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 * Function: scsi_unprep_request()
81 * Purpose: Remove all preparation done for a request, including its
82 * associated scsi_cmnd, so that it can be requeued.
84 * Arguments: req - request to unprepare
86 * Lock status: Assumed that no locks are held upon entry.
90 static void scsi_unprep_request(struct request *req)
92 struct scsi_cmnd *cmd = req->special;
94 blk_unprep_request(req);
97 scsi_put_command(cmd);
101 * __scsi_queue_insert - private queue insertion
102 * @cmd: The SCSI command being requeued
103 * @reason: The reason for the requeue
104 * @unbusy: Whether the queue should be unbusied
106 * This is a private queue insertion. The public interface
107 * scsi_queue_insert() always assumes the queue should be unbusied
108 * because it's always called before the completion. This function is
109 * for a requeue after completion, which should only occur in this
112 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
114 struct Scsi_Host *host = cmd->device->host;
115 struct scsi_device *device = cmd->device;
116 struct scsi_target *starget = scsi_target(device);
117 struct request_queue *q = device->request_queue;
121 printk("Inserting command %p into mlqueue\n", cmd));
124 * Set the appropriate busy bit for the device/host.
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
137 case SCSI_MLQUEUE_HOST_BUSY:
138 host->host_blocked = host->max_host_blocked;
140 case SCSI_MLQUEUE_DEVICE_BUSY:
141 case SCSI_MLQUEUE_EH_RETRY:
142 device->device_blocked = device->max_device_blocked;
144 case SCSI_MLQUEUE_TARGET_BUSY:
145 starget->target_blocked = starget->max_target_blocked;
150 * Decrement the counters, since these commands are no longer
151 * active on the host/device.
154 scsi_device_unbusy(device);
157 * Requeue this command. It will go before all other commands
158 * that are already in the queue.
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
164 kblockd_schedule_work(q, &device->requeue_work);
168 * Function: scsi_queue_insert()
170 * Purpose: Insert a command in the midlevel queue.
172 * Arguments: cmd - command that we are adding to queue.
173 * reason - why we are inserting command to queue.
175 * Lock status: Assumed that lock is not held upon entry.
179 * Notes: We do this for one of two cases. Either the host is busy
180 * and it cannot accept any more commands for the time being,
181 * or the device returned QUEUE_FULL and can accept no more
183 * Notes: This could be called either from an interrupt context or a
184 * normal process context.
186 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
188 __scsi_queue_insert(cmd, reason, 1);
191 * scsi_execute - insert request and wait for the result
194 * @data_direction: data direction
195 * @buffer: data buffer
196 * @bufflen: len of buffer
197 * @sense: optional sense buffer
198 * @timeout: request timeout in seconds
199 * @retries: number of times to retry request
200 * @flags: or into request flags;
201 * @resid: optional residual length
203 * returns the req->errors value which is the scsi_cmnd result
206 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
207 int data_direction, void *buffer, unsigned bufflen,
208 unsigned char *sense, int timeout, int retries, int flags,
212 int write = (data_direction == DMA_TO_DEVICE);
213 int ret = DRIVER_ERROR << 24;
215 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
219 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
220 buffer, bufflen, __GFP_WAIT))
223 req->cmd_len = COMMAND_SIZE(cmd[0]);
224 memcpy(req->cmd, cmd, req->cmd_len);
227 req->retries = retries;
228 req->timeout = timeout;
229 req->cmd_type = REQ_TYPE_BLOCK_PC;
230 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
233 * head injection *required* here otherwise quiesce won't work
235 blk_execute_rq(req->q, NULL, req, 1);
238 * Some devices (USB mass-storage in particular) may transfer
239 * garbage data together with a residue indicating that the data
240 * is invalid. Prevent the garbage from being misinterpreted
241 * and prevent security leaks by zeroing out the excess data.
243 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
244 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
247 *resid = req->resid_len;
250 blk_put_request(req);
254 EXPORT_SYMBOL(scsi_execute);
257 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
258 int data_direction, void *buffer, unsigned bufflen,
259 struct scsi_sense_hdr *sshdr, int timeout, int retries,
266 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
268 return DRIVER_ERROR << 24;
270 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
271 sense, timeout, retries, 0, resid);
273 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
278 EXPORT_SYMBOL(scsi_execute_req);
281 * Function: scsi_init_cmd_errh()
283 * Purpose: Initialize cmd fields related to error handling.
285 * Arguments: cmd - command that is ready to be queued.
287 * Notes: This function has the job of initializing a number of
288 * fields related to error handling. Typically this will
289 * be called once for each command, as required.
291 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
293 cmd->serial_number = 0;
294 scsi_set_resid(cmd, 0);
295 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
296 if (cmd->cmd_len == 0)
297 cmd->cmd_len = scsi_command_size(cmd->cmnd);
300 void scsi_device_unbusy(struct scsi_device *sdev)
302 struct Scsi_Host *shost = sdev->host;
303 struct scsi_target *starget = scsi_target(sdev);
306 spin_lock_irqsave(shost->host_lock, flags);
308 starget->target_busy--;
309 if (unlikely(scsi_host_in_recovery(shost) &&
310 (shost->host_failed || shost->host_eh_scheduled)))
311 scsi_eh_wakeup(shost);
312 spin_unlock(shost->host_lock);
313 spin_lock(sdev->request_queue->queue_lock);
315 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
319 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
320 * and call blk_run_queue for all the scsi_devices on the target -
321 * including current_sdev first.
323 * Called with *no* scsi locks held.
325 static void scsi_single_lun_run(struct scsi_device *current_sdev)
327 struct Scsi_Host *shost = current_sdev->host;
328 struct scsi_device *sdev, *tmp;
329 struct scsi_target *starget = scsi_target(current_sdev);
332 spin_lock_irqsave(shost->host_lock, flags);
333 starget->starget_sdev_user = NULL;
334 spin_unlock_irqrestore(shost->host_lock, flags);
337 * Call blk_run_queue for all LUNs on the target, starting with
338 * current_sdev. We race with others (to set starget_sdev_user),
339 * but in most cases, we will be first. Ideally, each LU on the
340 * target would get some limited time or requests on the target.
342 blk_run_queue(current_sdev->request_queue);
344 spin_lock_irqsave(shost->host_lock, flags);
345 if (starget->starget_sdev_user)
347 list_for_each_entry_safe(sdev, tmp, &starget->devices,
348 same_target_siblings) {
349 if (sdev == current_sdev)
351 if (scsi_device_get(sdev))
354 spin_unlock_irqrestore(shost->host_lock, flags);
355 blk_run_queue(sdev->request_queue);
356 spin_lock_irqsave(shost->host_lock, flags);
358 scsi_device_put(sdev);
361 spin_unlock_irqrestore(shost->host_lock, flags);
364 static inline int scsi_device_is_busy(struct scsi_device *sdev)
366 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
372 static inline int scsi_target_is_busy(struct scsi_target *starget)
374 return ((starget->can_queue > 0 &&
375 starget->target_busy >= starget->can_queue) ||
376 starget->target_blocked);
379 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
381 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
382 shost->host_blocked || shost->host_self_blocked)
389 * Function: scsi_run_queue()
391 * Purpose: Select a proper request queue to serve next
393 * Arguments: q - last request's queue
397 * Notes: The previous command was completely finished, start
398 * a new one if possible.
400 static void scsi_run_queue(struct request_queue *q)
402 struct scsi_device *sdev = q->queuedata;
403 struct Scsi_Host *shost;
404 LIST_HEAD(starved_list);
408 if (scsi_target(sdev)->single_lun)
409 scsi_single_lun_run(sdev);
411 spin_lock_irqsave(shost->host_lock, flags);
412 list_splice_init(&shost->starved_list, &starved_list);
414 while (!list_empty(&starved_list)) {
416 * As long as shost is accepting commands and we have
417 * starved queues, call blk_run_queue. scsi_request_fn
418 * drops the queue_lock and can add us back to the
421 * host_lock protects the starved_list and starved_entry.
422 * scsi_request_fn must get the host_lock before checking
423 * or modifying starved_list or starved_entry.
425 if (scsi_host_is_busy(shost))
428 sdev = list_entry(starved_list.next,
429 struct scsi_device, starved_entry);
430 list_del_init(&sdev->starved_entry);
431 if (scsi_target_is_busy(scsi_target(sdev))) {
432 list_move_tail(&sdev->starved_entry,
433 &shost->starved_list);
437 spin_unlock(shost->host_lock);
438 spin_lock(sdev->request_queue->queue_lock);
439 __blk_run_queue(sdev->request_queue);
440 spin_unlock(sdev->request_queue->queue_lock);
441 spin_lock(shost->host_lock);
443 /* put any unprocessed entries back */
444 list_splice(&starved_list, &shost->starved_list);
445 spin_unlock_irqrestore(shost->host_lock, flags);
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;
485 * We need to hold a reference on the device to avoid the queue being
486 * killed after the unlock and before scsi_run_queue is invoked which
487 * may happen because scsi_unprep_request() puts the command which
488 * releases its reference on the device.
490 get_device(&sdev->sdev_gendev);
492 spin_lock_irqsave(q->queue_lock, flags);
493 scsi_unprep_request(req);
494 blk_requeue_request(q, req);
495 spin_unlock_irqrestore(q->queue_lock, flags);
499 put_device(&sdev->sdev_gendev);
502 void scsi_next_command(struct scsi_cmnd *cmd)
504 struct scsi_device *sdev = cmd->device;
505 struct request_queue *q = sdev->request_queue;
507 /* need to hold a reference on the device before we let go of the cmd */
508 get_device(&sdev->sdev_gendev);
510 scsi_put_command(cmd);
513 /* ok to remove device now */
514 put_device(&sdev->sdev_gendev);
517 void scsi_run_host_queues(struct Scsi_Host *shost)
519 struct scsi_device *sdev;
521 shost_for_each_device(sdev, shost)
522 scsi_run_queue(sdev->request_queue);
525 static void __scsi_release_buffers(struct scsi_cmnd *, int);
528 * Function: scsi_end_request()
530 * Purpose: Post-processing of completed commands (usually invoked at end
531 * of upper level post-processing and scsi_io_completion).
533 * Arguments: cmd - command that is complete.
534 * error - 0 if I/O indicates success, < 0 for I/O error.
535 * bytes - number of bytes of completed I/O
536 * requeue - indicates whether we should requeue leftovers.
538 * Lock status: Assumed that lock is not held upon entry.
540 * Returns: cmd if requeue required, NULL otherwise.
542 * Notes: This is called for block device requests in order to
543 * mark some number of sectors as complete.
545 * We are guaranteeing that the request queue will be goosed
546 * at some point during this call.
547 * Notes: If cmd was requeued, upon return it will be a stale pointer.
549 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
550 int bytes, int requeue)
552 struct request_queue *q = cmd->device->request_queue;
553 struct request *req = cmd->request;
556 * If there are blocks left over at the end, set up the command
557 * to queue the remainder of them.
559 if (blk_end_request(req, error, bytes)) {
560 /* kill remainder if no retrys */
561 if (error && scsi_noretry_cmd(cmd))
562 blk_end_request_all(req, error);
566 * Bleah. Leftovers again. Stick the
567 * leftovers in the front of the
568 * queue, and goose the queue again.
570 scsi_release_buffers(cmd);
571 scsi_requeue_command(q, cmd);
579 * This will goose the queue request function at the end, so we don't
580 * need to worry about launching another command.
582 __scsi_release_buffers(cmd, 0);
583 scsi_next_command(cmd);
587 static inline unsigned int scsi_sgtable_index(unsigned short nents)
591 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
596 index = get_count_order(nents) - 3;
601 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
603 struct scsi_host_sg_pool *sgp;
605 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
606 mempool_free(sgl, sgp->pool);
609 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
611 struct scsi_host_sg_pool *sgp;
613 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
614 return mempool_alloc(sgp->pool, gfp_mask);
617 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
624 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
625 gfp_mask, scsi_sg_alloc);
627 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
633 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
635 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
638 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
641 if (cmd->sdb.table.nents)
642 scsi_free_sgtable(&cmd->sdb);
644 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
646 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
647 struct scsi_data_buffer *bidi_sdb =
648 cmd->request->next_rq->special;
649 scsi_free_sgtable(bidi_sdb);
650 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
651 cmd->request->next_rq->special = NULL;
654 if (scsi_prot_sg_count(cmd))
655 scsi_free_sgtable(cmd->prot_sdb);
659 * Function: scsi_release_buffers()
661 * Purpose: Completion processing for block device I/O requests.
663 * Arguments: cmd - command that we are bailing.
665 * Lock status: Assumed that no lock is held upon entry.
669 * Notes: In the event that an upper level driver rejects a
670 * command, we must release resources allocated during
671 * the __init_io() function. Primarily this would involve
672 * the scatter-gather table, and potentially any bounce
675 void scsi_release_buffers(struct scsi_cmnd *cmd)
677 __scsi_release_buffers(cmd, 1);
679 EXPORT_SYMBOL(scsi_release_buffers);
681 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
685 switch(host_byte(result)) {
686 case DID_TRANSPORT_FAILFAST:
689 case DID_TARGET_FAILURE:
690 set_host_byte(cmd, DID_OK);
693 case DID_NEXUS_FAILURE:
694 set_host_byte(cmd, DID_OK);
706 * Function: scsi_io_completion()
708 * Purpose: Completion processing for block device I/O requests.
710 * Arguments: cmd - command that is finished.
712 * Lock status: Assumed that no lock is held upon entry.
716 * Notes: This function is matched in terms of capabilities to
717 * the function that created the scatter-gather list.
718 * In other words, if there are no bounce buffers
719 * (the normal case for most drivers), we don't need
720 * the logic to deal with cleaning up afterwards.
722 * We must call scsi_end_request(). This will finish off
723 * the specified number of sectors. If we are done, the
724 * command block will be released and the queue function
725 * will be goosed. If we are not done then we have to
726 * figure out what to do next:
728 * a) We can call scsi_requeue_command(). The request
729 * will be unprepared and put back on the queue. Then
730 * a new command will be created for it. This should
731 * be used if we made forward progress, or if we want
732 * to switch from READ(10) to READ(6) for example.
734 * b) We can call scsi_queue_insert(). The request will
735 * be put back on the queue and retried using the same
736 * command as before, possibly after a delay.
738 * c) We can call blk_end_request() with -EIO to fail
739 * the remainder of the request.
741 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
743 int result = cmd->result;
744 struct request_queue *q = cmd->device->request_queue;
745 struct request *req = cmd->request;
747 struct scsi_sense_hdr sshdr;
749 int sense_deferred = 0;
750 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
751 ACTION_DELAYED_RETRY} action;
752 char *description = NULL;
755 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
757 sense_deferred = scsi_sense_is_deferred(&sshdr);
760 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
761 req->errors = result;
763 if (sense_valid && req->sense) {
765 * SG_IO wants current and deferred errors
767 int len = 8 + cmd->sense_buffer[7];
769 if (len > SCSI_SENSE_BUFFERSIZE)
770 len = SCSI_SENSE_BUFFERSIZE;
771 memcpy(req->sense, cmd->sense_buffer, len);
772 req->sense_len = len;
775 error = __scsi_error_from_host_byte(cmd, result);
778 req->resid_len = scsi_get_resid(cmd);
780 if (scsi_bidi_cmnd(cmd)) {
782 * Bidi commands Must be complete as a whole,
783 * both sides at once.
785 req->next_rq->resid_len = scsi_in(cmd)->resid;
787 scsi_release_buffers(cmd);
788 blk_end_request_all(req, 0);
790 scsi_next_command(cmd);
795 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
796 BUG_ON(blk_bidi_rq(req));
799 * Next deal with any sectors which we were able to correctly
802 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
804 blk_rq_sectors(req), good_bytes));
807 * Recovered errors need reporting, but they're always treated
808 * as success, so fiddle the result code here. For BLOCK_PC
809 * we already took a copy of the original into rq->errors which
810 * is what gets returned to the user
812 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
813 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
814 * print since caller wants ATA registers. Only occurs on
815 * SCSI ATA PASS_THROUGH commands when CK_COND=1
817 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
819 else if (!(req->cmd_flags & REQ_QUIET))
820 scsi_print_sense("", cmd);
822 /* BLOCK_PC may have set error */
827 * A number of bytes were successfully read. If there
828 * are leftovers and there is some kind of error
829 * (result != 0), retry the rest.
831 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
834 error = __scsi_error_from_host_byte(cmd, result);
836 if (host_byte(result) == DID_RESET) {
837 /* Third party bus reset or reset for error recovery
838 * reasons. Just retry the command and see what
841 action = ACTION_RETRY;
842 } else if (sense_valid && !sense_deferred) {
843 switch (sshdr.sense_key) {
845 if (cmd->device->removable) {
846 /* Detected disc change. Set a bit
847 * and quietly refuse further access.
849 cmd->device->changed = 1;
850 description = "Media Changed";
851 action = ACTION_FAIL;
853 /* Must have been a power glitch, or a
854 * bus reset. Could not have been a
855 * media change, so we just retry the
856 * command and see what happens.
858 action = ACTION_RETRY;
861 case ILLEGAL_REQUEST:
862 /* If we had an ILLEGAL REQUEST returned, then
863 * we may have performed an unsupported
864 * command. The only thing this should be
865 * would be a ten byte read where only a six
866 * byte read was supported. Also, on a system
867 * where READ CAPACITY failed, we may have
868 * read past the end of the disk.
870 if ((cmd->device->use_10_for_rw &&
871 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
872 (cmd->cmnd[0] == READ_10 ||
873 cmd->cmnd[0] == WRITE_10)) {
874 /* This will issue a new 6-byte command. */
875 cmd->device->use_10_for_rw = 0;
876 action = ACTION_REPREP;
877 } else if (sshdr.asc == 0x10) /* DIX */ {
878 description = "Host Data Integrity Failure";
879 action = ACTION_FAIL;
881 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
882 } else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
883 (cmd->cmnd[0] == UNMAP ||
884 cmd->cmnd[0] == WRITE_SAME_16 ||
885 cmd->cmnd[0] == WRITE_SAME)) {
886 description = "Discard failure";
887 action = ACTION_FAIL;
890 action = ACTION_FAIL;
892 case ABORTED_COMMAND:
893 action = ACTION_FAIL;
894 if (sshdr.asc == 0x10) { /* DIF */
895 description = "Target Data Integrity Failure";
900 /* If the device is in the process of becoming
901 * ready, or has a temporary blockage, retry.
903 if (sshdr.asc == 0x04) {
904 switch (sshdr.ascq) {
905 case 0x01: /* becoming ready */
906 case 0x04: /* format in progress */
907 case 0x05: /* rebuild in progress */
908 case 0x06: /* recalculation in progress */
909 case 0x07: /* operation in progress */
910 case 0x08: /* Long write in progress */
911 case 0x09: /* self test in progress */
912 case 0x14: /* space allocation in progress */
913 action = ACTION_DELAYED_RETRY;
916 description = "Device not ready";
917 action = ACTION_FAIL;
921 description = "Device not ready";
922 action = ACTION_FAIL;
925 case VOLUME_OVERFLOW:
926 /* See SSC3rXX or current. */
927 action = ACTION_FAIL;
930 description = "Unhandled sense code";
931 action = ACTION_FAIL;
935 description = "Unhandled error code";
936 action = ACTION_FAIL;
941 /* Give up and fail the remainder of the request */
942 scsi_release_buffers(cmd);
943 if (!(req->cmd_flags & REQ_QUIET)) {
945 scmd_printk(KERN_INFO, cmd, "%s\n",
947 scsi_print_result(cmd);
948 if (driver_byte(result) & DRIVER_SENSE)
949 scsi_print_sense("", cmd);
950 scsi_print_command(cmd);
952 if (blk_end_request_err(req, error))
953 scsi_requeue_command(q, cmd);
955 scsi_next_command(cmd);
958 /* Unprep the request and put it back at the head of the queue.
959 * A new command will be prepared and issued.
961 scsi_release_buffers(cmd);
962 scsi_requeue_command(q, cmd);
965 /* Retry the same command immediately */
966 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
968 case ACTION_DELAYED_RETRY:
969 /* Retry the same command after a delay */
970 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
975 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
981 * If sg table allocation fails, requeue request later.
983 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
985 return BLKPREP_DEFER;
991 * Next, walk the list, and fill in the addresses and sizes of
994 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
995 BUG_ON(count > sdb->table.nents);
996 sdb->table.nents = count;
997 sdb->length = blk_rq_bytes(req);
1002 * Function: scsi_init_io()
1004 * Purpose: SCSI I/O initialize function.
1006 * Arguments: cmd - Command descriptor we wish to initialize
1008 * Returns: 0 on success
1009 * BLKPREP_DEFER if the failure is retryable
1010 * BLKPREP_KILL if the failure is fatal
1012 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1014 struct request *rq = cmd->request;
1016 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1020 if (blk_bidi_rq(rq)) {
1021 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1022 scsi_sdb_cache, GFP_ATOMIC);
1024 error = BLKPREP_DEFER;
1028 rq->next_rq->special = bidi_sdb;
1029 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1034 if (blk_integrity_rq(rq)) {
1035 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1038 BUG_ON(prot_sdb == NULL);
1039 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1041 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1042 error = BLKPREP_DEFER;
1046 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1047 prot_sdb->table.sgl);
1048 BUG_ON(unlikely(count > ivecs));
1049 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1051 cmd->prot_sdb = prot_sdb;
1052 cmd->prot_sdb->table.nents = count;
1058 scsi_release_buffers(cmd);
1059 cmd->request->special = NULL;
1060 scsi_put_command(cmd);
1063 EXPORT_SYMBOL(scsi_init_io);
1065 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1066 struct request *req)
1068 struct scsi_cmnd *cmd;
1070 if (!req->special) {
1071 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1079 /* pull a tag out of the request if we have one */
1080 cmd->tag = req->tag;
1083 cmd->cmnd = req->cmd;
1084 cmd->prot_op = SCSI_PROT_NORMAL;
1089 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1091 struct scsi_cmnd *cmd;
1092 int ret = scsi_prep_state_check(sdev, req);
1094 if (ret != BLKPREP_OK)
1097 cmd = scsi_get_cmd_from_req(sdev, req);
1099 return BLKPREP_DEFER;
1102 * BLOCK_PC requests may transfer data, in which case they must
1103 * a bio attached to them. Or they might contain a SCSI command
1104 * that does not transfer data, in which case they may optionally
1105 * submit a request without an attached bio.
1110 BUG_ON(!req->nr_phys_segments);
1112 ret = scsi_init_io(cmd, GFP_ATOMIC);
1116 BUG_ON(blk_rq_bytes(req));
1118 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1122 cmd->cmd_len = req->cmd_len;
1123 if (!blk_rq_bytes(req))
1124 cmd->sc_data_direction = DMA_NONE;
1125 else if (rq_data_dir(req) == WRITE)
1126 cmd->sc_data_direction = DMA_TO_DEVICE;
1128 cmd->sc_data_direction = DMA_FROM_DEVICE;
1130 cmd->transfersize = blk_rq_bytes(req);
1131 cmd->allowed = req->retries;
1134 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1137 * Setup a REQ_TYPE_FS command. These are simple read/write request
1138 * from filesystems that still need to be translated to SCSI CDBs from
1141 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1143 struct scsi_cmnd *cmd;
1144 int ret = scsi_prep_state_check(sdev, req);
1146 if (ret != BLKPREP_OK)
1149 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1150 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1151 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1152 if (ret != BLKPREP_OK)
1157 * Filesystem requests must transfer data.
1159 BUG_ON(!req->nr_phys_segments);
1161 cmd = scsi_get_cmd_from_req(sdev, req);
1163 return BLKPREP_DEFER;
1165 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1166 return scsi_init_io(cmd, GFP_ATOMIC);
1168 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1170 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1172 int ret = BLKPREP_OK;
1175 * If the device is not in running state we will reject some
1178 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1179 switch (sdev->sdev_state) {
1181 case SDEV_TRANSPORT_OFFLINE:
1183 * If the device is offline we refuse to process any
1184 * commands. The device must be brought online
1185 * before trying any recovery commands.
1187 sdev_printk(KERN_ERR, sdev,
1188 "rejecting I/O to offline device\n");
1193 * If the device is fully deleted, we refuse to
1194 * process any commands as well.
1196 sdev_printk(KERN_ERR, sdev,
1197 "rejecting I/O to dead device\n");
1202 case SDEV_CREATED_BLOCK:
1204 * If the devices is blocked we defer normal commands.
1206 if (!(req->cmd_flags & REQ_PREEMPT))
1207 ret = BLKPREP_DEFER;
1211 * For any other not fully online state we only allow
1212 * special commands. In particular any user initiated
1213 * command is not allowed.
1215 if (!(req->cmd_flags & REQ_PREEMPT))
1222 EXPORT_SYMBOL(scsi_prep_state_check);
1224 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1226 struct scsi_device *sdev = q->queuedata;
1230 req->errors = DID_NO_CONNECT << 16;
1231 /* release the command and kill it */
1233 struct scsi_cmnd *cmd = req->special;
1234 scsi_release_buffers(cmd);
1235 scsi_put_command(cmd);
1236 req->special = NULL;
1241 * If we defer, the blk_peek_request() returns NULL, but the
1242 * queue must be restarted, so we schedule a callback to happen
1245 if (sdev->device_busy == 0)
1246 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1249 req->cmd_flags |= REQ_DONTPREP;
1254 EXPORT_SYMBOL(scsi_prep_return);
1256 int scsi_prep_fn(struct request_queue *q, struct request *req)
1258 struct scsi_device *sdev = q->queuedata;
1259 int ret = BLKPREP_KILL;
1261 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1262 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1263 return scsi_prep_return(q, req, ret);
1265 EXPORT_SYMBOL(scsi_prep_fn);
1268 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1271 * Called with the queue_lock held.
1273 static inline int scsi_dev_queue_ready(struct request_queue *q,
1274 struct scsi_device *sdev)
1276 if (sdev->device_busy == 0 && sdev->device_blocked) {
1278 * unblock after device_blocked iterates to zero
1280 if (--sdev->device_blocked == 0) {
1282 sdev_printk(KERN_INFO, sdev,
1283 "unblocking device at zero depth\n"));
1285 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1289 if (scsi_device_is_busy(sdev))
1297 * scsi_target_queue_ready: checks if there we can send commands to target
1298 * @sdev: scsi device on starget to check.
1300 * Called with the host lock held.
1302 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1303 struct scsi_device *sdev)
1305 struct scsi_target *starget = scsi_target(sdev);
1307 if (starget->single_lun) {
1308 if (starget->starget_sdev_user &&
1309 starget->starget_sdev_user != sdev)
1311 starget->starget_sdev_user = sdev;
1314 if (starget->target_busy == 0 && starget->target_blocked) {
1316 * unblock after target_blocked iterates to zero
1318 if (--starget->target_blocked == 0) {
1319 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1320 "unblocking target at zero depth\n"));
1325 if (scsi_target_is_busy(starget)) {
1326 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1334 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1335 * return 0. We must end up running the queue again whenever 0 is
1336 * returned, else IO can hang.
1338 * Called with host_lock held.
1340 static inline int scsi_host_queue_ready(struct request_queue *q,
1341 struct Scsi_Host *shost,
1342 struct scsi_device *sdev)
1344 if (scsi_host_in_recovery(shost))
1346 if (shost->host_busy == 0 && shost->host_blocked) {
1348 * unblock after host_blocked iterates to zero
1350 if (--shost->host_blocked == 0) {
1352 printk("scsi%d unblocking host at zero depth\n",
1358 if (scsi_host_is_busy(shost)) {
1359 if (list_empty(&sdev->starved_entry))
1360 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1364 /* We're OK to process the command, so we can't be starved */
1365 if (!list_empty(&sdev->starved_entry))
1366 list_del_init(&sdev->starved_entry);
1372 * Busy state exporting function for request stacking drivers.
1374 * For efficiency, no lock is taken to check the busy state of
1375 * shost/starget/sdev, since the returned value is not guaranteed and
1376 * may be changed after request stacking drivers call the function,
1377 * regardless of taking lock or not.
1379 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1380 * needs to return 'not busy'. Otherwise, request stacking drivers
1381 * may hold requests forever.
1383 static int scsi_lld_busy(struct request_queue *q)
1385 struct scsi_device *sdev = q->queuedata;
1386 struct Scsi_Host *shost;
1388 if (blk_queue_dead(q))
1394 * Ignore host/starget busy state.
1395 * Since block layer does not have a concept of fairness across
1396 * multiple queues, congestion of host/starget needs to be handled
1399 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1406 * Kill a request for a dead device
1408 static void scsi_kill_request(struct request *req, struct request_queue *q)
1410 struct scsi_cmnd *cmd = req->special;
1411 struct scsi_device *sdev;
1412 struct scsi_target *starget;
1413 struct Scsi_Host *shost;
1415 blk_start_request(req);
1417 scmd_printk(KERN_INFO, cmd, "killing request\n");
1420 starget = scsi_target(sdev);
1422 scsi_init_cmd_errh(cmd);
1423 cmd->result = DID_NO_CONNECT << 16;
1424 atomic_inc(&cmd->device->iorequest_cnt);
1427 * SCSI request completion path will do scsi_device_unbusy(),
1428 * bump busy counts. To bump the counters, we need to dance
1429 * with the locks as normal issue path does.
1431 sdev->device_busy++;
1432 spin_unlock(sdev->request_queue->queue_lock);
1433 spin_lock(shost->host_lock);
1435 starget->target_busy++;
1436 spin_unlock(shost->host_lock);
1437 spin_lock(sdev->request_queue->queue_lock);
1439 blk_complete_request(req);
1442 static void scsi_softirq_done(struct request *rq)
1444 struct scsi_cmnd *cmd = rq->special;
1445 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1448 INIT_LIST_HEAD(&cmd->eh_entry);
1450 atomic_inc(&cmd->device->iodone_cnt);
1452 atomic_inc(&cmd->device->ioerr_cnt);
1454 disposition = scsi_decide_disposition(cmd);
1455 if (disposition != SUCCESS &&
1456 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1457 sdev_printk(KERN_ERR, cmd->device,
1458 "timing out command, waited %lus\n",
1460 disposition = SUCCESS;
1463 scsi_log_completion(cmd, disposition);
1465 switch (disposition) {
1467 scsi_finish_command(cmd);
1470 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1472 case ADD_TO_MLQUEUE:
1473 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1476 if (!scsi_eh_scmd_add(cmd, 0))
1477 scsi_finish_command(cmd);
1482 * Function: scsi_request_fn()
1484 * Purpose: Main strategy routine for SCSI.
1486 * Arguments: q - Pointer to actual queue.
1490 * Lock status: IO request lock assumed to be held when called.
1492 static void scsi_request_fn(struct request_queue *q)
1494 struct scsi_device *sdev = q->queuedata;
1495 struct Scsi_Host *shost;
1496 struct scsi_cmnd *cmd;
1497 struct request *req;
1499 if(!get_device(&sdev->sdev_gendev))
1500 /* We must be tearing the block queue down already */
1504 * To start with, we keep looping until the queue is empty, or until
1505 * the host is no longer able to accept any more requests.
1511 * get next queueable request. We do this early to make sure
1512 * that the request is fully prepared even if we cannot
1515 req = blk_peek_request(q);
1516 if (!req || !scsi_dev_queue_ready(q, sdev))
1519 if (unlikely(!scsi_device_online(sdev))) {
1520 sdev_printk(KERN_ERR, sdev,
1521 "rejecting I/O to offline device\n");
1522 scsi_kill_request(req, q);
1528 * Remove the request from the request list.
1530 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1531 blk_start_request(req);
1532 sdev->device_busy++;
1534 spin_unlock(q->queue_lock);
1536 if (unlikely(cmd == NULL)) {
1537 printk(KERN_CRIT "impossible request in %s.\n"
1538 "please mail a stack trace to "
1539 "linux-scsi@vger.kernel.org\n",
1541 blk_dump_rq_flags(req, "foo");
1544 spin_lock(shost->host_lock);
1547 * We hit this when the driver is using a host wide
1548 * tag map. For device level tag maps the queue_depth check
1549 * in the device ready fn would prevent us from trying
1550 * to allocate a tag. Since the map is a shared host resource
1551 * we add the dev to the starved list so it eventually gets
1552 * a run when a tag is freed.
1554 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1555 if (list_empty(&sdev->starved_entry))
1556 list_add_tail(&sdev->starved_entry,
1557 &shost->starved_list);
1561 if (!scsi_target_queue_ready(shost, sdev))
1564 if (!scsi_host_queue_ready(q, shost, sdev))
1567 scsi_target(sdev)->target_busy++;
1571 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1572 * take the lock again.
1574 spin_unlock_irq(shost->host_lock);
1577 * Finally, initialize any error handling parameters, and set up
1578 * the timers for timeouts.
1580 scsi_init_cmd_errh(cmd);
1583 * Dispatch the command to the low-level driver.
1585 rtn = scsi_dispatch_cmd(cmd);
1586 spin_lock_irq(q->queue_lock);
1594 spin_unlock_irq(shost->host_lock);
1597 * lock q, handle tag, requeue req, and decrement device_busy. We
1598 * must return with queue_lock held.
1600 * Decrementing device_busy without checking it is OK, as all such
1601 * cases (host limits or settings) should run the queue at some
1604 spin_lock_irq(q->queue_lock);
1605 blk_requeue_request(q, req);
1606 sdev->device_busy--;
1608 if (sdev->device_busy == 0)
1609 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1611 /* must be careful here...if we trigger the ->remove() function
1612 * we cannot be holding the q lock */
1613 spin_unlock_irq(q->queue_lock);
1614 put_device(&sdev->sdev_gendev);
1615 spin_lock_irq(q->queue_lock);
1618 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1620 struct device *host_dev;
1621 u64 bounce_limit = 0xffffffff;
1623 if (shost->unchecked_isa_dma)
1624 return BLK_BOUNCE_ISA;
1626 * Platforms with virtual-DMA translation
1627 * hardware have no practical limit.
1629 if (!PCI_DMA_BUS_IS_PHYS)
1630 return BLK_BOUNCE_ANY;
1632 host_dev = scsi_get_device(shost);
1633 if (host_dev && host_dev->dma_mask)
1634 bounce_limit = *host_dev->dma_mask;
1636 return bounce_limit;
1638 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1640 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1641 request_fn_proc *request_fn)
1643 struct request_queue *q;
1644 struct device *dev = shost->dma_dev;
1646 q = blk_init_queue(request_fn, NULL);
1651 * this limit is imposed by hardware restrictions
1653 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1654 SCSI_MAX_SG_CHAIN_SEGMENTS));
1656 if (scsi_host_prot_dma(shost)) {
1657 shost->sg_prot_tablesize =
1658 min_not_zero(shost->sg_prot_tablesize,
1659 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1660 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1661 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1664 blk_queue_max_hw_sectors(q, shost->max_sectors);
1665 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1666 blk_queue_segment_boundary(q, shost->dma_boundary);
1667 dma_set_seg_boundary(dev, shost->dma_boundary);
1669 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1671 if (!shost->use_clustering)
1672 q->limits.cluster = 0;
1675 * set a reasonable default alignment on word boundaries: the
1676 * host and device may alter it using
1677 * blk_queue_update_dma_alignment() later.
1679 blk_queue_dma_alignment(q, 0x03);
1683 EXPORT_SYMBOL(__scsi_alloc_queue);
1685 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1687 struct request_queue *q;
1689 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1693 blk_queue_prep_rq(q, scsi_prep_fn);
1694 blk_queue_softirq_done(q, scsi_softirq_done);
1695 blk_queue_rq_timed_out(q, scsi_times_out);
1696 blk_queue_lld_busy(q, scsi_lld_busy);
1701 * Function: scsi_block_requests()
1703 * Purpose: Utility function used by low-level drivers to prevent further
1704 * commands from being queued to the device.
1706 * Arguments: shost - Host in question
1710 * Lock status: No locks are assumed held.
1712 * Notes: There is no timer nor any other means by which the requests
1713 * get unblocked other than the low-level driver calling
1714 * scsi_unblock_requests().
1716 void scsi_block_requests(struct Scsi_Host *shost)
1718 shost->host_self_blocked = 1;
1720 EXPORT_SYMBOL(scsi_block_requests);
1723 * Function: scsi_unblock_requests()
1725 * Purpose: Utility function used by low-level drivers to allow further
1726 * commands from being queued to the device.
1728 * Arguments: shost - Host in question
1732 * Lock status: No locks are assumed held.
1734 * Notes: There is no timer nor any other means by which the requests
1735 * get unblocked other than the low-level driver calling
1736 * scsi_unblock_requests().
1738 * This is done as an API function so that changes to the
1739 * internals of the scsi mid-layer won't require wholesale
1740 * changes to drivers that use this feature.
1742 void scsi_unblock_requests(struct Scsi_Host *shost)
1744 shost->host_self_blocked = 0;
1745 scsi_run_host_queues(shost);
1747 EXPORT_SYMBOL(scsi_unblock_requests);
1749 int __init scsi_init_queue(void)
1753 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1754 sizeof(struct scsi_data_buffer),
1756 if (!scsi_sdb_cache) {
1757 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1761 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1762 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1763 int size = sgp->size * sizeof(struct scatterlist);
1765 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1766 SLAB_HWCACHE_ALIGN, NULL);
1768 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1773 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1776 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1785 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1786 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1788 mempool_destroy(sgp->pool);
1790 kmem_cache_destroy(sgp->slab);
1792 kmem_cache_destroy(scsi_sdb_cache);
1797 void scsi_exit_queue(void)
1801 kmem_cache_destroy(scsi_sdb_cache);
1803 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1804 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1805 mempool_destroy(sgp->pool);
1806 kmem_cache_destroy(sgp->slab);
1811 * scsi_mode_select - issue a mode select
1812 * @sdev: SCSI device to be queried
1813 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1814 * @sp: Save page bit (0 == don't save, 1 == save)
1815 * @modepage: mode page being requested
1816 * @buffer: request buffer (may not be smaller than eight bytes)
1817 * @len: length of request buffer.
1818 * @timeout: command timeout
1819 * @retries: number of retries before failing
1820 * @data: returns a structure abstracting the mode header data
1821 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1822 * must be SCSI_SENSE_BUFFERSIZE big.
1824 * Returns zero if successful; negative error number or scsi
1829 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1830 unsigned char *buffer, int len, int timeout, int retries,
1831 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1833 unsigned char cmd[10];
1834 unsigned char *real_buffer;
1837 memset(cmd, 0, sizeof(cmd));
1838 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1840 if (sdev->use_10_for_ms) {
1843 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1846 memcpy(real_buffer + 8, buffer, len);
1850 real_buffer[2] = data->medium_type;
1851 real_buffer[3] = data->device_specific;
1852 real_buffer[4] = data->longlba ? 0x01 : 0;
1854 real_buffer[6] = data->block_descriptor_length >> 8;
1855 real_buffer[7] = data->block_descriptor_length;
1857 cmd[0] = MODE_SELECT_10;
1861 if (len > 255 || data->block_descriptor_length > 255 ||
1865 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1868 memcpy(real_buffer + 4, buffer, len);
1871 real_buffer[1] = data->medium_type;
1872 real_buffer[2] = data->device_specific;
1873 real_buffer[3] = data->block_descriptor_length;
1876 cmd[0] = MODE_SELECT;
1880 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1881 sshdr, timeout, retries, NULL);
1885 EXPORT_SYMBOL_GPL(scsi_mode_select);
1888 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1889 * @sdev: SCSI device to be queried
1890 * @dbd: set if mode sense will allow block descriptors to be returned
1891 * @modepage: mode page being requested
1892 * @buffer: request buffer (may not be smaller than eight bytes)
1893 * @len: length of request buffer.
1894 * @timeout: command timeout
1895 * @retries: number of retries before failing
1896 * @data: returns a structure abstracting the mode header data
1897 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1898 * must be SCSI_SENSE_BUFFERSIZE big.
1900 * Returns zero if unsuccessful, or the header offset (either 4
1901 * or 8 depending on whether a six or ten byte command was
1902 * issued) if successful.
1905 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1906 unsigned char *buffer, int len, int timeout, int retries,
1907 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1909 unsigned char cmd[12];
1913 struct scsi_sense_hdr my_sshdr;
1915 memset(data, 0, sizeof(*data));
1916 memset(&cmd[0], 0, 12);
1917 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1920 /* caller might not be interested in sense, but we need it */
1925 use_10_for_ms = sdev->use_10_for_ms;
1927 if (use_10_for_ms) {
1931 cmd[0] = MODE_SENSE_10;
1938 cmd[0] = MODE_SENSE;
1943 memset(buffer, 0, len);
1945 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1946 sshdr, timeout, retries, NULL);
1948 /* This code looks awful: what it's doing is making sure an
1949 * ILLEGAL REQUEST sense return identifies the actual command
1950 * byte as the problem. MODE_SENSE commands can return
1951 * ILLEGAL REQUEST if the code page isn't supported */
1953 if (use_10_for_ms && !scsi_status_is_good(result) &&
1954 (driver_byte(result) & DRIVER_SENSE)) {
1955 if (scsi_sense_valid(sshdr)) {
1956 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1957 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1959 * Invalid command operation code
1961 sdev->use_10_for_ms = 0;
1967 if(scsi_status_is_good(result)) {
1968 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1969 (modepage == 6 || modepage == 8))) {
1970 /* Initio breakage? */
1973 data->medium_type = 0;
1974 data->device_specific = 0;
1976 data->block_descriptor_length = 0;
1977 } else if(use_10_for_ms) {
1978 data->length = buffer[0]*256 + buffer[1] + 2;
1979 data->medium_type = buffer[2];
1980 data->device_specific = buffer[3];
1981 data->longlba = buffer[4] & 0x01;
1982 data->block_descriptor_length = buffer[6]*256
1985 data->length = buffer[0] + 1;
1986 data->medium_type = buffer[1];
1987 data->device_specific = buffer[2];
1988 data->block_descriptor_length = buffer[3];
1990 data->header_length = header_length;
1995 EXPORT_SYMBOL(scsi_mode_sense);
1998 * scsi_test_unit_ready - test if unit is ready
1999 * @sdev: scsi device to change the state of.
2000 * @timeout: command timeout
2001 * @retries: number of retries before failing
2002 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2003 * returning sense. Make sure that this is cleared before passing
2006 * Returns zero if unsuccessful or an error if TUR failed. For
2007 * removable media, UNIT_ATTENTION sets ->changed flag.
2010 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2011 struct scsi_sense_hdr *sshdr_external)
2014 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2016 struct scsi_sense_hdr *sshdr;
2019 if (!sshdr_external)
2020 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2022 sshdr = sshdr_external;
2024 /* try to eat the UNIT_ATTENTION if there are enough retries */
2026 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2027 timeout, retries, NULL);
2028 if (sdev->removable && scsi_sense_valid(sshdr) &&
2029 sshdr->sense_key == UNIT_ATTENTION)
2031 } while (scsi_sense_valid(sshdr) &&
2032 sshdr->sense_key == UNIT_ATTENTION && --retries);
2034 if (!sshdr_external)
2038 EXPORT_SYMBOL(scsi_test_unit_ready);
2041 * scsi_device_set_state - Take the given device through the device state model.
2042 * @sdev: scsi device to change the state of.
2043 * @state: state to change to.
2045 * Returns zero if unsuccessful or an error if the requested
2046 * transition is illegal.
2049 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2051 enum scsi_device_state oldstate = sdev->sdev_state;
2053 if (state == oldstate)
2059 case SDEV_CREATED_BLOCK:
2070 case SDEV_TRANSPORT_OFFLINE:
2083 case SDEV_TRANSPORT_OFFLINE:
2091 case SDEV_TRANSPORT_OFFLINE:
2106 case SDEV_CREATED_BLOCK:
2113 case SDEV_CREATED_BLOCK:
2128 case SDEV_TRANSPORT_OFFLINE:
2141 case SDEV_TRANSPORT_OFFLINE:
2150 sdev->sdev_state = state;
2154 SCSI_LOG_ERROR_RECOVERY(1,
2155 sdev_printk(KERN_ERR, sdev,
2156 "Illegal state transition %s->%s\n",
2157 scsi_device_state_name(oldstate),
2158 scsi_device_state_name(state))
2162 EXPORT_SYMBOL(scsi_device_set_state);
2165 * sdev_evt_emit - emit a single SCSI device uevent
2166 * @sdev: associated SCSI device
2167 * @evt: event to emit
2169 * Send a single uevent (scsi_event) to the associated scsi_device.
2171 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2176 switch (evt->evt_type) {
2177 case SDEV_EVT_MEDIA_CHANGE:
2178 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2188 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2192 * sdev_evt_thread - send a uevent for each scsi event
2193 * @work: work struct for scsi_device
2195 * Dispatch queued events to their associated scsi_device kobjects
2198 void scsi_evt_thread(struct work_struct *work)
2200 struct scsi_device *sdev;
2201 LIST_HEAD(event_list);
2203 sdev = container_of(work, struct scsi_device, event_work);
2206 struct scsi_event *evt;
2207 struct list_head *this, *tmp;
2208 unsigned long flags;
2210 spin_lock_irqsave(&sdev->list_lock, flags);
2211 list_splice_init(&sdev->event_list, &event_list);
2212 spin_unlock_irqrestore(&sdev->list_lock, flags);
2214 if (list_empty(&event_list))
2217 list_for_each_safe(this, tmp, &event_list) {
2218 evt = list_entry(this, struct scsi_event, node);
2219 list_del(&evt->node);
2220 scsi_evt_emit(sdev, evt);
2227 * sdev_evt_send - send asserted event to uevent thread
2228 * @sdev: scsi_device event occurred on
2229 * @evt: event to send
2231 * Assert scsi device event asynchronously.
2233 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2235 unsigned long flags;
2238 /* FIXME: currently this check eliminates all media change events
2239 * for polled devices. Need to update to discriminate between AN
2240 * and polled events */
2241 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2247 spin_lock_irqsave(&sdev->list_lock, flags);
2248 list_add_tail(&evt->node, &sdev->event_list);
2249 schedule_work(&sdev->event_work);
2250 spin_unlock_irqrestore(&sdev->list_lock, flags);
2252 EXPORT_SYMBOL_GPL(sdev_evt_send);
2255 * sdev_evt_alloc - allocate a new scsi event
2256 * @evt_type: type of event to allocate
2257 * @gfpflags: GFP flags for allocation
2259 * Allocates and returns a new scsi_event.
2261 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2264 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2268 evt->evt_type = evt_type;
2269 INIT_LIST_HEAD(&evt->node);
2271 /* evt_type-specific initialization, if any */
2273 case SDEV_EVT_MEDIA_CHANGE:
2281 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2284 * sdev_evt_send_simple - send asserted event to uevent thread
2285 * @sdev: scsi_device event occurred on
2286 * @evt_type: type of event to send
2287 * @gfpflags: GFP flags for allocation
2289 * Assert scsi device event asynchronously, given an event type.
2291 void sdev_evt_send_simple(struct scsi_device *sdev,
2292 enum scsi_device_event evt_type, gfp_t gfpflags)
2294 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2296 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2301 sdev_evt_send(sdev, evt);
2303 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2306 * scsi_device_quiesce - Block user issued commands.
2307 * @sdev: scsi device to quiesce.
2309 * This works by trying to transition to the SDEV_QUIESCE state
2310 * (which must be a legal transition). When the device is in this
2311 * state, only special requests will be accepted, all others will
2312 * be deferred. Since special requests may also be requeued requests,
2313 * a successful return doesn't guarantee the device will be
2314 * totally quiescent.
2316 * Must be called with user context, may sleep.
2318 * Returns zero if unsuccessful or an error if not.
2321 scsi_device_quiesce(struct scsi_device *sdev)
2323 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2327 scsi_run_queue(sdev->request_queue);
2328 while (sdev->device_busy) {
2329 msleep_interruptible(200);
2330 scsi_run_queue(sdev->request_queue);
2334 EXPORT_SYMBOL(scsi_device_quiesce);
2337 * scsi_device_resume - Restart user issued commands to a quiesced device.
2338 * @sdev: scsi device to resume.
2340 * Moves the device from quiesced back to running and restarts the
2343 * Must be called with user context, may sleep.
2345 void scsi_device_resume(struct scsi_device *sdev)
2347 /* check if the device state was mutated prior to resume, and if
2348 * so assume the state is being managed elsewhere (for example
2349 * device deleted during suspend)
2351 if (sdev->sdev_state != SDEV_QUIESCE ||
2352 scsi_device_set_state(sdev, SDEV_RUNNING))
2354 scsi_run_queue(sdev->request_queue);
2356 EXPORT_SYMBOL(scsi_device_resume);
2359 device_quiesce_fn(struct scsi_device *sdev, void *data)
2361 scsi_device_quiesce(sdev);
2365 scsi_target_quiesce(struct scsi_target *starget)
2367 starget_for_each_device(starget, NULL, device_quiesce_fn);
2369 EXPORT_SYMBOL(scsi_target_quiesce);
2372 device_resume_fn(struct scsi_device *sdev, void *data)
2374 scsi_device_resume(sdev);
2378 scsi_target_resume(struct scsi_target *starget)
2380 starget_for_each_device(starget, NULL, device_resume_fn);
2382 EXPORT_SYMBOL(scsi_target_resume);
2385 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2386 * @sdev: device to block
2388 * Block request made by scsi lld's to temporarily stop all
2389 * scsi commands on the specified device. Called from interrupt
2390 * or normal process context.
2392 * Returns zero if successful or error if not
2395 * This routine transitions the device to the SDEV_BLOCK state
2396 * (which must be a legal transition). When the device is in this
2397 * state, all commands are deferred until the scsi lld reenables
2398 * the device with scsi_device_unblock or device_block_tmo fires.
2401 scsi_internal_device_block(struct scsi_device *sdev)
2403 struct request_queue *q = sdev->request_queue;
2404 unsigned long flags;
2407 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2409 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2416 * The device has transitioned to SDEV_BLOCK. Stop the
2417 * block layer from calling the midlayer with this device's
2420 spin_lock_irqsave(q->queue_lock, flags);
2422 spin_unlock_irqrestore(q->queue_lock, flags);
2426 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2429 * scsi_internal_device_unblock - resume a device after a block request
2430 * @sdev: device to resume
2431 * @new_state: state to set devices to after unblocking
2433 * Called by scsi lld's or the midlayer to restart the device queue
2434 * for the previously suspended scsi device. Called from interrupt or
2435 * normal process context.
2437 * Returns zero if successful or error if not.
2440 * This routine transitions the device to the SDEV_RUNNING state
2441 * or to one of the offline states (which must be a legal transition)
2442 * allowing the midlayer to goose the queue for this device.
2445 scsi_internal_device_unblock(struct scsi_device *sdev,
2446 enum scsi_device_state new_state)
2448 struct request_queue *q = sdev->request_queue;
2449 unsigned long flags;
2452 * Try to transition the scsi device to SDEV_RUNNING or one of the
2453 * offlined states and goose the device queue if successful.
2455 if (sdev->sdev_state == SDEV_BLOCK)
2456 sdev->sdev_state = new_state;
2457 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2458 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2459 new_state == SDEV_OFFLINE)
2460 sdev->sdev_state = new_state;
2462 sdev->sdev_state = SDEV_CREATED;
2463 } else if (sdev->sdev_state != SDEV_CANCEL &&
2464 sdev->sdev_state != SDEV_OFFLINE)
2467 spin_lock_irqsave(q->queue_lock, flags);
2469 spin_unlock_irqrestore(q->queue_lock, flags);
2473 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2476 device_block(struct scsi_device *sdev, void *data)
2478 scsi_internal_device_block(sdev);
2482 target_block(struct device *dev, void *data)
2484 if (scsi_is_target_device(dev))
2485 starget_for_each_device(to_scsi_target(dev), NULL,
2491 scsi_target_block(struct device *dev)
2493 if (scsi_is_target_device(dev))
2494 starget_for_each_device(to_scsi_target(dev), NULL,
2497 device_for_each_child(dev, NULL, target_block);
2499 EXPORT_SYMBOL_GPL(scsi_target_block);
2502 device_unblock(struct scsi_device *sdev, void *data)
2504 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2508 target_unblock(struct device *dev, void *data)
2510 if (scsi_is_target_device(dev))
2511 starget_for_each_device(to_scsi_target(dev), data,
2517 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2519 if (scsi_is_target_device(dev))
2520 starget_for_each_device(to_scsi_target(dev), &new_state,
2523 device_for_each_child(dev, &new_state, target_unblock);
2525 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2528 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2529 * @sgl: scatter-gather list
2530 * @sg_count: number of segments in sg
2531 * @offset: offset in bytes into sg, on return offset into the mapped area
2532 * @len: bytes to map, on return number of bytes mapped
2534 * Returns virtual address of the start of the mapped page
2536 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2537 size_t *offset, size_t *len)
2540 size_t sg_len = 0, len_complete = 0;
2541 struct scatterlist *sg;
2544 WARN_ON(!irqs_disabled());
2546 for_each_sg(sgl, sg, sg_count, i) {
2547 len_complete = sg_len; /* Complete sg-entries */
2548 sg_len += sg->length;
2549 if (sg_len > *offset)
2553 if (unlikely(i == sg_count)) {
2554 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2556 __func__, sg_len, *offset, sg_count);
2561 /* Offset starting from the beginning of first page in this sg-entry */
2562 *offset = *offset - len_complete + sg->offset;
2564 /* Assumption: contiguous pages can be accessed as "page + i" */
2565 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2566 *offset &= ~PAGE_MASK;
2568 /* Bytes in this sg-entry from *offset to the end of the page */
2569 sg_len = PAGE_SIZE - *offset;
2573 return kmap_atomic(page);
2575 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2578 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2579 * @virt: virtual address to be unmapped
2581 void scsi_kunmap_atomic_sg(void *virt)
2583 kunmap_atomic(virt);
2585 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);