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 #include <acpi/acpi_bus.h>
74 static bool acpi_scsi_bus_match(struct device *dev)
76 return dev->bus == &scsi_bus_type;
79 int scsi_register_acpi_bus_type(struct acpi_bus_type *bus)
81 bus->match = acpi_scsi_bus_match;
82 return register_acpi_bus_type(bus);
84 EXPORT_SYMBOL_GPL(scsi_register_acpi_bus_type);
86 void scsi_unregister_acpi_bus_type(struct acpi_bus_type *bus)
88 unregister_acpi_bus_type(bus);
90 EXPORT_SYMBOL_GPL(scsi_unregister_acpi_bus_type);
94 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
95 * not change behaviour from the previous unplug mechanism, experimentation
96 * may prove this needs changing.
98 #define SCSI_QUEUE_DELAY 3
101 * Function: scsi_unprep_request()
103 * Purpose: Remove all preparation done for a request, including its
104 * associated scsi_cmnd, so that it can be requeued.
106 * Arguments: req - request to unprepare
108 * Lock status: Assumed that no locks are held upon entry.
112 static void scsi_unprep_request(struct request *req)
114 struct scsi_cmnd *cmd = req->special;
116 blk_unprep_request(req);
119 scsi_put_command(cmd);
123 * __scsi_queue_insert - private queue insertion
124 * @cmd: The SCSI command being requeued
125 * @reason: The reason for the requeue
126 * @unbusy: Whether the queue should be unbusied
128 * This is a private queue insertion. The public interface
129 * scsi_queue_insert() always assumes the queue should be unbusied
130 * because it's always called before the completion. This function is
131 * for a requeue after completion, which should only occur in this
134 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
136 struct Scsi_Host *host = cmd->device->host;
137 struct scsi_device *device = cmd->device;
138 struct scsi_target *starget = scsi_target(device);
139 struct request_queue *q = device->request_queue;
143 printk("Inserting command %p into mlqueue\n", cmd));
146 * Set the appropriate busy bit for the device/host.
148 * If the host/device isn't busy, assume that something actually
149 * completed, and that we should be able to queue a command now.
151 * Note that the prior mid-layer assumption that any host could
152 * always queue at least one command is now broken. The mid-layer
153 * will implement a user specifiable stall (see
154 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
155 * if a command is requeued with no other commands outstanding
156 * either for the device or for the host.
159 case SCSI_MLQUEUE_HOST_BUSY:
160 host->host_blocked = host->max_host_blocked;
162 case SCSI_MLQUEUE_DEVICE_BUSY:
163 case SCSI_MLQUEUE_EH_RETRY:
164 device->device_blocked = device->max_device_blocked;
166 case SCSI_MLQUEUE_TARGET_BUSY:
167 starget->target_blocked = starget->max_target_blocked;
172 * Decrement the counters, since these commands are no longer
173 * active on the host/device.
176 scsi_device_unbusy(device);
179 * Requeue this command. It will go before all other commands
180 * that are already in the queue. Schedule requeue work under
181 * lock such that the kblockd_schedule_work() call happens
182 * before blk_cleanup_queue() finishes.
184 spin_lock_irqsave(q->queue_lock, flags);
185 blk_requeue_request(q, cmd->request);
186 kblockd_schedule_work(q, &device->requeue_work);
187 spin_unlock_irqrestore(q->queue_lock, flags);
191 * Function: scsi_queue_insert()
193 * Purpose: Insert a command in the midlevel queue.
195 * Arguments: cmd - command that we are adding to queue.
196 * reason - why we are inserting command to queue.
198 * Lock status: Assumed that lock is not held upon entry.
202 * Notes: We do this for one of two cases. Either the host is busy
203 * and it cannot accept any more commands for the time being,
204 * or the device returned QUEUE_FULL and can accept no more
206 * Notes: This could be called either from an interrupt context or a
207 * normal process context.
209 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
211 __scsi_queue_insert(cmd, reason, 1);
214 * scsi_execute - insert request and wait for the result
217 * @data_direction: data direction
218 * @buffer: data buffer
219 * @bufflen: len of buffer
220 * @sense: optional sense buffer
221 * @timeout: request timeout in seconds
222 * @retries: number of times to retry request
223 * @flags: or into request flags;
224 * @resid: optional residual length
226 * returns the req->errors value which is the scsi_cmnd result
229 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
230 int data_direction, void *buffer, unsigned bufflen,
231 unsigned char *sense, int timeout, int retries, int flags,
235 int write = (data_direction == DMA_TO_DEVICE);
236 int ret = DRIVER_ERROR << 24;
238 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
242 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
243 buffer, bufflen, __GFP_WAIT))
246 req->cmd_len = COMMAND_SIZE(cmd[0]);
247 memcpy(req->cmd, cmd, req->cmd_len);
250 req->retries = retries;
251 req->timeout = timeout;
252 req->cmd_type = REQ_TYPE_BLOCK_PC;
253 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
256 * head injection *required* here otherwise quiesce won't work
258 blk_execute_rq(req->q, NULL, req, 1);
261 * Some devices (USB mass-storage in particular) may transfer
262 * garbage data together with a residue indicating that the data
263 * is invalid. Prevent the garbage from being misinterpreted
264 * and prevent security leaks by zeroing out the excess data.
266 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
267 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
270 *resid = req->resid_len;
273 blk_put_request(req);
277 EXPORT_SYMBOL(scsi_execute);
279 int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
280 int data_direction, void *buffer, unsigned bufflen,
281 struct scsi_sense_hdr *sshdr, int timeout, int retries,
282 int *resid, int flags)
288 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
290 return DRIVER_ERROR << 24;
292 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
293 sense, timeout, retries, flags, resid);
295 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
300 EXPORT_SYMBOL(scsi_execute_req_flags);
303 * Function: scsi_init_cmd_errh()
305 * Purpose: Initialize cmd fields related to error handling.
307 * Arguments: cmd - command that is ready to be queued.
309 * Notes: This function has the job of initializing a number of
310 * fields related to error handling. Typically this will
311 * be called once for each command, as required.
313 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
315 cmd->serial_number = 0;
316 scsi_set_resid(cmd, 0);
317 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
318 if (cmd->cmd_len == 0)
319 cmd->cmd_len = scsi_command_size(cmd->cmnd);
322 void scsi_device_unbusy(struct scsi_device *sdev)
324 struct Scsi_Host *shost = sdev->host;
325 struct scsi_target *starget = scsi_target(sdev);
328 spin_lock_irqsave(shost->host_lock, flags);
330 starget->target_busy--;
331 if (unlikely(scsi_host_in_recovery(shost) &&
332 (shost->host_failed || shost->host_eh_scheduled)))
333 scsi_eh_wakeup(shost);
334 spin_unlock(shost->host_lock);
335 spin_lock(sdev->request_queue->queue_lock);
337 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
341 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
342 * and call blk_run_queue for all the scsi_devices on the target -
343 * including current_sdev first.
345 * Called with *no* scsi locks held.
347 static void scsi_single_lun_run(struct scsi_device *current_sdev)
349 struct Scsi_Host *shost = current_sdev->host;
350 struct scsi_device *sdev, *tmp;
351 struct scsi_target *starget = scsi_target(current_sdev);
354 spin_lock_irqsave(shost->host_lock, flags);
355 starget->starget_sdev_user = NULL;
356 spin_unlock_irqrestore(shost->host_lock, flags);
359 * Call blk_run_queue for all LUNs on the target, starting with
360 * current_sdev. We race with others (to set starget_sdev_user),
361 * but in most cases, we will be first. Ideally, each LU on the
362 * target would get some limited time or requests on the target.
364 blk_run_queue(current_sdev->request_queue);
366 spin_lock_irqsave(shost->host_lock, flags);
367 if (starget->starget_sdev_user)
369 list_for_each_entry_safe(sdev, tmp, &starget->devices,
370 same_target_siblings) {
371 if (sdev == current_sdev)
373 if (scsi_device_get(sdev))
376 spin_unlock_irqrestore(shost->host_lock, flags);
377 blk_run_queue(sdev->request_queue);
378 spin_lock_irqsave(shost->host_lock, flags);
380 scsi_device_put(sdev);
383 spin_unlock_irqrestore(shost->host_lock, flags);
386 static inline int scsi_device_is_busy(struct scsi_device *sdev)
388 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
394 static inline int scsi_target_is_busy(struct scsi_target *starget)
396 return ((starget->can_queue > 0 &&
397 starget->target_busy >= starget->can_queue) ||
398 starget->target_blocked);
401 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
403 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
404 shost->host_blocked || shost->host_self_blocked)
411 * Function: scsi_run_queue()
413 * Purpose: Select a proper request queue to serve next
415 * Arguments: q - last request's queue
419 * Notes: The previous command was completely finished, start
420 * a new one if possible.
422 static void scsi_run_queue(struct request_queue *q)
424 struct scsi_device *sdev = q->queuedata;
425 struct Scsi_Host *shost;
426 LIST_HEAD(starved_list);
430 if (scsi_target(sdev)->single_lun)
431 scsi_single_lun_run(sdev);
433 spin_lock_irqsave(shost->host_lock, flags);
434 list_splice_init(&shost->starved_list, &starved_list);
436 while (!list_empty(&starved_list)) {
438 * As long as shost is accepting commands and we have
439 * starved queues, call blk_run_queue. scsi_request_fn
440 * drops the queue_lock and can add us back to the
443 * host_lock protects the starved_list and starved_entry.
444 * scsi_request_fn must get the host_lock before checking
445 * or modifying starved_list or starved_entry.
447 if (scsi_host_is_busy(shost))
450 sdev = list_entry(starved_list.next,
451 struct scsi_device, starved_entry);
452 list_del_init(&sdev->starved_entry);
453 if (scsi_target_is_busy(scsi_target(sdev))) {
454 list_move_tail(&sdev->starved_entry,
455 &shost->starved_list);
459 spin_unlock(shost->host_lock);
460 spin_lock(sdev->request_queue->queue_lock);
461 __blk_run_queue(sdev->request_queue);
462 spin_unlock(sdev->request_queue->queue_lock);
463 spin_lock(shost->host_lock);
465 /* put any unprocessed entries back */
466 list_splice(&starved_list, &shost->starved_list);
467 spin_unlock_irqrestore(shost->host_lock, flags);
472 void scsi_requeue_run_queue(struct work_struct *work)
474 struct scsi_device *sdev;
475 struct request_queue *q;
477 sdev = container_of(work, struct scsi_device, requeue_work);
478 q = sdev->request_queue;
483 * Function: scsi_requeue_command()
485 * Purpose: Handle post-processing of completed commands.
487 * Arguments: q - queue to operate on
488 * cmd - command that may need to be requeued.
492 * Notes: After command completion, there may be blocks left
493 * over which weren't finished by the previous command
494 * this can be for a number of reasons - the main one is
495 * I/O errors in the middle of the request, in which case
496 * we need to request the blocks that come after the bad
498 * Notes: Upon return, cmd is a stale pointer.
500 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
502 struct scsi_device *sdev = cmd->device;
503 struct request *req = cmd->request;
507 * We need to hold a reference on the device to avoid the queue being
508 * killed after the unlock and before scsi_run_queue is invoked which
509 * may happen because scsi_unprep_request() puts the command which
510 * releases its reference on the device.
512 get_device(&sdev->sdev_gendev);
514 spin_lock_irqsave(q->queue_lock, flags);
515 scsi_unprep_request(req);
516 blk_requeue_request(q, req);
517 spin_unlock_irqrestore(q->queue_lock, flags);
521 put_device(&sdev->sdev_gendev);
524 void scsi_next_command(struct scsi_cmnd *cmd)
526 struct scsi_device *sdev = cmd->device;
527 struct request_queue *q = sdev->request_queue;
529 /* need to hold a reference on the device before we let go of the cmd */
530 get_device(&sdev->sdev_gendev);
532 scsi_put_command(cmd);
535 /* ok to remove device now */
536 put_device(&sdev->sdev_gendev);
539 void scsi_run_host_queues(struct Scsi_Host *shost)
541 struct scsi_device *sdev;
543 shost_for_each_device(sdev, shost)
544 scsi_run_queue(sdev->request_queue);
547 static void __scsi_release_buffers(struct scsi_cmnd *, int);
550 * Function: scsi_end_request()
552 * Purpose: Post-processing of completed commands (usually invoked at end
553 * of upper level post-processing and scsi_io_completion).
555 * Arguments: cmd - command that is complete.
556 * error - 0 if I/O indicates success, < 0 for I/O error.
557 * bytes - number of bytes of completed I/O
558 * requeue - indicates whether we should requeue leftovers.
560 * Lock status: Assumed that lock is not held upon entry.
562 * Returns: cmd if requeue required, NULL otherwise.
564 * Notes: This is called for block device requests in order to
565 * mark some number of sectors as complete.
567 * We are guaranteeing that the request queue will be goosed
568 * at some point during this call.
569 * Notes: If cmd was requeued, upon return it will be a stale pointer.
571 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
572 int bytes, int requeue)
574 struct request_queue *q = cmd->device->request_queue;
575 struct request *req = cmd->request;
578 * If there are blocks left over at the end, set up the command
579 * to queue the remainder of them.
581 if (blk_end_request(req, error, bytes)) {
582 /* kill remainder if no retrys */
583 if (error && scsi_noretry_cmd(cmd))
584 blk_end_request_all(req, error);
588 * Bleah. Leftovers again. Stick the
589 * leftovers in the front of the
590 * queue, and goose the queue again.
592 scsi_release_buffers(cmd);
593 scsi_requeue_command(q, cmd);
601 * This will goose the queue request function at the end, so we don't
602 * need to worry about launching another command.
604 __scsi_release_buffers(cmd, 0);
605 scsi_next_command(cmd);
609 static inline unsigned int scsi_sgtable_index(unsigned short nents)
613 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
618 index = get_count_order(nents) - 3;
623 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
625 struct scsi_host_sg_pool *sgp;
627 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
628 mempool_free(sgl, sgp->pool);
631 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
633 struct scsi_host_sg_pool *sgp;
635 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
636 return mempool_alloc(sgp->pool, gfp_mask);
639 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
646 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
647 gfp_mask, scsi_sg_alloc);
649 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
655 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
657 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
660 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
663 if (cmd->sdb.table.nents)
664 scsi_free_sgtable(&cmd->sdb);
666 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
668 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
669 struct scsi_data_buffer *bidi_sdb =
670 cmd->request->next_rq->special;
671 scsi_free_sgtable(bidi_sdb);
672 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
673 cmd->request->next_rq->special = NULL;
676 if (scsi_prot_sg_count(cmd))
677 scsi_free_sgtable(cmd->prot_sdb);
681 * Function: scsi_release_buffers()
683 * Purpose: Completion processing for block device I/O requests.
685 * Arguments: cmd - command that we are bailing.
687 * Lock status: Assumed that no lock is held upon entry.
691 * Notes: In the event that an upper level driver rejects a
692 * command, we must release resources allocated during
693 * the __init_io() function. Primarily this would involve
694 * the scatter-gather table, and potentially any bounce
697 void scsi_release_buffers(struct scsi_cmnd *cmd)
699 __scsi_release_buffers(cmd, 1);
701 EXPORT_SYMBOL(scsi_release_buffers);
703 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
707 switch(host_byte(result)) {
708 case DID_TRANSPORT_FAILFAST:
711 case DID_TARGET_FAILURE:
712 set_host_byte(cmd, DID_OK);
715 case DID_NEXUS_FAILURE:
716 set_host_byte(cmd, DID_OK);
728 * Function: scsi_io_completion()
730 * Purpose: Completion processing for block device I/O requests.
732 * Arguments: cmd - command that is finished.
734 * Lock status: Assumed that no lock is held upon entry.
738 * Notes: This function is matched in terms of capabilities to
739 * the function that created the scatter-gather list.
740 * In other words, if there are no bounce buffers
741 * (the normal case for most drivers), we don't need
742 * the logic to deal with cleaning up afterwards.
744 * We must call scsi_end_request(). This will finish off
745 * the specified number of sectors. If we are done, the
746 * command block will be released and the queue function
747 * will be goosed. If we are not done then we have to
748 * figure out what to do next:
750 * a) We can call scsi_requeue_command(). The request
751 * will be unprepared and put back on the queue. Then
752 * a new command will be created for it. This should
753 * be used if we made forward progress, or if we want
754 * to switch from READ(10) to READ(6) for example.
756 * b) We can call scsi_queue_insert(). The request will
757 * be put back on the queue and retried using the same
758 * command as before, possibly after a delay.
760 * c) We can call blk_end_request() with -EIO to fail
761 * the remainder of the request.
763 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
765 int result = cmd->result;
766 struct request_queue *q = cmd->device->request_queue;
767 struct request *req = cmd->request;
769 struct scsi_sense_hdr sshdr;
771 int sense_deferred = 0;
772 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
773 ACTION_DELAYED_RETRY} action;
774 char *description = NULL;
777 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
779 sense_deferred = scsi_sense_is_deferred(&sshdr);
782 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
784 if (sense_valid && req->sense) {
786 * SG_IO wants current and deferred errors
788 int len = 8 + cmd->sense_buffer[7];
790 if (len > SCSI_SENSE_BUFFERSIZE)
791 len = SCSI_SENSE_BUFFERSIZE;
792 memcpy(req->sense, cmd->sense_buffer, len);
793 req->sense_len = len;
796 error = __scsi_error_from_host_byte(cmd, result);
799 * __scsi_error_from_host_byte may have reset the host_byte
801 req->errors = cmd->result;
803 req->resid_len = scsi_get_resid(cmd);
805 if (scsi_bidi_cmnd(cmd)) {
807 * Bidi commands Must be complete as a whole,
808 * both sides at once.
810 req->next_rq->resid_len = scsi_in(cmd)->resid;
812 scsi_release_buffers(cmd);
813 blk_end_request_all(req, 0);
815 scsi_next_command(cmd);
818 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
820 * Certain non BLOCK_PC requests are commands that don't
821 * actually transfer anything (FLUSH), so cannot use
822 * good_bytes != blk_rq_bytes(req) as the signal for an error.
823 * This sets the error explicitly for the problem case.
825 error = __scsi_error_from_host_byte(cmd, result);
828 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
829 BUG_ON(blk_bidi_rq(req));
832 * Next deal with any sectors which we were able to correctly
835 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
837 blk_rq_sectors(req), good_bytes));
840 * Recovered errors need reporting, but they're always treated
841 * as success, so fiddle the result code here. For BLOCK_PC
842 * we already took a copy of the original into rq->errors which
843 * is what gets returned to the user
845 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
846 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
847 * print since caller wants ATA registers. Only occurs on
848 * SCSI ATA PASS_THROUGH commands when CK_COND=1
850 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
852 else if (!(req->cmd_flags & REQ_QUIET))
853 scsi_print_sense("", cmd);
855 /* BLOCK_PC may have set error */
860 * A number of bytes were successfully read. If there
861 * are leftovers and there is some kind of error
862 * (result != 0), retry the rest.
864 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
867 error = __scsi_error_from_host_byte(cmd, result);
869 if (host_byte(result) == DID_RESET) {
870 /* Third party bus reset or reset for error recovery
871 * reasons. Just retry the command and see what
874 action = ACTION_RETRY;
875 } else if (sense_valid && !sense_deferred) {
876 switch (sshdr.sense_key) {
878 if (cmd->device->removable) {
879 /* Detected disc change. Set a bit
880 * and quietly refuse further access.
882 cmd->device->changed = 1;
883 description = "Media Changed";
884 action = ACTION_FAIL;
886 /* Must have been a power glitch, or a
887 * bus reset. Could not have been a
888 * media change, so we just retry the
889 * command and see what happens.
891 action = ACTION_RETRY;
894 case ILLEGAL_REQUEST:
895 /* If we had an ILLEGAL REQUEST returned, then
896 * we may have performed an unsupported
897 * command. The only thing this should be
898 * would be a ten byte read where only a six
899 * byte read was supported. Also, on a system
900 * where READ CAPACITY failed, we may have
901 * read past the end of the disk.
903 if ((cmd->device->use_10_for_rw &&
904 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
905 (cmd->cmnd[0] == READ_10 ||
906 cmd->cmnd[0] == WRITE_10)) {
907 /* This will issue a new 6-byte command. */
908 cmd->device->use_10_for_rw = 0;
909 action = ACTION_REPREP;
910 } else if (sshdr.asc == 0x10) /* DIX */ {
911 description = "Host Data Integrity Failure";
912 action = ACTION_FAIL;
914 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
915 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
916 switch (cmd->cmnd[0]) {
918 description = "Discard failure";
922 if (cmd->cmnd[1] & 0x8)
923 description = "Discard failure";
926 "Write same failure";
929 description = "Invalid command failure";
932 action = ACTION_FAIL;
935 action = ACTION_FAIL;
937 case ABORTED_COMMAND:
938 action = ACTION_FAIL;
939 if (sshdr.asc == 0x10) { /* DIF */
940 description = "Target Data Integrity Failure";
945 /* If the device is in the process of becoming
946 * ready, or has a temporary blockage, retry.
948 if (sshdr.asc == 0x04) {
949 switch (sshdr.ascq) {
950 case 0x01: /* becoming ready */
951 case 0x04: /* format in progress */
952 case 0x05: /* rebuild in progress */
953 case 0x06: /* recalculation in progress */
954 case 0x07: /* operation in progress */
955 case 0x08: /* Long write in progress */
956 case 0x09: /* self test in progress */
957 case 0x14: /* space allocation in progress */
958 action = ACTION_DELAYED_RETRY;
961 description = "Device not ready";
962 action = ACTION_FAIL;
966 description = "Device not ready";
967 action = ACTION_FAIL;
970 case VOLUME_OVERFLOW:
971 /* See SSC3rXX or current. */
972 action = ACTION_FAIL;
975 description = "Unhandled sense code";
976 action = ACTION_FAIL;
980 description = "Unhandled error code";
981 action = ACTION_FAIL;
986 /* Give up and fail the remainder of the request */
987 scsi_release_buffers(cmd);
988 if (!(req->cmd_flags & REQ_QUIET)) {
990 scmd_printk(KERN_INFO, cmd, "%s\n",
992 scsi_print_result(cmd);
993 if (driver_byte(result) & DRIVER_SENSE)
994 scsi_print_sense("", cmd);
995 scsi_print_command(cmd);
997 if (blk_end_request_err(req, error))
998 scsi_requeue_command(q, cmd);
1000 scsi_next_command(cmd);
1003 /* Unprep the request and put it back at the head of the queue.
1004 * A new command will be prepared and issued.
1006 scsi_release_buffers(cmd);
1007 scsi_requeue_command(q, cmd);
1010 /* Retry the same command immediately */
1011 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1013 case ACTION_DELAYED_RETRY:
1014 /* Retry the same command after a delay */
1015 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1020 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1026 * If sg table allocation fails, requeue request later.
1028 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1030 return BLKPREP_DEFER;
1036 * Next, walk the list, and fill in the addresses and sizes of
1039 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1040 BUG_ON(count > sdb->table.nents);
1041 sdb->table.nents = count;
1042 sdb->length = blk_rq_bytes(req);
1047 * Function: scsi_init_io()
1049 * Purpose: SCSI I/O initialize function.
1051 * Arguments: cmd - Command descriptor we wish to initialize
1053 * Returns: 0 on success
1054 * BLKPREP_DEFER if the failure is retryable
1055 * BLKPREP_KILL if the failure is fatal
1057 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1059 struct request *rq = cmd->request;
1061 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1065 if (blk_bidi_rq(rq)) {
1066 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1067 scsi_sdb_cache, GFP_ATOMIC);
1069 error = BLKPREP_DEFER;
1073 rq->next_rq->special = bidi_sdb;
1074 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1079 if (blk_integrity_rq(rq)) {
1080 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1083 BUG_ON(prot_sdb == NULL);
1084 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1086 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1087 error = BLKPREP_DEFER;
1091 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1092 prot_sdb->table.sgl);
1093 BUG_ON(unlikely(count > ivecs));
1094 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1096 cmd->prot_sdb = prot_sdb;
1097 cmd->prot_sdb->table.nents = count;
1103 scsi_release_buffers(cmd);
1104 cmd->request->special = NULL;
1105 scsi_put_command(cmd);
1108 EXPORT_SYMBOL(scsi_init_io);
1110 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1111 struct request *req)
1113 struct scsi_cmnd *cmd;
1115 if (!req->special) {
1116 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1124 /* pull a tag out of the request if we have one */
1125 cmd->tag = req->tag;
1128 cmd->cmnd = req->cmd;
1129 cmd->prot_op = SCSI_PROT_NORMAL;
1134 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1136 struct scsi_cmnd *cmd;
1137 int ret = scsi_prep_state_check(sdev, req);
1139 if (ret != BLKPREP_OK)
1142 cmd = scsi_get_cmd_from_req(sdev, req);
1144 return BLKPREP_DEFER;
1147 * BLOCK_PC requests may transfer data, in which case they must
1148 * a bio attached to them. Or they might contain a SCSI command
1149 * that does not transfer data, in which case they may optionally
1150 * submit a request without an attached bio.
1155 BUG_ON(!req->nr_phys_segments);
1157 ret = scsi_init_io(cmd, GFP_ATOMIC);
1161 BUG_ON(blk_rq_bytes(req));
1163 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1167 cmd->cmd_len = req->cmd_len;
1168 if (!blk_rq_bytes(req))
1169 cmd->sc_data_direction = DMA_NONE;
1170 else if (rq_data_dir(req) == WRITE)
1171 cmd->sc_data_direction = DMA_TO_DEVICE;
1173 cmd->sc_data_direction = DMA_FROM_DEVICE;
1175 cmd->transfersize = blk_rq_bytes(req);
1176 cmd->allowed = req->retries;
1179 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1182 * Setup a REQ_TYPE_FS command. These are simple read/write request
1183 * from filesystems that still need to be translated to SCSI CDBs from
1186 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1188 struct scsi_cmnd *cmd;
1189 int ret = scsi_prep_state_check(sdev, req);
1191 if (ret != BLKPREP_OK)
1194 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1195 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1196 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1197 if (ret != BLKPREP_OK)
1202 * Filesystem requests must transfer data.
1204 BUG_ON(!req->nr_phys_segments);
1206 cmd = scsi_get_cmd_from_req(sdev, req);
1208 return BLKPREP_DEFER;
1210 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1211 return scsi_init_io(cmd, GFP_ATOMIC);
1213 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1215 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1217 int ret = BLKPREP_OK;
1220 * If the device is not in running state we will reject some
1223 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1224 switch (sdev->sdev_state) {
1226 case SDEV_TRANSPORT_OFFLINE:
1228 * If the device is offline we refuse to process any
1229 * commands. The device must be brought online
1230 * before trying any recovery commands.
1232 sdev_printk(KERN_ERR, sdev,
1233 "rejecting I/O to offline device\n");
1238 * If the device is fully deleted, we refuse to
1239 * process any commands as well.
1241 sdev_printk(KERN_ERR, sdev,
1242 "rejecting I/O to dead device\n");
1246 case SDEV_CREATED_BLOCK:
1247 ret = BLKPREP_DEFER;
1251 * If the devices is blocked we defer normal commands.
1253 if (!(req->cmd_flags & REQ_PREEMPT))
1254 ret = BLKPREP_DEFER;
1258 * For any other not fully online state we only allow
1259 * special commands. In particular any user initiated
1260 * command is not allowed.
1262 if (!(req->cmd_flags & REQ_PREEMPT))
1269 EXPORT_SYMBOL(scsi_prep_state_check);
1271 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1273 struct scsi_device *sdev = q->queuedata;
1277 req->errors = DID_NO_CONNECT << 16;
1278 /* release the command and kill it */
1280 struct scsi_cmnd *cmd = req->special;
1281 scsi_release_buffers(cmd);
1282 scsi_put_command(cmd);
1283 req->special = NULL;
1288 * If we defer, the blk_peek_request() returns NULL, but the
1289 * queue must be restarted, so we schedule a callback to happen
1292 if (sdev->device_busy == 0)
1293 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1296 req->cmd_flags |= REQ_DONTPREP;
1301 EXPORT_SYMBOL(scsi_prep_return);
1303 int scsi_prep_fn(struct request_queue *q, struct request *req)
1305 struct scsi_device *sdev = q->queuedata;
1306 int ret = BLKPREP_KILL;
1308 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1309 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1310 return scsi_prep_return(q, req, ret);
1312 EXPORT_SYMBOL(scsi_prep_fn);
1315 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1318 * Called with the queue_lock held.
1320 static inline int scsi_dev_queue_ready(struct request_queue *q,
1321 struct scsi_device *sdev)
1323 if (sdev->device_busy == 0 && sdev->device_blocked) {
1325 * unblock after device_blocked iterates to zero
1327 if (--sdev->device_blocked == 0) {
1329 sdev_printk(KERN_INFO, sdev,
1330 "unblocking device at zero depth\n"));
1332 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1336 if (scsi_device_is_busy(sdev))
1344 * scsi_target_queue_ready: checks if there we can send commands to target
1345 * @sdev: scsi device on starget to check.
1347 * Called with the host lock held.
1349 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1350 struct scsi_device *sdev)
1352 struct scsi_target *starget = scsi_target(sdev);
1354 if (starget->single_lun) {
1355 if (starget->starget_sdev_user &&
1356 starget->starget_sdev_user != sdev)
1358 starget->starget_sdev_user = sdev;
1361 if (starget->target_busy == 0 && starget->target_blocked) {
1363 * unblock after target_blocked iterates to zero
1365 if (--starget->target_blocked == 0) {
1366 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1367 "unblocking target at zero depth\n"));
1372 if (scsi_target_is_busy(starget)) {
1373 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1381 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1382 * return 0. We must end up running the queue again whenever 0 is
1383 * returned, else IO can hang.
1385 * Called with host_lock held.
1387 static inline int scsi_host_queue_ready(struct request_queue *q,
1388 struct Scsi_Host *shost,
1389 struct scsi_device *sdev)
1391 if (scsi_host_in_recovery(shost))
1393 if (shost->host_busy == 0 && shost->host_blocked) {
1395 * unblock after host_blocked iterates to zero
1397 if (--shost->host_blocked == 0) {
1399 printk("scsi%d unblocking host at zero depth\n",
1405 if (scsi_host_is_busy(shost)) {
1406 if (list_empty(&sdev->starved_entry))
1407 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1411 /* We're OK to process the command, so we can't be starved */
1412 if (!list_empty(&sdev->starved_entry))
1413 list_del_init(&sdev->starved_entry);
1419 * Busy state exporting function for request stacking drivers.
1421 * For efficiency, no lock is taken to check the busy state of
1422 * shost/starget/sdev, since the returned value is not guaranteed and
1423 * may be changed after request stacking drivers call the function,
1424 * regardless of taking lock or not.
1426 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1427 * needs to return 'not busy'. Otherwise, request stacking drivers
1428 * may hold requests forever.
1430 static int scsi_lld_busy(struct request_queue *q)
1432 struct scsi_device *sdev = q->queuedata;
1433 struct Scsi_Host *shost;
1435 if (blk_queue_dying(q))
1441 * Ignore host/starget busy state.
1442 * Since block layer does not have a concept of fairness across
1443 * multiple queues, congestion of host/starget needs to be handled
1446 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1453 * Kill a request for a dead device
1455 static void scsi_kill_request(struct request *req, struct request_queue *q)
1457 struct scsi_cmnd *cmd = req->special;
1458 struct scsi_device *sdev;
1459 struct scsi_target *starget;
1460 struct Scsi_Host *shost;
1462 blk_start_request(req);
1464 scmd_printk(KERN_INFO, cmd, "killing request\n");
1467 starget = scsi_target(sdev);
1469 scsi_init_cmd_errh(cmd);
1470 cmd->result = DID_NO_CONNECT << 16;
1471 atomic_inc(&cmd->device->iorequest_cnt);
1474 * SCSI request completion path will do scsi_device_unbusy(),
1475 * bump busy counts. To bump the counters, we need to dance
1476 * with the locks as normal issue path does.
1478 sdev->device_busy++;
1479 spin_unlock(sdev->request_queue->queue_lock);
1480 spin_lock(shost->host_lock);
1482 starget->target_busy++;
1483 spin_unlock(shost->host_lock);
1484 spin_lock(sdev->request_queue->queue_lock);
1486 blk_complete_request(req);
1489 static void scsi_softirq_done(struct request *rq)
1491 struct scsi_cmnd *cmd = rq->special;
1492 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1495 INIT_LIST_HEAD(&cmd->eh_entry);
1497 atomic_inc(&cmd->device->iodone_cnt);
1499 atomic_inc(&cmd->device->ioerr_cnt);
1501 disposition = scsi_decide_disposition(cmd);
1502 if (disposition != SUCCESS &&
1503 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1504 sdev_printk(KERN_ERR, cmd->device,
1505 "timing out command, waited %lus\n",
1507 disposition = SUCCESS;
1510 scsi_log_completion(cmd, disposition);
1512 switch (disposition) {
1514 scsi_finish_command(cmd);
1517 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1519 case ADD_TO_MLQUEUE:
1520 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1523 if (!scsi_eh_scmd_add(cmd, 0))
1524 scsi_finish_command(cmd);
1529 * Function: scsi_request_fn()
1531 * Purpose: Main strategy routine for SCSI.
1533 * Arguments: q - Pointer to actual queue.
1537 * Lock status: IO request lock assumed to be held when called.
1539 static void scsi_request_fn(struct request_queue *q)
1541 struct scsi_device *sdev = q->queuedata;
1542 struct Scsi_Host *shost;
1543 struct scsi_cmnd *cmd;
1544 struct request *req;
1546 if(!get_device(&sdev->sdev_gendev))
1547 /* We must be tearing the block queue down already */
1551 * To start with, we keep looping until the queue is empty, or until
1552 * the host is no longer able to accept any more requests.
1558 * get next queueable request. We do this early to make sure
1559 * that the request is fully prepared even if we cannot
1562 req = blk_peek_request(q);
1563 if (!req || !scsi_dev_queue_ready(q, sdev))
1566 if (unlikely(!scsi_device_online(sdev))) {
1567 sdev_printk(KERN_ERR, sdev,
1568 "rejecting I/O to offline device\n");
1569 scsi_kill_request(req, q);
1575 * Remove the request from the request list.
1577 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1578 blk_start_request(req);
1579 sdev->device_busy++;
1581 spin_unlock(q->queue_lock);
1583 if (unlikely(cmd == NULL)) {
1584 printk(KERN_CRIT "impossible request in %s.\n"
1585 "please mail a stack trace to "
1586 "linux-scsi@vger.kernel.org\n",
1588 blk_dump_rq_flags(req, "foo");
1591 spin_lock(shost->host_lock);
1594 * We hit this when the driver is using a host wide
1595 * tag map. For device level tag maps the queue_depth check
1596 * in the device ready fn would prevent us from trying
1597 * to allocate a tag. Since the map is a shared host resource
1598 * we add the dev to the starved list so it eventually gets
1599 * a run when a tag is freed.
1601 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1602 if (list_empty(&sdev->starved_entry))
1603 list_add_tail(&sdev->starved_entry,
1604 &shost->starved_list);
1608 if (!scsi_target_queue_ready(shost, sdev))
1611 if (!scsi_host_queue_ready(q, shost, sdev))
1614 scsi_target(sdev)->target_busy++;
1618 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1619 * take the lock again.
1621 spin_unlock_irq(shost->host_lock);
1624 * Finally, initialize any error handling parameters, and set up
1625 * the timers for timeouts.
1627 scsi_init_cmd_errh(cmd);
1630 * Dispatch the command to the low-level driver.
1632 rtn = scsi_dispatch_cmd(cmd);
1633 spin_lock_irq(q->queue_lock);
1641 spin_unlock_irq(shost->host_lock);
1644 * lock q, handle tag, requeue req, and decrement device_busy. We
1645 * must return with queue_lock held.
1647 * Decrementing device_busy without checking it is OK, as all such
1648 * cases (host limits or settings) should run the queue at some
1651 spin_lock_irq(q->queue_lock);
1652 blk_requeue_request(q, req);
1653 sdev->device_busy--;
1655 if (sdev->device_busy == 0)
1656 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1658 /* must be careful here...if we trigger the ->remove() function
1659 * we cannot be holding the q lock */
1660 spin_unlock_irq(q->queue_lock);
1661 put_device(&sdev->sdev_gendev);
1662 spin_lock_irq(q->queue_lock);
1665 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1667 struct device *host_dev;
1668 u64 bounce_limit = 0xffffffff;
1670 if (shost->unchecked_isa_dma)
1671 return BLK_BOUNCE_ISA;
1673 * Platforms with virtual-DMA translation
1674 * hardware have no practical limit.
1676 if (!PCI_DMA_BUS_IS_PHYS)
1677 return BLK_BOUNCE_ANY;
1679 host_dev = scsi_get_device(shost);
1680 if (host_dev && host_dev->dma_mask)
1681 bounce_limit = *host_dev->dma_mask;
1683 return bounce_limit;
1685 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1687 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1688 request_fn_proc *request_fn)
1690 struct request_queue *q;
1691 struct device *dev = shost->dma_dev;
1693 q = blk_init_queue(request_fn, NULL);
1698 * this limit is imposed by hardware restrictions
1700 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1701 SCSI_MAX_SG_CHAIN_SEGMENTS));
1703 if (scsi_host_prot_dma(shost)) {
1704 shost->sg_prot_tablesize =
1705 min_not_zero(shost->sg_prot_tablesize,
1706 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1707 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1708 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1711 blk_queue_max_hw_sectors(q, shost->max_sectors);
1712 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1713 blk_queue_segment_boundary(q, shost->dma_boundary);
1714 dma_set_seg_boundary(dev, shost->dma_boundary);
1716 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1718 if (!shost->use_clustering)
1719 q->limits.cluster = 0;
1722 * set a reasonable default alignment on word boundaries: the
1723 * host and device may alter it using
1724 * blk_queue_update_dma_alignment() later.
1726 blk_queue_dma_alignment(q, 0x03);
1730 EXPORT_SYMBOL(__scsi_alloc_queue);
1732 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1734 struct request_queue *q;
1736 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1740 blk_queue_prep_rq(q, scsi_prep_fn);
1741 blk_queue_softirq_done(q, scsi_softirq_done);
1742 blk_queue_rq_timed_out(q, scsi_times_out);
1743 blk_queue_lld_busy(q, scsi_lld_busy);
1748 * Function: scsi_block_requests()
1750 * Purpose: Utility function used by low-level drivers to prevent further
1751 * commands from being queued to the device.
1753 * Arguments: shost - Host in question
1757 * Lock status: No locks are assumed held.
1759 * Notes: There is no timer nor any other means by which the requests
1760 * get unblocked other than the low-level driver calling
1761 * scsi_unblock_requests().
1763 void scsi_block_requests(struct Scsi_Host *shost)
1765 shost->host_self_blocked = 1;
1767 EXPORT_SYMBOL(scsi_block_requests);
1770 * Function: scsi_unblock_requests()
1772 * Purpose: Utility function used by low-level drivers to allow further
1773 * commands from being queued to the device.
1775 * Arguments: shost - Host in question
1779 * Lock status: No locks are assumed held.
1781 * Notes: There is no timer nor any other means by which the requests
1782 * get unblocked other than the low-level driver calling
1783 * scsi_unblock_requests().
1785 * This is done as an API function so that changes to the
1786 * internals of the scsi mid-layer won't require wholesale
1787 * changes to drivers that use this feature.
1789 void scsi_unblock_requests(struct Scsi_Host *shost)
1791 shost->host_self_blocked = 0;
1792 scsi_run_host_queues(shost);
1794 EXPORT_SYMBOL(scsi_unblock_requests);
1796 int __init scsi_init_queue(void)
1800 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1801 sizeof(struct scsi_data_buffer),
1803 if (!scsi_sdb_cache) {
1804 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1808 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1809 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1810 int size = sgp->size * sizeof(struct scatterlist);
1812 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1813 SLAB_HWCACHE_ALIGN, NULL);
1815 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1820 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1823 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1832 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1833 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1835 mempool_destroy(sgp->pool);
1837 kmem_cache_destroy(sgp->slab);
1839 kmem_cache_destroy(scsi_sdb_cache);
1844 void scsi_exit_queue(void)
1848 kmem_cache_destroy(scsi_sdb_cache);
1850 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1851 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1852 mempool_destroy(sgp->pool);
1853 kmem_cache_destroy(sgp->slab);
1858 * scsi_mode_select - issue a mode select
1859 * @sdev: SCSI device to be queried
1860 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1861 * @sp: Save page bit (0 == don't save, 1 == save)
1862 * @modepage: mode page being requested
1863 * @buffer: request buffer (may not be smaller than eight bytes)
1864 * @len: length of request buffer.
1865 * @timeout: command timeout
1866 * @retries: number of retries before failing
1867 * @data: returns a structure abstracting the mode header data
1868 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1869 * must be SCSI_SENSE_BUFFERSIZE big.
1871 * Returns zero if successful; negative error number or scsi
1876 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1877 unsigned char *buffer, int len, int timeout, int retries,
1878 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1880 unsigned char cmd[10];
1881 unsigned char *real_buffer;
1884 memset(cmd, 0, sizeof(cmd));
1885 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1887 if (sdev->use_10_for_ms) {
1890 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1893 memcpy(real_buffer + 8, buffer, len);
1897 real_buffer[2] = data->medium_type;
1898 real_buffer[3] = data->device_specific;
1899 real_buffer[4] = data->longlba ? 0x01 : 0;
1901 real_buffer[6] = data->block_descriptor_length >> 8;
1902 real_buffer[7] = data->block_descriptor_length;
1904 cmd[0] = MODE_SELECT_10;
1908 if (len > 255 || data->block_descriptor_length > 255 ||
1912 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1915 memcpy(real_buffer + 4, buffer, len);
1918 real_buffer[1] = data->medium_type;
1919 real_buffer[2] = data->device_specific;
1920 real_buffer[3] = data->block_descriptor_length;
1923 cmd[0] = MODE_SELECT;
1927 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1928 sshdr, timeout, retries, NULL);
1932 EXPORT_SYMBOL_GPL(scsi_mode_select);
1935 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1936 * @sdev: SCSI device to be queried
1937 * @dbd: set if mode sense will allow block descriptors to be returned
1938 * @modepage: mode page being requested
1939 * @buffer: request buffer (may not be smaller than eight bytes)
1940 * @len: length of request buffer.
1941 * @timeout: command timeout
1942 * @retries: number of retries before failing
1943 * @data: returns a structure abstracting the mode header data
1944 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1945 * must be SCSI_SENSE_BUFFERSIZE big.
1947 * Returns zero if unsuccessful, or the header offset (either 4
1948 * or 8 depending on whether a six or ten byte command was
1949 * issued) if successful.
1952 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1953 unsigned char *buffer, int len, int timeout, int retries,
1954 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1956 unsigned char cmd[12];
1960 struct scsi_sense_hdr my_sshdr;
1962 memset(data, 0, sizeof(*data));
1963 memset(&cmd[0], 0, 12);
1964 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1967 /* caller might not be interested in sense, but we need it */
1972 use_10_for_ms = sdev->use_10_for_ms;
1974 if (use_10_for_ms) {
1978 cmd[0] = MODE_SENSE_10;
1985 cmd[0] = MODE_SENSE;
1990 memset(buffer, 0, len);
1992 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1993 sshdr, timeout, retries, NULL);
1995 /* This code looks awful: what it's doing is making sure an
1996 * ILLEGAL REQUEST sense return identifies the actual command
1997 * byte as the problem. MODE_SENSE commands can return
1998 * ILLEGAL REQUEST if the code page isn't supported */
2000 if (use_10_for_ms && !scsi_status_is_good(result) &&
2001 (driver_byte(result) & DRIVER_SENSE)) {
2002 if (scsi_sense_valid(sshdr)) {
2003 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2004 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2006 * Invalid command operation code
2008 sdev->use_10_for_ms = 0;
2014 if(scsi_status_is_good(result)) {
2015 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2016 (modepage == 6 || modepage == 8))) {
2017 /* Initio breakage? */
2020 data->medium_type = 0;
2021 data->device_specific = 0;
2023 data->block_descriptor_length = 0;
2024 } else if(use_10_for_ms) {
2025 data->length = buffer[0]*256 + buffer[1] + 2;
2026 data->medium_type = buffer[2];
2027 data->device_specific = buffer[3];
2028 data->longlba = buffer[4] & 0x01;
2029 data->block_descriptor_length = buffer[6]*256
2032 data->length = buffer[0] + 1;
2033 data->medium_type = buffer[1];
2034 data->device_specific = buffer[2];
2035 data->block_descriptor_length = buffer[3];
2037 data->header_length = header_length;
2042 EXPORT_SYMBOL(scsi_mode_sense);
2045 * scsi_test_unit_ready - test if unit is ready
2046 * @sdev: scsi device to change the state of.
2047 * @timeout: command timeout
2048 * @retries: number of retries before failing
2049 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2050 * returning sense. Make sure that this is cleared before passing
2053 * Returns zero if unsuccessful or an error if TUR failed. For
2054 * removable media, UNIT_ATTENTION sets ->changed flag.
2057 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2058 struct scsi_sense_hdr *sshdr_external)
2061 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2063 struct scsi_sense_hdr *sshdr;
2066 if (!sshdr_external)
2067 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2069 sshdr = sshdr_external;
2071 /* try to eat the UNIT_ATTENTION if there are enough retries */
2073 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2074 timeout, retries, NULL);
2075 if (sdev->removable && scsi_sense_valid(sshdr) &&
2076 sshdr->sense_key == UNIT_ATTENTION)
2078 } while (scsi_sense_valid(sshdr) &&
2079 sshdr->sense_key == UNIT_ATTENTION && --retries);
2081 if (!sshdr_external)
2085 EXPORT_SYMBOL(scsi_test_unit_ready);
2088 * scsi_device_set_state - Take the given device through the device state model.
2089 * @sdev: scsi device to change the state of.
2090 * @state: state to change to.
2092 * Returns zero if unsuccessful or an error if the requested
2093 * transition is illegal.
2096 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2098 enum scsi_device_state oldstate = sdev->sdev_state;
2100 if (state == oldstate)
2106 case SDEV_CREATED_BLOCK:
2117 case SDEV_TRANSPORT_OFFLINE:
2130 case SDEV_TRANSPORT_OFFLINE:
2138 case SDEV_TRANSPORT_OFFLINE:
2153 case SDEV_CREATED_BLOCK:
2160 case SDEV_CREATED_BLOCK:
2175 case SDEV_TRANSPORT_OFFLINE:
2188 case SDEV_TRANSPORT_OFFLINE:
2197 sdev->sdev_state = state;
2201 SCSI_LOG_ERROR_RECOVERY(1,
2202 sdev_printk(KERN_ERR, sdev,
2203 "Illegal state transition %s->%s\n",
2204 scsi_device_state_name(oldstate),
2205 scsi_device_state_name(state))
2209 EXPORT_SYMBOL(scsi_device_set_state);
2212 * sdev_evt_emit - emit a single SCSI device uevent
2213 * @sdev: associated SCSI device
2214 * @evt: event to emit
2216 * Send a single uevent (scsi_event) to the associated scsi_device.
2218 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2223 switch (evt->evt_type) {
2224 case SDEV_EVT_MEDIA_CHANGE:
2225 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2235 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2239 * sdev_evt_thread - send a uevent for each scsi event
2240 * @work: work struct for scsi_device
2242 * Dispatch queued events to their associated scsi_device kobjects
2245 void scsi_evt_thread(struct work_struct *work)
2247 struct scsi_device *sdev;
2248 LIST_HEAD(event_list);
2250 sdev = container_of(work, struct scsi_device, event_work);
2253 struct scsi_event *evt;
2254 struct list_head *this, *tmp;
2255 unsigned long flags;
2257 spin_lock_irqsave(&sdev->list_lock, flags);
2258 list_splice_init(&sdev->event_list, &event_list);
2259 spin_unlock_irqrestore(&sdev->list_lock, flags);
2261 if (list_empty(&event_list))
2264 list_for_each_safe(this, tmp, &event_list) {
2265 evt = list_entry(this, struct scsi_event, node);
2266 list_del(&evt->node);
2267 scsi_evt_emit(sdev, evt);
2274 * sdev_evt_send - send asserted event to uevent thread
2275 * @sdev: scsi_device event occurred on
2276 * @evt: event to send
2278 * Assert scsi device event asynchronously.
2280 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2282 unsigned long flags;
2285 /* FIXME: currently this check eliminates all media change events
2286 * for polled devices. Need to update to discriminate between AN
2287 * and polled events */
2288 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2294 spin_lock_irqsave(&sdev->list_lock, flags);
2295 list_add_tail(&evt->node, &sdev->event_list);
2296 schedule_work(&sdev->event_work);
2297 spin_unlock_irqrestore(&sdev->list_lock, flags);
2299 EXPORT_SYMBOL_GPL(sdev_evt_send);
2302 * sdev_evt_alloc - allocate a new scsi event
2303 * @evt_type: type of event to allocate
2304 * @gfpflags: GFP flags for allocation
2306 * Allocates and returns a new scsi_event.
2308 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2311 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2315 evt->evt_type = evt_type;
2316 INIT_LIST_HEAD(&evt->node);
2318 /* evt_type-specific initialization, if any */
2320 case SDEV_EVT_MEDIA_CHANGE:
2328 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2331 * sdev_evt_send_simple - send asserted event to uevent thread
2332 * @sdev: scsi_device event occurred on
2333 * @evt_type: type of event to send
2334 * @gfpflags: GFP flags for allocation
2336 * Assert scsi device event asynchronously, given an event type.
2338 void sdev_evt_send_simple(struct scsi_device *sdev,
2339 enum scsi_device_event evt_type, gfp_t gfpflags)
2341 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2343 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2348 sdev_evt_send(sdev, evt);
2350 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2353 * scsi_device_quiesce - Block user issued commands.
2354 * @sdev: scsi device to quiesce.
2356 * This works by trying to transition to the SDEV_QUIESCE state
2357 * (which must be a legal transition). When the device is in this
2358 * state, only special requests will be accepted, all others will
2359 * be deferred. Since special requests may also be requeued requests,
2360 * a successful return doesn't guarantee the device will be
2361 * totally quiescent.
2363 * Must be called with user context, may sleep.
2365 * Returns zero if unsuccessful or an error if not.
2368 scsi_device_quiesce(struct scsi_device *sdev)
2370 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2374 scsi_run_queue(sdev->request_queue);
2375 while (sdev->device_busy) {
2376 msleep_interruptible(200);
2377 scsi_run_queue(sdev->request_queue);
2381 EXPORT_SYMBOL(scsi_device_quiesce);
2384 * scsi_device_resume - Restart user issued commands to a quiesced device.
2385 * @sdev: scsi device to resume.
2387 * Moves the device from quiesced back to running and restarts the
2390 * Must be called with user context, may sleep.
2392 void scsi_device_resume(struct scsi_device *sdev)
2394 /* check if the device state was mutated prior to resume, and if
2395 * so assume the state is being managed elsewhere (for example
2396 * device deleted during suspend)
2398 if (sdev->sdev_state != SDEV_QUIESCE ||
2399 scsi_device_set_state(sdev, SDEV_RUNNING))
2401 scsi_run_queue(sdev->request_queue);
2403 EXPORT_SYMBOL(scsi_device_resume);
2406 device_quiesce_fn(struct scsi_device *sdev, void *data)
2408 scsi_device_quiesce(sdev);
2412 scsi_target_quiesce(struct scsi_target *starget)
2414 starget_for_each_device(starget, NULL, device_quiesce_fn);
2416 EXPORT_SYMBOL(scsi_target_quiesce);
2419 device_resume_fn(struct scsi_device *sdev, void *data)
2421 scsi_device_resume(sdev);
2425 scsi_target_resume(struct scsi_target *starget)
2427 starget_for_each_device(starget, NULL, device_resume_fn);
2429 EXPORT_SYMBOL(scsi_target_resume);
2432 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2433 * @sdev: device to block
2435 * Block request made by scsi lld's to temporarily stop all
2436 * scsi commands on the specified device. Called from interrupt
2437 * or normal process context.
2439 * Returns zero if successful or error if not
2442 * This routine transitions the device to the SDEV_BLOCK state
2443 * (which must be a legal transition). When the device is in this
2444 * state, all commands are deferred until the scsi lld reenables
2445 * the device with scsi_device_unblock or device_block_tmo fires.
2448 scsi_internal_device_block(struct scsi_device *sdev)
2450 struct request_queue *q = sdev->request_queue;
2451 unsigned long flags;
2454 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2456 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2463 * The device has transitioned to SDEV_BLOCK. Stop the
2464 * block layer from calling the midlayer with this device's
2467 spin_lock_irqsave(q->queue_lock, flags);
2469 spin_unlock_irqrestore(q->queue_lock, flags);
2473 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2476 * scsi_internal_device_unblock - resume a device after a block request
2477 * @sdev: device to resume
2478 * @new_state: state to set devices to after unblocking
2480 * Called by scsi lld's or the midlayer to restart the device queue
2481 * for the previously suspended scsi device. Called from interrupt or
2482 * normal process context.
2484 * Returns zero if successful or error if not.
2487 * This routine transitions the device to the SDEV_RUNNING state
2488 * or to one of the offline states (which must be a legal transition)
2489 * allowing the midlayer to goose the queue for this device.
2492 scsi_internal_device_unblock(struct scsi_device *sdev,
2493 enum scsi_device_state new_state)
2495 struct request_queue *q = sdev->request_queue;
2496 unsigned long flags;
2499 * Try to transition the scsi device to SDEV_RUNNING or one of the
2500 * offlined states and goose the device queue if successful.
2502 if ((sdev->sdev_state == SDEV_BLOCK) ||
2503 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2504 sdev->sdev_state = new_state;
2505 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2506 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2507 new_state == SDEV_OFFLINE)
2508 sdev->sdev_state = new_state;
2510 sdev->sdev_state = SDEV_CREATED;
2511 } else if (sdev->sdev_state != SDEV_CANCEL &&
2512 sdev->sdev_state != SDEV_OFFLINE)
2515 spin_lock_irqsave(q->queue_lock, flags);
2517 spin_unlock_irqrestore(q->queue_lock, flags);
2521 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2524 device_block(struct scsi_device *sdev, void *data)
2526 scsi_internal_device_block(sdev);
2530 target_block(struct device *dev, void *data)
2532 if (scsi_is_target_device(dev))
2533 starget_for_each_device(to_scsi_target(dev), NULL,
2539 scsi_target_block(struct device *dev)
2541 if (scsi_is_target_device(dev))
2542 starget_for_each_device(to_scsi_target(dev), NULL,
2545 device_for_each_child(dev, NULL, target_block);
2547 EXPORT_SYMBOL_GPL(scsi_target_block);
2550 device_unblock(struct scsi_device *sdev, void *data)
2552 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2556 target_unblock(struct device *dev, void *data)
2558 if (scsi_is_target_device(dev))
2559 starget_for_each_device(to_scsi_target(dev), data,
2565 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2567 if (scsi_is_target_device(dev))
2568 starget_for_each_device(to_scsi_target(dev), &new_state,
2571 device_for_each_child(dev, &new_state, target_unblock);
2573 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2576 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2577 * @sgl: scatter-gather list
2578 * @sg_count: number of segments in sg
2579 * @offset: offset in bytes into sg, on return offset into the mapped area
2580 * @len: bytes to map, on return number of bytes mapped
2582 * Returns virtual address of the start of the mapped page
2584 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2585 size_t *offset, size_t *len)
2588 size_t sg_len = 0, len_complete = 0;
2589 struct scatterlist *sg;
2592 WARN_ON(!irqs_disabled());
2594 for_each_sg(sgl, sg, sg_count, i) {
2595 len_complete = sg_len; /* Complete sg-entries */
2596 sg_len += sg->length;
2597 if (sg_len > *offset)
2601 if (unlikely(i == sg_count)) {
2602 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2604 __func__, sg_len, *offset, sg_count);
2609 /* Offset starting from the beginning of first page in this sg-entry */
2610 *offset = *offset - len_complete + sg->offset;
2612 /* Assumption: contiguous pages can be accessed as "page + i" */
2613 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2614 *offset &= ~PAGE_MASK;
2616 /* Bytes in this sg-entry from *offset to the end of the page */
2617 sg_len = PAGE_SIZE - *offset;
2621 return kmap_atomic(page);
2623 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2626 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2627 * @virt: virtual address to be unmapped
2629 void scsi_kunmap_atomic_sg(void *virt)
2631 kunmap_atomic(virt);
2633 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2635 void sdev_disable_disk_events(struct scsi_device *sdev)
2637 atomic_inc(&sdev->disk_events_disable_depth);
2639 EXPORT_SYMBOL(sdev_disable_disk_events);
2641 void sdev_enable_disk_events(struct scsi_device *sdev)
2643 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2645 atomic_dec(&sdev->disk_events_disable_depth);
2647 EXPORT_SYMBOL(sdev_enable_disk_events);