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/blkdev.h>
12 #include <linux/completion.h>
13 #include <linux/kernel.h>
14 #include <linux/mempool.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/pci.h>
19 #include <scsi/scsi.h>
20 #include <scsi/scsi_dbg.h>
21 #include <scsi/scsi_device.h>
22 #include <scsi/scsi_driver.h>
23 #include <scsi/scsi_eh.h>
24 #include <scsi/scsi_host.h>
25 #include <scsi/scsi_request.h>
27 #include "scsi_priv.h"
28 #include "scsi_logging.h"
31 #define SG_MEMPOOL_NR (sizeof(scsi_sg_pools)/sizeof(struct scsi_host_sg_pool))
32 #define SG_MEMPOOL_SIZE 32
34 struct scsi_host_sg_pool {
41 #if (SCSI_MAX_PHYS_SEGMENTS < 32)
42 #error SCSI_MAX_PHYS_SEGMENTS is too small
45 #define SP(x) { x, "sgpool-" #x }
46 struct scsi_host_sg_pool scsi_sg_pools[] = {
50 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
52 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
54 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
56 #if (SCSI_MAX_PHYS_SEGMENTS > 256)
57 #error SCSI_MAX_PHYS_SEGMENTS is too large
67 * Function: scsi_insert_special_req()
69 * Purpose: Insert pre-formed request into request queue.
71 * Arguments: sreq - request that is ready to be queued.
72 * at_head - boolean. True if we should insert at head
73 * of queue, false if we should insert at tail.
75 * Lock status: Assumed that lock is not held upon entry.
79 * Notes: This function is called from character device and from
80 * ioctl types of functions where the caller knows exactly
81 * what SCSI command needs to be issued. The idea is that
82 * we merely inject the command into the queue (at the head
83 * for now), and then call the queue request function to actually
86 int scsi_insert_special_req(struct scsi_request *sreq, int at_head)
89 * Because users of this function are apt to reuse requests with no
90 * modification, we have to sanitise the request flags here
92 sreq->sr_request->flags &= ~REQ_DONTPREP;
93 blk_insert_request(sreq->sr_device->request_queue, sreq->sr_request,
99 * Function: scsi_queue_insert()
101 * Purpose: Insert a command in the midlevel queue.
103 * Arguments: cmd - command that we are adding to queue.
104 * reason - why we are inserting command to queue.
106 * Lock status: Assumed that lock is not held upon entry.
110 * Notes: We do this for one of two cases. Either the host is busy
111 * and it cannot accept any more commands for the time being,
112 * or the device returned QUEUE_FULL and can accept no more
114 * Notes: This could be called either from an interrupt context or a
115 * normal process context.
117 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
119 struct Scsi_Host *host = cmd->device->host;
120 struct scsi_device *device = cmd->device;
123 printk("Inserting command %p into mlqueue\n", cmd));
126 * We are inserting the command into the ml queue. First, we
127 * cancel the timer, so it doesn't time out.
129 scsi_delete_timer(cmd);
132 * Next, set the appropriate busy bit for the device/host.
134 * If the host/device isn't busy, assume that something actually
135 * completed, and that we should be able to queue a command now.
137 * Note that the prior mid-layer assumption that any host could
138 * always queue at least one command is now broken. The mid-layer
139 * will implement a user specifiable stall (see
140 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
141 * if a command is requeued with no other commands outstanding
142 * either for the device or for the host.
144 if (reason == SCSI_MLQUEUE_HOST_BUSY)
145 host->host_blocked = host->max_host_blocked;
146 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
147 device->device_blocked = device->max_device_blocked;
150 * Register the fact that we own the thing for now.
152 cmd->state = SCSI_STATE_MLQUEUE;
153 cmd->owner = SCSI_OWNER_MIDLEVEL;
156 * Decrement the counters, since these commands are no longer
157 * active on the host/device.
159 scsi_device_unbusy(device);
162 * Insert this command at the head of the queue for it's device.
163 * It will go before all other commands that are already in the queue.
165 * NOTE: there is magic here about the way the queue is plugged if
166 * we have no outstanding commands.
168 * Although this *doesn't* plug the queue, it does call the request
169 * function. The SCSI request function detects the blocked condition
170 * and plugs the queue appropriately.
172 blk_insert_request(device->request_queue, cmd->request, 1, cmd, 1);
177 * Function: scsi_do_req
179 * Purpose: Queue a SCSI request
181 * Arguments: sreq - command descriptor.
182 * cmnd - actual SCSI command to be performed.
183 * buffer - data buffer.
184 * bufflen - size of data buffer.
185 * done - completion function to be run.
186 * timeout - how long to let it run before timeout.
187 * retries - number of retries we allow.
189 * Lock status: No locks held upon entry.
193 * Notes: This function is only used for queueing requests for things
194 * like ioctls and character device requests - this is because
195 * we essentially just inject a request into the queue for the
198 * In order to support the scsi_device_quiesce function, we
199 * now inject requests on the *head* of the device queue
200 * rather than the tail.
202 void scsi_do_req(struct scsi_request *sreq, const void *cmnd,
203 void *buffer, unsigned bufflen,
204 void (*done)(struct scsi_cmnd *),
205 int timeout, int retries)
208 * If the upper level driver is reusing these things, then
209 * we should release the low-level block now. Another one will
210 * be allocated later when this request is getting queued.
212 __scsi_release_request(sreq);
215 * Our own function scsi_done (which marks the host as not busy,
216 * disables the timeout counter, etc) will be called by us or by the
217 * scsi_hosts[host].queuecommand() function needs to also call
218 * the completion function for the high level driver.
220 memcpy(sreq->sr_cmnd, cmnd, sizeof(sreq->sr_cmnd));
221 sreq->sr_bufflen = bufflen;
222 sreq->sr_buffer = buffer;
223 sreq->sr_allowed = retries;
224 sreq->sr_done = done;
225 sreq->sr_timeout_per_command = timeout;
227 if (sreq->sr_cmd_len == 0)
228 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]);
231 * head injection *required* here otherwise quiesce won't work
233 scsi_insert_special_req(sreq, 1);
236 static void scsi_wait_done(struct scsi_cmnd *cmd)
238 struct request *req = cmd->request;
239 struct request_queue *q = cmd->device->request_queue;
242 req->rq_status = RQ_SCSI_DONE; /* Busy, but indicate request done */
244 spin_lock_irqsave(q->queue_lock, flags);
245 if (blk_rq_tagged(req))
246 blk_queue_end_tag(q, req);
247 spin_unlock_irqrestore(q->queue_lock, flags);
250 complete(req->waiting);
253 void scsi_wait_req(struct scsi_request *sreq, const void *cmnd, void *buffer,
254 unsigned bufflen, int timeout, int retries)
256 DECLARE_COMPLETION(wait);
258 sreq->sr_request->waiting = &wait;
259 sreq->sr_request->rq_status = RQ_SCSI_BUSY;
260 scsi_do_req(sreq, cmnd, buffer, bufflen, scsi_wait_done,
262 wait_for_completion(&wait);
263 sreq->sr_request->waiting = NULL;
264 if (sreq->sr_request->rq_status != RQ_SCSI_DONE)
265 sreq->sr_result |= (DRIVER_ERROR << 24);
267 __scsi_release_request(sreq);
271 * Function: scsi_init_cmd_errh()
273 * Purpose: Initialize cmd fields related to error handling.
275 * Arguments: cmd - command that is ready to be queued.
279 * Notes: This function has the job of initializing a number of
280 * fields related to error handling. Typically this will
281 * be called once for each command, as required.
283 static int scsi_init_cmd_errh(struct scsi_cmnd *cmd)
285 cmd->owner = SCSI_OWNER_MIDLEVEL;
286 cmd->serial_number = 0;
287 cmd->serial_number_at_timeout = 0;
288 cmd->abort_reason = 0;
290 memset(cmd->sense_buffer, 0, sizeof cmd->sense_buffer);
292 if (cmd->cmd_len == 0)
293 cmd->cmd_len = COMMAND_SIZE(cmd->cmnd[0]);
296 * We need saved copies of a number of fields - this is because
297 * error handling may need to overwrite these with different values
298 * to run different commands, and once error handling is complete,
299 * we will need to restore these values prior to running the actual
302 cmd->old_use_sg = cmd->use_sg;
303 cmd->old_cmd_len = cmd->cmd_len;
304 cmd->sc_old_data_direction = cmd->sc_data_direction;
305 cmd->old_underflow = cmd->underflow;
306 memcpy(cmd->data_cmnd, cmd->cmnd, sizeof(cmd->cmnd));
307 cmd->buffer = cmd->request_buffer;
308 cmd->bufflen = cmd->request_bufflen;
309 cmd->internal_timeout = NORMAL_TIMEOUT;
310 cmd->abort_reason = 0;
316 * Function: scsi_setup_cmd_retry()
318 * Purpose: Restore the command state for a retry
320 * Arguments: cmd - command to be restored
324 * Notes: Immediately prior to retrying a command, we need
325 * to restore certain fields that we saved above.
327 void scsi_setup_cmd_retry(struct scsi_cmnd *cmd)
329 memcpy(cmd->cmnd, cmd->data_cmnd, sizeof(cmd->data_cmnd));
330 cmd->request_buffer = cmd->buffer;
331 cmd->request_bufflen = cmd->bufflen;
332 cmd->use_sg = cmd->old_use_sg;
333 cmd->cmd_len = cmd->old_cmd_len;
334 cmd->sc_data_direction = cmd->sc_old_data_direction;
335 cmd->underflow = cmd->old_underflow;
338 void scsi_device_unbusy(struct scsi_device *sdev)
340 struct Scsi_Host *shost = sdev->host;
343 spin_lock_irqsave(shost->host_lock, flags);
345 if (unlikely(test_bit(SHOST_RECOVERY, &shost->shost_state) &&
347 scsi_eh_wakeup(shost);
348 spin_unlock(shost->host_lock);
349 spin_lock(&sdev->sdev_lock);
351 spin_unlock_irqrestore(&sdev->sdev_lock, flags);
355 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
356 * and call blk_run_queue for all the scsi_devices on the target -
357 * including current_sdev first.
359 * Called with *no* scsi locks held.
361 static void scsi_single_lun_run(struct scsi_device *current_sdev)
363 struct Scsi_Host *shost = current_sdev->host;
364 struct scsi_device *sdev, *tmp;
367 spin_lock_irqsave(shost->host_lock, flags);
368 current_sdev->sdev_target->starget_sdev_user = NULL;
369 spin_unlock_irqrestore(shost->host_lock, flags);
372 * Call blk_run_queue for all LUNs on the target, starting with
373 * current_sdev. We race with others (to set starget_sdev_user),
374 * but in most cases, we will be first. Ideally, each LU on the
375 * target would get some limited time or requests on the target.
377 blk_run_queue(current_sdev->request_queue);
379 spin_lock_irqsave(shost->host_lock, flags);
380 if (current_sdev->sdev_target->starget_sdev_user)
382 list_for_each_entry_safe(sdev, tmp, ¤t_sdev->same_target_siblings,
383 same_target_siblings) {
384 if (scsi_device_get(sdev))
387 spin_unlock_irqrestore(shost->host_lock, flags);
388 blk_run_queue(sdev->request_queue);
389 spin_lock_irqsave(shost->host_lock, flags);
391 scsi_device_put(sdev);
394 spin_unlock_irqrestore(shost->host_lock, flags);
398 * Function: scsi_run_queue()
400 * Purpose: Select a proper request queue to serve next
402 * Arguments: q - last request's queue
406 * Notes: The previous command was completely finished, start
407 * a new one if possible.
409 static void scsi_run_queue(struct request_queue *q)
411 struct scsi_device *sdev = q->queuedata;
412 struct Scsi_Host *shost = sdev->host;
415 if (sdev->single_lun)
416 scsi_single_lun_run(sdev);
418 spin_lock_irqsave(shost->host_lock, flags);
419 while (!list_empty(&shost->starved_list) &&
420 !shost->host_blocked && !shost->host_self_blocked &&
421 !((shost->can_queue > 0) &&
422 (shost->host_busy >= shost->can_queue))) {
424 * As long as shost is accepting commands and we have
425 * starved queues, call blk_run_queue. scsi_request_fn
426 * drops the queue_lock and can add us back to the
429 * host_lock protects the starved_list and starved_entry.
430 * scsi_request_fn must get the host_lock before checking
431 * or modifying starved_list or starved_entry.
433 sdev = list_entry(shost->starved_list.next,
434 struct scsi_device, starved_entry);
435 list_del_init(&sdev->starved_entry);
436 spin_unlock_irqrestore(shost->host_lock, flags);
438 blk_run_queue(sdev->request_queue);
440 spin_lock_irqsave(shost->host_lock, flags);
441 if (unlikely(!list_empty(&sdev->starved_entry)))
443 * sdev lost a race, and was put back on the
444 * starved list. This is unlikely but without this
445 * in theory we could loop forever.
449 spin_unlock_irqrestore(shost->host_lock, flags);
455 * Function: scsi_requeue_command()
457 * Purpose: Handle post-processing of completed commands.
459 * Arguments: q - queue to operate on
460 * cmd - command that may need to be requeued.
464 * Notes: After command completion, there may be blocks left
465 * over which weren't finished by the previous command
466 * this can be for a number of reasons - the main one is
467 * I/O errors in the middle of the request, in which case
468 * we need to request the blocks that come after the bad
471 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
473 cmd->request->flags &= ~REQ_DONTPREP;
474 blk_insert_request(q, cmd->request, 1, cmd, 1);
479 void scsi_next_command(struct scsi_cmnd *cmd)
481 struct request_queue *q = cmd->device->request_queue;
483 scsi_put_command(cmd);
487 void scsi_run_host_queues(struct Scsi_Host *shost)
489 struct scsi_device *sdev;
491 shost_for_each_device(sdev, shost)
492 scsi_run_queue(sdev->request_queue);
496 * Function: scsi_end_request()
498 * Purpose: Post-processing of completed commands called from interrupt
499 * handler or a bottom-half handler.
501 * Arguments: cmd - command that is complete.
502 * uptodate - 1 if I/O indicates success, 0 for I/O error.
503 * sectors - number of sectors we want to mark.
504 * requeue - indicates whether we should requeue leftovers.
505 * frequeue - indicates that if we release the command block
506 * that the queue request function should be called.
508 * Lock status: Assumed that lock is not held upon entry.
512 * Notes: This is called for block device requests in order to
513 * mark some number of sectors as complete.
515 * We are guaranteeing that the request queue will be goosed
516 * at some point during this call.
518 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int uptodate,
519 int bytes, int requeue)
521 request_queue_t *q = cmd->device->request_queue;
522 struct request *req = cmd->request;
526 * If there are blocks left over at the end, set up the command
527 * to queue the remainder of them.
529 if (end_that_request_chunk(req, uptodate, bytes)) {
530 int leftover = (req->hard_nr_sectors << 9);
532 if (blk_pc_request(req))
533 leftover = req->data_len;
535 /* kill remainder if no retrys */
536 if (!uptodate && blk_noretry_request(req))
537 end_that_request_chunk(req, 0, leftover);
541 * Bleah. Leftovers again. Stick the
542 * leftovers in the front of the
543 * queue, and goose the queue again.
545 scsi_requeue_command(q, cmd);
551 add_disk_randomness(req->rq_disk);
553 spin_lock_irqsave(q->queue_lock, flags);
554 if (blk_rq_tagged(req))
555 blk_queue_end_tag(q, req);
556 end_that_request_last(req);
557 spin_unlock_irqrestore(q->queue_lock, flags);
560 * This will goose the queue request function at the end, so we don't
561 * need to worry about launching another command.
563 scsi_next_command(cmd);
567 static struct scatterlist *scsi_alloc_sgtable(struct scsi_cmnd *cmd, int gfp_mask)
569 struct scsi_host_sg_pool *sgp;
570 struct scatterlist *sgl;
572 BUG_ON(!cmd->use_sg);
574 switch (cmd->use_sg) {
584 #if (SCSI_MAX_PHYS_SEGMENTS > 32)
588 #if (SCSI_MAX_PHYS_SEGMENTS > 64)
592 #if (SCSI_MAX_PHYS_SEGMENTS > 128)
603 sgp = scsi_sg_pools + cmd->sglist_len;
604 sgl = mempool_alloc(sgp->pool, gfp_mask);
606 memset(sgl, 0, sgp->size);
610 static void scsi_free_sgtable(struct scatterlist *sgl, int index)
612 struct scsi_host_sg_pool *sgp;
614 BUG_ON(index > SG_MEMPOOL_NR);
616 sgp = scsi_sg_pools + index;
617 mempool_free(sgl, sgp->pool);
621 * Function: scsi_release_buffers()
623 * Purpose: Completion processing for block device I/O requests.
625 * Arguments: cmd - command that we are bailing.
627 * Lock status: Assumed that no lock is held upon entry.
631 * Notes: In the event that an upper level driver rejects a
632 * command, we must release resources allocated during
633 * the __init_io() function. Primarily this would involve
634 * the scatter-gather table, and potentially any bounce
637 static void scsi_release_buffers(struct scsi_cmnd *cmd)
639 struct request *req = cmd->request;
642 * Free up any indirection buffers we allocated for DMA purposes.
645 scsi_free_sgtable(cmd->request_buffer, cmd->sglist_len);
646 else if (cmd->request_buffer != req->buffer)
647 kfree(cmd->request_buffer);
650 * Zero these out. They now point to freed memory, and it is
651 * dangerous to hang onto the pointers.
655 cmd->request_buffer = NULL;
656 cmd->request_bufflen = 0;
660 * Function: scsi_io_completion()
662 * Purpose: Completion processing for block device I/O requests.
664 * Arguments: cmd - command that is finished.
666 * Lock status: Assumed that no lock is held upon entry.
670 * Notes: This function is matched in terms of capabilities to
671 * the function that created the scatter-gather list.
672 * In other words, if there are no bounce buffers
673 * (the normal case for most drivers), we don't need
674 * the logic to deal with cleaning up afterwards.
676 * We must do one of several things here:
678 * a) Call scsi_end_request. This will finish off the
679 * specified number of sectors. If we are done, the
680 * command block will be released, and the queue
681 * function will be goosed. If we are not done, then
682 * scsi_end_request will directly goose the queue.
684 * b) We can just use scsi_requeue_command() here. This would
685 * be used if we just wanted to retry, for example.
687 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes,
688 unsigned int block_bytes)
690 int result = cmd->result;
691 int this_count = cmd->bufflen;
692 request_queue_t *q = cmd->device->request_queue;
693 struct request *req = cmd->request;
694 int clear_errors = 1;
697 * Free up any indirection buffers we allocated for DMA purposes.
698 * For the case of a READ, we need to copy the data out of the
699 * bounce buffer and into the real buffer.
702 scsi_free_sgtable(cmd->buffer, cmd->sglist_len);
703 else if (cmd->buffer != req->buffer) {
704 if (rq_data_dir(req) == READ) {
706 char *to = bio_kmap_irq(req->bio, &flags);
707 memcpy(to, cmd->buffer, cmd->bufflen);
708 bio_kunmap_irq(to, &flags);
713 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
714 req->errors = (driver_byte(result) & DRIVER_SENSE) ?
715 (CHECK_CONDITION << 1) : (result & 0xff);
718 if (cmd->sense_buffer[0] & 0x70) {
719 int len = 8 + cmd->sense_buffer[7];
721 if (len > SCSI_SENSE_BUFFERSIZE)
722 len = SCSI_SENSE_BUFFERSIZE;
723 memcpy(req->sense, cmd->sense_buffer, len);
724 req->sense_len = len;
727 req->data_len -= cmd->bufflen;
731 * Zero these out. They now point to freed memory, and it is
732 * dangerous to hang onto the pointers.
736 cmd->request_buffer = NULL;
737 cmd->request_bufflen = 0;
740 * Next deal with any sectors which we were able to correctly
743 if (good_bytes >= 0) {
744 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, %d bytes done.\n",
745 req->nr_sectors, good_bytes));
746 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd->use_sg));
751 * If multiple sectors are requested in one buffer, then
752 * they will have been finished off by the first command.
753 * If not, then we have a multi-buffer command.
755 * If block_bytes != 0, it means we had a medium error
756 * of some sort, and that we want to mark some number of
757 * sectors as not uptodate. Thus we want to inhibit
758 * requeueing right here - we will requeue down below
759 * when we handle the bad sectors.
761 cmd = scsi_end_request(cmd, 1, good_bytes, result == 0);
764 * If the command completed without error, then either finish off the
765 * rest of the command, or start a new one.
767 if (result == 0 || cmd == NULL ) {
772 * Now, if we were good little boys and girls, Santa left us a request
773 * sense buffer. We can extract information from this, so we
774 * can choose a block to remap, etc.
776 if (driver_byte(result) != 0) {
777 if ((cmd->sense_buffer[0] & 0x7f) == 0x70) {
779 * If the device is in the process of becoming ready,
782 if (cmd->sense_buffer[12] == 0x04 &&
783 cmd->sense_buffer[13] == 0x01) {
784 scsi_requeue_command(q, cmd);
787 if ((cmd->sense_buffer[2] & 0xf) == UNIT_ATTENTION) {
788 if (cmd->device->removable) {
789 /* detected disc change. set a bit
790 * and quietly refuse further access.
792 cmd->device->changed = 1;
793 cmd = scsi_end_request(cmd, 0,
798 * Must have been a power glitch, or a
799 * bus reset. Could not have been a
800 * media change, so we just retry the
801 * request and see what happens.
803 scsi_requeue_command(q, cmd);
809 * If we had an ILLEGAL REQUEST returned, then we may have
810 * performed an unsupported command. The only thing this
811 * should be would be a ten byte read where only a six byte
812 * read was supported. Also, on a system where READ CAPACITY
813 * failed, we may have read past the end of the disk.
816 switch (cmd->sense_buffer[2]) {
817 case ILLEGAL_REQUEST:
818 if (cmd->device->use_10_for_rw &&
819 (cmd->cmnd[0] == READ_10 ||
820 cmd->cmnd[0] == WRITE_10)) {
821 cmd->device->use_10_for_rw = 0;
823 * This will cause a retry with a 6-byte
826 scsi_requeue_command(q, cmd);
829 cmd = scsi_end_request(cmd, 0, this_count, 1);
834 printk(KERN_INFO "Device %s not ready.\n",
835 req->rq_disk ? req->rq_disk->disk_name : "");
836 cmd = scsi_end_request(cmd, 0, this_count, 1);
840 case VOLUME_OVERFLOW:
841 printk("scsi%d: ERROR on channel %d, id %d, lun %d, CDB: ",
842 cmd->device->host->host_no, (int) cmd->device->channel,
843 (int) cmd->device->id, (int) cmd->device->lun);
844 __scsi_print_command(cmd->data_cmnd);
845 scsi_print_sense("", cmd);
846 cmd = scsi_end_request(cmd, 0, block_bytes, 1);
851 } /* driver byte != 0 */
852 if (host_byte(result) == DID_RESET) {
854 * Third party bus reset or reset for error
855 * recovery reasons. Just retry the request
856 * and see what happens.
858 scsi_requeue_command(q, cmd);
862 printk("SCSI error : <%d %d %d %d> return code = 0x%x\n",
863 cmd->device->host->host_no,
864 cmd->device->channel,
866 cmd->device->lun, result);
868 if (driver_byte(result) & DRIVER_SENSE)
869 scsi_print_sense("", cmd);
871 * Mark a single buffer as not uptodate. Queue the remainder.
872 * We sometimes get this cruft in the event that a medium error
873 * isn't properly reported.
875 block_bytes = req->hard_cur_sectors << 9;
877 block_bytes = req->data_len;
878 cmd = scsi_end_request(cmd, 0, block_bytes, 1);
883 * Function: scsi_init_io()
885 * Purpose: SCSI I/O initialize function.
887 * Arguments: cmd - Command descriptor we wish to initialize
889 * Returns: 0 on success
890 * BLKPREP_DEFER if the failure is retryable
891 * BLKPREP_KILL if the failure is fatal
893 static int scsi_init_io(struct scsi_cmnd *cmd)
895 struct request *req = cmd->request;
896 struct scatterlist *sgpnt;
900 * if this is a rq->data based REQ_BLOCK_PC, setup for a non-sg xfer
902 if ((req->flags & REQ_BLOCK_PC) && !req->bio) {
903 cmd->request_bufflen = req->data_len;
904 cmd->request_buffer = req->data;
905 req->buffer = req->data;
911 * we used to not use scatter-gather for single segment request,
912 * but now we do (it makes highmem I/O easier to support without
915 cmd->use_sg = req->nr_phys_segments;
918 * if sg table allocation fails, requeue request later.
920 sgpnt = scsi_alloc_sgtable(cmd, GFP_ATOMIC);
921 if (unlikely(!sgpnt)) {
922 req->flags |= REQ_SPECIAL;
923 return BLKPREP_DEFER;
926 cmd->request_buffer = (char *) sgpnt;
927 cmd->request_bufflen = req->nr_sectors << 9;
928 if (blk_pc_request(req))
929 cmd->request_bufflen = req->data_len;
933 * Next, walk the list, and fill in the addresses and sizes of
936 count = blk_rq_map_sg(req->q, req, cmd->request_buffer);
939 * mapped well, send it off
941 if (likely(count <= cmd->use_sg)) {
946 printk(KERN_ERR "Incorrect number of segments after building list\n");
947 printk(KERN_ERR "counted %d, received %d\n", count, cmd->use_sg);
948 printk(KERN_ERR "req nr_sec %lu, cur_nr_sec %u\n", req->nr_sectors,
949 req->current_nr_sectors);
951 /* release the command and kill it */
952 scsi_release_buffers(cmd);
953 scsi_put_command(cmd);
957 static int scsi_prep_fn(struct request_queue *q, struct request *req)
959 struct scsi_device *sdev = q->queuedata;
960 struct scsi_cmnd *cmd;
961 int specials_only = 0;
964 * Just check to see if the device is online. If it isn't, we
965 * refuse to process any commands. The device must be brought
966 * online before trying any recovery commands
968 if (unlikely(!scsi_device_online(sdev))) {
969 printk(KERN_ERR "scsi%d (%d:%d): rejecting I/O to offline device\n",
970 sdev->host->host_no, sdev->id, sdev->lun);
973 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
974 /* OK, we're not in a running state don't prep
976 if (sdev->sdev_state == SDEV_DEL) {
977 /* Device is fully deleted, no commands
978 * at all allowed down */
979 printk(KERN_ERR "scsi%d (%d:%d): rejecting I/O to dead device\n",
980 sdev->host->host_no, sdev->id, sdev->lun);
983 /* OK, we only allow special commands (i.e. not
984 * user initiated ones */
985 specials_only = sdev->sdev_state;
989 * Find the actual device driver associated with this command.
990 * The SPECIAL requests are things like character device or
991 * ioctls, which did not originate from ll_rw_blk. Note that
992 * the special field is also used to indicate the cmd for
993 * the remainder of a partially fulfilled request that can
994 * come up when there is a medium error. We have to treat
995 * these two cases differently. We differentiate by looking
996 * at request->cmd, as this tells us the real story.
998 if (req->flags & REQ_SPECIAL) {
999 struct scsi_request *sreq = req->special;
1001 if (sreq->sr_magic == SCSI_REQ_MAGIC) {
1002 cmd = scsi_get_command(sreq->sr_device, GFP_ATOMIC);
1005 scsi_init_cmd_from_req(cmd, sreq);
1008 } else if (req->flags & (REQ_CMD | REQ_BLOCK_PC)) {
1010 if(unlikely(specials_only)) {
1011 if(specials_only == SDEV_QUIESCE)
1012 return BLKPREP_DEFER;
1014 printk(KERN_ERR "scsi%d (%d:%d): rejecting I/O to device being removed\n",
1015 sdev->host->host_no, sdev->id, sdev->lun);
1016 return BLKPREP_KILL;
1021 * Now try and find a command block that we can use.
1023 if (!req->special) {
1024 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1030 /* pull a tag out of the request if we have one */
1031 cmd->tag = req->tag;
1033 blk_dump_rq_flags(req, "SCSI bad req");
1034 return BLKPREP_KILL;
1037 /* note the overloading of req->special. When the tag
1038 * is active it always means cmd. If the tag goes
1039 * back for re-queueing, it may be reset */
1044 * FIXME: drop the lock here because the functions below
1045 * expect to be called without the queue lock held. Also,
1046 * previously, we dequeued the request before dropping the
1047 * lock. We hope REQ_STARTED prevents anything untoward from
1050 if (req->flags & (REQ_CMD | REQ_BLOCK_PC)) {
1051 struct scsi_driver *drv;
1055 * This will do a couple of things:
1056 * 1) Fill in the actual SCSI command.
1057 * 2) Fill in any other upper-level specific fields
1060 * If this returns 0, it means that the request failed
1061 * (reading past end of disk, reading offline device,
1062 * etc). This won't actually talk to the device, but
1063 * some kinds of consistency checking may cause the
1064 * request to be rejected immediately.
1068 * This sets up the scatter-gather table (allocating if
1071 ret = scsi_init_io(cmd);
1072 if (ret) /* BLKPREP_KILL return also releases the command */
1076 * Initialize the actual SCSI command for this request.
1078 drv = *(struct scsi_driver **)req->rq_disk->private_data;
1079 if (unlikely(!drv->init_command(cmd))) {
1080 scsi_release_buffers(cmd);
1081 scsi_put_command(cmd);
1082 return BLKPREP_KILL;
1087 * The request is now prepped, no need to come back here
1089 req->flags |= REQ_DONTPREP;
1093 /* If we defer, the elv_next_request() returns NULL, but the
1094 * queue must be restarted, so we plug here if no returning
1095 * command will automatically do that. */
1096 if (sdev->device_busy == 0)
1098 return BLKPREP_DEFER;
1102 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1105 * Called with the queue_lock held.
1107 static inline int scsi_dev_queue_ready(struct request_queue *q,
1108 struct scsi_device *sdev)
1110 if (sdev->device_busy >= sdev->queue_depth)
1112 if (sdev->device_busy == 0 && sdev->device_blocked) {
1114 * unblock after device_blocked iterates to zero
1116 if (--sdev->device_blocked == 0) {
1118 printk("scsi%d (%d:%d) unblocking device at"
1119 " zero depth\n", sdev->host->host_no,
1120 sdev->id, sdev->lun));
1126 if (sdev->device_blocked)
1133 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1134 * return 0. We must end up running the queue again whenever 0 is
1135 * returned, else IO can hang.
1137 * Called with host_lock held.
1139 static inline int scsi_host_queue_ready(struct request_queue *q,
1140 struct Scsi_Host *shost,
1141 struct scsi_device *sdev)
1143 if (test_bit(SHOST_RECOVERY, &shost->shost_state))
1145 if (shost->host_busy == 0 && shost->host_blocked) {
1147 * unblock after host_blocked iterates to zero
1149 if (--shost->host_blocked == 0) {
1151 printk("scsi%d unblocking host at zero depth\n",
1158 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1159 shost->host_blocked || shost->host_self_blocked) {
1160 if (list_empty(&sdev->starved_entry))
1161 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1165 /* We're OK to process the command, so we can't be starved */
1166 if (!list_empty(&sdev->starved_entry))
1167 list_del_init(&sdev->starved_entry);
1173 * Function: scsi_request_fn()
1175 * Purpose: Main strategy routine for SCSI.
1177 * Arguments: q - Pointer to actual queue.
1181 * Lock status: IO request lock assumed to be held when called.
1183 static void scsi_request_fn(struct request_queue *q)
1185 struct scsi_device *sdev = q->queuedata;
1186 struct Scsi_Host *shost = sdev->host;
1187 struct scsi_cmnd *cmd;
1188 struct request *req;
1190 if(!get_device(&sdev->sdev_gendev))
1191 /* We must be tearing the block queue down already */
1195 * To start with, we keep looping until the queue is empty, or until
1196 * the host is no longer able to accept any more requests.
1198 while (!blk_queue_plugged(q)) {
1201 * get next queueable request. We do this early to make sure
1202 * that the request is fully prepared even if we cannot
1205 req = elv_next_request(q);
1206 if (!req || !scsi_dev_queue_ready(q, sdev))
1209 if (unlikely(!scsi_device_online(sdev))) {
1210 printk(KERN_ERR "scsi%d (%d:%d): rejecting I/O to offline device\n",
1211 sdev->host->host_no, sdev->id, sdev->lun);
1212 blkdev_dequeue_request(req);
1213 req->flags |= REQ_QUIET;
1214 while (end_that_request_first(req, 0, req->nr_sectors))
1216 end_that_request_last(req);
1222 * Remove the request from the request list.
1224 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1225 blkdev_dequeue_request(req);
1226 sdev->device_busy++;
1228 spin_unlock(q->queue_lock);
1229 spin_lock(shost->host_lock);
1231 if (!scsi_host_queue_ready(q, shost, sdev))
1233 if (sdev->single_lun) {
1234 if (sdev->sdev_target->starget_sdev_user &&
1235 sdev->sdev_target->starget_sdev_user != sdev)
1237 sdev->sdev_target->starget_sdev_user = sdev;
1242 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1243 * take the lock again.
1245 spin_unlock_irq(shost->host_lock);
1248 if (unlikely(cmd == NULL)) {
1249 printk(KERN_CRIT "impossible request in %s.\n"
1250 "please mail a stack trace to "
1251 "linux-scsi@vger.kernel.org",
1257 * Finally, initialize any error handling parameters, and set up
1258 * the timers for timeouts.
1260 scsi_init_cmd_errh(cmd);
1263 * Dispatch the command to the low-level driver.
1265 rtn = scsi_dispatch_cmd(cmd);
1266 spin_lock_irq(q->queue_lock);
1268 /* we're refusing the command; because of
1269 * the way locks get dropped, we need to
1270 * check here if plugging is required */
1271 if(sdev->device_busy == 0)
1281 spin_unlock_irq(shost->host_lock);
1284 * lock q, handle tag, requeue req, and decrement device_busy. We
1285 * must return with queue_lock held.
1287 * Decrementing device_busy without checking it is OK, as all such
1288 * cases (host limits or settings) should run the queue at some
1291 spin_lock_irq(q->queue_lock);
1292 blk_requeue_request(q, req);
1293 sdev->device_busy--;
1294 if(sdev->device_busy == 0)
1297 /* must be careful here...if we trigger the ->remove() function
1298 * we cannot be holding the q lock */
1299 spin_unlock_irq(q->queue_lock);
1300 put_device(&sdev->sdev_gendev);
1301 spin_lock_irq(q->queue_lock);
1304 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1306 struct device *host_dev;
1308 if (shost->unchecked_isa_dma)
1309 return BLK_BOUNCE_ISA;
1311 host_dev = scsi_get_device(shost);
1312 if (PCI_DMA_BUS_IS_PHYS && host_dev && host_dev->dma_mask)
1313 return *host_dev->dma_mask;
1316 * Platforms with virtual-DMA translation
1317 * hardware have no practical limit.
1319 return BLK_BOUNCE_ANY;
1322 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1324 struct Scsi_Host *shost = sdev->host;
1325 struct request_queue *q;
1327 q = blk_init_queue(scsi_request_fn, &sdev->sdev_lock);
1331 blk_queue_prep_rq(q, scsi_prep_fn);
1333 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1334 blk_queue_max_phys_segments(q, SCSI_MAX_PHYS_SEGMENTS);
1335 blk_queue_max_sectors(q, shost->max_sectors);
1336 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1337 blk_queue_segment_boundary(q, shost->dma_boundary);
1339 if (!shost->use_clustering)
1340 clear_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
1344 void scsi_free_queue(struct request_queue *q)
1346 blk_cleanup_queue(q);
1350 * Function: scsi_block_requests()
1352 * Purpose: Utility function used by low-level drivers to prevent further
1353 * commands from being queued to the device.
1355 * Arguments: shost - Host in question
1359 * Lock status: No locks are assumed held.
1361 * Notes: There is no timer nor any other means by which the requests
1362 * get unblocked other than the low-level driver calling
1363 * scsi_unblock_requests().
1365 void scsi_block_requests(struct Scsi_Host *shost)
1367 shost->host_self_blocked = 1;
1371 * Function: scsi_unblock_requests()
1373 * Purpose: Utility function used by low-level drivers to allow further
1374 * commands from being queued to the device.
1376 * Arguments: shost - Host in question
1380 * Lock status: No locks are assumed held.
1382 * Notes: There is no timer nor any other means by which the requests
1383 * get unblocked other than the low-level driver calling
1384 * scsi_unblock_requests().
1386 * This is done as an API function so that changes to the
1387 * internals of the scsi mid-layer won't require wholesale
1388 * changes to drivers that use this feature.
1390 void scsi_unblock_requests(struct Scsi_Host *shost)
1392 shost->host_self_blocked = 0;
1393 scsi_run_host_queues(shost);
1396 int __init scsi_init_queue(void)
1400 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1401 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1402 int size = sgp->size * sizeof(struct scatterlist);
1404 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1405 SLAB_HWCACHE_ALIGN, NULL, NULL);
1407 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1411 sgp->pool = mempool_create(SG_MEMPOOL_SIZE,
1412 mempool_alloc_slab, mempool_free_slab,
1415 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1423 void scsi_exit_queue(void)
1427 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1428 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1429 mempool_destroy(sgp->pool);
1430 kmem_cache_destroy(sgp->slab);
1434 * __scsi_mode_sense - issue a mode sense, falling back from 10 to
1435 * six bytes if necessary.
1436 * @sreq: SCSI request to fill in with the MODE_SENSE
1437 * @dbd: set if mode sense will allow block descriptors to be returned
1438 * @modepage: mode page being requested
1439 * @buffer: request buffer (may not be smaller than eight bytes)
1440 * @len: length of request buffer.
1441 * @timeout: command timeout
1442 * @retries: number of retries before failing
1443 * @data: returns a structure abstracting the mode header data
1445 * Returns zero if unsuccessful, or the header offset (either 4
1446 * or 8 depending on whether a six or ten byte command was
1447 * issued) if successful.
1450 __scsi_mode_sense(struct scsi_request *sreq, int dbd, int modepage,
1451 unsigned char *buffer, int len, int timeout, int retries,
1452 struct scsi_mode_data *data) {
1453 unsigned char cmd[12];
1457 memset(data, 0, sizeof(*data));
1458 memset(&cmd[0], 0, 12);
1459 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1463 use_10_for_ms = sreq->sr_device->use_10_for_ms;
1465 if (use_10_for_ms) {
1469 cmd[0] = MODE_SENSE_10;
1476 cmd[0] = MODE_SENSE;
1481 sreq->sr_cmd_len = 0;
1482 sreq->sr_sense_buffer[0] = 0;
1483 sreq->sr_sense_buffer[2] = 0;
1484 sreq->sr_data_direction = DMA_FROM_DEVICE;
1486 memset(buffer, 0, len);
1488 scsi_wait_req(sreq, cmd, buffer, len, timeout, retries);
1490 /* This code looks awful: what it's doing is making sure an
1491 * ILLEGAL REQUEST sense return identifies the actual command
1492 * byte as the problem. MODE_SENSE commands can return
1493 * ILLEGAL REQUEST if the code page isn't supported */
1494 if (use_10_for_ms && ! scsi_status_is_good(sreq->sr_result) &&
1495 (driver_byte(sreq->sr_result) & DRIVER_SENSE) &&
1496 sreq->sr_sense_buffer[2] == ILLEGAL_REQUEST &&
1497 (sreq->sr_sense_buffer[4] & 0x40) == 0x40 &&
1498 sreq->sr_sense_buffer[5] == 0 &&
1499 sreq->sr_sense_buffer[6] == 0 ) {
1500 sreq->sr_device->use_10_for_ms = 0;
1504 if(scsi_status_is_good(sreq->sr_result)) {
1505 data->header_length = header_length;
1507 data->length = buffer[0]*256 + buffer[1] + 2;
1508 data->medium_type = buffer[2];
1509 data->device_specific = buffer[3];
1510 data->longlba = buffer[4] & 0x01;
1511 data->block_descriptor_length = buffer[6]*256
1514 data->length = buffer[0] + 1;
1515 data->medium_type = buffer[1];
1516 data->device_specific = buffer[2];
1517 data->block_descriptor_length = buffer[3];
1521 return sreq->sr_result;
1525 * scsi_mode_sense - issue a mode sense, falling back from 10 to
1526 * six bytes if necessary.
1527 * @sdev: scsi device to send command to.
1528 * @dbd: set if mode sense will disable block descriptors in the return
1529 * @modepage: mode page being requested
1530 * @buffer: request buffer (may not be smaller than eight bytes)
1531 * @len: length of request buffer.
1532 * @timeout: command timeout
1533 * @retries: number of retries before failing
1535 * Returns zero if unsuccessful, or the header offset (either 4
1536 * or 8 depending on whether a six or ten byte command was
1537 * issued) if successful.
1540 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1541 unsigned char *buffer, int len, int timeout, int retries,
1542 struct scsi_mode_data *data)
1544 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL);
1550 ret = __scsi_mode_sense(sreq, dbd, modepage, buffer, len,
1551 timeout, retries, data);
1553 scsi_release_request(sreq);
1559 * scsi_device_set_state - Take the given device through the device
1561 * @sdev: scsi device to change the state of.
1562 * @state: state to change to.
1564 * Returns zero if unsuccessful or an error if the requested
1565 * transition is illegal.
1568 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
1570 enum scsi_device_state oldstate = sdev->sdev_state;
1572 if (state == oldstate)
1577 /* There are no legal states that come back to
1578 * created. This is the manually initialised start
1635 sdev->sdev_state = state;
1639 dev_printk(KERN_ERR, &sdev->sdev_gendev,
1640 "Illegal state transition %s->%s\n",
1641 scsi_device_state_name(oldstate),
1642 scsi_device_state_name(state));
1646 EXPORT_SYMBOL(scsi_device_set_state);
1649 * scsi_device_quiesce - Block user issued commands.
1650 * @sdev: scsi device to quiesce.
1652 * This works by trying to transition to the SDEV_QUIESCE state
1653 * (which must be a legal transition). When the device is in this
1654 * state, only special requests will be accepted, all others will
1655 * be deferred. Since special requests may also be requeued requests,
1656 * a successful return doesn't guarantee the device will be
1657 * totally quiescent.
1659 * Must be called with user context, may sleep.
1661 * Returns zero if unsuccessful or an error if not.
1664 scsi_device_quiesce(struct scsi_device *sdev)
1666 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
1670 scsi_run_queue(sdev->request_queue);
1671 while (sdev->device_busy) {
1672 schedule_timeout(HZ/5);
1673 scsi_run_queue(sdev->request_queue);
1677 EXPORT_SYMBOL(scsi_device_quiesce);
1680 * scsi_device_resume - Restart user issued commands to a quiesced device.
1681 * @sdev: scsi device to resume.
1683 * Moves the device from quiesced back to running and restarts the
1686 * Must be called with user context, may sleep.
1689 scsi_device_resume(struct scsi_device *sdev)
1691 if(scsi_device_set_state(sdev, SDEV_RUNNING))
1693 scsi_run_queue(sdev->request_queue);
1695 EXPORT_SYMBOL(scsi_device_resume);