Must be called with user context, may sleep.
2391 void scsi_device_resume(struct scsi_device *sdev)
2393 /* check if the device state was mutated prior to resume, and if
2394 * so assume the state is being managed elsewhere (for example
2395 * device deleted during suspend)
2397 if (sdev->sdev_state != SDEV_QUIESCE ||
2398 scsi_device_set_state(sdev, SDEV_RUNNING))
2400 scsi_run_queue(sdev->request_queue);
2402 EXPORT_SYMBOL(scsi_device_resume);
2405 device_quiesce_fn(struct scsi_device *sdev, void *data)
2407 scsi_device_quiesce(sdev);
2411 scsi_target_quiesce(struct scsi_target *starget)
2413 starget_for_each_device(starget, NULL, device_quiesce_fn);
2415 EXPORT_SYMBOL(scsi_target_quiesce);
2418 device_resume_fn(struct scsi_device *sdev, void *data)
2420 scsi_device_resume(sdev);
2424 scsi_target_resume(struct scsi_target *starget)
2426 starget_for_each_device(starget, NULL, device_resume_fn);
2428 EXPORT_SYMBOL(scsi_target_resume);
2431 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2432 * @sdev: device to block
2434 * Block request made by scsi lld's to temporarily stop all
2435 * scsi commands on the specified device. Called from interrupt
2436 * or normal process context.
2438 * Returns zero if successful or error if not
2441 * This routine transitions the device to the SDEV_BLOCK state
2442 * (which must be a legal transition). When the device is in this
2443 * state, all commands are deferred until the scsi lld reenables
2444 * the device with scsi_device_unblock or device_block_tmo fires.
2447 scsi_internal_device_block(struct scsi_device *sdev)
2449 struct request_queue *q = sdev->request_queue;
2450 unsigned long flags;
2453 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2455 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2462 * The device has transitioned to SDEV_BLOCK. Stop the
2463 * block layer from calling the midlayer with this device's
2466 spin_lock_irqsave(q->queue_lock, flags);
2468 spin_unlock_irqrestore(q->queue_lock, flags);
2472 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2475 * scsi_internal_device_unblock - resume a device after a block request
2476 * @sdev: device to resume
2477 * @new_state: state to set devices to after unblocking
2479 * Called by scsi lld's or the midlayer to restart the device queue
2480 * for the previously suspended scsi device. Called from interrupt or
2481 * normal process context.
2483 * Returns zero if successful or error if not.
2486 * This routine transitions the device to the SDEV_RUNNING state
2487 * or to one of the offline states (which must be a legal transition)
2488 * allowing the midlayer to goose the queue for this device.
2491 scsi_internal_device_unblock(struct scsi_device *sdev,
2492 enum scsi_device_state new_state)
2494 struct request_queue *q = sdev->request_queue;
2495 unsigned long flags;
2498 * Try to transition the scsi device to SDEV_RUNNING or one of the
2499 * offlined states and goose the device queue if successful.
2501 if ((sdev->sdev_state == SDEV_BLOCK) ||
2502 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2503 sdev->sdev_state = new_state;
2504 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2505 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2506 new_state == SDEV_OFFLINE)
2507 sdev->sdev_state = new_state;
2509 sdev->sdev_state = SDEV_CREATED;
2510 } else if (sdev->sdev_state != SDEV_CANCEL &&
2511 sdev->sdev_state != SDEV_OFFLINE)
2514 spin_lock_irqsave(q->queue_lock, flags);
2516 spin_unlock_irqrestore(q->queue_lock, flags);
2520 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2523 device_block(struct scsi_device *sdev, void *data)
2525 scsi_internal_device_block(sdev);
2529 target_block(struct device *dev, void *data)
2531 if (scsi_is_target_device(dev))
2532 starget_for_each_device(to_scsi_target(dev), NULL,
2538 scsi_target_block(struct device *dev)
2540 if (scsi_is_target_device(dev))
2541 starget_for_each_device(to_scsi_target(dev), NULL,
2544 device_for_each_child(dev, NULL, target_block);
2546 EXPORT_SYMBOL_GPL(scsi_target_block);
2549 device_unblock(struct scsi_device *sdev, void *data)
2551 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2555 target_unblock(struct device *dev, void *data)
2557 if (scsi_is_target_device(dev))
2558 starget_for_each_device(to_scsi_target(dev), data,
2564 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2566 if (scsi_is_target_device(dev))
2567 starget_for_each_device(to_scsi_target(dev), &new_state,
2570 device_for_each_child(dev, &new_state, target_unblock);
2572 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2575 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2576 * @sgl: scatter-gather list
2577 * @sg_count: number of segments in sg
2578 * @offset: offset in bytes into sg, on return offset into the mapped area
2579 * @len: bytes to map, on return number of bytes mapped
2581 * Returns virtual address of the start of the mapped page
2583 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2584 size_t *offset, size_t *len)
2587 size_t sg_len = 0, len_complete = 0;
2588 struct scatterlist *sg;
2591 WARN_ON(!irqs_disabled());
2593 for_each_sg(sgl, sg, sg_count, i) {
2594 len_complete = sg_len; /* Complete sg-entries */
2595 sg_len += sg->length;
2596 if (sg_len > *offset)
2600 if (unlikely(i == sg_count)) {
2601 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2603 __func__, sg_len, *offset, sg_count);
2608 /* Offset starting from the beginning of first page in this sg-entry */
2609 *offset = *offset - len_complete + sg->offset;
2611 /* Assumption: contiguous pages can be accessed as "page + i" */
2612 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2613 *offset &= ~PAGE_MASK;
2615 /* Bytes in this sg-entry from *offset to the end of the page */
2616 sg_len = PAGE_SIZE - *offset;
2620 return kmap_atomic(page);
2622 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2625 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2626 * @virt: virtual address to be unmapped
2628 void scsi_kunmap_atomic_sg(void *virt)
2630 kunmap_atomic(virt);
2632 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2634 void sdev_disable_disk_events(struct scsi_device *sdev)
2636 atomic_inc(&sdev->disk_events_disable_depth);
2638 EXPORT_SYMBOL(sdev_disable_disk_events);
2640 void sdev_enable_disk_events(struct scsi_device *sdev)
2642 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2644 atomic_dec(&sdev->disk_events_disable_depth);
2646 EXPORT_SYMBOL(sdev_enable_disk_events);