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/mempool.h>
16 #include <linux/slab.h>
17 #include <linux/init.h>
18 #include <linux/pci.h>
19 #include <linux/delay.h>
20 #include <linux/hardirq.h>
21 #include <linux/scatterlist.h>
23 #include <scsi/scsi.h>
24 #include <scsi/scsi_cmnd.h>
25 #include <scsi/scsi_dbg.h>
26 #include <scsi/scsi_device.h>
27 #include <scsi/scsi_driver.h>
28 #include <scsi/scsi_eh.h>
29 #include <scsi/scsi_host.h>
31 #include "scsi_priv.h"
32 #include "scsi_logging.h"
35 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
36 #define SG_MEMPOOL_SIZE 2
38 struct scsi_host_sg_pool {
41 struct kmem_cache *slab;
45 #define SP(x) { x, "sgpool-" __stringify(x) }
46 #if (SCSI_MAX_SG_SEGMENTS < 32)
47 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
49 static struct scsi_host_sg_pool scsi_sg_pools[] = {
52 #if (SCSI_MAX_SG_SEGMENTS > 32)
54 #if (SCSI_MAX_SG_SEGMENTS > 64)
56 #if (SCSI_MAX_SG_SEGMENTS > 128)
58 #if (SCSI_MAX_SG_SEGMENTS > 256)
59 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
64 SP(SCSI_MAX_SG_SEGMENTS)
68 struct kmem_cache *scsi_sdb_cache;
70 static void scsi_run_queue(struct request_queue *q);
73 * Function: scsi_unprep_request()
75 * Purpose: Remove all preparation done for a request, including its
76 * associated scsi_cmnd, so that it can be requeued.
78 * Arguments: req - request to unprepare
80 * Lock status: Assumed that no locks are held upon entry.
84 static void scsi_unprep_request(struct request *req)
86 struct scsi_cmnd *cmd = req->special;
88 req->cmd_flags &= ~REQ_DONTPREP;
91 scsi_put_command(cmd);
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
106 static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
108 struct Scsi_Host *host = cmd->device->host;
109 struct scsi_device *device = cmd->device;
110 struct scsi_target *starget = scsi_target(device);
111 struct request_queue *q = device->request_queue;
115 printk("Inserting command %p into mlqueue\n", cmd));
118 * Set the appropriate busy bit for the device/host.
120 * If the host/device isn't busy, assume that something actually
121 * completed, and that we should be able to queue a command now.
123 * Note that the prior mid-layer assumption that any host could
124 * always queue at least one command is now broken. The mid-layer
125 * will implement a user specifiable stall (see
126 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
127 * if a command is requeued with no other commands outstanding
128 * either for the device or for the host.
131 case SCSI_MLQUEUE_HOST_BUSY:
132 host->host_blocked = host->max_host_blocked;
134 case SCSI_MLQUEUE_DEVICE_BUSY:
135 device->device_blocked = device->max_device_blocked;
137 case SCSI_MLQUEUE_TARGET_BUSY:
138 starget->target_blocked = starget->max_target_blocked;
143 * Decrement the counters, since these commands are no longer
144 * active on the host/device.
147 scsi_device_unbusy(device);
150 * Requeue this command. It will go before all other commands
151 * that are already in the queue.
153 * NOTE: there is magic here about the way the queue is plugged if
154 * we have no outstanding commands.
156 * Although we *don't* plug the queue, we call the request
157 * function. The SCSI request function detects the blocked condition
158 * and plugs the queue appropriately.
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
170 * Function: scsi_queue_insert()
172 * Purpose: Insert a command in the midlevel queue.
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
177 * Lock status: Assumed that lock is not held upon entry.
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
188 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
190 return __scsi_queue_insert(cmd, reason, 1);
193 * scsi_execute - insert request and wait for the result
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
205 * returns the req->errors value which is the scsi_cmnd result
208 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209 int data_direction, void *buffer, unsigned bufflen,
210 unsigned char *sense, int timeout, int retries, int flags,
214 int write = (data_direction == DMA_TO_DEVICE);
215 int ret = DRIVER_ERROR << 24;
217 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 = sdev->host;
404 LIST_HEAD(starved_list);
407 if (scsi_target(sdev)->single_lun)
408 scsi_single_lun_run(sdev);
410 spin_lock_irqsave(shost->host_lock, flags);
411 list_splice_init(&shost->starved_list, &starved_list);
413 while (!list_empty(&starved_list)) {
417 * As long as shost is accepting commands and we have
418 * starved queues, call blk_run_queue. scsi_request_fn
419 * drops the queue_lock and can add us back to the
422 * host_lock protects the starved_list and starved_entry.
423 * scsi_request_fn must get the host_lock before checking
424 * or modifying starved_list or starved_entry.
426 if (scsi_host_is_busy(shost))
429 sdev = list_entry(starved_list.next,
430 struct scsi_device, starved_entry);
431 list_del_init(&sdev->starved_entry);
432 if (scsi_target_is_busy(scsi_target(sdev))) {
433 list_move_tail(&sdev->starved_entry,
434 &shost->starved_list);
438 spin_unlock(shost->host_lock);
440 spin_lock(sdev->request_queue->queue_lock);
441 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
442 !test_bit(QUEUE_FLAG_REENTER,
443 &sdev->request_queue->queue_flags);
445 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
446 __blk_run_queue(sdev->request_queue);
448 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
449 spin_unlock(sdev->request_queue->queue_lock);
451 spin_lock(shost->host_lock);
453 /* put any unprocessed entries back */
454 list_splice(&starved_list, &shost->starved_list);
455 spin_unlock_irqrestore(shost->host_lock, flags);
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 request *req = cmd->request;
483 spin_lock_irqsave(q->queue_lock, flags);
484 scsi_unprep_request(req);
485 blk_requeue_request(q, req);
486 spin_unlock_irqrestore(q->queue_lock, flags);
491 void scsi_next_command(struct scsi_cmnd *cmd)
493 struct scsi_device *sdev = cmd->device;
494 struct request_queue *q = sdev->request_queue;
496 /* need to hold a reference on the device before we let go of the cmd */
497 get_device(&sdev->sdev_gendev);
499 scsi_put_command(cmd);
502 /* ok to remove device now */
503 put_device(&sdev->sdev_gendev);
506 void scsi_run_host_queues(struct Scsi_Host *shost)
508 struct scsi_device *sdev;
510 shost_for_each_device(sdev, shost)
511 scsi_run_queue(sdev->request_queue);
514 static void __scsi_release_buffers(struct scsi_cmnd *, int);
517 * Function: scsi_end_request()
519 * Purpose: Post-processing of completed commands (usually invoked at end
520 * of upper level post-processing and scsi_io_completion).
522 * Arguments: cmd - command that is complete.
523 * error - 0 if I/O indicates success, < 0 for I/O error.
524 * bytes - number of bytes of completed I/O
525 * requeue - indicates whether we should requeue leftovers.
527 * Lock status: Assumed that lock is not held upon entry.
529 * Returns: cmd if requeue required, NULL otherwise.
531 * Notes: This is called for block device requests in order to
532 * mark some number of sectors as complete.
534 * We are guaranteeing that the request queue will be goosed
535 * at some point during this call.
536 * Notes: If cmd was requeued, upon return it will be a stale pointer.
538 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
539 int bytes, int requeue)
541 struct request_queue *q = cmd->device->request_queue;
542 struct request *req = cmd->request;
545 * If there are blocks left over at the end, set up the command
546 * to queue the remainder of them.
548 if (blk_end_request(req, error, bytes)) {
549 /* kill remainder if no retrys */
550 if (error && scsi_noretry_cmd(cmd))
551 blk_end_request_all(req, error);
555 * Bleah. Leftovers again. Stick the
556 * leftovers in the front of the
557 * queue, and goose the queue again.
559 scsi_release_buffers(cmd);
560 scsi_requeue_command(q, cmd);
568 * This will goose the queue request function at the end, so we don't
569 * need to worry about launching another command.
571 __scsi_release_buffers(cmd, 0);
572 scsi_next_command(cmd);
576 static inline unsigned int scsi_sgtable_index(unsigned short nents)
580 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
585 index = get_count_order(nents) - 3;
590 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
592 struct scsi_host_sg_pool *sgp;
594 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
595 mempool_free(sgl, sgp->pool);
598 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
600 struct scsi_host_sg_pool *sgp;
602 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
603 return mempool_alloc(sgp->pool, gfp_mask);
606 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
613 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
614 gfp_mask, scsi_sg_alloc);
616 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
622 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
624 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
627 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
630 if (cmd->sdb.table.nents)
631 scsi_free_sgtable(&cmd->sdb);
633 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
635 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
636 struct scsi_data_buffer *bidi_sdb =
637 cmd->request->next_rq->special;
638 scsi_free_sgtable(bidi_sdb);
639 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
640 cmd->request->next_rq->special = NULL;
643 if (scsi_prot_sg_count(cmd))
644 scsi_free_sgtable(cmd->prot_sdb);
648 * Function: scsi_release_buffers()
650 * Purpose: Completion processing for block device I/O requests.
652 * Arguments: cmd - command that we are bailing.
654 * Lock status: Assumed that no lock is held upon entry.
658 * Notes: In the event that an upper level driver rejects a
659 * command, we must release resources allocated during
660 * the __init_io() function. Primarily this would involve
661 * the scatter-gather table, and potentially any bounce
664 void scsi_release_buffers(struct scsi_cmnd *cmd)
666 __scsi_release_buffers(cmd, 1);
668 EXPORT_SYMBOL(scsi_release_buffers);
671 * Function: scsi_io_completion()
673 * Purpose: Completion processing for block device I/O requests.
675 * Arguments: cmd - command that is finished.
677 * Lock status: Assumed that no lock is held upon entry.
681 * Notes: This function is matched in terms of capabilities to
682 * the function that created the scatter-gather list.
683 * In other words, if there are no bounce buffers
684 * (the normal case for most drivers), we don't need
685 * the logic to deal with cleaning up afterwards.
687 * We must call scsi_end_request(). This will finish off
688 * the specified number of sectors. If we are done, the
689 * command block will be released and the queue function
690 * will be goosed. If we are not done then we have to
691 * figure out what to do next:
693 * a) We can call scsi_requeue_command(). The request
694 * will be unprepared and put back on the queue. Then
695 * a new command will be created for it. This should
696 * be used if we made forward progress, or if we want
697 * to switch from READ(10) to READ(6) for example.
699 * b) We can call scsi_queue_insert(). The request will
700 * be put back on the queue and retried using the same
701 * command as before, possibly after a delay.
703 * c) We can call blk_end_request() with -EIO to fail
704 * the remainder of the request.
706 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
708 int result = cmd->result;
709 struct request_queue *q = cmd->device->request_queue;
710 struct request *req = cmd->request;
712 struct scsi_sense_hdr sshdr;
714 int sense_deferred = 0;
715 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
716 ACTION_DELAYED_RETRY} action;
717 char *description = NULL;
720 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
722 sense_deferred = scsi_sense_is_deferred(&sshdr);
725 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
726 req->errors = result;
728 if (sense_valid && req->sense) {
730 * SG_IO wants current and deferred errors
732 int len = 8 + cmd->sense_buffer[7];
734 if (len > SCSI_SENSE_BUFFERSIZE)
735 len = SCSI_SENSE_BUFFERSIZE;
736 memcpy(req->sense, cmd->sense_buffer, len);
737 req->sense_len = len;
743 req->resid_len = scsi_get_resid(cmd);
745 if (scsi_bidi_cmnd(cmd)) {
747 * Bidi commands Must be complete as a whole,
748 * both sides at once.
750 req->next_rq->resid_len = scsi_in(cmd)->resid;
752 scsi_release_buffers(cmd);
753 blk_end_request_all(req, 0);
755 scsi_next_command(cmd);
760 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
763 * Next deal with any sectors which we were able to correctly
766 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
768 blk_rq_sectors(req), good_bytes));
771 * Recovered errors need reporting, but they're always treated
772 * as success, so fiddle the result code here. For BLOCK_PC
773 * we already took a copy of the original into rq->errors which
774 * is what gets returned to the user
776 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
777 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
778 * print since caller wants ATA registers. Only occurs on
779 * SCSI ATA PASS_THROUGH commands when CK_COND=1
781 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
783 else if (!(req->cmd_flags & REQ_QUIET))
784 scsi_print_sense("", cmd);
786 /* BLOCK_PC may have set error */
791 * A number of bytes were successfully read. If there
792 * are leftovers and there is some kind of error
793 * (result != 0), retry the rest.
795 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
800 if (host_byte(result) == DID_RESET) {
801 /* Third party bus reset or reset for error recovery
802 * reasons. Just retry the command and see what
805 action = ACTION_RETRY;
806 } else if (sense_valid && !sense_deferred) {
807 switch (sshdr.sense_key) {
809 if (cmd->device->removable) {
810 /* Detected disc change. Set a bit
811 * and quietly refuse further access.
813 cmd->device->changed = 1;
814 description = "Media Changed";
815 action = ACTION_FAIL;
817 /* Must have been a power glitch, or a
818 * bus reset. Could not have been a
819 * media change, so we just retry the
820 * command and see what happens.
822 action = ACTION_RETRY;
825 case ILLEGAL_REQUEST:
826 /* If we had an ILLEGAL REQUEST returned, then
827 * we may have performed an unsupported
828 * command. The only thing this should be
829 * would be a ten byte read where only a six
830 * byte read was supported. Also, on a system
831 * where READ CAPACITY failed, we may have
832 * read past the end of the disk.
834 if ((cmd->device->use_10_for_rw &&
835 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
836 (cmd->cmnd[0] == READ_10 ||
837 cmd->cmnd[0] == WRITE_10)) {
838 /* This will issue a new 6-byte command. */
839 cmd->device->use_10_for_rw = 0;
840 action = ACTION_REPREP;
841 } else if (sshdr.asc == 0x10) /* DIX */ {
842 description = "Host Data Integrity Failure";
843 action = ACTION_FAIL;
846 action = ACTION_FAIL;
848 case ABORTED_COMMAND:
849 action = ACTION_FAIL;
850 if (sshdr.asc == 0x10) { /* DIF */
851 description = "Target Data Integrity Failure";
856 /* If the device is in the process of becoming
857 * ready, or has a temporary blockage, retry.
859 if (sshdr.asc == 0x04) {
860 switch (sshdr.ascq) {
861 case 0x01: /* becoming ready */
862 case 0x04: /* format in progress */
863 case 0x05: /* rebuild in progress */
864 case 0x06: /* recalculation in progress */
865 case 0x07: /* operation in progress */
866 case 0x08: /* Long write in progress */
867 case 0x09: /* self test in progress */
868 action = ACTION_DELAYED_RETRY;
871 description = "Device not ready";
872 action = ACTION_FAIL;
876 description = "Device not ready";
877 action = ACTION_FAIL;
880 case VOLUME_OVERFLOW:
881 /* See SSC3rXX or current. */
882 action = ACTION_FAIL;
885 description = "Unhandled sense code";
886 action = ACTION_FAIL;
890 description = "Unhandled error code";
891 action = ACTION_FAIL;
896 /* Give up and fail the remainder of the request */
897 scsi_release_buffers(cmd);
898 if (!(req->cmd_flags & REQ_QUIET)) {
900 scmd_printk(KERN_INFO, cmd, "%s\n",
902 scsi_print_result(cmd);
903 if (driver_byte(result) & DRIVER_SENSE)
904 scsi_print_sense("", cmd);
905 scsi_print_command(cmd);
907 if (blk_end_request_err(req, -EIO))
908 scsi_requeue_command(q, cmd);
910 scsi_next_command(cmd);
913 /* Unprep the request and put it back at the head of the queue.
914 * A new command will be prepared and issued.
916 scsi_release_buffers(cmd);
917 scsi_requeue_command(q, cmd);
920 /* Retry the same command immediately */
921 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
923 case ACTION_DELAYED_RETRY:
924 /* Retry the same command after a delay */
925 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
930 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
936 * If sg table allocation fails, requeue request later.
938 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
940 return BLKPREP_DEFER;
946 * Next, walk the list, and fill in the addresses and sizes of
949 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
950 BUG_ON(count > sdb->table.nents);
951 sdb->table.nents = count;
952 sdb->length = blk_rq_bytes(req);
957 * Function: scsi_init_io()
959 * Purpose: SCSI I/O initialize function.
961 * Arguments: cmd - Command descriptor we wish to initialize
963 * Returns: 0 on success
964 * BLKPREP_DEFER if the failure is retryable
965 * BLKPREP_KILL if the failure is fatal
967 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
969 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
973 if (blk_bidi_rq(cmd->request)) {
974 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
975 scsi_sdb_cache, GFP_ATOMIC);
977 error = BLKPREP_DEFER;
981 cmd->request->next_rq->special = bidi_sdb;
982 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
988 if (blk_integrity_rq(cmd->request)) {
989 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
992 BUG_ON(prot_sdb == NULL);
993 ivecs = blk_rq_count_integrity_sg(cmd->request);
995 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
996 error = BLKPREP_DEFER;
1000 count = blk_rq_map_integrity_sg(cmd->request,
1001 prot_sdb->table.sgl);
1002 BUG_ON(unlikely(count > ivecs));
1004 cmd->prot_sdb = prot_sdb;
1005 cmd->prot_sdb->table.nents = count;
1011 scsi_release_buffers(cmd);
1012 if (error == BLKPREP_KILL)
1013 scsi_put_command(cmd);
1014 else /* BLKPREP_DEFER */
1015 scsi_unprep_request(cmd->request);
1019 EXPORT_SYMBOL(scsi_init_io);
1021 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1022 struct request *req)
1024 struct scsi_cmnd *cmd;
1026 if (!req->special) {
1027 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1035 /* pull a tag out of the request if we have one */
1036 cmd->tag = req->tag;
1039 cmd->cmnd = req->cmd;
1044 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1046 struct scsi_cmnd *cmd;
1047 int ret = scsi_prep_state_check(sdev, req);
1049 if (ret != BLKPREP_OK)
1052 cmd = scsi_get_cmd_from_req(sdev, req);
1054 return BLKPREP_DEFER;
1057 * BLOCK_PC requests may transfer data, in which case they must
1058 * a bio attached to them. Or they might contain a SCSI command
1059 * that does not transfer data, in which case they may optionally
1060 * submit a request without an attached bio.
1065 BUG_ON(!req->nr_phys_segments);
1067 ret = scsi_init_io(cmd, GFP_ATOMIC);
1071 BUG_ON(blk_rq_bytes(req));
1073 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1077 cmd->cmd_len = req->cmd_len;
1078 if (!blk_rq_bytes(req))
1079 cmd->sc_data_direction = DMA_NONE;
1080 else if (rq_data_dir(req) == WRITE)
1081 cmd->sc_data_direction = DMA_TO_DEVICE;
1083 cmd->sc_data_direction = DMA_FROM_DEVICE;
1085 cmd->transfersize = blk_rq_bytes(req);
1086 cmd->allowed = req->retries;
1089 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1092 * Setup a REQ_TYPE_FS command. These are simple read/write request
1093 * from filesystems that still need to be translated to SCSI CDBs from
1096 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1098 struct scsi_cmnd *cmd;
1099 int ret = scsi_prep_state_check(sdev, req);
1101 if (ret != BLKPREP_OK)
1104 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1105 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1106 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1107 if (ret != BLKPREP_OK)
1112 * Filesystem requests must transfer data.
1114 BUG_ON(!req->nr_phys_segments);
1116 cmd = scsi_get_cmd_from_req(sdev, req);
1118 return BLKPREP_DEFER;
1120 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1121 return scsi_init_io(cmd, GFP_ATOMIC);
1123 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1125 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1127 int ret = BLKPREP_OK;
1130 * If the device is not in running state we will reject some
1133 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1134 switch (sdev->sdev_state) {
1137 * If the device is offline we refuse to process any
1138 * commands. The device must be brought online
1139 * before trying any recovery commands.
1141 sdev_printk(KERN_ERR, sdev,
1142 "rejecting I/O to offline device\n");
1147 * If the device is fully deleted, we refuse to
1148 * process any commands as well.
1150 sdev_printk(KERN_ERR, sdev,
1151 "rejecting I/O to dead device\n");
1156 case SDEV_CREATED_BLOCK:
1158 * If the devices is blocked we defer normal commands.
1160 if (!(req->cmd_flags & REQ_PREEMPT))
1161 ret = BLKPREP_DEFER;
1165 * For any other not fully online state we only allow
1166 * special commands. In particular any user initiated
1167 * command is not allowed.
1169 if (!(req->cmd_flags & REQ_PREEMPT))
1176 EXPORT_SYMBOL(scsi_prep_state_check);
1178 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1180 struct scsi_device *sdev = q->queuedata;
1184 req->errors = DID_NO_CONNECT << 16;
1185 /* release the command and kill it */
1187 struct scsi_cmnd *cmd = req->special;
1188 scsi_release_buffers(cmd);
1189 scsi_put_command(cmd);
1190 req->special = NULL;
1195 * If we defer, the blk_peek_request() returns NULL, but the
1196 * queue must be restarted, so we plug here if no returning
1197 * command will automatically do that.
1199 if (sdev->device_busy == 0)
1203 req->cmd_flags |= REQ_DONTPREP;
1208 EXPORT_SYMBOL(scsi_prep_return);
1210 int scsi_prep_fn(struct request_queue *q, struct request *req)
1212 struct scsi_device *sdev = q->queuedata;
1213 int ret = BLKPREP_KILL;
1215 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1216 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1217 return scsi_prep_return(q, req, ret);
1219 EXPORT_SYMBOL(scsi_prep_fn);
1222 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1225 * Called with the queue_lock held.
1227 static inline int scsi_dev_queue_ready(struct request_queue *q,
1228 struct scsi_device *sdev)
1230 if (sdev->device_busy == 0 && sdev->device_blocked) {
1232 * unblock after device_blocked iterates to zero
1234 if (--sdev->device_blocked == 0) {
1236 sdev_printk(KERN_INFO, sdev,
1237 "unblocking device at zero depth\n"));
1243 if (scsi_device_is_busy(sdev))
1251 * scsi_target_queue_ready: checks if there we can send commands to target
1252 * @sdev: scsi device on starget to check.
1254 * Called with the host lock held.
1256 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1257 struct scsi_device *sdev)
1259 struct scsi_target *starget = scsi_target(sdev);
1261 if (starget->single_lun) {
1262 if (starget->starget_sdev_user &&
1263 starget->starget_sdev_user != sdev)
1265 starget->starget_sdev_user = sdev;
1268 if (starget->target_busy == 0 && starget->target_blocked) {
1270 * unblock after target_blocked iterates to zero
1272 if (--starget->target_blocked == 0) {
1273 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1274 "unblocking target at zero depth\n"));
1279 if (scsi_target_is_busy(starget)) {
1280 if (list_empty(&sdev->starved_entry)) {
1281 list_add_tail(&sdev->starved_entry,
1282 &shost->starved_list);
1287 /* We're OK to process the command, so we can't be starved */
1288 if (!list_empty(&sdev->starved_entry))
1289 list_del_init(&sdev->starved_entry);
1294 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1295 * return 0. We must end up running the queue again whenever 0 is
1296 * returned, else IO can hang.
1298 * Called with host_lock held.
1300 static inline int scsi_host_queue_ready(struct request_queue *q,
1301 struct Scsi_Host *shost,
1302 struct scsi_device *sdev)
1304 if (scsi_host_in_recovery(shost))
1306 if (shost->host_busy == 0 && shost->host_blocked) {
1308 * unblock after host_blocked iterates to zero
1310 if (--shost->host_blocked == 0) {
1312 printk("scsi%d unblocking host at zero depth\n",
1318 if (scsi_host_is_busy(shost)) {
1319 if (list_empty(&sdev->starved_entry))
1320 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1324 /* We're OK to process the command, so we can't be starved */
1325 if (!list_empty(&sdev->starved_entry))
1326 list_del_init(&sdev->starved_entry);
1332 * Busy state exporting function for request stacking drivers.
1334 * For efficiency, no lock is taken to check the busy state of
1335 * shost/starget/sdev, since the returned value is not guaranteed and
1336 * may be changed after request stacking drivers call the function,
1337 * regardless of taking lock or not.
1339 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1340 * (e.g. !sdev), scsi needs to return 'not busy'.
1341 * Otherwise, request stacking drivers may hold requests forever.
1343 static int scsi_lld_busy(struct request_queue *q)
1345 struct scsi_device *sdev = q->queuedata;
1346 struct Scsi_Host *shost;
1347 struct scsi_target *starget;
1353 starget = scsi_target(sdev);
1355 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1356 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1363 * Kill a request for a dead device
1365 static void scsi_kill_request(struct request *req, struct request_queue *q)
1367 struct scsi_cmnd *cmd = req->special;
1368 struct scsi_device *sdev = cmd->device;
1369 struct scsi_target *starget = scsi_target(sdev);
1370 struct Scsi_Host *shost = sdev->host;
1372 blk_start_request(req);
1374 if (unlikely(cmd == NULL)) {
1375 printk(KERN_CRIT "impossible request in %s.\n",
1380 scsi_init_cmd_errh(cmd);
1381 cmd->result = DID_NO_CONNECT << 16;
1382 atomic_inc(&cmd->device->iorequest_cnt);
1385 * SCSI request completion path will do scsi_device_unbusy(),
1386 * bump busy counts. To bump the counters, we need to dance
1387 * with the locks as normal issue path does.
1389 sdev->device_busy++;
1390 spin_unlock(sdev->request_queue->queue_lock);
1391 spin_lock(shost->host_lock);
1393 starget->target_busy++;
1394 spin_unlock(shost->host_lock);
1395 spin_lock(sdev->request_queue->queue_lock);
1397 blk_complete_request(req);
1400 static void scsi_softirq_done(struct request *rq)
1402 struct scsi_cmnd *cmd = rq->special;
1403 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1406 INIT_LIST_HEAD(&cmd->eh_entry);
1409 * Set the serial numbers back to zero
1411 cmd->serial_number = 0;
1413 atomic_inc(&cmd->device->iodone_cnt);
1415 atomic_inc(&cmd->device->ioerr_cnt);
1417 disposition = scsi_decide_disposition(cmd);
1418 if (disposition != SUCCESS &&
1419 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1420 sdev_printk(KERN_ERR, cmd->device,
1421 "timing out command, waited %lus\n",
1423 disposition = SUCCESS;
1426 scsi_log_completion(cmd, disposition);
1428 switch (disposition) {
1430 scsi_finish_command(cmd);
1433 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1435 case ADD_TO_MLQUEUE:
1436 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1439 if (!scsi_eh_scmd_add(cmd, 0))
1440 scsi_finish_command(cmd);
1445 * Function: scsi_request_fn()
1447 * Purpose: Main strategy routine for SCSI.
1449 * Arguments: q - Pointer to actual queue.
1453 * Lock status: IO request lock assumed to be held when called.
1455 static void scsi_request_fn(struct request_queue *q)
1457 struct scsi_device *sdev = q->queuedata;
1458 struct Scsi_Host *shost;
1459 struct scsi_cmnd *cmd;
1460 struct request *req;
1463 printk("scsi: killing requests for dead queue\n");
1464 while ((req = blk_peek_request(q)) != NULL)
1465 scsi_kill_request(req, q);
1469 if(!get_device(&sdev->sdev_gendev))
1470 /* We must be tearing the block queue down already */
1474 * To start with, we keep looping until the queue is empty, or until
1475 * the host is no longer able to accept any more requests.
1478 while (!blk_queue_plugged(q)) {
1481 * get next queueable request. We do this early to make sure
1482 * that the request is fully prepared even if we cannot
1485 req = blk_peek_request(q);
1486 if (!req || !scsi_dev_queue_ready(q, sdev))
1489 if (unlikely(!scsi_device_online(sdev))) {
1490 sdev_printk(KERN_ERR, sdev,
1491 "rejecting I/O to offline device\n");
1492 scsi_kill_request(req, q);
1498 * Remove the request from the request list.
1500 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1501 blk_start_request(req);
1502 sdev->device_busy++;
1504 spin_unlock(q->queue_lock);
1506 if (unlikely(cmd == NULL)) {
1507 printk(KERN_CRIT "impossible request in %s.\n"
1508 "please mail a stack trace to "
1509 "linux-scsi@vger.kernel.org\n",
1511 blk_dump_rq_flags(req, "foo");
1514 spin_lock(shost->host_lock);
1517 * We hit this when the driver is using a host wide
1518 * tag map. For device level tag maps the queue_depth check
1519 * in the device ready fn would prevent us from trying
1520 * to allocate a tag. Since the map is a shared host resource
1521 * we add the dev to the starved list so it eventually gets
1522 * a run when a tag is freed.
1524 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1525 if (list_empty(&sdev->starved_entry))
1526 list_add_tail(&sdev->starved_entry,
1527 &shost->starved_list);
1531 if (!scsi_target_queue_ready(shost, sdev))
1534 if (!scsi_host_queue_ready(q, shost, sdev))
1537 scsi_target(sdev)->target_busy++;
1541 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1542 * take the lock again.
1544 spin_unlock_irq(shost->host_lock);
1547 * Finally, initialize any error handling parameters, and set up
1548 * the timers for timeouts.
1550 scsi_init_cmd_errh(cmd);
1553 * Dispatch the command to the low-level driver.
1555 rtn = scsi_dispatch_cmd(cmd);
1556 spin_lock_irq(q->queue_lock);
1558 /* we're refusing the command; because of
1559 * the way locks get dropped, we need to
1560 * check here if plugging is required */
1561 if(sdev->device_busy == 0)
1571 spin_unlock_irq(shost->host_lock);
1574 * lock q, handle tag, requeue req, and decrement device_busy. We
1575 * must return with queue_lock held.
1577 * Decrementing device_busy without checking it is OK, as all such
1578 * cases (host limits or settings) should run the queue at some
1581 spin_lock_irq(q->queue_lock);
1582 blk_requeue_request(q, req);
1583 sdev->device_busy--;
1584 if(sdev->device_busy == 0)
1587 /* must be careful here...if we trigger the ->remove() function
1588 * we cannot be holding the q lock */
1589 spin_unlock_irq(q->queue_lock);
1590 put_device(&sdev->sdev_gendev);
1591 spin_lock_irq(q->queue_lock);
1594 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1596 struct device *host_dev;
1597 u64 bounce_limit = 0xffffffff;
1599 if (shost->unchecked_isa_dma)
1600 return BLK_BOUNCE_ISA;
1602 * Platforms with virtual-DMA translation
1603 * hardware have no practical limit.
1605 if (!PCI_DMA_BUS_IS_PHYS)
1606 return BLK_BOUNCE_ANY;
1608 host_dev = scsi_get_device(shost);
1609 if (host_dev && host_dev->dma_mask)
1610 bounce_limit = *host_dev->dma_mask;
1612 return bounce_limit;
1614 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1616 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1617 request_fn_proc *request_fn)
1619 struct request_queue *q;
1620 struct device *dev = shost->shost_gendev.parent;
1622 q = blk_init_queue(request_fn, NULL);
1627 * this limit is imposed by hardware restrictions
1629 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1630 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
1632 blk_queue_max_sectors(q, shost->max_sectors);
1633 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1634 blk_queue_segment_boundary(q, shost->dma_boundary);
1635 dma_set_seg_boundary(dev, shost->dma_boundary);
1637 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1639 /* New queue, no concurrency on queue_flags */
1640 if (!shost->use_clustering)
1641 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1644 * set a reasonable default alignment on word boundaries: the
1645 * host and device may alter it using
1646 * blk_queue_update_dma_alignment() later.
1648 blk_queue_dma_alignment(q, 0x03);
1652 EXPORT_SYMBOL(__scsi_alloc_queue);
1654 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1656 struct request_queue *q;
1658 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1662 blk_queue_prep_rq(q, scsi_prep_fn);
1663 blk_queue_softirq_done(q, scsi_softirq_done);
1664 blk_queue_rq_timed_out(q, scsi_times_out);
1665 blk_queue_lld_busy(q, scsi_lld_busy);
1669 void scsi_free_queue(struct request_queue *q)
1671 blk_cleanup_queue(q);
1675 * Function: scsi_block_requests()
1677 * Purpose: Utility function used by low-level drivers to prevent further
1678 * commands from being queued to the device.
1680 * Arguments: shost - Host in question
1684 * Lock status: No locks are assumed held.
1686 * Notes: There is no timer nor any other means by which the requests
1687 * get unblocked other than the low-level driver calling
1688 * scsi_unblock_requests().
1690 void scsi_block_requests(struct Scsi_Host *shost)
1692 shost->host_self_blocked = 1;
1694 EXPORT_SYMBOL(scsi_block_requests);
1697 * Function: scsi_unblock_requests()
1699 * Purpose: Utility function used by low-level drivers to allow further
1700 * commands from being queued to the device.
1702 * Arguments: shost - Host in question
1706 * Lock status: No locks are assumed held.
1708 * Notes: There is no timer nor any other means by which the requests
1709 * get unblocked other than the low-level driver calling
1710 * scsi_unblock_requests().
1712 * This is done as an API function so that changes to the
1713 * internals of the scsi mid-layer won't require wholesale
1714 * changes to drivers that use this feature.
1716 void scsi_unblock_requests(struct Scsi_Host *shost)
1718 shost->host_self_blocked = 0;
1719 scsi_run_host_queues(shost);
1721 EXPORT_SYMBOL(scsi_unblock_requests);
1723 int __init scsi_init_queue(void)
1727 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1728 sizeof(struct scsi_data_buffer),
1730 if (!scsi_sdb_cache) {
1731 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1735 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1736 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1737 int size = sgp->size * sizeof(struct scatterlist);
1739 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1740 SLAB_HWCACHE_ALIGN, NULL);
1742 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1747 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1750 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1759 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1760 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1762 mempool_destroy(sgp->pool);
1764 kmem_cache_destroy(sgp->slab);
1766 kmem_cache_destroy(scsi_sdb_cache);
1771 void scsi_exit_queue(void)
1775 kmem_cache_destroy(scsi_sdb_cache);
1777 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1778 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1779 mempool_destroy(sgp->pool);
1780 kmem_cache_destroy(sgp->slab);
1785 * scsi_mode_select - issue a mode select
1786 * @sdev: SCSI device to be queried
1787 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1788 * @sp: Save page bit (0 == don't save, 1 == save)
1789 * @modepage: mode page being requested
1790 * @buffer: request buffer (may not be smaller than eight bytes)
1791 * @len: length of request buffer.
1792 * @timeout: command timeout
1793 * @retries: number of retries before failing
1794 * @data: returns a structure abstracting the mode header data
1795 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1796 * must be SCSI_SENSE_BUFFERSIZE big.
1798 * Returns zero if successful; negative error number or scsi
1803 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1804 unsigned char *buffer, int len, int timeout, int retries,
1805 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1807 unsigned char cmd[10];
1808 unsigned char *real_buffer;
1811 memset(cmd, 0, sizeof(cmd));
1812 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1814 if (sdev->use_10_for_ms) {
1817 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1820 memcpy(real_buffer + 8, buffer, len);
1824 real_buffer[2] = data->medium_type;
1825 real_buffer[3] = data->device_specific;
1826 real_buffer[4] = data->longlba ? 0x01 : 0;
1828 real_buffer[6] = data->block_descriptor_length >> 8;
1829 real_buffer[7] = data->block_descriptor_length;
1831 cmd[0] = MODE_SELECT_10;
1835 if (len > 255 || data->block_descriptor_length > 255 ||
1839 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1842 memcpy(real_buffer + 4, buffer, len);
1845 real_buffer[1] = data->medium_type;
1846 real_buffer[2] = data->device_specific;
1847 real_buffer[3] = data->block_descriptor_length;
1850 cmd[0] = MODE_SELECT;
1854 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1855 sshdr, timeout, retries, NULL);
1859 EXPORT_SYMBOL_GPL(scsi_mode_select);
1862 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1863 * @sdev: SCSI device to be queried
1864 * @dbd: set if mode sense will allow block descriptors to be returned
1865 * @modepage: mode page being requested
1866 * @buffer: request buffer (may not be smaller than eight bytes)
1867 * @len: length of request buffer.
1868 * @timeout: command timeout
1869 * @retries: number of retries before failing
1870 * @data: returns a structure abstracting the mode header data
1871 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1872 * must be SCSI_SENSE_BUFFERSIZE big.
1874 * Returns zero if unsuccessful, or the header offset (either 4
1875 * or 8 depending on whether a six or ten byte command was
1876 * issued) if successful.
1879 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1880 unsigned char *buffer, int len, int timeout, int retries,
1881 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1883 unsigned char cmd[12];
1887 struct scsi_sense_hdr my_sshdr;
1889 memset(data, 0, sizeof(*data));
1890 memset(&cmd[0], 0, 12);
1891 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1894 /* caller might not be interested in sense, but we need it */
1899 use_10_for_ms = sdev->use_10_for_ms;
1901 if (use_10_for_ms) {
1905 cmd[0] = MODE_SENSE_10;
1912 cmd[0] = MODE_SENSE;
1917 memset(buffer, 0, len);
1919 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1920 sshdr, timeout, retries, NULL);
1922 /* This code looks awful: what it's doing is making sure an
1923 * ILLEGAL REQUEST sense return identifies the actual command
1924 * byte as the problem. MODE_SENSE commands can return
1925 * ILLEGAL REQUEST if the code page isn't supported */
1927 if (use_10_for_ms && !scsi_status_is_good(result) &&
1928 (driver_byte(result) & DRIVER_SENSE)) {
1929 if (scsi_sense_valid(sshdr)) {
1930 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1931 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1933 * Invalid command operation code
1935 sdev->use_10_for_ms = 0;
1941 if(scsi_status_is_good(result)) {
1942 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1943 (modepage == 6 || modepage == 8))) {
1944 /* Initio breakage? */
1947 data->medium_type = 0;
1948 data->device_specific = 0;
1950 data->block_descriptor_length = 0;
1951 } else if(use_10_for_ms) {
1952 data->length = buffer[0]*256 + buffer[1] + 2;
1953 data->medium_type = buffer[2];
1954 data->device_specific = buffer[3];
1955 data->longlba = buffer[4] & 0x01;
1956 data->block_descriptor_length = buffer[6]*256
1959 data->length = buffer[0] + 1;
1960 data->medium_type = buffer[1];
1961 data->device_specific = buffer[2];
1962 data->block_descriptor_length = buffer[3];
1964 data->header_length = header_length;
1969 EXPORT_SYMBOL(scsi_mode_sense);
1972 * scsi_test_unit_ready - test if unit is ready
1973 * @sdev: scsi device to change the state of.
1974 * @timeout: command timeout
1975 * @retries: number of retries before failing
1976 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1977 * returning sense. Make sure that this is cleared before passing
1980 * Returns zero if unsuccessful or an error if TUR failed. For
1981 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1982 * translated to success, with the ->changed flag updated.
1985 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1986 struct scsi_sense_hdr *sshdr_external)
1989 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1991 struct scsi_sense_hdr *sshdr;
1994 if (!sshdr_external)
1995 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1997 sshdr = sshdr_external;
1999 /* try to eat the UNIT_ATTENTION if there are enough retries */
2001 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2002 timeout, retries, NULL);
2003 if (sdev->removable && scsi_sense_valid(sshdr) &&
2004 sshdr->sense_key == UNIT_ATTENTION)
2006 } while (scsi_sense_valid(sshdr) &&
2007 sshdr->sense_key == UNIT_ATTENTION && --retries);
2010 /* could not allocate sense buffer, so can't process it */
2013 if (sdev->removable && scsi_sense_valid(sshdr) &&
2014 (sshdr->sense_key == UNIT_ATTENTION ||
2015 sshdr->sense_key == NOT_READY)) {
2019 if (!sshdr_external)
2023 EXPORT_SYMBOL(scsi_test_unit_ready);
2026 * scsi_device_set_state - Take the given device through the device state model.
2027 * @sdev: scsi device to change the state of.
2028 * @state: state to change to.
2030 * Returns zero if unsuccessful or an error if the requested
2031 * transition is illegal.
2034 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2036 enum scsi_device_state oldstate = sdev->sdev_state;
2038 if (state == oldstate)
2044 case SDEV_CREATED_BLOCK:
2088 case SDEV_CREATED_BLOCK:
2095 case SDEV_CREATED_BLOCK:
2130 sdev->sdev_state = state;
2134 SCSI_LOG_ERROR_RECOVERY(1,
2135 sdev_printk(KERN_ERR, sdev,
2136 "Illegal state transition %s->%s\n",
2137 scsi_device_state_name(oldstate),
2138 scsi_device_state_name(state))
2142 EXPORT_SYMBOL(scsi_device_set_state);
2145 * sdev_evt_emit - emit a single SCSI device uevent
2146 * @sdev: associated SCSI device
2147 * @evt: event to emit
2149 * Send a single uevent (scsi_event) to the associated scsi_device.
2151 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2156 switch (evt->evt_type) {
2157 case SDEV_EVT_MEDIA_CHANGE:
2158 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2168 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2172 * sdev_evt_thread - send a uevent for each scsi event
2173 * @work: work struct for scsi_device
2175 * Dispatch queued events to their associated scsi_device kobjects
2178 void scsi_evt_thread(struct work_struct *work)
2180 struct scsi_device *sdev;
2181 LIST_HEAD(event_list);
2183 sdev = container_of(work, struct scsi_device, event_work);
2186 struct scsi_event *evt;
2187 struct list_head *this, *tmp;
2188 unsigned long flags;
2190 spin_lock_irqsave(&sdev->list_lock, flags);
2191 list_splice_init(&sdev->event_list, &event_list);
2192 spin_unlock_irqrestore(&sdev->list_lock, flags);
2194 if (list_empty(&event_list))
2197 list_for_each_safe(this, tmp, &event_list) {
2198 evt = list_entry(this, struct scsi_event, node);
2199 list_del(&evt->node);
2200 scsi_evt_emit(sdev, evt);
2207 * sdev_evt_send - send asserted event to uevent thread
2208 * @sdev: scsi_device event occurred on
2209 * @evt: event to send
2211 * Assert scsi device event asynchronously.
2213 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2215 unsigned long flags;
2218 /* FIXME: currently this check eliminates all media change events
2219 * for polled devices. Need to update to discriminate between AN
2220 * and polled events */
2221 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2227 spin_lock_irqsave(&sdev->list_lock, flags);
2228 list_add_tail(&evt->node, &sdev->event_list);
2229 schedule_work(&sdev->event_work);
2230 spin_unlock_irqrestore(&sdev->list_lock, flags);
2232 EXPORT_SYMBOL_GPL(sdev_evt_send);
2235 * sdev_evt_alloc - allocate a new scsi event
2236 * @evt_type: type of event to allocate
2237 * @gfpflags: GFP flags for allocation
2239 * Allocates and returns a new scsi_event.
2241 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2244 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2248 evt->evt_type = evt_type;
2249 INIT_LIST_HEAD(&evt->node);
2251 /* evt_type-specific initialization, if any */
2253 case SDEV_EVT_MEDIA_CHANGE:
2261 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2264 * sdev_evt_send_simple - send asserted event to uevent thread
2265 * @sdev: scsi_device event occurred on
2266 * @evt_type: type of event to send
2267 * @gfpflags: GFP flags for allocation
2269 * Assert scsi device event asynchronously, given an event type.
2271 void sdev_evt_send_simple(struct scsi_device *sdev,
2272 enum scsi_device_event evt_type, gfp_t gfpflags)
2274 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2276 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2281 sdev_evt_send(sdev, evt);
2283 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2286 * scsi_device_quiesce - Block user issued commands.
2287 * @sdev: scsi device to quiesce.
2289 * This works by trying to transition to the SDEV_QUIESCE state
2290 * (which must be a legal transition). When the device is in this
2291 * state, only special requests will be accepted, all others will
2292 * be deferred. Since special requests may also be requeued requests,
2293 * a successful return doesn't guarantee the device will be
2294 * totally quiescent.
2296 * Must be called with user context, may sleep.
2298 * Returns zero if unsuccessful or an error if not.
2301 scsi_device_quiesce(struct scsi_device *sdev)
2303 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2307 scsi_run_queue(sdev->request_queue);
2308 while (sdev->device_busy) {
2309 msleep_interruptible(200);
2310 scsi_run_queue(sdev->request_queue);
2314 EXPORT_SYMBOL(scsi_device_quiesce);
2317 * scsi_device_resume - Restart user issued commands to a quiesced device.
2318 * @sdev: scsi device to resume.
2320 * Moves the device from quiesced back to running and restarts the
2323 * Must be called with user context, may sleep.
2326 scsi_device_resume(struct scsi_device *sdev)
2328 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2330 scsi_run_queue(sdev->request_queue);
2332 EXPORT_SYMBOL(scsi_device_resume);
2335 device_quiesce_fn(struct scsi_device *sdev, void *data)
2337 scsi_device_quiesce(sdev);
2341 scsi_target_quiesce(struct scsi_target *starget)
2343 starget_for_each_device(starget, NULL, device_quiesce_fn);
2345 EXPORT_SYMBOL(scsi_target_quiesce);
2348 device_resume_fn(struct scsi_device *sdev, void *data)
2350 scsi_device_resume(sdev);
2354 scsi_target_resume(struct scsi_target *starget)
2356 starget_for_each_device(starget, NULL, device_resume_fn);
2358 EXPORT_SYMBOL(scsi_target_resume);
2361 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2362 * @sdev: device to block
2364 * Block request made by scsi lld's to temporarily stop all
2365 * scsi commands on the specified device. Called from interrupt
2366 * or normal process context.
2368 * Returns zero if successful or error if not
2371 * This routine transitions the device to the SDEV_BLOCK state
2372 * (which must be a legal transition). When the device is in this
2373 * state, all commands are deferred until the scsi lld reenables
2374 * the device with scsi_device_unblock or device_block_tmo fires.
2375 * This routine assumes the host_lock is held on entry.
2378 scsi_internal_device_block(struct scsi_device *sdev)
2380 struct request_queue *q = sdev->request_queue;
2381 unsigned long flags;
2384 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2386 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2393 * The device has transitioned to SDEV_BLOCK. Stop the
2394 * block layer from calling the midlayer with this device's
2397 spin_lock_irqsave(q->queue_lock, flags);
2399 spin_unlock_irqrestore(q->queue_lock, flags);
2403 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2406 * scsi_internal_device_unblock - resume a device after a block request
2407 * @sdev: device to resume
2409 * Called by scsi lld's or the midlayer to restart the device queue
2410 * for the previously suspended scsi device. Called from interrupt or
2411 * normal process context.
2413 * Returns zero if successful or error if not.
2416 * This routine transitions the device to the SDEV_RUNNING state
2417 * (which must be a legal transition) allowing the midlayer to
2418 * goose the queue for this device. This routine assumes the
2419 * host_lock is held upon entry.
2422 scsi_internal_device_unblock(struct scsi_device *sdev)
2424 struct request_queue *q = sdev->request_queue;
2425 unsigned long flags;
2428 * Try to transition the scsi device to SDEV_RUNNING
2429 * and goose the device queue if successful.
2431 if (sdev->sdev_state == SDEV_BLOCK)
2432 sdev->sdev_state = SDEV_RUNNING;
2433 else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2434 sdev->sdev_state = SDEV_CREATED;
2435 else if (sdev->sdev_state != SDEV_CANCEL &&
2436 sdev->sdev_state != SDEV_OFFLINE)
2439 spin_lock_irqsave(q->queue_lock, flags);
2441 spin_unlock_irqrestore(q->queue_lock, flags);
2445 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2448 device_block(struct scsi_device *sdev, void *data)
2450 scsi_internal_device_block(sdev);
2454 target_block(struct device *dev, void *data)
2456 if (scsi_is_target_device(dev))
2457 starget_for_each_device(to_scsi_target(dev), NULL,
2463 scsi_target_block(struct device *dev)
2465 if (scsi_is_target_device(dev))
2466 starget_for_each_device(to_scsi_target(dev), NULL,
2469 device_for_each_child(dev, NULL, target_block);
2471 EXPORT_SYMBOL_GPL(scsi_target_block);
2474 device_unblock(struct scsi_device *sdev, void *data)
2476 scsi_internal_device_unblock(sdev);
2480 target_unblock(struct device *dev, void *data)
2482 if (scsi_is_target_device(dev))
2483 starget_for_each_device(to_scsi_target(dev), NULL,
2489 scsi_target_unblock(struct device *dev)
2491 if (scsi_is_target_device(dev))
2492 starget_for_each_device(to_scsi_target(dev), NULL,
2495 device_for_each_child(dev, NULL, target_unblock);
2497 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2500 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2501 * @sgl: scatter-gather list
2502 * @sg_count: number of segments in sg
2503 * @offset: offset in bytes into sg, on return offset into the mapped area
2504 * @len: bytes to map, on return number of bytes mapped
2506 * Returns virtual address of the start of the mapped page
2508 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2509 size_t *offset, size_t *len)
2512 size_t sg_len = 0, len_complete = 0;
2513 struct scatterlist *sg;
2516 WARN_ON(!irqs_disabled());
2518 for_each_sg(sgl, sg, sg_count, i) {
2519 len_complete = sg_len; /* Complete sg-entries */
2520 sg_len += sg->length;
2521 if (sg_len > *offset)
2525 if (unlikely(i == sg_count)) {
2526 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2528 __func__, sg_len, *offset, sg_count);
2533 /* Offset starting from the beginning of first page in this sg-entry */
2534 *offset = *offset - len_complete + sg->offset;
2536 /* Assumption: contiguous pages can be accessed as "page + i" */
2537 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2538 *offset &= ~PAGE_MASK;
2540 /* Bytes in this sg-entry from *offset to the end of the page */
2541 sg_len = PAGE_SIZE - *offset;
2545 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2547 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2550 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2551 * @virt: virtual address to be unmapped
2553 void scsi_kunmap_atomic_sg(void *virt)
2555 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2557 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);