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. Schedule requeue work under
159 * lock such that the kblockd_schedule_work() call happens
160 * before blk_cleanup_queue() finishes.
162 spin_lock_irqsave(q->queue_lock, flags);
163 blk_requeue_request(q, cmd->request);
164 kblockd_schedule_work(q, &device->requeue_work);
165 spin_unlock_irqrestore(q->queue_lock, flags);
169 * Function: scsi_queue_insert()
171 * Purpose: Insert a command in the midlevel queue.
173 * Arguments: cmd - command that we are adding to queue.
174 * reason - why we are inserting command to queue.
176 * Lock status: Assumed that lock is not held upon entry.
180 * Notes: We do this for one of two cases. Either the host is busy
181 * and it cannot accept any more commands for the time being,
182 * or the device returned QUEUE_FULL and can accept no more
184 * Notes: This could be called either from an interrupt context or a
185 * normal process context.
187 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
189 __scsi_queue_insert(cmd, reason, 1);
192 * scsi_execute - insert request and wait for the result
195 * @data_direction: data direction
196 * @buffer: data buffer
197 * @bufflen: len of buffer
198 * @sense: optional sense buffer
199 * @timeout: request timeout in seconds
200 * @retries: number of times to retry request
201 * @flags: or into request flags;
202 * @resid: optional residual length
204 * returns the req->errors value which is the scsi_cmnd result
207 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
208 int data_direction, void *buffer, unsigned bufflen,
209 unsigned char *sense, int timeout, int retries, int flags,
213 int write = (data_direction == DMA_TO_DEVICE);
214 int ret = DRIVER_ERROR << 24;
216 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
220 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
221 buffer, bufflen, __GFP_WAIT))
224 req->cmd_len = COMMAND_SIZE(cmd[0]);
225 memcpy(req->cmd, cmd, req->cmd_len);
228 req->retries = retries;
229 req->timeout = timeout;
230 req->cmd_type = REQ_TYPE_BLOCK_PC;
231 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
234 * head injection *required* here otherwise quiesce won't work
236 blk_execute_rq(req->q, NULL, req, 1);
239 * Some devices (USB mass-storage in particular) may transfer
240 * garbage data together with a residue indicating that the data
241 * is invalid. Prevent the garbage from being misinterpreted
242 * and prevent security leaks by zeroing out the excess data.
244 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
245 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
248 *resid = req->resid_len;
251 blk_put_request(req);
255 EXPORT_SYMBOL(scsi_execute);
258 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
259 int data_direction, void *buffer, unsigned bufflen,
260 struct scsi_sense_hdr *sshdr, int timeout, int retries,
267 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
269 return DRIVER_ERROR << 24;
271 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
272 sense, timeout, retries, 0, resid);
274 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
279 EXPORT_SYMBOL(scsi_execute_req);
282 * Function: scsi_init_cmd_errh()
284 * Purpose: Initialize cmd fields related to error handling.
286 * Arguments: cmd - command that is ready to be queued.
288 * Notes: This function has the job of initializing a number of
289 * fields related to error handling. Typically this will
290 * be called once for each command, as required.
292 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
294 cmd->serial_number = 0;
295 scsi_set_resid(cmd, 0);
296 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
297 if (cmd->cmd_len == 0)
298 cmd->cmd_len = scsi_command_size(cmd->cmnd);
301 void scsi_device_unbusy(struct scsi_device *sdev)
303 struct Scsi_Host *shost = sdev->host;
304 struct scsi_target *starget = scsi_target(sdev);
307 spin_lock_irqsave(shost->host_lock, flags);
309 starget->target_busy--;
310 if (unlikely(scsi_host_in_recovery(shost) &&
311 (shost->host_failed || shost->host_eh_scheduled)))
312 scsi_eh_wakeup(shost);
313 spin_unlock(shost->host_lock);
314 spin_lock(sdev->request_queue->queue_lock);
316 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
320 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
321 * and call blk_run_queue for all the scsi_devices on the target -
322 * including current_sdev first.
324 * Called with *no* scsi locks held.
326 static void scsi_single_lun_run(struct scsi_device *current_sdev)
328 struct Scsi_Host *shost = current_sdev->host;
329 struct scsi_device *sdev, *tmp;
330 struct scsi_target *starget = scsi_target(current_sdev);
333 spin_lock_irqsave(shost->host_lock, flags);
334 starget->starget_sdev_user = NULL;
335 spin_unlock_irqrestore(shost->host_lock, flags);
338 * Call blk_run_queue for all LUNs on the target, starting with
339 * current_sdev. We race with others (to set starget_sdev_user),
340 * but in most cases, we will be first. Ideally, each LU on the
341 * target would get some limited time or requests on the target.
343 blk_run_queue(current_sdev->request_queue);
345 spin_lock_irqsave(shost->host_lock, flags);
346 if (starget->starget_sdev_user)
348 list_for_each_entry_safe(sdev, tmp, &starget->devices,
349 same_target_siblings) {
350 if (sdev == current_sdev)
352 if (scsi_device_get(sdev))
355 spin_unlock_irqrestore(shost->host_lock, flags);
356 blk_run_queue(sdev->request_queue);
357 spin_lock_irqsave(shost->host_lock, flags);
359 scsi_device_put(sdev);
362 spin_unlock_irqrestore(shost->host_lock, flags);
365 static inline int scsi_device_is_busy(struct scsi_device *sdev)
367 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
373 static inline int scsi_target_is_busy(struct scsi_target *starget)
375 return ((starget->can_queue > 0 &&
376 starget->target_busy >= starget->can_queue) ||
377 starget->target_blocked);
380 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
382 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
383 shost->host_blocked || shost->host_self_blocked)
390 * Function: scsi_run_queue()
392 * Purpose: Select a proper request queue to serve next
394 * Arguments: q - last request's queue
398 * Notes: The previous command was completely finished, start
399 * a new one if possible.
401 static void scsi_run_queue(struct request_queue *q)
403 struct scsi_device *sdev = q->queuedata;
404 struct Scsi_Host *shost;
405 LIST_HEAD(starved_list);
409 if (scsi_target(sdev)->single_lun)
410 scsi_single_lun_run(sdev);
412 spin_lock_irqsave(shost->host_lock, flags);
413 list_splice_init(&shost->starved_list, &starved_list);
415 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);
439 spin_lock(sdev->request_queue->queue_lock);
440 __blk_run_queue(sdev->request_queue);
441 spin_unlock(sdev->request_queue->queue_lock);
442 spin_lock(shost->host_lock);
444 /* put any unprocessed entries back */
445 list_splice(&starved_list, &shost->starved_list);
446 spin_unlock_irqrestore(shost->host_lock, flags);
451 void scsi_requeue_run_queue(struct work_struct *work)
453 struct scsi_device *sdev;
454 struct request_queue *q;
456 sdev = container_of(work, struct scsi_device, requeue_work);
457 q = sdev->request_queue;
462 * Function: scsi_requeue_command()
464 * Purpose: Handle post-processing of completed commands.
466 * Arguments: q - queue to operate on
467 * cmd - command that may need to be requeued.
471 * Notes: After command completion, there may be blocks left
472 * over which weren't finished by the previous command
473 * this can be for a number of reasons - the main one is
474 * I/O errors in the middle of the request, in which case
475 * we need to request the blocks that come after the bad
477 * Notes: Upon return, cmd is a stale pointer.
479 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
481 struct scsi_device *sdev = cmd->device;
482 struct request *req = cmd->request;
486 * We need to hold a reference on the device to avoid the queue being
487 * killed after the unlock and before scsi_run_queue is invoked which
488 * may happen because scsi_unprep_request() puts the command which
489 * releases its reference on the device.
491 get_device(&sdev->sdev_gendev);
493 spin_lock_irqsave(q->queue_lock, flags);
494 scsi_unprep_request(req);
495 blk_requeue_request(q, req);
496 spin_unlock_irqrestore(q->queue_lock, flags);
500 put_device(&sdev->sdev_gendev);
503 void scsi_next_command(struct scsi_cmnd *cmd)
505 struct scsi_device *sdev = cmd->device;
506 struct request_queue *q = sdev->request_queue;
508 /* need to hold a reference on the device before we let go of the cmd */
509 get_device(&sdev->sdev_gendev);
511 scsi_put_command(cmd);
514 /* ok to remove device now */
515 put_device(&sdev->sdev_gendev);
518 void scsi_run_host_queues(struct Scsi_Host *shost)
520 struct scsi_device *sdev;
522 shost_for_each_device(sdev, shost)
523 scsi_run_queue(sdev->request_queue);
526 static void __scsi_release_buffers(struct scsi_cmnd *, int);
529 * Function: scsi_end_request()
531 * Purpose: Post-processing of completed commands (usually invoked at end
532 * of upper level post-processing and scsi_io_completion).
534 * Arguments: cmd - command that is complete.
535 * error - 0 if I/O indicates success, < 0 for I/O error.
536 * bytes - number of bytes of completed I/O
537 * requeue - indicates whether we should requeue leftovers.
539 * Lock status: Assumed that lock is not held upon entry.
541 * Returns: cmd if requeue required, NULL otherwise.
543 * Notes: This is called for block device requests in order to
544 * mark some number of sectors as complete.
546 * We are guaranteeing that the request queue will be goosed
547 * at some point during this call.
548 * Notes: If cmd was requeued, upon return it will be a stale pointer.
550 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
551 int bytes, int requeue)
553 struct request_queue *q = cmd->device->request_queue;
554 struct request *req = cmd->request;
557 * If there are blocks left over at the end, set up the command
558 * to queue the remainder of them.
560 if (blk_end_request(req, error, bytes)) {
561 /* kill remainder if no retrys */
562 if (error && scsi_noretry_cmd(cmd))
563 blk_end_request_all(req, error);
567 * Bleah. Leftovers again. Stick the
568 * leftovers in the front of the
569 * queue, and goose the queue again.
571 scsi_release_buffers(cmd);
572 scsi_requeue_command(q, cmd);
580 * This will goose the queue request function at the end, so we don't
581 * need to worry about launching another command.
583 __scsi_release_buffers(cmd, 0);
584 scsi_next_command(cmd);
588 static inline unsigned int scsi_sgtable_index(unsigned short nents)
592 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
597 index = get_count_order(nents) - 3;
602 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
604 struct scsi_host_sg_pool *sgp;
606 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
607 mempool_free(sgl, sgp->pool);
610 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
612 struct scsi_host_sg_pool *sgp;
614 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
615 return mempool_alloc(sgp->pool, gfp_mask);
618 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
625 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
626 gfp_mask, scsi_sg_alloc);
628 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
634 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
636 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
639 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
642 if (cmd->sdb.table.nents)
643 scsi_free_sgtable(&cmd->sdb);
645 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
647 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
648 struct scsi_data_buffer *bidi_sdb =
649 cmd->request->next_rq->special;
650 scsi_free_sgtable(bidi_sdb);
651 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
652 cmd->request->next_rq->special = NULL;
655 if (scsi_prot_sg_count(cmd))
656 scsi_free_sgtable(cmd->prot_sdb);
660 * Function: scsi_release_buffers()
662 * Purpose: Completion processing for block device I/O requests.
664 * Arguments: cmd - command that we are bailing.
666 * Lock status: Assumed that no lock is held upon entry.
670 * Notes: In the event that an upper level driver rejects a
671 * command, we must release resources allocated during
672 * the __init_io() function. Primarily this would involve
673 * the scatter-gather table, and potentially any bounce
676 void scsi_release_buffers(struct scsi_cmnd *cmd)
678 __scsi_release_buffers(cmd, 1);
680 EXPORT_SYMBOL(scsi_release_buffers);
682 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
686 switch(host_byte(result)) {
687 case DID_TRANSPORT_FAILFAST:
690 case DID_TARGET_FAILURE:
691 set_host_byte(cmd, DID_OK);
694 case DID_NEXUS_FAILURE:
695 set_host_byte(cmd, DID_OK);
707 * Function: scsi_io_completion()
709 * Purpose: Completion processing for block device I/O requests.
711 * Arguments: cmd - command that is finished.
713 * Lock status: Assumed that no lock is held upon entry.
717 * Notes: This function is matched in terms of capabilities to
718 * the function that created the scatter-gather list.
719 * In other words, if there are no bounce buffers
720 * (the normal case for most drivers), we don't need
721 * the logic to deal with cleaning up afterwards.
723 * We must call scsi_end_request(). This will finish off
724 * the specified number of sectors. If we are done, the
725 * command block will be released and the queue function
726 * will be goosed. If we are not done then we have to
727 * figure out what to do next:
729 * a) We can call scsi_requeue_command(). The request
730 * will be unprepared and put back on the queue. Then
731 * a new command will be created for it. This should
732 * be used if we made forward progress, or if we want
733 * to switch from READ(10) to READ(6) for example.
735 * b) We can call scsi_queue_insert(). The request will
736 * be put back on the queue and retried using the same
737 * command as before, possibly after a delay.
739 * c) We can call blk_end_request() with -EIO to fail
740 * the remainder of the request.
742 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
744 int result = cmd->result;
745 struct request_queue *q = cmd->device->request_queue;
746 struct request *req = cmd->request;
748 struct scsi_sense_hdr sshdr;
750 int sense_deferred = 0;
751 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
752 ACTION_DELAYED_RETRY} action;
753 char *description = NULL;
756 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
758 sense_deferred = scsi_sense_is_deferred(&sshdr);
761 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
762 req->errors = result;
764 if (sense_valid && req->sense) {
766 * SG_IO wants current and deferred errors
768 int len = 8 + cmd->sense_buffer[7];
770 if (len > SCSI_SENSE_BUFFERSIZE)
771 len = SCSI_SENSE_BUFFERSIZE;
772 memcpy(req->sense, cmd->sense_buffer, len);
773 req->sense_len = len;
776 error = __scsi_error_from_host_byte(cmd, result);
779 req->resid_len = scsi_get_resid(cmd);
781 if (scsi_bidi_cmnd(cmd)) {
783 * Bidi commands Must be complete as a whole,
784 * both sides at once.
786 req->next_rq->resid_len = scsi_in(cmd)->resid;
788 scsi_release_buffers(cmd);
789 blk_end_request_all(req, 0);
791 scsi_next_command(cmd);
796 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
797 BUG_ON(blk_bidi_rq(req));
800 * Next deal with any sectors which we were able to correctly
803 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
805 blk_rq_sectors(req), good_bytes));
808 * Recovered errors need reporting, but they're always treated
809 * as success, so fiddle the result code here. For BLOCK_PC
810 * we already took a copy of the original into rq->errors which
811 * is what gets returned to the user
813 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
814 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
815 * print since caller wants ATA registers. Only occurs on
816 * SCSI ATA PASS_THROUGH commands when CK_COND=1
818 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
820 else if (!(req->cmd_flags & REQ_QUIET))
821 scsi_print_sense("", cmd);
823 /* BLOCK_PC may have set error */
828 * A number of bytes were successfully read. If there
829 * are leftovers and there is some kind of error
830 * (result != 0), retry the rest.
832 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
835 error = __scsi_error_from_host_byte(cmd, result);
837 if (host_byte(result) == DID_RESET) {
838 /* Third party bus reset or reset for error recovery
839 * reasons. Just retry the command and see what
842 action = ACTION_RETRY;
843 } else if (sense_valid && !sense_deferred) {
844 switch (sshdr.sense_key) {
846 if (cmd->device->removable) {
847 /* Detected disc change. Set a bit
848 * and quietly refuse further access.
850 cmd->device->changed = 1;
851 description = "Media Changed";
852 action = ACTION_FAIL;
854 /* Must have been a power glitch, or a
855 * bus reset. Could not have been a
856 * media change, so we just retry the
857 * command and see what happens.
859 action = ACTION_RETRY;
862 case ILLEGAL_REQUEST:
863 /* If we had an ILLEGAL REQUEST returned, then
864 * we may have performed an unsupported
865 * command. The only thing this should be
866 * would be a ten byte read where only a six
867 * byte read was supported. Also, on a system
868 * where READ CAPACITY failed, we may have
869 * read past the end of the disk.
871 if ((cmd->device->use_10_for_rw &&
872 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
873 (cmd->cmnd[0] == READ_10 ||
874 cmd->cmnd[0] == WRITE_10)) {
875 /* This will issue a new 6-byte command. */
876 cmd->device->use_10_for_rw = 0;
877 action = ACTION_REPREP;
878 } else if (sshdr.asc == 0x10) /* DIX */ {
879 description = "Host Data Integrity Failure";
880 action = ACTION_FAIL;
882 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
883 } else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
884 (cmd->cmnd[0] == UNMAP ||
885 cmd->cmnd[0] == WRITE_SAME_16 ||
886 cmd->cmnd[0] == WRITE_SAME)) {
887 description = "Discard failure";
888 action = ACTION_FAIL;
891 action = ACTION_FAIL;
893 case ABORTED_COMMAND:
894 action = ACTION_FAIL;
895 if (sshdr.asc == 0x10) { /* DIF */
896 description = "Target Data Integrity Failure";
901 /* If the device is in the process of becoming
902 * ready, or has a temporary blockage, retry.
904 if (sshdr.asc == 0x04) {
905 switch (sshdr.ascq) {
906 case 0x01: /* becoming ready */
907 case 0x04: /* format in progress */
908 case 0x05: /* rebuild in progress */
909 case 0x06: /* recalculation in progress */
910 case 0x07: /* operation in progress */
911 case 0x08: /* Long write in progress */
912 case 0x09: /* self test in progress */
913 case 0x14: /* space allocation in progress */
914 action = ACTION_DELAYED_RETRY;
917 description = "Device not ready";
918 action = ACTION_FAIL;
922 description = "Device not ready";
923 action = ACTION_FAIL;
926 case VOLUME_OVERFLOW:
927 /* See SSC3rXX or current. */
928 action = ACTION_FAIL;
931 description = "Unhandled sense code";
932 action = ACTION_FAIL;
936 description = "Unhandled error code";
937 action = ACTION_FAIL;
942 /* Give up and fail the remainder of the request */
943 scsi_release_buffers(cmd);
944 if (!(req->cmd_flags & REQ_QUIET)) {
946 scmd_printk(KERN_INFO, cmd, "%s\n",
948 scsi_print_result(cmd);
949 if (driver_byte(result) & DRIVER_SENSE)
950 scsi_print_sense("", cmd);
951 scsi_print_command(cmd);
953 if (blk_end_request_err(req, error))
954 scsi_requeue_command(q, cmd);
956 scsi_next_command(cmd);
959 /* Unprep the request and put it back at the head of the queue.
960 * A new command will be prepared and issued.
962 scsi_release_buffers(cmd);
963 scsi_requeue_command(q, cmd);
966 /* Retry the same command immediately */
967 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
969 case ACTION_DELAYED_RETRY:
970 /* Retry the same command after a delay */
971 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
976 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
982 * If sg table allocation fails, requeue request later.
984 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
986 return BLKPREP_DEFER;
992 * Next, walk the list, and fill in the addresses and sizes of
995 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
996 BUG_ON(count > sdb->table.nents);
997 sdb->table.nents = count;
998 sdb->length = blk_rq_bytes(req);
1003 * Function: scsi_init_io()
1005 * Purpose: SCSI I/O initialize function.
1007 * Arguments: cmd - Command descriptor we wish to initialize
1009 * Returns: 0 on success
1010 * BLKPREP_DEFER if the failure is retryable
1011 * BLKPREP_KILL if the failure is fatal
1013 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1015 struct request *rq = cmd->request;
1017 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1021 if (blk_bidi_rq(rq)) {
1022 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1023 scsi_sdb_cache, GFP_ATOMIC);
1025 error = BLKPREP_DEFER;
1029 rq->next_rq->special = bidi_sdb;
1030 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1035 if (blk_integrity_rq(rq)) {
1036 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1039 BUG_ON(prot_sdb == NULL);
1040 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1042 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1043 error = BLKPREP_DEFER;
1047 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1048 prot_sdb->table.sgl);
1049 BUG_ON(unlikely(count > ivecs));
1050 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1052 cmd->prot_sdb = prot_sdb;
1053 cmd->prot_sdb->table.nents = count;
1059 scsi_release_buffers(cmd);
1060 cmd->request->special = NULL;
1061 scsi_put_command(cmd);
1064 EXPORT_SYMBOL(scsi_init_io);
1066 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1067 struct request *req)
1069 struct scsi_cmnd *cmd;
1071 if (!req->special) {
1072 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1080 /* pull a tag out of the request if we have one */
1081 cmd->tag = req->tag;
1084 cmd->cmnd = req->cmd;
1085 cmd->prot_op = SCSI_PROT_NORMAL;
1090 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1092 struct scsi_cmnd *cmd;
1093 int ret = scsi_prep_state_check(sdev, req);
1095 if (ret != BLKPREP_OK)
1098 cmd = scsi_get_cmd_from_req(sdev, req);
1100 return BLKPREP_DEFER;
1103 * BLOCK_PC requests may transfer data, in which case they must
1104 * a bio attached to them. Or they might contain a SCSI command
1105 * that does not transfer data, in which case they may optionally
1106 * submit a request without an attached bio.
1111 BUG_ON(!req->nr_phys_segments);
1113 ret = scsi_init_io(cmd, GFP_ATOMIC);
1117 BUG_ON(blk_rq_bytes(req));
1119 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1123 cmd->cmd_len = req->cmd_len;
1124 if (!blk_rq_bytes(req))
1125 cmd->sc_data_direction = DMA_NONE;
1126 else if (rq_data_dir(req) == WRITE)
1127 cmd->sc_data_direction = DMA_TO_DEVICE;
1129 cmd->sc_data_direction = DMA_FROM_DEVICE;
1131 cmd->transfersize = blk_rq_bytes(req);
1132 cmd->allowed = req->retries;
1135 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1138 * Setup a REQ_TYPE_FS command. These are simple read/write request
1139 * from filesystems that still need to be translated to SCSI CDBs from
1142 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1144 struct scsi_cmnd *cmd;
1145 int ret = scsi_prep_state_check(sdev, req);
1147 if (ret != BLKPREP_OK)
1150 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1151 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1152 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1153 if (ret != BLKPREP_OK)
1158 * Filesystem requests must transfer data.
1160 BUG_ON(!req->nr_phys_segments);
1162 cmd = scsi_get_cmd_from_req(sdev, req);
1164 return BLKPREP_DEFER;
1166 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1167 return scsi_init_io(cmd, GFP_ATOMIC);
1169 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1171 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1173 int ret = BLKPREP_OK;
1176 * If the device is not in running state we will reject some
1179 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1180 switch (sdev->sdev_state) {
1182 case SDEV_TRANSPORT_OFFLINE:
1184 * If the device is offline we refuse to process any
1185 * commands. The device must be brought online
1186 * before trying any recovery commands.
1188 sdev_printk(KERN_ERR, sdev,
1189 "rejecting I/O to offline device\n");
1194 * If the device is fully deleted, we refuse to
1195 * process any commands as well.
1197 sdev_printk(KERN_ERR, sdev,
1198 "rejecting I/O to dead device\n");
1203 case SDEV_CREATED_BLOCK:
1205 * If the devices is blocked we defer normal commands.
1207 if (!(req->cmd_flags & REQ_PREEMPT))
1208 ret = BLKPREP_DEFER;
1212 * For any other not fully online state we only allow
1213 * special commands. In particular any user initiated
1214 * command is not allowed.
1216 if (!(req->cmd_flags & REQ_PREEMPT))
1223 EXPORT_SYMBOL(scsi_prep_state_check);
1225 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1227 struct scsi_device *sdev = q->queuedata;
1231 req->errors = DID_NO_CONNECT << 16;
1232 /* release the command and kill it */
1234 struct scsi_cmnd *cmd = req->special;
1235 scsi_release_buffers(cmd);
1236 scsi_put_command(cmd);
1237 req->special = NULL;
1242 * If we defer, the blk_peek_request() returns NULL, but the
1243 * queue must be restarted, so we schedule a callback to happen
1246 if (sdev->device_busy == 0)
1247 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1250 req->cmd_flags |= REQ_DONTPREP;
1255 EXPORT_SYMBOL(scsi_prep_return);
1257 int scsi_prep_fn(struct request_queue *q, struct request *req)
1259 struct scsi_device *sdev = q->queuedata;
1260 int ret = BLKPREP_KILL;
1262 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1263 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1264 return scsi_prep_return(q, req, ret);
1266 EXPORT_SYMBOL(scsi_prep_fn);
1269 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1272 * Called with the queue_lock held.
1274 static inline int scsi_dev_queue_ready(struct request_queue *q,
1275 struct scsi_device *sdev)
1277 if (sdev->device_busy == 0 && sdev->device_blocked) {
1279 * unblock after device_blocked iterates to zero
1281 if (--sdev->device_blocked == 0) {
1283 sdev_printk(KERN_INFO, sdev,
1284 "unblocking device at zero depth\n"));
1286 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 if (scsi_device_is_busy(sdev))
1298 * scsi_target_queue_ready: checks if there we can send commands to target
1299 * @sdev: scsi device on starget to check.
1301 * Called with the host lock held.
1303 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1304 struct scsi_device *sdev)
1306 struct scsi_target *starget = scsi_target(sdev);
1308 if (starget->single_lun) {
1309 if (starget->starget_sdev_user &&
1310 starget->starget_sdev_user != sdev)
1312 starget->starget_sdev_user = sdev;
1315 if (starget->target_busy == 0 && starget->target_blocked) {
1317 * unblock after target_blocked iterates to zero
1319 if (--starget->target_blocked == 0) {
1320 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1321 "unblocking target at zero depth\n"));
1326 if (scsi_target_is_busy(starget)) {
1327 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1335 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1336 * return 0. We must end up running the queue again whenever 0 is
1337 * returned, else IO can hang.
1339 * Called with host_lock held.
1341 static inline int scsi_host_queue_ready(struct request_queue *q,
1342 struct Scsi_Host *shost,
1343 struct scsi_device *sdev)
1345 if (scsi_host_in_recovery(shost))
1347 if (shost->host_busy == 0 && shost->host_blocked) {
1349 * unblock after host_blocked iterates to zero
1351 if (--shost->host_blocked == 0) {
1353 printk("scsi%d unblocking host at zero depth\n",
1359 if (scsi_host_is_busy(shost)) {
1360 if (list_empty(&sdev->starved_entry))
1361 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1365 /* We're OK to process the command, so we can't be starved */
1366 if (!list_empty(&sdev->starved_entry))
1367 list_del_init(&sdev->starved_entry);
1373 * Busy state exporting function for request stacking drivers.
1375 * For efficiency, no lock is taken to check the busy state of
1376 * shost/starget/sdev, since the returned value is not guaranteed and
1377 * may be changed after request stacking drivers call the function,
1378 * regardless of taking lock or not.
1380 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1381 * needs to return 'not busy'. Otherwise, request stacking drivers
1382 * may hold requests forever.
1384 static int scsi_lld_busy(struct request_queue *q)
1386 struct scsi_device *sdev = q->queuedata;
1387 struct Scsi_Host *shost;
1389 if (blk_queue_dead(q))
1395 * Ignore host/starget busy state.
1396 * Since block layer does not have a concept of fairness across
1397 * multiple queues, congestion of host/starget needs to be handled
1400 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1407 * Kill a request for a dead device
1409 static void scsi_kill_request(struct request *req, struct request_queue *q)
1411 struct scsi_cmnd *cmd = req->special;
1412 struct scsi_device *sdev;
1413 struct scsi_target *starget;
1414 struct Scsi_Host *shost;
1416 blk_start_request(req);
1418 scmd_printk(KERN_INFO, cmd, "killing request\n");
1421 starget = scsi_target(sdev);
1423 scsi_init_cmd_errh(cmd);
1424 cmd->result = DID_NO_CONNECT << 16;
1425 atomic_inc(&cmd->device->iorequest_cnt);
1428 * SCSI request completion path will do scsi_device_unbusy(),
1429 * bump busy counts. To bump the counters, we need to dance
1430 * with the locks as normal issue path does.
1432 sdev->device_busy++;
1433 spin_unlock(sdev->request_queue->queue_lock);
1434 spin_lock(shost->host_lock);
1436 starget->target_busy++;
1437 spin_unlock(shost->host_lock);
1438 spin_lock(sdev->request_queue->queue_lock);
1440 blk_complete_request(req);
1443 static void scsi_softirq_done(struct request *rq)
1445 struct scsi_cmnd *cmd = rq->special;
1446 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1449 INIT_LIST_HEAD(&cmd->eh_entry);
1451 atomic_inc(&cmd->device->iodone_cnt);
1453 atomic_inc(&cmd->device->ioerr_cnt);
1455 disposition = scsi_decide_disposition(cmd);
1456 if (disposition != SUCCESS &&
1457 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1458 sdev_printk(KERN_ERR, cmd->device,
1459 "timing out command, waited %lus\n",
1461 disposition = SUCCESS;
1464 scsi_log_completion(cmd, disposition);
1466 switch (disposition) {
1468 scsi_finish_command(cmd);
1471 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1473 case ADD_TO_MLQUEUE:
1474 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1477 if (!scsi_eh_scmd_add(cmd, 0))
1478 scsi_finish_command(cmd);
1483 * Function: scsi_request_fn()
1485 * Purpose: Main strategy routine for SCSI.
1487 * Arguments: q - Pointer to actual queue.
1491 * Lock status: IO request lock assumed to be held when called.
1493 static void scsi_request_fn(struct request_queue *q)
1495 struct scsi_device *sdev = q->queuedata;
1496 struct Scsi_Host *shost;
1497 struct scsi_cmnd *cmd;
1498 struct request *req;
1500 if(!get_device(&sdev->sdev_gendev))
1501 /* We must be tearing the block queue down already */
1505 * To start with, we keep looping until the queue is empty, or until
1506 * the host is no longer able to accept any more requests.
1512 * get next queueable request. We do this early to make sure
1513 * that the request is fully prepared even if we cannot
1516 req = blk_peek_request(q);
1517 if (!req || !scsi_dev_queue_ready(q, sdev))
1520 if (unlikely(!scsi_device_online(sdev))) {
1521 sdev_printk(KERN_ERR, sdev,
1522 "rejecting I/O to offline device\n");
1523 scsi_kill_request(req, q);
1529 * Remove the request from the request list.
1531 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1532 blk_start_request(req);
1533 sdev->device_busy++;
1535 spin_unlock(q->queue_lock);
1537 if (unlikely(cmd == NULL)) {
1538 printk(KERN_CRIT "impossible request in %s.\n"
1539 "please mail a stack trace to "
1540 "linux-scsi@vger.kernel.org\n",
1542 blk_dump_rq_flags(req, "foo");
1545 spin_lock(shost->host_lock);
1548 * We hit this when the driver is using a host wide
1549 * tag map. For device level tag maps the queue_depth check
1550 * in the device ready fn would prevent us from trying
1551 * to allocate a tag. Since the map is a shared host resource
1552 * we add the dev to the starved list so it eventually gets
1553 * a run when a tag is freed.
1555 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1556 if (list_empty(&sdev->starved_entry))
1557 list_add_tail(&sdev->starved_entry,
1558 &shost->starved_list);
1562 if (!scsi_target_queue_ready(shost, sdev))
1565 if (!scsi_host_queue_ready(q, shost, sdev))
1568 scsi_target(sdev)->target_busy++;
1572 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1573 * take the lock again.
1575 spin_unlock_irq(shost->host_lock);
1578 * Finally, initialize any error handling parameters, and set up
1579 * the timers for timeouts.
1581 scsi_init_cmd_errh(cmd);
1584 * Dispatch the command to the low-level driver.
1586 rtn = scsi_dispatch_cmd(cmd);
1587 spin_lock_irq(q->queue_lock);
1595 spin_unlock_irq(shost->host_lock);
1598 * lock q, handle tag, requeue req, and decrement device_busy. We
1599 * must return with queue_lock held.
1601 * Decrementing device_busy without checking it is OK, as all such
1602 * cases (host limits or settings) should run the queue at some
1605 spin_lock_irq(q->queue_lock);
1606 blk_requeue_request(q, req);
1607 sdev->device_busy--;
1609 if (sdev->device_busy == 0)
1610 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1612 /* must be careful here...if we trigger the ->remove() function
1613 * we cannot be holding the q lock */
1614 spin_unlock_irq(q->queue_lock);
1615 put_device(&sdev->sdev_gendev);
1616 spin_lock_irq(q->queue_lock);
1619 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1621 struct device *host_dev;
1622 u64 bounce_limit = 0xffffffff;
1624 if (shost->unchecked_isa_dma)
1625 return BLK_BOUNCE_ISA;
1627 * Platforms with virtual-DMA translation
1628 * hardware have no practical limit.
1630 if (!PCI_DMA_BUS_IS_PHYS)
1631 return BLK_BOUNCE_ANY;
1633 host_dev = scsi_get_device(shost);
1634 if (host_dev && host_dev->dma_mask)
1635 bounce_limit = *host_dev->dma_mask;
1637 return bounce_limit;
1639 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1641 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1642 request_fn_proc *request_fn)
1644 struct request_queue *q;
1645 struct device *dev = shost->dma_dev;
1647 q = blk_init_queue(request_fn, NULL);
1652 * this limit is imposed by hardware restrictions
1654 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1655 SCSI_MAX_SG_CHAIN_SEGMENTS));
1657 if (scsi_host_prot_dma(shost)) {
1658 shost->sg_prot_tablesize =
1659 min_not_zero(shost->sg_prot_tablesize,
1660 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1661 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1662 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1665 blk_queue_max_hw_sectors(q, shost->max_sectors);
1666 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1667 blk_queue_segment_boundary(q, shost->dma_boundary);
1668 dma_set_seg_boundary(dev, shost->dma_boundary);
1670 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1672 if (!shost->use_clustering)
1673 q->limits.cluster = 0;
1676 * set a reasonable default alignment on word boundaries: the
1677 * host and device may alter it using
1678 * blk_queue_update_dma_alignment() later.
1680 blk_queue_dma_alignment(q, 0x03);
1684 EXPORT_SYMBOL(__scsi_alloc_queue);
1686 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1688 struct request_queue *q;
1690 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1694 blk_queue_prep_rq(q, scsi_prep_fn);
1695 blk_queue_softirq_done(q, scsi_softirq_done);
1696 blk_queue_rq_timed_out(q, scsi_times_out);
1697 blk_queue_lld_busy(q, scsi_lld_busy);
1702 * Function: scsi_block_requests()
1704 * Purpose: Utility function used by low-level drivers to prevent further
1705 * commands from being queued to the device.
1707 * Arguments: shost - Host in question
1711 * Lock status: No locks are assumed held.
1713 * Notes: There is no timer nor any other means by which the requests
1714 * get unblocked other than the low-level driver calling
1715 * scsi_unblock_requests().
1717 void scsi_block_requests(struct Scsi_Host *shost)
1719 shost->host_self_blocked = 1;
1721 EXPORT_SYMBOL(scsi_block_requests);
1724 * Function: scsi_unblock_requests()
1726 * Purpose: Utility function used by low-level drivers to allow further
1727 * commands from being queued to the device.
1729 * Arguments: shost - Host in question
1733 * Lock status: No locks are assumed held.
1735 * Notes: There is no timer nor any other means by which the requests
1736 * get unblocked other than the low-level driver calling
1737 * scsi_unblock_requests().
1739 * This is done as an API function so that changes to the
1740 * internals of the scsi mid-layer won't require wholesale
1741 * changes to drivers that use this feature.
1743 void scsi_unblock_requests(struct Scsi_Host *shost)
1745 shost->host_self_blocked = 0;
1746 scsi_run_host_queues(shost);
1748 EXPORT_SYMBOL(scsi_unblock_requests);
1750 int __init scsi_init_queue(void)
1754 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1755 sizeof(struct scsi_data_buffer),
1757 if (!scsi_sdb_cache) {
1758 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1762 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1763 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1764 int size = sgp->size * sizeof(struct scatterlist);
1766 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1767 SLAB_HWCACHE_ALIGN, NULL);
1769 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1774 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1777 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1786 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1787 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1789 mempool_destroy(sgp->pool);
1791 kmem_cache_destroy(sgp->slab);
1793 kmem_cache_destroy(scsi_sdb_cache);
1798 void scsi_exit_queue(void)
1802 kmem_cache_destroy(scsi_sdb_cache);
1804 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1805 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1806 mempool_destroy(sgp->pool);
1807 kmem_cache_destroy(sgp->slab);
1812 * scsi_mode_select - issue a mode select
1813 * @sdev: SCSI device to be queried
1814 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1815 * @sp: Save page bit (0 == don't save, 1 == save)
1816 * @modepage: mode page being requested
1817 * @buffer: request buffer (may not be smaller than eight bytes)
1818 * @len: length of request buffer.
1819 * @timeout: command timeout
1820 * @retries: number of retries before failing
1821 * @data: returns a structure abstracting the mode header data
1822 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1823 * must be SCSI_SENSE_BUFFERSIZE big.
1825 * Returns zero if successful; negative error number or scsi
1830 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1831 unsigned char *buffer, int len, int timeout, int retries,
1832 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1834 unsigned char cmd[10];
1835 unsigned char *real_buffer;
1838 memset(cmd, 0, sizeof(cmd));
1839 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1841 if (sdev->use_10_for_ms) {
1844 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1847 memcpy(real_buffer + 8, buffer, len);
1851 real_buffer[2] = data->medium_type;
1852 real_buffer[3] = data->device_specific;
1853 real_buffer[4] = data->longlba ? 0x01 : 0;
1855 real_buffer[6] = data->block_descriptor_length >> 8;
1856 real_buffer[7] = data->block_descriptor_length;
1858 cmd[0] = MODE_SELECT_10;
1862 if (len > 255 || data->block_descriptor_length > 255 ||
1866 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1869 memcpy(real_buffer + 4, buffer, len);
1872 real_buffer[1] = data->medium_type;
1873 real_buffer[2] = data->device_specific;
1874 real_buffer[3] = data->block_descriptor_length;
1877 cmd[0] = MODE_SELECT;
1881 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1882 sshdr, timeout, retries, NULL);
1886 EXPORT_SYMBOL_GPL(scsi_mode_select);
1889 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1890 * @sdev: SCSI device to be queried
1891 * @dbd: set if mode sense will allow block descriptors to be returned
1892 * @modepage: mode page being requested
1893 * @buffer: request buffer (may not be smaller than eight bytes)
1894 * @len: length of request buffer.
1895 * @timeout: command timeout
1896 * @retries: number of retries before failing
1897 * @data: returns a structure abstracting the mode header data
1898 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1899 * must be SCSI_SENSE_BUFFERSIZE big.
1901 * Returns zero if unsuccessful, or the header offset (either 4
1902 * or 8 depending on whether a six or ten byte command was
1903 * issued) if successful.
1906 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1907 unsigned char *buffer, int len, int timeout, int retries,
1908 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1910 unsigned char cmd[12];
1914 struct scsi_sense_hdr my_sshdr;
1916 memset(data, 0, sizeof(*data));
1917 memset(&cmd[0], 0, 12);
1918 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1921 /* caller might not be interested in sense, but we need it */
1926 use_10_for_ms = sdev->use_10_for_ms;
1928 if (use_10_for_ms) {
1932 cmd[0] = MODE_SENSE_10;
1939 cmd[0] = MODE_SENSE;
1944 memset(buffer, 0, len);
1946 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1947 sshdr, timeout, retries, NULL);
1949 /* This code looks awful: what it's doing is making sure an
1950 * ILLEGAL REQUEST sense return identifies the actual command
1951 * byte as the problem. MODE_SENSE commands can return
1952 * ILLEGAL REQUEST if the code page isn't supported */
1954 if (use_10_for_ms && !scsi_status_is_good(result) &&
1955 (driver_byte(result) & DRIVER_SENSE)) {
1956 if (scsi_sense_valid(sshdr)) {
1957 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1958 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1960 * Invalid command operation code
1962 sdev->use_10_for_ms = 0;
1968 if(scsi_status_is_good(result)) {
1969 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1970 (modepage == 6 || modepage == 8))) {
1971 /* Initio breakage? */
1974 data->medium_type = 0;
1975 data->device_specific = 0;
1977 data->block_descriptor_length = 0;
1978 } else if(use_10_for_ms) {
1979 data->length = buffer[0]*256 + buffer[1] + 2;
1980 data->medium_type = buffer[2];
1981 data->device_specific = buffer[3];
1982 data->longlba = buffer[4] & 0x01;
1983 data->block_descriptor_length = buffer[6]*256
1986 data->length = buffer[0] + 1;
1987 data->medium_type = buffer[1];
1988 data->device_specific = buffer[2];
1989 data->block_descriptor_length = buffer[3];
1991 data->header_length = header_length;
1996 EXPORT_SYMBOL(scsi_mode_sense);
1999 * scsi_test_unit_ready - test if unit is ready
2000 * @sdev: scsi device to change the state of.
2001 * @timeout: command timeout
2002 * @retries: number of retries before failing
2003 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2004 * returning sense. Make sure that this is cleared before passing
2007 * Returns zero if unsuccessful or an error if TUR failed. For
2008 * removable media, UNIT_ATTENTION sets ->changed flag.
2011 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2012 struct scsi_sense_hdr *sshdr_external)
2015 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2017 struct scsi_sense_hdr *sshdr;
2020 if (!sshdr_external)
2021 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2023 sshdr = sshdr_external;
2025 /* try to eat the UNIT_ATTENTION if there are enough retries */
2027 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2028 timeout, retries, NULL);
2029 if (sdev->removable && scsi_sense_valid(sshdr) &&
2030 sshdr->sense_key == UNIT_ATTENTION)
2032 } while (scsi_sense_valid(sshdr) &&
2033 sshdr->sense_key == UNIT_ATTENTION && --retries);
2035 if (!sshdr_external)
2039 EXPORT_SYMBOL(scsi_test_unit_ready);
2042 * scsi_device_set_state - Take the given device through the device state model.
2043 * @sdev: scsi device to change the state of.
2044 * @state: state to change to.
2046 * Returns zero if unsuccessful or an error if the requested
2047 * transition is illegal.
2050 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2052 enum scsi_device_state oldstate = sdev->sdev_state;
2054 if (state == oldstate)
2060 case SDEV_CREATED_BLOCK:
2071 case SDEV_TRANSPORT_OFFLINE:
2084 case SDEV_TRANSPORT_OFFLINE:
2092 case SDEV_TRANSPORT_OFFLINE:
2107 case SDEV_CREATED_BLOCK:
2114 case SDEV_CREATED_BLOCK:
2129 case SDEV_TRANSPORT_OFFLINE:
2142 case SDEV_TRANSPORT_OFFLINE:
2151 sdev->sdev_state = state;
2155 SCSI_LOG_ERROR_RECOVERY(1,
2156 sdev_printk(KERN_ERR, sdev,
2157 "Illegal state transition %s->%s\n",
2158 scsi_device_state_name(oldstate),
2159 scsi_device_state_name(state))
2163 EXPORT_SYMBOL(scsi_device_set_state);
2166 * sdev_evt_emit - emit a single SCSI device uevent
2167 * @sdev: associated SCSI device
2168 * @evt: event to emit
2170 * Send a single uevent (scsi_event) to the associated scsi_device.
2172 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2177 switch (evt->evt_type) {
2178 case SDEV_EVT_MEDIA_CHANGE:
2179 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2189 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2193 * sdev_evt_thread - send a uevent for each scsi event
2194 * @work: work struct for scsi_device
2196 * Dispatch queued events to their associated scsi_device kobjects
2199 void scsi_evt_thread(struct work_struct *work)
2201 struct scsi_device *sdev;
2202 LIST_HEAD(event_list);
2204 sdev = container_of(work, struct scsi_device, event_work);
2207 struct scsi_event *evt;
2208 struct list_head *this, *tmp;
2209 unsigned long flags;
2211 spin_lock_irqsave(&sdev->list_lock, flags);
2212 list_splice_init(&sdev->event_list, &event_list);
2213 spin_unlock_irqrestore(&sdev->list_lock, flags);
2215 if (list_empty(&event_list))
2218 list_for_each_safe(this, tmp, &event_list) {
2219 evt = list_entry(this, struct scsi_event, node);
2220 list_del(&evt->node);
2221 scsi_evt_emit(sdev, evt);
2228 * sdev_evt_send - send asserted event to uevent thread
2229 * @sdev: scsi_device event occurred on
2230 * @evt: event to send
2232 * Assert scsi device event asynchronously.
2234 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2236 unsigned long flags;
2239 /* FIXME: currently this check eliminates all media change events
2240 * for polled devices. Need to update to discriminate between AN
2241 * and polled events */
2242 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2248 spin_lock_irqsave(&sdev->list_lock, flags);
2249 list_add_tail(&evt->node, &sdev->event_list);
2250 schedule_work(&sdev->event_work);
2251 spin_unlock_irqrestore(&sdev->list_lock, flags);
2253 EXPORT_SYMBOL_GPL(sdev_evt_send);
2256 * sdev_evt_alloc - allocate a new scsi event
2257 * @evt_type: type of event to allocate
2258 * @gfpflags: GFP flags for allocation
2260 * Allocates and returns a new scsi_event.
2262 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2265 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2269 evt->evt_type = evt_type;
2270 INIT_LIST_HEAD(&evt->node);
2272 /* evt_type-specific initialization, if any */
2274 case SDEV_EVT_MEDIA_CHANGE:
2282 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2285 * sdev_evt_send_simple - send asserted event to uevent thread
2286 * @sdev: scsi_device event occurred on
2287 * @evt_type: type of event to send
2288 * @gfpflags: GFP flags for allocation
2290 * Assert scsi device event asynchronously, given an event type.
2292 void sdev_evt_send_simple(struct scsi_device *sdev,
2293 enum scsi_device_event evt_type, gfp_t gfpflags)
2295 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2297 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2302 sdev_evt_send(sdev, evt);
2304 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2307 * scsi_device_quiesce - Block user issued commands.
2308 * @sdev: scsi device to quiesce.
2310 * This works by trying to transition to the SDEV_QUIESCE state
2311 * (which must be a legal transition). When the device is in this
2312 * state, only special requests will be accepted, all others will
2313 * be deferred. Since special requests may also be requeued requests,
2314 * a successful return doesn't guarantee the device will be
2315 * totally quiescent.
2317 * Must be called with user context, may sleep.
2319 * Returns zero if unsuccessful or an error if not.
2322 scsi_device_quiesce(struct scsi_device *sdev)
2324 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2328 scsi_run_queue(sdev->request_queue);
2329 while (sdev->device_busy) {
2330 msleep_interruptible(200);
2331 scsi_run_queue(sdev->request_queue);
2335 EXPORT_SYMBOL(scsi_device_quiesce);
2338 * scsi_device_resume - Restart user issued commands to a quiesced device.
2339 * @sdev: scsi device to resume.
2341 * Moves the device from quiesced back to running and restarts the
2344 * Must be called with user context, may sleep.
2346 void scsi_device_resume(struct scsi_device *sdev)
2348 /* check if the device state was mutated prior to resume, and if
2349 * so assume the state is being managed elsewhere (for example
2350 * device deleted during suspend)
2352 if (sdev->sdev_state != SDEV_QUIESCE ||
2353 scsi_device_set_state(sdev, SDEV_RUNNING))
2355 scsi_run_queue(sdev->request_queue);
2357 EXPORT_SYMBOL(scsi_device_resume);
2360 device_quiesce_fn(struct scsi_device *sdev, void *data)
2362 scsi_device_quiesce(sdev);
2366 scsi_target_quiesce(struct scsi_target *starget)
2368 starget_for_each_device(starget, NULL, device_quiesce_fn);
2370 EXPORT_SYMBOL(scsi_target_quiesce);
2373 device_resume_fn(struct scsi_device *sdev, void *data)
2375 scsi_device_resume(sdev);
2379 scsi_target_resume(struct scsi_target *starget)
2381 starget_for_each_device(starget, NULL, device_resume_fn);
2383 EXPORT_SYMBOL(scsi_target_resume);
2386 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2387 * @sdev: device to block
2389 * Block request made by scsi lld's to temporarily stop all
2390 * scsi commands on the specified device. Called from interrupt
2391 * or normal process context.
2393 * Returns zero if successful or error if not
2396 * This routine transitions the device to the SDEV_BLOCK state
2397 * (which must be a legal transition). When the device is in this
2398 * state, all commands are deferred until the scsi lld reenables
2399 * the device with scsi_device_unblock or device_block_tmo fires.
2402 scsi_internal_device_block(struct scsi_device *sdev)
2404 struct request_queue *q = sdev->request_queue;
2405 unsigned long flags;
2408 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2410 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2417 * The device has transitioned to SDEV_BLOCK. Stop the
2418 * block layer from calling the midlayer with this device's
2421 spin_lock_irqsave(q->queue_lock, flags);
2423 spin_unlock_irqrestore(q->queue_lock, flags);
2427 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2430 * scsi_internal_device_unblock - resume a device after a block request
2431 * @sdev: device to resume
2432 * @new_state: state to set devices to after unblocking
2434 * Called by scsi lld's or the midlayer to restart the device queue
2435 * for the previously suspended scsi device. Called from interrupt or
2436 * normal process context.
2438 * Returns zero if successful or error if not.
2441 * This routine transitions the device to the SDEV_RUNNING state
2442 * or to one of the offline states (which must be a legal transition)
2443 * allowing the midlayer to goose the queue for this device.
2446 scsi_internal_device_unblock(struct scsi_device *sdev,
2447 enum scsi_device_state new_state)
2449 struct request_queue *q = sdev->request_queue;
2450 unsigned long flags;
2453 * Try to transition the scsi device to SDEV_RUNNING or one of the
2454 * offlined states and goose the device queue if successful.
2456 if (sdev->sdev_state == SDEV_BLOCK)
2457 sdev->sdev_state = new_state;
2458 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2459 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2460 new_state == SDEV_OFFLINE)
2461 sdev->sdev_state = new_state;
2463 sdev->sdev_state = SDEV_CREATED;
2464 } else if (sdev->sdev_state != SDEV_CANCEL &&
2465 sdev->sdev_state != SDEV_OFFLINE)
2468 spin_lock_irqsave(q->queue_lock, flags);
2470 spin_unlock_irqrestore(q->queue_lock, flags);
2474 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2477 device_block(struct scsi_device *sdev, void *data)
2479 scsi_internal_device_block(sdev);
2483 target_block(struct device *dev, void *data)
2485 if (scsi_is_target_device(dev))
2486 starget_for_each_device(to_scsi_target(dev), NULL,
2492 scsi_target_block(struct device *dev)
2494 if (scsi_is_target_device(dev))
2495 starget_for_each_device(to_scsi_target(dev), NULL,
2498 device_for_each_child(dev, NULL, target_block);
2500 EXPORT_SYMBOL_GPL(scsi_target_block);
2503 device_unblock(struct scsi_device *sdev, void *data)
2505 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2509 target_unblock(struct device *dev, void *data)
2511 if (scsi_is_target_device(dev))
2512 starget_for_each_device(to_scsi_target(dev), data,
2518 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2520 if (scsi_is_target_device(dev))
2521 starget_for_each_device(to_scsi_target(dev), &new_state,
2524 device_for_each_child(dev, &new_state, target_unblock);
2526 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2529 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2530 * @sgl: scatter-gather list
2531 * @sg_count: number of segments in sg
2532 * @offset: offset in bytes into sg, on return offset into the mapped area
2533 * @len: bytes to map, on return number of bytes mapped
2535 * Returns virtual address of the start of the mapped page
2537 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2538 size_t *offset, size_t *len)
2541 size_t sg_len = 0, len_complete = 0;
2542 struct scatterlist *sg;
2545 WARN_ON(!irqs_disabled());
2547 for_each_sg(sgl, sg, sg_count, i) {
2548 len_complete = sg_len; /* Complete sg-entries */
2549 sg_len += sg->length;
2550 if (sg_len > *offset)
2554 if (unlikely(i == sg_count)) {
2555 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2557 __func__, sg_len, *offset, sg_count);
2562 /* Offset starting from the beginning of first page in this sg-entry */
2563 *offset = *offset - len_complete + sg->offset;
2565 /* Assumption: contiguous pages can be accessed as "page + i" */
2566 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2567 *offset &= ~PAGE_MASK;
2569 /* Bytes in this sg-entry from *offset to the end of the page */
2570 sg_len = PAGE_SIZE - *offset;
2574 return kmap_atomic(page);
2576 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2579 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2580 * @virt: virtual address to be unmapped
2582 void scsi_kunmap_atomic_sg(void *virt)
2584 kunmap_atomic(virt);
2586 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);