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(q, &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);
199 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
200 buffer, bufflen, __GFP_WAIT))
203 req->cmd_len = COMMAND_SIZE(cmd[0]);
204 memcpy(req->cmd, cmd, req->cmd_len);
207 req->retries = retries;
208 req->timeout = timeout;
209 req->cmd_type = REQ_TYPE_BLOCK_PC;
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 void __scsi_release_buffers(struct scsi_cmnd *, int);
518 * Function: scsi_end_request()
520 * Purpose: Post-processing of completed commands (usually invoked at end
521 * of upper level post-processing and scsi_io_completion).
523 * Arguments: cmd - command that is complete.
524 * error - 0 if I/O indicates success, < 0 for I/O error.
525 * bytes - number of bytes of completed I/O
526 * requeue - indicates whether we should requeue leftovers.
528 * Lock status: Assumed that lock is not held upon entry.
530 * Returns: cmd if requeue required, NULL otherwise.
532 * Notes: This is called for block device requests in order to
533 * mark some number of sectors as complete.
535 * We are guaranteeing that the request queue will be goosed
536 * at some point during this call.
537 * Notes: If cmd was requeued, upon return it will be a stale pointer.
539 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
540 int bytes, int requeue)
542 struct request_queue *q = cmd->device->request_queue;
543 struct request *req = cmd->request;
546 * If there are blocks left over at the end, set up the command
547 * to queue the remainder of them.
549 if (blk_end_request(req, error, bytes)) {
550 /* kill remainder if no retrys */
551 if (error && scsi_noretry_cmd(cmd))
552 blk_end_request_all(req, error);
556 * Bleah. Leftovers again. Stick the
557 * leftovers in the front of the
558 * queue, and goose the queue again.
560 scsi_release_buffers(cmd);
561 scsi_requeue_command(q, cmd);
569 * This will goose the queue request function at the end, so we don't
570 * need to worry about launching another command.
572 __scsi_release_buffers(cmd, 0);
573 scsi_next_command(cmd);
577 static inline unsigned int scsi_sgtable_index(unsigned short nents)
581 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
586 index = get_count_order(nents) - 3;
591 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
593 struct scsi_host_sg_pool *sgp;
595 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
596 mempool_free(sgl, sgp->pool);
599 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
601 struct scsi_host_sg_pool *sgp;
603 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
604 return mempool_alloc(sgp->pool, gfp_mask);
607 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
614 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
615 gfp_mask, scsi_sg_alloc);
617 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
623 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
625 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
628 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
631 if (cmd->sdb.table.nents)
632 scsi_free_sgtable(&cmd->sdb);
634 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
636 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
637 struct scsi_data_buffer *bidi_sdb =
638 cmd->request->next_rq->special;
639 scsi_free_sgtable(bidi_sdb);
640 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
641 cmd->request->next_rq->special = NULL;
644 if (scsi_prot_sg_count(cmd))
645 scsi_free_sgtable(cmd->prot_sdb);
649 * Function: scsi_release_buffers()
651 * Purpose: Completion processing for block device I/O requests.
653 * Arguments: cmd - command that we are bailing.
655 * Lock status: Assumed that no lock is held upon entry.
659 * Notes: In the event that an upper level driver rejects a
660 * command, we must release resources allocated during
661 * the __init_io() function. Primarily this would involve
662 * the scatter-gather table, and potentially any bounce
665 void scsi_release_buffers(struct scsi_cmnd *cmd)
667 __scsi_release_buffers(cmd, 1);
669 EXPORT_SYMBOL(scsi_release_buffers);
672 * __scsi_error_from_host_byte - translate SCSI error code into errno
673 * @cmd: SCSI command (unused)
674 * @result: scsi error code
676 * Translate SCSI error code into standard UNIX errno.
678 * -ENOLINK temporary transport failure
679 * -EREMOTEIO permanent target failure, do not retry
680 * -EBADE permanent nexus failure, retry on other path
681 * -ENOSPC No write space available
682 * -ENODATA Medium error
683 * -EIO unspecified I/O error
685 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
689 switch(host_byte(result)) {
690 case DID_TRANSPORT_FAILFAST:
693 case DID_TARGET_FAILURE:
694 set_host_byte(cmd, DID_OK);
697 case DID_NEXUS_FAILURE:
698 set_host_byte(cmd, DID_OK);
701 case DID_ALLOC_FAILURE:
702 set_host_byte(cmd, DID_OK);
705 case DID_MEDIUM_ERROR:
706 set_host_byte(cmd, DID_OK);
718 * Function: scsi_io_completion()
720 * Purpose: Completion processing for block device I/O requests.
722 * Arguments: cmd - command that is finished.
724 * Lock status: Assumed that no lock is held upon entry.
728 * Notes: This function is matched in terms of capabilities to
729 * the function that created the scatter-gather list.
730 * In other words, if there are no bounce buffers
731 * (the normal case for most drivers), we don't need
732 * the logic to deal with cleaning up afterwards.
734 * We must call scsi_end_request(). This will finish off
735 * the specified number of sectors. If we are done, the
736 * command block will be released and the queue function
737 * will be goosed. If we are not done then we have to
738 * figure out what to do next:
740 * a) We can call scsi_requeue_command(). The request
741 * will be unprepared and put back on the queue. Then
742 * a new command will be created for it. This should
743 * be used if we made forward progress, or if we want
744 * to switch from READ(10) to READ(6) for example.
746 * b) We can call scsi_queue_insert(). The request will
747 * be put back on the queue and retried using the same
748 * command as before, possibly after a delay.
750 * c) We can call blk_end_request() with -EIO to fail
751 * the remainder of the request.
753 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
755 int result = cmd->result;
756 struct request_queue *q = cmd->device->request_queue;
757 struct request *req = cmd->request;
759 struct scsi_sense_hdr sshdr;
761 int sense_deferred = 0;
762 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
763 ACTION_DELAYED_RETRY} action;
764 char *description = NULL;
765 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
768 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
770 sense_deferred = scsi_sense_is_deferred(&sshdr);
773 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
775 if (sense_valid && req->sense) {
777 * SG_IO wants current and deferred errors
779 int len = 8 + cmd->sense_buffer[7];
781 if (len > SCSI_SENSE_BUFFERSIZE)
782 len = SCSI_SENSE_BUFFERSIZE;
783 memcpy(req->sense, cmd->sense_buffer, len);
784 req->sense_len = len;
787 error = __scsi_error_from_host_byte(cmd, result);
790 * __scsi_error_from_host_byte may have reset the host_byte
792 req->errors = cmd->result;
794 req->resid_len = scsi_get_resid(cmd);
796 if (scsi_bidi_cmnd(cmd)) {
798 * Bidi commands Must be complete as a whole,
799 * both sides at once.
801 req->next_rq->resid_len = scsi_in(cmd)->resid;
803 scsi_release_buffers(cmd);
804 blk_end_request_all(req, 0);
806 scsi_next_command(cmd);
811 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
812 BUG_ON(blk_bidi_rq(req));
815 * Next deal with any sectors which we were able to correctly
818 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
820 blk_rq_sectors(req), good_bytes));
823 * Recovered errors need reporting, but they're always treated
824 * as success, so fiddle the result code here. For BLOCK_PC
825 * we already took a copy of the original into rq->errors which
826 * is what gets returned to the user
828 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
829 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
830 * print since caller wants ATA registers. Only occurs on
831 * SCSI ATA PASS_THROUGH commands when CK_COND=1
833 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
835 else if (!(req->cmd_flags & REQ_QUIET))
836 scsi_print_sense("", cmd);
838 /* BLOCK_PC may have set error */
843 * A number of bytes were successfully read. If there
844 * are leftovers and there is some kind of error
845 * (result != 0), retry the rest.
847 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
850 error = __scsi_error_from_host_byte(cmd, result);
852 if (host_byte(result) == DID_RESET) {
853 /* Third party bus reset or reset for error recovery
854 * reasons. Just retry the command and see what
857 action = ACTION_RETRY;
858 } else if (sense_valid && !sense_deferred) {
859 switch (sshdr.sense_key) {
861 if (cmd->device->removable) {
862 /* Detected disc change. Set a bit
863 * and quietly refuse further access.
865 cmd->device->changed = 1;
866 description = "Media Changed";
867 action = ACTION_FAIL;
869 /* Must have been a power glitch, or a
870 * bus reset. Could not have been a
871 * media change, so we just retry the
872 * command and see what happens.
874 action = ACTION_RETRY;
877 case ILLEGAL_REQUEST:
878 /* If we had an ILLEGAL REQUEST returned, then
879 * we may have performed an unsupported
880 * command. The only thing this should be
881 * would be a ten byte read where only a six
882 * byte read was supported. Also, on a system
883 * where READ CAPACITY failed, we may have
884 * read past the end of the disk.
886 if ((cmd->device->use_10_for_rw &&
887 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
888 (cmd->cmnd[0] == READ_10 ||
889 cmd->cmnd[0] == WRITE_10)) {
890 /* This will issue a new 6-byte command. */
891 cmd->device->use_10_for_rw = 0;
892 action = ACTION_REPREP;
893 } else if (sshdr.asc == 0x10) /* DIX */ {
894 description = "Host Data Integrity Failure";
895 action = ACTION_FAIL;
897 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
898 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
899 switch (cmd->cmnd[0]) {
901 description = "Discard failure";
905 if (cmd->cmnd[1] & 0x8)
906 description = "Discard failure";
909 "Write same failure";
912 description = "Invalid command failure";
915 action = ACTION_FAIL;
918 action = ACTION_FAIL;
920 case ABORTED_COMMAND:
921 action = ACTION_FAIL;
922 if (sshdr.asc == 0x10) { /* DIF */
923 description = "Target Data Integrity Failure";
928 /* If the device is in the process of becoming
929 * ready, or has a temporary blockage, retry.
931 if (sshdr.asc == 0x04) {
932 switch (sshdr.ascq) {
933 case 0x01: /* becoming ready */
934 case 0x04: /* format in progress */
935 case 0x05: /* rebuild in progress */
936 case 0x06: /* recalculation in progress */
937 case 0x07: /* operation in progress */
938 case 0x08: /* Long write in progress */
939 case 0x09: /* self test in progress */
940 case 0x14: /* space allocation in progress */
941 action = ACTION_DELAYED_RETRY;
944 description = "Device not ready";
945 action = ACTION_FAIL;
949 description = "Device not ready";
950 action = ACTION_FAIL;
953 case VOLUME_OVERFLOW:
954 /* See SSC3rXX or current. */
955 action = ACTION_FAIL;
958 description = "Unhandled sense code";
959 action = ACTION_FAIL;
963 description = "Unhandled error code";
964 action = ACTION_FAIL;
967 if (action != ACTION_FAIL &&
968 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
969 action = ACTION_FAIL;
970 description = "Command timed out";
975 /* Give up and fail the remainder of the request */
976 scsi_release_buffers(cmd);
977 if (!(req->cmd_flags & REQ_QUIET)) {
979 scmd_printk(KERN_INFO, cmd, "%s\n",
981 scsi_print_result(cmd);
982 if (driver_byte(result) & DRIVER_SENSE)
983 scsi_print_sense("", cmd);
984 scsi_print_command(cmd);
986 if (blk_end_request_err(req, error))
987 scsi_requeue_command(q, cmd);
989 scsi_next_command(cmd);
992 /* Unprep the request and put it back at the head of the queue.
993 * A new command will be prepared and issued.
995 scsi_release_buffers(cmd);
996 scsi_requeue_command(q, cmd);
999 /* Retry the same command immediately */
1000 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1002 case ACTION_DELAYED_RETRY:
1003 /* Retry the same command after a delay */
1004 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1009 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1015 * If sg table allocation fails, requeue request later.
1017 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1019 return BLKPREP_DEFER;
1025 * Next, walk the list, and fill in the addresses and sizes of
1028 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1029 BUG_ON(count > sdb->table.nents);
1030 sdb->table.nents = count;
1031 sdb->length = blk_rq_bytes(req);
1036 * Function: scsi_init_io()
1038 * Purpose: SCSI I/O initialize function.
1040 * Arguments: cmd - Command descriptor we wish to initialize
1042 * Returns: 0 on success
1043 * BLKPREP_DEFER if the failure is retryable
1044 * BLKPREP_KILL if the failure is fatal
1046 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1048 struct scsi_device *sdev = cmd->device;
1049 struct request *rq = cmd->request;
1051 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1055 if (blk_bidi_rq(rq)) {
1056 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1057 scsi_sdb_cache, GFP_ATOMIC);
1059 error = BLKPREP_DEFER;
1063 rq->next_rq->special = bidi_sdb;
1064 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1069 if (blk_integrity_rq(rq)) {
1070 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1073 BUG_ON(prot_sdb == NULL);
1074 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1076 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1077 error = BLKPREP_DEFER;
1081 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1082 prot_sdb->table.sgl);
1083 BUG_ON(unlikely(count > ivecs));
1084 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1086 cmd->prot_sdb = prot_sdb;
1087 cmd->prot_sdb->table.nents = count;
1093 scsi_release_buffers(cmd);
1094 cmd->request->special = NULL;
1095 scsi_put_command(cmd);
1096 put_device(&sdev->sdev_gendev);
1099 EXPORT_SYMBOL(scsi_init_io);
1101 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1102 struct request *req)
1104 struct scsi_cmnd *cmd;
1106 if (!req->special) {
1107 /* Bail if we can't get a reference to the device */
1108 if (!get_device(&sdev->sdev_gendev))
1111 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1112 if (unlikely(!cmd)) {
1113 put_device(&sdev->sdev_gendev);
1121 /* pull a tag out of the request if we have one */
1122 cmd->tag = req->tag;
1125 cmd->cmnd = req->cmd;
1126 cmd->prot_op = SCSI_PROT_NORMAL;
1131 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1133 struct scsi_cmnd *cmd;
1134 int ret = scsi_prep_state_check(sdev, req);
1136 if (ret != BLKPREP_OK)
1139 cmd = scsi_get_cmd_from_req(sdev, req);
1141 return BLKPREP_DEFER;
1144 * BLOCK_PC requests may transfer data, in which case they must
1145 * a bio attached to them. Or they might contain a SCSI command
1146 * that does not transfer data, in which case they may optionally
1147 * submit a request without an attached bio.
1152 BUG_ON(!req->nr_phys_segments);
1154 ret = scsi_init_io(cmd, GFP_ATOMIC);
1158 BUG_ON(blk_rq_bytes(req));
1160 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1164 cmd->cmd_len = req->cmd_len;
1165 if (!blk_rq_bytes(req))
1166 cmd->sc_data_direction = DMA_NONE;
1167 else if (rq_data_dir(req) == WRITE)
1168 cmd->sc_data_direction = DMA_TO_DEVICE;
1170 cmd->sc_data_direction = DMA_FROM_DEVICE;
1172 cmd->transfersize = blk_rq_bytes(req);
1173 cmd->allowed = req->retries;
1176 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1179 * Setup a REQ_TYPE_FS command. These are simple read/write request
1180 * from filesystems that still need to be translated to SCSI CDBs from
1183 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1185 struct scsi_cmnd *cmd;
1186 int ret = scsi_prep_state_check(sdev, req);
1188 if (ret != BLKPREP_OK)
1191 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1192 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1193 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1194 if (ret != BLKPREP_OK)
1199 * Filesystem requests must transfer data.
1201 BUG_ON(!req->nr_phys_segments);
1203 cmd = scsi_get_cmd_from_req(sdev, req);
1205 return BLKPREP_DEFER;
1207 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1208 return scsi_init_io(cmd, GFP_ATOMIC);
1210 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1212 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1214 int ret = BLKPREP_OK;
1217 * If the device is not in running state we will reject some
1220 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1221 switch (sdev->sdev_state) {
1223 case SDEV_TRANSPORT_OFFLINE:
1225 * If the device is offline we refuse to process any
1226 * commands. The device must be brought online
1227 * before trying any recovery commands.
1229 sdev_printk(KERN_ERR, sdev,
1230 "rejecting I/O to offline device\n");
1235 * If the device is fully deleted, we refuse to
1236 * process any commands as well.
1238 sdev_printk(KERN_ERR, sdev,
1239 "rejecting I/O to dead device\n");
1244 case SDEV_CREATED_BLOCK:
1246 * If the devices is blocked we defer normal commands.
1248 if (!(req->cmd_flags & REQ_PREEMPT))
1249 ret = BLKPREP_DEFER;
1253 * For any other not fully online state we only allow
1254 * special commands. In particular any user initiated
1255 * command is not allowed.
1257 if (!(req->cmd_flags & REQ_PREEMPT))
1264 EXPORT_SYMBOL(scsi_prep_state_check);
1266 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1268 struct scsi_device *sdev = q->queuedata;
1272 req->errors = DID_NO_CONNECT << 16;
1273 /* release the command and kill it */
1275 struct scsi_cmnd *cmd = req->special;
1276 scsi_release_buffers(cmd);
1277 scsi_put_command(cmd);
1278 put_device(&sdev->sdev_gendev);
1279 req->special = NULL;
1284 * If we defer, the blk_peek_request() returns NULL, but the
1285 * queue must be restarted, so we schedule a callback to happen
1288 if (sdev->device_busy == 0)
1289 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1292 req->cmd_flags |= REQ_DONTPREP;
1297 EXPORT_SYMBOL(scsi_prep_return);
1299 int scsi_prep_fn(struct request_queue *q, struct request *req)
1301 struct scsi_device *sdev = q->queuedata;
1302 int ret = BLKPREP_KILL;
1304 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1305 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1306 return scsi_prep_return(q, req, ret);
1308 EXPORT_SYMBOL(scsi_prep_fn);
1311 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1314 * Called with the queue_lock held.
1316 static inline int scsi_dev_queue_ready(struct request_queue *q,
1317 struct scsi_device *sdev)
1319 if (sdev->device_busy == 0 && sdev->device_blocked) {
1321 * unblock after device_blocked iterates to zero
1323 if (--sdev->device_blocked == 0) {
1325 sdev_printk(KERN_INFO, sdev,
1326 "unblocking device at zero depth\n"));
1328 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1332 if (scsi_device_is_busy(sdev))
1340 * scsi_target_queue_ready: checks if there we can send commands to target
1341 * @sdev: scsi device on starget to check.
1343 * Called with the host lock held.
1345 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1346 struct scsi_device *sdev)
1348 struct scsi_target *starget = scsi_target(sdev);
1350 if (starget->single_lun) {
1351 if (starget->starget_sdev_user &&
1352 starget->starget_sdev_user != sdev)
1354 starget->starget_sdev_user = sdev;
1357 if (starget->target_busy == 0 && starget->target_blocked) {
1359 * unblock after target_blocked iterates to zero
1361 if (--starget->target_blocked == 0) {
1362 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1363 "unblocking target at zero depth\n"));
1368 if (scsi_target_is_busy(starget)) {
1369 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1377 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1378 * return 0. We must end up running the queue again whenever 0 is
1379 * returned, else IO can hang.
1381 * Called with host_lock held.
1383 static inline int scsi_host_queue_ready(struct request_queue *q,
1384 struct Scsi_Host *shost,
1385 struct scsi_device *sdev)
1387 if (scsi_host_in_recovery(shost))
1389 if (shost->host_busy == 0 && shost->host_blocked) {
1391 * unblock after host_blocked iterates to zero
1393 if (--shost->host_blocked == 0) {
1395 printk("scsi%d unblocking host at zero depth\n",
1401 if (scsi_host_is_busy(shost)) {
1402 if (list_empty(&sdev->starved_entry))
1403 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1407 /* We're OK to process the command, so we can't be starved */
1408 if (!list_empty(&sdev->starved_entry))
1409 list_del_init(&sdev->starved_entry);
1415 * Busy state exporting function for request stacking drivers.
1417 * For efficiency, no lock is taken to check the busy state of
1418 * shost/starget/sdev, since the returned value is not guaranteed and
1419 * may be changed after request stacking drivers call the function,
1420 * regardless of taking lock or not.
1422 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1423 * needs to return 'not busy'. Otherwise, request stacking drivers
1424 * may hold requests forever.
1426 static int scsi_lld_busy(struct request_queue *q)
1428 struct scsi_device *sdev = q->queuedata;
1429 struct Scsi_Host *shost;
1431 if (blk_queue_dying(q))
1437 * Ignore host/starget busy state.
1438 * Since block layer does not have a concept of fairness across
1439 * multiple queues, congestion of host/starget needs to be handled
1442 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1449 * Kill a request for a dead device
1451 static void scsi_kill_request(struct request *req, struct request_queue *q)
1453 struct scsi_cmnd *cmd = req->special;
1454 struct scsi_device *sdev;
1455 struct scsi_target *starget;
1456 struct Scsi_Host *shost;
1458 blk_start_request(req);
1460 scmd_printk(KERN_INFO, cmd, "killing request\n");
1463 starget = scsi_target(sdev);
1465 scsi_init_cmd_errh(cmd);
1466 cmd->result = DID_NO_CONNECT << 16;
1467 atomic_inc(&cmd->device->iorequest_cnt);
1470 * SCSI request completion path will do scsi_device_unbusy(),
1471 * bump busy counts. To bump the counters, we need to dance
1472 * with the locks as normal issue path does.
1474 sdev->device_busy++;
1475 spin_unlock(sdev->request_queue->queue_lock);
1476 spin_lock(shost->host_lock);
1478 starget->target_busy++;
1479 spin_unlock(shost->host_lock);
1480 spin_lock(sdev->request_queue->queue_lock);
1482 blk_complete_request(req);
1485 static void scsi_softirq_done(struct request *rq)
1487 struct scsi_cmnd *cmd = rq->special;
1488 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1491 INIT_LIST_HEAD(&cmd->eh_entry);
1493 atomic_inc(&cmd->device->iodone_cnt);
1495 atomic_inc(&cmd->device->ioerr_cnt);
1497 disposition = scsi_decide_disposition(cmd);
1498 if (disposition != SUCCESS &&
1499 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1500 sdev_printk(KERN_ERR, cmd->device,
1501 "timing out command, waited %lus\n",
1503 disposition = SUCCESS;
1506 scsi_log_completion(cmd, disposition);
1508 switch (disposition) {
1510 scsi_finish_command(cmd);
1513 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1515 case ADD_TO_MLQUEUE:
1516 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1519 if (!scsi_eh_scmd_add(cmd, 0))
1520 scsi_finish_command(cmd);
1525 * Function: scsi_request_fn()
1527 * Purpose: Main strategy routine for SCSI.
1529 * Arguments: q - Pointer to actual queue.
1533 * Lock status: IO request lock assumed to be held when called.
1535 static void scsi_request_fn(struct request_queue *q)
1536 __releases(q->queue_lock)
1537 __acquires(q->queue_lock)
1539 struct scsi_device *sdev = q->queuedata;
1540 struct Scsi_Host *shost;
1541 struct scsi_cmnd *cmd;
1542 struct request *req;
1545 * To start with, we keep looping until the queue is empty, or until
1546 * the host is no longer able to accept any more requests.
1552 * get next queueable request. We do this early to make sure
1553 * that the request is fully prepared even if we cannot
1556 req = blk_peek_request(q);
1557 if (!req || !scsi_dev_queue_ready(q, sdev))
1560 if (unlikely(!scsi_device_online(sdev))) {
1561 sdev_printk(KERN_ERR, sdev,
1562 "rejecting I/O to offline device\n");
1563 scsi_kill_request(req, q);
1569 * Remove the request from the request list.
1571 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1572 blk_start_request(req);
1573 sdev->device_busy++;
1575 spin_unlock(q->queue_lock);
1577 if (unlikely(cmd == NULL)) {
1578 printk(KERN_CRIT "impossible request in %s.\n"
1579 "please mail a stack trace to "
1580 "linux-scsi@vger.kernel.org\n",
1582 blk_dump_rq_flags(req, "foo");
1585 spin_lock(shost->host_lock);
1588 * We hit this when the driver is using a host wide
1589 * tag map. For device level tag maps the queue_depth check
1590 * in the device ready fn would prevent us from trying
1591 * to allocate a tag. Since the map is a shared host resource
1592 * we add the dev to the starved list so it eventually gets
1593 * a run when a tag is freed.
1595 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1596 if (list_empty(&sdev->starved_entry))
1597 list_add_tail(&sdev->starved_entry,
1598 &shost->starved_list);
1602 if (!scsi_target_queue_ready(shost, sdev))
1605 if (!scsi_host_queue_ready(q, shost, sdev))
1608 scsi_target(sdev)->target_busy++;
1612 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1613 * take the lock again.
1615 spin_unlock_irq(shost->host_lock);
1618 * Finally, initialize any error handling parameters, and set up
1619 * the timers for timeouts.
1621 scsi_init_cmd_errh(cmd);
1624 * Dispatch the command to the low-level driver.
1626 rtn = scsi_dispatch_cmd(cmd);
1627 spin_lock_irq(q->queue_lock);
1635 spin_unlock_irq(shost->host_lock);
1638 * lock q, handle tag, requeue req, and decrement device_busy. We
1639 * must return with queue_lock held.
1641 * Decrementing device_busy without checking it is OK, as all such
1642 * cases (host limits or settings) should run the queue at some
1645 spin_lock_irq(q->queue_lock);
1646 blk_requeue_request(q, req);
1647 sdev->device_busy--;
1649 if (sdev->device_busy == 0)
1650 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1653 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1655 struct device *host_dev;
1656 u64 bounce_limit = 0xffffffff;
1658 if (shost->unchecked_isa_dma)
1659 return BLK_BOUNCE_ISA;
1661 * Platforms with virtual-DMA translation
1662 * hardware have no practical limit.
1664 if (!PCI_DMA_BUS_IS_PHYS)
1665 return BLK_BOUNCE_ANY;
1667 host_dev = scsi_get_device(shost);
1668 if (host_dev && host_dev->dma_mask)
1669 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1671 return bounce_limit;
1673 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1675 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1676 request_fn_proc *request_fn)
1678 struct request_queue *q;
1679 struct device *dev = shost->dma_dev;
1681 q = blk_init_queue(request_fn, NULL);
1686 * this limit is imposed by hardware restrictions
1688 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1689 SCSI_MAX_SG_CHAIN_SEGMENTS));
1691 if (scsi_host_prot_dma(shost)) {
1692 shost->sg_prot_tablesize =
1693 min_not_zero(shost->sg_prot_tablesize,
1694 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1695 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1696 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1699 blk_queue_max_hw_sectors(q, shost->max_sectors);
1700 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1701 blk_queue_segment_boundary(q, shost->dma_boundary);
1702 dma_set_seg_boundary(dev, shost->dma_boundary);
1704 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1706 if (!shost->use_clustering)
1707 q->limits.cluster = 0;
1710 * set a reasonable default alignment on word boundaries: the
1711 * host and device may alter it using
1712 * blk_queue_update_dma_alignment() later.
1714 blk_queue_dma_alignment(q, 0x03);
1718 EXPORT_SYMBOL(__scsi_alloc_queue);
1720 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1722 struct request_queue *q;
1724 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1728 blk_queue_prep_rq(q, scsi_prep_fn);
1729 blk_queue_softirq_done(q, scsi_softirq_done);
1730 blk_queue_rq_timed_out(q, scsi_times_out);
1731 blk_queue_lld_busy(q, scsi_lld_busy);
1736 * Function: scsi_block_requests()
1738 * Purpose: Utility function used by low-level drivers to prevent further
1739 * commands from being queued to the device.
1741 * Arguments: shost - Host in question
1745 * Lock status: No locks are assumed held.
1747 * Notes: There is no timer nor any other means by which the requests
1748 * get unblocked other than the low-level driver calling
1749 * scsi_unblock_requests().
1751 void scsi_block_requests(struct Scsi_Host *shost)
1753 shost->host_self_blocked = 1;
1755 EXPORT_SYMBOL(scsi_block_requests);
1758 * Function: scsi_unblock_requests()
1760 * Purpose: Utility function used by low-level drivers to allow further
1761 * commands from being queued to the device.
1763 * Arguments: shost - Host in question
1767 * Lock status: No locks are assumed held.
1769 * Notes: There is no timer nor any other means by which the requests
1770 * get unblocked other than the low-level driver calling
1771 * scsi_unblock_requests().
1773 * This is done as an API function so that changes to the
1774 * internals of the scsi mid-layer won't require wholesale
1775 * changes to drivers that use this feature.
1777 void scsi_unblock_requests(struct Scsi_Host *shost)
1779 shost->host_self_blocked = 0;
1780 scsi_run_host_queues(shost);
1782 EXPORT_SYMBOL(scsi_unblock_requests);
1784 int __init scsi_init_queue(void)
1788 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1789 sizeof(struct scsi_data_buffer),
1791 if (!scsi_sdb_cache) {
1792 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1796 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1797 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1798 int size = sgp->size * sizeof(struct scatterlist);
1800 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1801 SLAB_HWCACHE_ALIGN, NULL);
1803 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1808 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1811 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1820 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1821 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1823 mempool_destroy(sgp->pool);
1825 kmem_cache_destroy(sgp->slab);
1827 kmem_cache_destroy(scsi_sdb_cache);
1832 void scsi_exit_queue(void)
1836 kmem_cache_destroy(scsi_sdb_cache);
1838 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1839 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1840 mempool_destroy(sgp->pool);
1841 kmem_cache_destroy(sgp->slab);
1846 * scsi_mode_select - issue a mode select
1847 * @sdev: SCSI device to be queried
1848 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1849 * @sp: Save page bit (0 == don't save, 1 == save)
1850 * @modepage: mode page being requested
1851 * @buffer: request buffer (may not be smaller than eight bytes)
1852 * @len: length of request buffer.
1853 * @timeout: command timeout
1854 * @retries: number of retries before failing
1855 * @data: returns a structure abstracting the mode header data
1856 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1857 * must be SCSI_SENSE_BUFFERSIZE big.
1859 * Returns zero if successful; negative error number or scsi
1864 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1865 unsigned char *buffer, int len, int timeout, int retries,
1866 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1868 unsigned char cmd[10];
1869 unsigned char *real_buffer;
1872 memset(cmd, 0, sizeof(cmd));
1873 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1875 if (sdev->use_10_for_ms) {
1878 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1881 memcpy(real_buffer + 8, buffer, len);
1885 real_buffer[2] = data->medium_type;
1886 real_buffer[3] = data->device_specific;
1887 real_buffer[4] = data->longlba ? 0x01 : 0;
1889 real_buffer[6] = data->block_descriptor_length >> 8;
1890 real_buffer[7] = data->block_descriptor_length;
1892 cmd[0] = MODE_SELECT_10;
1896 if (len > 255 || data->block_descriptor_length > 255 ||
1900 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1903 memcpy(real_buffer + 4, buffer, len);
1906 real_buffer[1] = data->medium_type;
1907 real_buffer[2] = data->device_specific;
1908 real_buffer[3] = data->block_descriptor_length;
1911 cmd[0] = MODE_SELECT;
1915 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1916 sshdr, timeout, retries, NULL);
1920 EXPORT_SYMBOL_GPL(scsi_mode_select);
1923 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1924 * @sdev: SCSI device to be queried
1925 * @dbd: set if mode sense will allow block descriptors to be returned
1926 * @modepage: mode page being requested
1927 * @buffer: request buffer (may not be smaller than eight bytes)
1928 * @len: length of request buffer.
1929 * @timeout: command timeout
1930 * @retries: number of retries before failing
1931 * @data: returns a structure abstracting the mode header data
1932 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1933 * must be SCSI_SENSE_BUFFERSIZE big.
1935 * Returns zero if unsuccessful, or the header offset (either 4
1936 * or 8 depending on whether a six or ten byte command was
1937 * issued) if successful.
1940 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1941 unsigned char *buffer, int len, int timeout, int retries,
1942 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1944 unsigned char cmd[12];
1948 struct scsi_sense_hdr my_sshdr;
1950 memset(data, 0, sizeof(*data));
1951 memset(&cmd[0], 0, 12);
1952 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1955 /* caller might not be interested in sense, but we need it */
1960 use_10_for_ms = sdev->use_10_for_ms;
1962 if (use_10_for_ms) {
1966 cmd[0] = MODE_SENSE_10;
1973 cmd[0] = MODE_SENSE;
1978 memset(buffer, 0, len);
1980 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1981 sshdr, timeout, retries, NULL);
1983 /* This code looks awful: what it's doing is making sure an
1984 * ILLEGAL REQUEST sense return identifies the actual command
1985 * byte as the problem. MODE_SENSE commands can return
1986 * ILLEGAL REQUEST if the code page isn't supported */
1988 if (use_10_for_ms && !scsi_status_is_good(result) &&
1989 (driver_byte(result) & DRIVER_SENSE)) {
1990 if (scsi_sense_valid(sshdr)) {
1991 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1992 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1994 * Invalid command operation code
1996 sdev->use_10_for_ms = 0;
2002 if(scsi_status_is_good(result)) {
2003 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2004 (modepage == 6 || modepage == 8))) {
2005 /* Initio breakage? */
2008 data->medium_type = 0;
2009 data->device_specific = 0;
2011 data->block_descriptor_length = 0;
2012 } else if(use_10_for_ms) {
2013 data->length = buffer[0]*256 + buffer[1] + 2;
2014 data->medium_type = buffer[2];
2015 data->device_specific = buffer[3];
2016 data->longlba = buffer[4] & 0x01;
2017 data->block_descriptor_length = buffer[6]*256
2020 data->length = buffer[0] + 1;
2021 data->medium_type = buffer[1];
2022 data->device_specific = buffer[2];
2023 data->block_descriptor_length = buffer[3];
2025 data->header_length = header_length;
2030 EXPORT_SYMBOL(scsi_mode_sense);
2033 * scsi_test_unit_ready - test if unit is ready
2034 * @sdev: scsi device to change the state of.
2035 * @timeout: command timeout
2036 * @retries: number of retries before failing
2037 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2038 * returning sense. Make sure that this is cleared before passing
2041 * Returns zero if unsuccessful or an error if TUR failed. For
2042 * removable media, UNIT_ATTENTION sets ->changed flag.
2045 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2046 struct scsi_sense_hdr *sshdr_external)
2049 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2051 struct scsi_sense_hdr *sshdr;
2054 if (!sshdr_external)
2055 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2057 sshdr = sshdr_external;
2059 /* try to eat the UNIT_ATTENTION if there are enough retries */
2061 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2062 timeout, retries, NULL);
2063 if (sdev->removable && scsi_sense_valid(sshdr) &&
2064 sshdr->sense_key == UNIT_ATTENTION)
2066 } while (scsi_sense_valid(sshdr) &&
2067 sshdr->sense_key == UNIT_ATTENTION && --retries);
2069 if (!sshdr_external)
2073 EXPORT_SYMBOL(scsi_test_unit_ready);
2076 * scsi_device_set_state - Take the given device through the device state model.
2077 * @sdev: scsi device to change the state of.
2078 * @state: state to change to.
2080 * Returns zero if unsuccessful or an error if the requested
2081 * transition is illegal.
2084 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2086 enum scsi_device_state oldstate = sdev->sdev_state;
2088 if (state == oldstate)
2094 case SDEV_CREATED_BLOCK:
2105 case SDEV_TRANSPORT_OFFLINE:
2118 case SDEV_TRANSPORT_OFFLINE:
2126 case SDEV_TRANSPORT_OFFLINE:
2141 case SDEV_CREATED_BLOCK:
2148 case SDEV_CREATED_BLOCK:
2163 case SDEV_TRANSPORT_OFFLINE:
2176 case SDEV_TRANSPORT_OFFLINE:
2178 case SDEV_CREATED_BLOCK:
2186 sdev->sdev_state = state;
2190 SCSI_LOG_ERROR_RECOVERY(1,
2191 sdev_printk(KERN_ERR, sdev,
2192 "Illegal state transition %s->%s\n",
2193 scsi_device_state_name(oldstate),
2194 scsi_device_state_name(state))
2198 EXPORT_SYMBOL(scsi_device_set_state);
2201 * sdev_evt_emit - emit a single SCSI device uevent
2202 * @sdev: associated SCSI device
2203 * @evt: event to emit
2205 * Send a single uevent (scsi_event) to the associated scsi_device.
2207 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2212 switch (evt->evt_type) {
2213 case SDEV_EVT_MEDIA_CHANGE:
2214 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2216 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2217 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2219 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2220 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2222 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2223 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2225 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2226 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2228 case SDEV_EVT_LUN_CHANGE_REPORTED:
2229 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2238 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2242 * sdev_evt_thread - send a uevent for each scsi event
2243 * @work: work struct for scsi_device
2245 * Dispatch queued events to their associated scsi_device kobjects
2248 void scsi_evt_thread(struct work_struct *work)
2250 struct scsi_device *sdev;
2251 enum scsi_device_event evt_type;
2252 LIST_HEAD(event_list);
2254 sdev = container_of(work, struct scsi_device, event_work);
2256 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2257 if (test_and_clear_bit(evt_type, sdev->pending_events))
2258 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2261 struct scsi_event *evt;
2262 struct list_head *this, *tmp;
2263 unsigned long flags;
2265 spin_lock_irqsave(&sdev->list_lock, flags);
2266 list_splice_init(&sdev->event_list, &event_list);
2267 spin_unlock_irqrestore(&sdev->list_lock, flags);
2269 if (list_empty(&event_list))
2272 list_for_each_safe(this, tmp, &event_list) {
2273 evt = list_entry(this, struct scsi_event, node);
2274 list_del(&evt->node);
2275 scsi_evt_emit(sdev, evt);
2282 * sdev_evt_send - send asserted event to uevent thread
2283 * @sdev: scsi_device event occurred on
2284 * @evt: event to send
2286 * Assert scsi device event asynchronously.
2288 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2290 unsigned long flags;
2293 /* FIXME: currently this check eliminates all media change events
2294 * for polled devices. Need to update to discriminate between AN
2295 * and polled events */
2296 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2302 spin_lock_irqsave(&sdev->list_lock, flags);
2303 list_add_tail(&evt->node, &sdev->event_list);
2304 schedule_work(&sdev->event_work);
2305 spin_unlock_irqrestore(&sdev->list_lock, flags);
2307 EXPORT_SYMBOL_GPL(sdev_evt_send);
2310 * sdev_evt_alloc - allocate a new scsi event
2311 * @evt_type: type of event to allocate
2312 * @gfpflags: GFP flags for allocation
2314 * Allocates and returns a new scsi_event.
2316 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2319 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2323 evt->evt_type = evt_type;
2324 INIT_LIST_HEAD(&evt->node);
2326 /* evt_type-specific initialization, if any */
2328 case SDEV_EVT_MEDIA_CHANGE:
2329 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2330 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2331 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2332 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2333 case SDEV_EVT_LUN_CHANGE_REPORTED:
2341 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2344 * sdev_evt_send_simple - send asserted event to uevent thread
2345 * @sdev: scsi_device event occurred on
2346 * @evt_type: type of event to send
2347 * @gfpflags: GFP flags for allocation
2349 * Assert scsi device event asynchronously, given an event type.
2351 void sdev_evt_send_simple(struct scsi_device *sdev,
2352 enum scsi_device_event evt_type, gfp_t gfpflags)
2354 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2356 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2361 sdev_evt_send(sdev, evt);
2363 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2366 * scsi_device_quiesce - Block user issued commands.
2367 * @sdev: scsi device to quiesce.
2369 * This works by trying to transition to the SDEV_QUIESCE state
2370 * (which must be a legal transition). When the device is in this
2371 * state, only special requests will be accepted, all others will
2372 * be deferred. Since special requests may also be requeued requests,
2373 * a successful return doesn't guarantee the device will be
2374 * totally quiescent.
2376 * Must be called with user context, may sleep.
2378 * Returns zero if unsuccessful or an error if not.
2381 scsi_device_quiesce(struct scsi_device *sdev)
2383 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2387 scsi_run_queue(sdev->request_queue);
2388 while (sdev->device_busy) {
2389 msleep_interruptible(200);
2390 scsi_run_queue(sdev->request_queue);
2394 EXPORT_SYMBOL(scsi_device_quiesce);
2397 * scsi_device_resume - Restart user issued commands to a quiesced device.
2398 * @sdev: scsi device to resume.
2400 * Moves the device from quiesced back to running and restarts the
2403 * Must be called with user context, may sleep.
2405 void scsi_device_resume(struct scsi_device *sdev)
2407 /* check if the device state was mutated prior to resume, and if
2408 * so assume the state is being managed elsewhere (for example
2409 * device deleted during suspend)
2411 if (sdev->sdev_state != SDEV_QUIESCE ||
2412 scsi_device_set_state(sdev, SDEV_RUNNING))
2414 scsi_run_queue(sdev->request_queue);
2416 EXPORT_SYMBOL(scsi_device_resume);
2419 device_quiesce_fn(struct scsi_device *sdev, void *data)
2421 scsi_device_quiesce(sdev);
2425 scsi_target_quiesce(struct scsi_target *starget)
2427 starget_for_each_device(starget, NULL, device_quiesce_fn);
2429 EXPORT_SYMBOL(scsi_target_quiesce);
2432 device_resume_fn(struct scsi_device *sdev, void *data)
2434 scsi_device_resume(sdev);
2438 scsi_target_resume(struct scsi_target *starget)
2440 starget_for_each_device(starget, NULL, device_resume_fn);
2442 EXPORT_SYMBOL(scsi_target_resume);
2445 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2446 * @sdev: device to block
2448 * Block request made by scsi lld's to temporarily stop all
2449 * scsi commands on the specified device. Called from interrupt
2450 * or normal process context.
2452 * Returns zero if successful or error if not
2455 * This routine transitions the device to the SDEV_BLOCK state
2456 * (which must be a legal transition). When the device is in this
2457 * state, all commands are deferred until the scsi lld reenables
2458 * the device with scsi_device_unblock or device_block_tmo fires.
2461 scsi_internal_device_block(struct scsi_device *sdev)
2463 struct request_queue *q = sdev->request_queue;
2464 unsigned long flags;
2467 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2469 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2476 * The device has transitioned to SDEV_BLOCK. Stop the
2477 * block layer from calling the midlayer with this device's
2480 spin_lock_irqsave(q->queue_lock, flags);
2482 spin_unlock_irqrestore(q->queue_lock, flags);
2486 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2489 * scsi_internal_device_unblock - resume a device after a block request
2490 * @sdev: device to resume
2491 * @new_state: state to set devices to after unblocking
2493 * Called by scsi lld's or the midlayer to restart the device queue
2494 * for the previously suspended scsi device. Called from interrupt or
2495 * normal process context.
2497 * Returns zero if successful or error if not.
2500 * This routine transitions the device to the SDEV_RUNNING state
2501 * or to one of the offline states (which must be a legal transition)
2502 * allowing the midlayer to goose the queue for this device.
2505 scsi_internal_device_unblock(struct scsi_device *sdev,
2506 enum scsi_device_state new_state)
2508 struct request_queue *q = sdev->request_queue;
2509 unsigned long flags;
2512 * Try to transition the scsi device to SDEV_RUNNING or one of the
2513 * offlined states and goose the device queue if successful.
2515 if ((sdev->sdev_state == SDEV_BLOCK) ||
2516 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2517 sdev->sdev_state = new_state;
2518 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2519 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2520 new_state == SDEV_OFFLINE)
2521 sdev->sdev_state = new_state;
2523 sdev->sdev_state = SDEV_CREATED;
2524 } else if (sdev->sdev_state != SDEV_CANCEL &&
2525 sdev->sdev_state != SDEV_OFFLINE)
2528 spin_lock_irqsave(q->queue_lock, flags);
2530 spin_unlock_irqrestore(q->queue_lock, flags);
2534 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2537 device_block(struct scsi_device *sdev, void *data)
2539 scsi_internal_device_block(sdev);
2543 target_block(struct device *dev, void *data)
2545 if (scsi_is_target_device(dev))
2546 starget_for_each_device(to_scsi_target(dev), NULL,
2552 scsi_target_block(struct device *dev)
2554 if (scsi_is_target_device(dev))
2555 starget_for_each_device(to_scsi_target(dev), NULL,
2558 device_for_each_child(dev, NULL, target_block);
2560 EXPORT_SYMBOL_GPL(scsi_target_block);
2563 device_unblock(struct scsi_device *sdev, void *data)
2565 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2569 target_unblock(struct device *dev, void *data)
2571 if (scsi_is_target_device(dev))
2572 starget_for_each_device(to_scsi_target(dev), data,
2578 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2580 if (scsi_is_target_device(dev))
2581 starget_for_each_device(to_scsi_target(dev), &new_state,
2584 device_for_each_child(dev, &new_state, target_unblock);
2586 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2589 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2590 * @sgl: scatter-gather list
2591 * @sg_count: number of segments in sg
2592 * @offset: offset in bytes into sg, on return offset into the mapped area
2593 * @len: bytes to map, on return number of bytes mapped
2595 * Returns virtual address of the start of the mapped page
2597 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2598 size_t *offset, size_t *len)
2601 size_t sg_len = 0, len_complete = 0;
2602 struct scatterlist *sg;
2605 WARN_ON(!irqs_disabled());
2607 for_each_sg(sgl, sg, sg_count, i) {
2608 len_complete = sg_len; /* Complete sg-entries */
2609 sg_len += sg->length;
2610 if (sg_len > *offset)
2614 if (unlikely(i == sg_count)) {
2615 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2617 __func__, sg_len, *offset, sg_count);
2622 /* Offset starting from the beginning of first page in this sg-entry */
2623 *offset = *offset - len_complete + sg->offset;
2625 /* Assumption: contiguous pages can be accessed as "page + i" */
2626 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2627 *offset &= ~PAGE_MASK;
2629 /* Bytes in this sg-entry from *offset to the end of the page */
2630 sg_len = PAGE_SIZE - *offset;
2634 return kmap_atomic(page);
2636 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2639 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2640 * @virt: virtual address to be unmapped
2642 void scsi_kunmap_atomic_sg(void *virt)
2644 kunmap_atomic(virt);
2646 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2648 void sdev_disable_disk_events(struct scsi_device *sdev)
2650 atomic_inc(&sdev->disk_events_disable_depth);
2652 EXPORT_SYMBOL(sdev_disable_disk_events);
2654 void sdev_enable_disk_events(struct scsi_device *sdev)
2656 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2658 atomic_dec(&sdev->disk_events_disable_depth);
2660 EXPORT_SYMBOL(sdev_enable_disk_events);