Loading...
1/*
2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
4 *
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
9 */
10
11#include <linux/bio.h>
12#include <linux/bitops.h>
13#include <linux/blkdev.h>
14#include <linux/completion.h>
15#include <linux/kernel.h>
16#include <linux/export.h>
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
20#include <linux/hardirq.h>
21#include <linux/scatterlist.h>
22#include <linux/blk-mq.h>
23#include <linux/ratelimit.h>
24#include <asm/unaligned.h>
25
26#include <scsi/scsi.h>
27#include <scsi/scsi_cmnd.h>
28#include <scsi/scsi_dbg.h>
29#include <scsi/scsi_device.h>
30#include <scsi/scsi_driver.h>
31#include <scsi/scsi_eh.h>
32#include <scsi/scsi_host.h>
33#include <scsi/scsi_dh.h>
34
35#include <trace/events/scsi.h>
36
37#include "scsi_priv.h"
38#include "scsi_logging.h"
39
40
41struct kmem_cache *scsi_sdb_cache;
42
43/*
44 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
45 * not change behaviour from the previous unplug mechanism, experimentation
46 * may prove this needs changing.
47 */
48#define SCSI_QUEUE_DELAY 3
49
50static void
51scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
52{
53 struct Scsi_Host *host = cmd->device->host;
54 struct scsi_device *device = cmd->device;
55 struct scsi_target *starget = scsi_target(device);
56
57 /*
58 * Set the appropriate busy bit for the device/host.
59 *
60 * If the host/device isn't busy, assume that something actually
61 * completed, and that we should be able to queue a command now.
62 *
63 * Note that the prior mid-layer assumption that any host could
64 * always queue at least one command is now broken. The mid-layer
65 * will implement a user specifiable stall (see
66 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
67 * if a command is requeued with no other commands outstanding
68 * either for the device or for the host.
69 */
70 switch (reason) {
71 case SCSI_MLQUEUE_HOST_BUSY:
72 atomic_set(&host->host_blocked, host->max_host_blocked);
73 break;
74 case SCSI_MLQUEUE_DEVICE_BUSY:
75 case SCSI_MLQUEUE_EH_RETRY:
76 atomic_set(&device->device_blocked,
77 device->max_device_blocked);
78 break;
79 case SCSI_MLQUEUE_TARGET_BUSY:
80 atomic_set(&starget->target_blocked,
81 starget->max_target_blocked);
82 break;
83 }
84}
85
86static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
87{
88 struct scsi_device *sdev = cmd->device;
89
90 blk_mq_requeue_request(cmd->request, true);
91 put_device(&sdev->sdev_gendev);
92}
93
94/**
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
99 *
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
104 * file.
105 */
106static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
107{
108 struct scsi_device *device = cmd->device;
109 struct request_queue *q = device->request_queue;
110 unsigned long flags;
111
112 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
113 "Inserting command %p into mlqueue\n", cmd));
114
115 scsi_set_blocked(cmd, reason);
116
117 /*
118 * Decrement the counters, since these commands are no longer
119 * active on the host/device.
120 */
121 if (unbusy)
122 scsi_device_unbusy(device);
123
124 /*
125 * Requeue this command. It will go before all other commands
126 * that are already in the queue. Schedule requeue work under
127 * lock such that the kblockd_schedule_work() call happens
128 * before blk_cleanup_queue() finishes.
129 */
130 cmd->result = 0;
131 if (q->mq_ops) {
132 scsi_mq_requeue_cmd(cmd);
133 return;
134 }
135 spin_lock_irqsave(q->queue_lock, flags);
136 blk_requeue_request(q, cmd->request);
137 kblockd_schedule_work(&device->requeue_work);
138 spin_unlock_irqrestore(q->queue_lock, flags);
139}
140
141/*
142 * Function: scsi_queue_insert()
143 *
144 * Purpose: Insert a command in the midlevel queue.
145 *
146 * Arguments: cmd - command that we are adding to queue.
147 * reason - why we are inserting command to queue.
148 *
149 * Lock status: Assumed that lock is not held upon entry.
150 *
151 * Returns: Nothing.
152 *
153 * Notes: We do this for one of two cases. Either the host is busy
154 * and it cannot accept any more commands for the time being,
155 * or the device returned QUEUE_FULL and can accept no more
156 * commands.
157 * Notes: This could be called either from an interrupt context or a
158 * normal process context.
159 */
160void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
161{
162 __scsi_queue_insert(cmd, reason, 1);
163}
164
165static int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
166 int data_direction, void *buffer, unsigned bufflen,
167 unsigned char *sense, int timeout, int retries, u64 flags,
168 req_flags_t rq_flags, int *resid)
169{
170 struct request *req;
171 int write = (data_direction == DMA_TO_DEVICE);
172 int ret = DRIVER_ERROR << 24;
173
174 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
175 if (IS_ERR(req))
176 return ret;
177 blk_rq_set_block_pc(req);
178
179 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
180 buffer, bufflen, __GFP_RECLAIM))
181 goto out;
182
183 req->cmd_len = COMMAND_SIZE(cmd[0]);
184 memcpy(req->cmd, cmd, req->cmd_len);
185 req->sense = sense;
186 req->sense_len = 0;
187 req->retries = retries;
188 req->timeout = timeout;
189 req->cmd_flags |= flags;
190 req->rq_flags |= rq_flags | RQF_QUIET | RQF_PREEMPT;
191
192 /*
193 * head injection *required* here otherwise quiesce won't work
194 */
195 blk_execute_rq(req->q, NULL, req, 1);
196
197 /*
198 * Some devices (USB mass-storage in particular) may transfer
199 * garbage data together with a residue indicating that the data
200 * is invalid. Prevent the garbage from being misinterpreted
201 * and prevent security leaks by zeroing out the excess data.
202 */
203 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
204 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
205
206 if (resid)
207 *resid = req->resid_len;
208 ret = req->errors;
209 out:
210 blk_put_request(req);
211
212 return ret;
213}
214
215/**
216 * scsi_execute - insert request and wait for the result
217 * @sdev: scsi device
218 * @cmd: scsi command
219 * @data_direction: data direction
220 * @buffer: data buffer
221 * @bufflen: len of buffer
222 * @sense: optional sense buffer
223 * @timeout: request timeout in seconds
224 * @retries: number of times to retry request
225 * @flags: or into request flags;
226 * @resid: optional residual length
227 *
228 * returns the req->errors value which is the scsi_cmnd result
229 * field.
230 */
231int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
232 int data_direction, void *buffer, unsigned bufflen,
233 unsigned char *sense, int timeout, int retries, u64 flags,
234 int *resid)
235{
236 return __scsi_execute(sdev, cmd, data_direction, buffer, bufflen, sense,
237 timeout, retries, flags, 0, resid);
238}
239EXPORT_SYMBOL(scsi_execute);
240
241int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
242 int data_direction, void *buffer, unsigned bufflen,
243 struct scsi_sense_hdr *sshdr, int timeout, int retries,
244 int *resid, u64 flags, req_flags_t rq_flags)
245{
246 char *sense = NULL;
247 int result;
248
249 if (sshdr) {
250 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
251 if (!sense)
252 return DRIVER_ERROR << 24;
253 }
254 result = __scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
255 sense, timeout, retries, flags, rq_flags, resid);
256 if (sshdr)
257 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
258
259 kfree(sense);
260 return result;
261}
262EXPORT_SYMBOL(scsi_execute_req_flags);
263
264/*
265 * Function: scsi_init_cmd_errh()
266 *
267 * Purpose: Initialize cmd fields related to error handling.
268 *
269 * Arguments: cmd - command that is ready to be queued.
270 *
271 * Notes: This function has the job of initializing a number of
272 * fields related to error handling. Typically this will
273 * be called once for each command, as required.
274 */
275static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
276{
277 cmd->serial_number = 0;
278 scsi_set_resid(cmd, 0);
279 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
280 if (cmd->cmd_len == 0)
281 cmd->cmd_len = scsi_command_size(cmd->cmnd);
282}
283
284void scsi_device_unbusy(struct scsi_device *sdev)
285{
286 struct Scsi_Host *shost = sdev->host;
287 struct scsi_target *starget = scsi_target(sdev);
288 unsigned long flags;
289
290 atomic_dec(&shost->host_busy);
291 if (starget->can_queue > 0)
292 atomic_dec(&starget->target_busy);
293
294 if (unlikely(scsi_host_in_recovery(shost) &&
295 (shost->host_failed || shost->host_eh_scheduled))) {
296 spin_lock_irqsave(shost->host_lock, flags);
297 scsi_eh_wakeup(shost);
298 spin_unlock_irqrestore(shost->host_lock, flags);
299 }
300
301 atomic_dec(&sdev->device_busy);
302}
303
304static void scsi_kick_queue(struct request_queue *q)
305{
306 if (q->mq_ops)
307 blk_mq_start_hw_queues(q);
308 else
309 blk_run_queue(q);
310}
311
312/*
313 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
314 * and call blk_run_queue for all the scsi_devices on the target -
315 * including current_sdev first.
316 *
317 * Called with *no* scsi locks held.
318 */
319static void scsi_single_lun_run(struct scsi_device *current_sdev)
320{
321 struct Scsi_Host *shost = current_sdev->host;
322 struct scsi_device *sdev, *tmp;
323 struct scsi_target *starget = scsi_target(current_sdev);
324 unsigned long flags;
325
326 spin_lock_irqsave(shost->host_lock, flags);
327 starget->starget_sdev_user = NULL;
328 spin_unlock_irqrestore(shost->host_lock, flags);
329
330 /*
331 * Call blk_run_queue for all LUNs on the target, starting with
332 * current_sdev. We race with others (to set starget_sdev_user),
333 * but in most cases, we will be first. Ideally, each LU on the
334 * target would get some limited time or requests on the target.
335 */
336 scsi_kick_queue(current_sdev->request_queue);
337
338 spin_lock_irqsave(shost->host_lock, flags);
339 if (starget->starget_sdev_user)
340 goto out;
341 list_for_each_entry_safe(sdev, tmp, &starget->devices,
342 same_target_siblings) {
343 if (sdev == current_sdev)
344 continue;
345 if (scsi_device_get(sdev))
346 continue;
347
348 spin_unlock_irqrestore(shost->host_lock, flags);
349 scsi_kick_queue(sdev->request_queue);
350 spin_lock_irqsave(shost->host_lock, flags);
351
352 scsi_device_put(sdev);
353 }
354 out:
355 spin_unlock_irqrestore(shost->host_lock, flags);
356}
357
358static inline bool scsi_device_is_busy(struct scsi_device *sdev)
359{
360 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
361 return true;
362 if (atomic_read(&sdev->device_blocked) > 0)
363 return true;
364 return false;
365}
366
367static inline bool scsi_target_is_busy(struct scsi_target *starget)
368{
369 if (starget->can_queue > 0) {
370 if (atomic_read(&starget->target_busy) >= starget->can_queue)
371 return true;
372 if (atomic_read(&starget->target_blocked) > 0)
373 return true;
374 }
375 return false;
376}
377
378static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
379{
380 if (shost->can_queue > 0 &&
381 atomic_read(&shost->host_busy) >= shost->can_queue)
382 return true;
383 if (atomic_read(&shost->host_blocked) > 0)
384 return true;
385 if (shost->host_self_blocked)
386 return true;
387 return false;
388}
389
390static void scsi_starved_list_run(struct Scsi_Host *shost)
391{
392 LIST_HEAD(starved_list);
393 struct scsi_device *sdev;
394 unsigned long flags;
395
396 spin_lock_irqsave(shost->host_lock, flags);
397 list_splice_init(&shost->starved_list, &starved_list);
398
399 while (!list_empty(&starved_list)) {
400 struct request_queue *slq;
401
402 /*
403 * As long as shost is accepting commands and we have
404 * starved queues, call blk_run_queue. scsi_request_fn
405 * drops the queue_lock and can add us back to the
406 * starved_list.
407 *
408 * host_lock protects the starved_list and starved_entry.
409 * scsi_request_fn must get the host_lock before checking
410 * or modifying starved_list or starved_entry.
411 */
412 if (scsi_host_is_busy(shost))
413 break;
414
415 sdev = list_entry(starved_list.next,
416 struct scsi_device, starved_entry);
417 list_del_init(&sdev->starved_entry);
418 if (scsi_target_is_busy(scsi_target(sdev))) {
419 list_move_tail(&sdev->starved_entry,
420 &shost->starved_list);
421 continue;
422 }
423
424 /*
425 * Once we drop the host lock, a racing scsi_remove_device()
426 * call may remove the sdev from the starved list and destroy
427 * it and the queue. Mitigate by taking a reference to the
428 * queue and never touching the sdev again after we drop the
429 * host lock. Note: if __scsi_remove_device() invokes
430 * blk_cleanup_queue() before the queue is run from this
431 * function then blk_run_queue() will return immediately since
432 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
433 */
434 slq = sdev->request_queue;
435 if (!blk_get_queue(slq))
436 continue;
437 spin_unlock_irqrestore(shost->host_lock, flags);
438
439 scsi_kick_queue(slq);
440 blk_put_queue(slq);
441
442 spin_lock_irqsave(shost->host_lock, flags);
443 }
444 /* put any unprocessed entries back */
445 list_splice(&starved_list, &shost->starved_list);
446 spin_unlock_irqrestore(shost->host_lock, flags);
447}
448
449/*
450 * Function: scsi_run_queue()
451 *
452 * Purpose: Select a proper request queue to serve next
453 *
454 * Arguments: q - last request's queue
455 *
456 * Returns: Nothing
457 *
458 * Notes: The previous command was completely finished, start
459 * a new one if possible.
460 */
461static void scsi_run_queue(struct request_queue *q)
462{
463 struct scsi_device *sdev = q->queuedata;
464
465 if (scsi_target(sdev)->single_lun)
466 scsi_single_lun_run(sdev);
467 if (!list_empty(&sdev->host->starved_list))
468 scsi_starved_list_run(sdev->host);
469
470 if (q->mq_ops)
471 blk_mq_start_stopped_hw_queues(q, false);
472 else
473 blk_run_queue(q);
474}
475
476void scsi_requeue_run_queue(struct work_struct *work)
477{
478 struct scsi_device *sdev;
479 struct request_queue *q;
480
481 sdev = container_of(work, struct scsi_device, requeue_work);
482 q = sdev->request_queue;
483 scsi_run_queue(q);
484}
485
486/*
487 * Function: scsi_requeue_command()
488 *
489 * Purpose: Handle post-processing of completed commands.
490 *
491 * Arguments: q - queue to operate on
492 * cmd - command that may need to be requeued.
493 *
494 * Returns: Nothing
495 *
496 * Notes: After command completion, there may be blocks left
497 * over which weren't finished by the previous command
498 * this can be for a number of reasons - the main one is
499 * I/O errors in the middle of the request, in which case
500 * we need to request the blocks that come after the bad
501 * sector.
502 * Notes: Upon return, cmd is a stale pointer.
503 */
504static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
505{
506 struct scsi_device *sdev = cmd->device;
507 struct request *req = cmd->request;
508 unsigned long flags;
509
510 spin_lock_irqsave(q->queue_lock, flags);
511 blk_unprep_request(req);
512 req->special = NULL;
513 scsi_put_command(cmd);
514 blk_requeue_request(q, req);
515 spin_unlock_irqrestore(q->queue_lock, flags);
516
517 scsi_run_queue(q);
518
519 put_device(&sdev->sdev_gendev);
520}
521
522void scsi_run_host_queues(struct Scsi_Host *shost)
523{
524 struct scsi_device *sdev;
525
526 shost_for_each_device(sdev, shost)
527 scsi_run_queue(sdev->request_queue);
528}
529
530static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
531{
532 if (cmd->request->cmd_type == REQ_TYPE_FS) {
533 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
534
535 if (drv->uninit_command)
536 drv->uninit_command(cmd);
537 }
538}
539
540static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
541{
542 struct scsi_data_buffer *sdb;
543
544 if (cmd->sdb.table.nents)
545 sg_free_table_chained(&cmd->sdb.table, true);
546 if (cmd->request->next_rq) {
547 sdb = cmd->request->next_rq->special;
548 if (sdb)
549 sg_free_table_chained(&sdb->table, true);
550 }
551 if (scsi_prot_sg_count(cmd))
552 sg_free_table_chained(&cmd->prot_sdb->table, true);
553}
554
555static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
556{
557 struct scsi_device *sdev = cmd->device;
558 struct Scsi_Host *shost = sdev->host;
559 unsigned long flags;
560
561 scsi_mq_free_sgtables(cmd);
562 scsi_uninit_cmd(cmd);
563
564 if (shost->use_cmd_list) {
565 BUG_ON(list_empty(&cmd->list));
566 spin_lock_irqsave(&sdev->list_lock, flags);
567 list_del_init(&cmd->list);
568 spin_unlock_irqrestore(&sdev->list_lock, flags);
569 }
570}
571
572/*
573 * Function: scsi_release_buffers()
574 *
575 * Purpose: Free resources allocate for a scsi_command.
576 *
577 * Arguments: cmd - command that we are bailing.
578 *
579 * Lock status: Assumed that no lock is held upon entry.
580 *
581 * Returns: Nothing
582 *
583 * Notes: In the event that an upper level driver rejects a
584 * command, we must release resources allocated during
585 * the __init_io() function. Primarily this would involve
586 * the scatter-gather table.
587 */
588static void scsi_release_buffers(struct scsi_cmnd *cmd)
589{
590 if (cmd->sdb.table.nents)
591 sg_free_table_chained(&cmd->sdb.table, false);
592
593 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
594
595 if (scsi_prot_sg_count(cmd))
596 sg_free_table_chained(&cmd->prot_sdb->table, false);
597}
598
599static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
600{
601 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
602
603 sg_free_table_chained(&bidi_sdb->table, false);
604 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
605 cmd->request->next_rq->special = NULL;
606}
607
608static bool scsi_end_request(struct request *req, int error,
609 unsigned int bytes, unsigned int bidi_bytes)
610{
611 struct scsi_cmnd *cmd = req->special;
612 struct scsi_device *sdev = cmd->device;
613 struct request_queue *q = sdev->request_queue;
614
615 if (blk_update_request(req, error, bytes))
616 return true;
617
618 /* Bidi request must be completed as a whole */
619 if (unlikely(bidi_bytes) &&
620 blk_update_request(req->next_rq, error, bidi_bytes))
621 return true;
622
623 if (blk_queue_add_random(q))
624 add_disk_randomness(req->rq_disk);
625
626 if (req->mq_ctx) {
627 /*
628 * In the MQ case the command gets freed by __blk_mq_end_request,
629 * so we have to do all cleanup that depends on it earlier.
630 *
631 * We also can't kick the queues from irq context, so we
632 * will have to defer it to a workqueue.
633 */
634 scsi_mq_uninit_cmd(cmd);
635
636 __blk_mq_end_request(req, error);
637
638 if (scsi_target(sdev)->single_lun ||
639 !list_empty(&sdev->host->starved_list))
640 kblockd_schedule_work(&sdev->requeue_work);
641 else
642 blk_mq_start_stopped_hw_queues(q, true);
643 } else {
644 unsigned long flags;
645
646 if (bidi_bytes)
647 scsi_release_bidi_buffers(cmd);
648
649 spin_lock_irqsave(q->queue_lock, flags);
650 blk_finish_request(req, error);
651 spin_unlock_irqrestore(q->queue_lock, flags);
652
653 scsi_release_buffers(cmd);
654
655 scsi_put_command(cmd);
656 scsi_run_queue(q);
657 }
658
659 put_device(&sdev->sdev_gendev);
660 return false;
661}
662
663/**
664 * __scsi_error_from_host_byte - translate SCSI error code into errno
665 * @cmd: SCSI command (unused)
666 * @result: scsi error code
667 *
668 * Translate SCSI error code into standard UNIX errno.
669 * Return values:
670 * -ENOLINK temporary transport failure
671 * -EREMOTEIO permanent target failure, do not retry
672 * -EBADE permanent nexus failure, retry on other path
673 * -ENOSPC No write space available
674 * -ENODATA Medium error
675 * -EIO unspecified I/O error
676 */
677static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
678{
679 int error = 0;
680
681 switch(host_byte(result)) {
682 case DID_TRANSPORT_FAILFAST:
683 error = -ENOLINK;
684 break;
685 case DID_TARGET_FAILURE:
686 set_host_byte(cmd, DID_OK);
687 error = -EREMOTEIO;
688 break;
689 case DID_NEXUS_FAILURE:
690 set_host_byte(cmd, DID_OK);
691 error = -EBADE;
692 break;
693 case DID_ALLOC_FAILURE:
694 set_host_byte(cmd, DID_OK);
695 error = -ENOSPC;
696 break;
697 case DID_MEDIUM_ERROR:
698 set_host_byte(cmd, DID_OK);
699 error = -ENODATA;
700 break;
701 default:
702 error = -EIO;
703 break;
704 }
705
706 return error;
707}
708
709/*
710 * Function: scsi_io_completion()
711 *
712 * Purpose: Completion processing for block device I/O requests.
713 *
714 * Arguments: cmd - command that is finished.
715 *
716 * Lock status: Assumed that no lock is held upon entry.
717 *
718 * Returns: Nothing
719 *
720 * Notes: We will finish off the specified number of sectors. If we
721 * are done, the command block will be released and the queue
722 * function will be goosed. If we are not done then we have to
723 * figure out what to do next:
724 *
725 * a) We can call scsi_requeue_command(). The request
726 * will be unprepared and put back on the queue. Then
727 * a new command will be created for it. This should
728 * be used if we made forward progress, or if we want
729 * to switch from READ(10) to READ(6) for example.
730 *
731 * b) We can call __scsi_queue_insert(). The request will
732 * be put back on the queue and retried using the same
733 * command as before, possibly after a delay.
734 *
735 * c) We can call scsi_end_request() with -EIO to fail
736 * the remainder of the request.
737 */
738void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
739{
740 int result = cmd->result;
741 struct request_queue *q = cmd->device->request_queue;
742 struct request *req = cmd->request;
743 int error = 0;
744 struct scsi_sense_hdr sshdr;
745 bool sense_valid = false;
746 int sense_deferred = 0, level = 0;
747 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
748 ACTION_DELAYED_RETRY} action;
749 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
750
751 if (result) {
752 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
753 if (sense_valid)
754 sense_deferred = scsi_sense_is_deferred(&sshdr);
755 }
756
757 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
758 if (result) {
759 if (sense_valid && req->sense) {
760 /*
761 * SG_IO wants current and deferred errors
762 */
763 int len = 8 + cmd->sense_buffer[7];
764
765 if (len > SCSI_SENSE_BUFFERSIZE)
766 len = SCSI_SENSE_BUFFERSIZE;
767 memcpy(req->sense, cmd->sense_buffer, len);
768 req->sense_len = len;
769 }
770 if (!sense_deferred)
771 error = __scsi_error_from_host_byte(cmd, result);
772 }
773 /*
774 * __scsi_error_from_host_byte may have reset the host_byte
775 */
776 req->errors = cmd->result;
777
778 req->resid_len = scsi_get_resid(cmd);
779
780 if (scsi_bidi_cmnd(cmd)) {
781 /*
782 * Bidi commands Must be complete as a whole,
783 * both sides at once.
784 */
785 req->next_rq->resid_len = scsi_in(cmd)->resid;
786 if (scsi_end_request(req, 0, blk_rq_bytes(req),
787 blk_rq_bytes(req->next_rq)))
788 BUG();
789 return;
790 }
791 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
792 /*
793 * Certain non BLOCK_PC requests are commands that don't
794 * actually transfer anything (FLUSH), so cannot use
795 * good_bytes != blk_rq_bytes(req) as the signal for an error.
796 * This sets the error explicitly for the problem case.
797 */
798 error = __scsi_error_from_host_byte(cmd, result);
799 }
800
801 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
802 BUG_ON(blk_bidi_rq(req));
803
804 /*
805 * Next deal with any sectors which we were able to correctly
806 * handle.
807 */
808 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
809 "%u sectors total, %d bytes done.\n",
810 blk_rq_sectors(req), good_bytes));
811
812 /*
813 * Recovered errors need reporting, but they're always treated
814 * as success, so fiddle the result code here. For BLOCK_PC
815 * we already took a copy of the original into rq->errors which
816 * is what gets returned to the user
817 */
818 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
819 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
820 * print since caller wants ATA registers. Only occurs on
821 * SCSI ATA PASS_THROUGH commands when CK_COND=1
822 */
823 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
824 ;
825 else if (!(req->rq_flags & RQF_QUIET))
826 scsi_print_sense(cmd);
827 result = 0;
828 /* BLOCK_PC may have set error */
829 error = 0;
830 }
831
832 /*
833 * special case: failed zero length commands always need to
834 * drop down into the retry code. Otherwise, if we finished
835 * all bytes in the request we are done now.
836 */
837 if (!(blk_rq_bytes(req) == 0 && error) &&
838 !scsi_end_request(req, error, good_bytes, 0))
839 return;
840
841 /*
842 * Kill remainder if no retrys.
843 */
844 if (error && scsi_noretry_cmd(cmd)) {
845 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
846 BUG();
847 return;
848 }
849
850 /*
851 * If there had been no error, but we have leftover bytes in the
852 * requeues just queue the command up again.
853 */
854 if (result == 0)
855 goto requeue;
856
857 error = __scsi_error_from_host_byte(cmd, result);
858
859 if (host_byte(result) == DID_RESET) {
860 /* Third party bus reset or reset for error recovery
861 * reasons. Just retry the command and see what
862 * happens.
863 */
864 action = ACTION_RETRY;
865 } else if (sense_valid && !sense_deferred) {
866 switch (sshdr.sense_key) {
867 case UNIT_ATTENTION:
868 if (cmd->device->removable) {
869 /* Detected disc change. Set a bit
870 * and quietly refuse further access.
871 */
872 cmd->device->changed = 1;
873 action = ACTION_FAIL;
874 } else {
875 /* Must have been a power glitch, or a
876 * bus reset. Could not have been a
877 * media change, so we just retry the
878 * command and see what happens.
879 */
880 action = ACTION_RETRY;
881 }
882 break;
883 case ILLEGAL_REQUEST:
884 /* If we had an ILLEGAL REQUEST returned, then
885 * we may have performed an unsupported
886 * command. The only thing this should be
887 * would be a ten byte read where only a six
888 * byte read was supported. Also, on a system
889 * where READ CAPACITY failed, we may have
890 * read past the end of the disk.
891 */
892 if ((cmd->device->use_10_for_rw &&
893 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
894 (cmd->cmnd[0] == READ_10 ||
895 cmd->cmnd[0] == WRITE_10)) {
896 /* This will issue a new 6-byte command. */
897 cmd->device->use_10_for_rw = 0;
898 action = ACTION_REPREP;
899 } else if (sshdr.asc == 0x10) /* DIX */ {
900 action = ACTION_FAIL;
901 error = -EILSEQ;
902 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
903 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
904 action = ACTION_FAIL;
905 error = -EREMOTEIO;
906 } else
907 action = ACTION_FAIL;
908 break;
909 case ABORTED_COMMAND:
910 action = ACTION_FAIL;
911 if (sshdr.asc == 0x10) /* DIF */
912 error = -EILSEQ;
913 break;
914 case NOT_READY:
915 /* If the device is in the process of becoming
916 * ready, or has a temporary blockage, retry.
917 */
918 if (sshdr.asc == 0x04) {
919 switch (sshdr.ascq) {
920 case 0x01: /* becoming ready */
921 case 0x04: /* format in progress */
922 case 0x05: /* rebuild in progress */
923 case 0x06: /* recalculation in progress */
924 case 0x07: /* operation in progress */
925 case 0x08: /* Long write in progress */
926 case 0x09: /* self test in progress */
927 case 0x14: /* space allocation in progress */
928 action = ACTION_DELAYED_RETRY;
929 break;
930 default:
931 action = ACTION_FAIL;
932 break;
933 }
934 } else
935 action = ACTION_FAIL;
936 break;
937 case VOLUME_OVERFLOW:
938 /* See SSC3rXX or current. */
939 action = ACTION_FAIL;
940 break;
941 default:
942 action = ACTION_FAIL;
943 break;
944 }
945 } else
946 action = ACTION_FAIL;
947
948 if (action != ACTION_FAIL &&
949 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
950 action = ACTION_FAIL;
951
952 switch (action) {
953 case ACTION_FAIL:
954 /* Give up and fail the remainder of the request */
955 if (!(req->rq_flags & RQF_QUIET)) {
956 static DEFINE_RATELIMIT_STATE(_rs,
957 DEFAULT_RATELIMIT_INTERVAL,
958 DEFAULT_RATELIMIT_BURST);
959
960 if (unlikely(scsi_logging_level))
961 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
962 SCSI_LOG_MLCOMPLETE_BITS);
963
964 /*
965 * if logging is enabled the failure will be printed
966 * in scsi_log_completion(), so avoid duplicate messages
967 */
968 if (!level && __ratelimit(&_rs)) {
969 scsi_print_result(cmd, NULL, FAILED);
970 if (driver_byte(result) & DRIVER_SENSE)
971 scsi_print_sense(cmd);
972 scsi_print_command(cmd);
973 }
974 }
975 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
976 return;
977 /*FALLTHRU*/
978 case ACTION_REPREP:
979 requeue:
980 /* Unprep the request and put it back at the head of the queue.
981 * A new command will be prepared and issued.
982 */
983 if (q->mq_ops) {
984 cmd->request->rq_flags &= ~RQF_DONTPREP;
985 scsi_mq_uninit_cmd(cmd);
986 scsi_mq_requeue_cmd(cmd);
987 } else {
988 scsi_release_buffers(cmd);
989 scsi_requeue_command(q, cmd);
990 }
991 break;
992 case ACTION_RETRY:
993 /* Retry the same command immediately */
994 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
995 break;
996 case ACTION_DELAYED_RETRY:
997 /* Retry the same command after a delay */
998 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
999 break;
1000 }
1001}
1002
1003static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1004{
1005 int count;
1006
1007 /*
1008 * If sg table allocation fails, requeue request later.
1009 */
1010 if (unlikely(sg_alloc_table_chained(&sdb->table,
1011 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1012 return BLKPREP_DEFER;
1013
1014 /*
1015 * Next, walk the list, and fill in the addresses and sizes of
1016 * each segment.
1017 */
1018 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1019 BUG_ON(count > sdb->table.nents);
1020 sdb->table.nents = count;
1021 sdb->length = blk_rq_payload_bytes(req);
1022 return BLKPREP_OK;
1023}
1024
1025/*
1026 * Function: scsi_init_io()
1027 *
1028 * Purpose: SCSI I/O initialize function.
1029 *
1030 * Arguments: cmd - Command descriptor we wish to initialize
1031 *
1032 * Returns: 0 on success
1033 * BLKPREP_DEFER if the failure is retryable
1034 * BLKPREP_KILL if the failure is fatal
1035 */
1036int scsi_init_io(struct scsi_cmnd *cmd)
1037{
1038 struct scsi_device *sdev = cmd->device;
1039 struct request *rq = cmd->request;
1040 bool is_mq = (rq->mq_ctx != NULL);
1041 int error;
1042
1043 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1044 return -EINVAL;
1045
1046 error = scsi_init_sgtable(rq, &cmd->sdb);
1047 if (error)
1048 goto err_exit;
1049
1050 if (blk_bidi_rq(rq)) {
1051 if (!rq->q->mq_ops) {
1052 struct scsi_data_buffer *bidi_sdb =
1053 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1054 if (!bidi_sdb) {
1055 error = BLKPREP_DEFER;
1056 goto err_exit;
1057 }
1058
1059 rq->next_rq->special = bidi_sdb;
1060 }
1061
1062 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1063 if (error)
1064 goto err_exit;
1065 }
1066
1067 if (blk_integrity_rq(rq)) {
1068 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1069 int ivecs, count;
1070
1071 if (prot_sdb == NULL) {
1072 /*
1073 * This can happen if someone (e.g. multipath)
1074 * queues a command to a device on an adapter
1075 * that does not support DIX.
1076 */
1077 WARN_ON_ONCE(1);
1078 error = BLKPREP_KILL;
1079 goto err_exit;
1080 }
1081
1082 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1083
1084 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1085 prot_sdb->table.sgl)) {
1086 error = BLKPREP_DEFER;
1087 goto err_exit;
1088 }
1089
1090 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1091 prot_sdb->table.sgl);
1092 BUG_ON(unlikely(count > ivecs));
1093 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1094
1095 cmd->prot_sdb = prot_sdb;
1096 cmd->prot_sdb->table.nents = count;
1097 }
1098
1099 return BLKPREP_OK;
1100err_exit:
1101 if (is_mq) {
1102 scsi_mq_free_sgtables(cmd);
1103 } else {
1104 scsi_release_buffers(cmd);
1105 cmd->request->special = NULL;
1106 scsi_put_command(cmd);
1107 put_device(&sdev->sdev_gendev);
1108 }
1109 return error;
1110}
1111EXPORT_SYMBOL(scsi_init_io);
1112
1113static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1114 struct request *req)
1115{
1116 struct scsi_cmnd *cmd;
1117
1118 if (!req->special) {
1119 /* Bail if we can't get a reference to the device */
1120 if (!get_device(&sdev->sdev_gendev))
1121 return NULL;
1122
1123 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1124 if (unlikely(!cmd)) {
1125 put_device(&sdev->sdev_gendev);
1126 return NULL;
1127 }
1128 req->special = cmd;
1129 } else {
1130 cmd = req->special;
1131 }
1132
1133 /* pull a tag out of the request if we have one */
1134 cmd->tag = req->tag;
1135 cmd->request = req;
1136
1137 cmd->cmnd = req->cmd;
1138 cmd->prot_op = SCSI_PROT_NORMAL;
1139
1140 return cmd;
1141}
1142
1143static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1144{
1145 struct scsi_cmnd *cmd = req->special;
1146
1147 /*
1148 * BLOCK_PC requests may transfer data, in which case they must
1149 * a bio attached to them. Or they might contain a SCSI command
1150 * that does not transfer data, in which case they may optionally
1151 * submit a request without an attached bio.
1152 */
1153 if (req->bio) {
1154 int ret = scsi_init_io(cmd);
1155 if (unlikely(ret))
1156 return ret;
1157 } else {
1158 BUG_ON(blk_rq_bytes(req));
1159
1160 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1161 }
1162
1163 cmd->cmd_len = req->cmd_len;
1164 cmd->transfersize = blk_rq_bytes(req);
1165 cmd->allowed = req->retries;
1166 return BLKPREP_OK;
1167}
1168
1169/*
1170 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1171 * that still need to be translated to SCSI CDBs from the ULD.
1172 */
1173static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1174{
1175 struct scsi_cmnd *cmd = req->special;
1176
1177 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1178 int ret = sdev->handler->prep_fn(sdev, req);
1179 if (ret != BLKPREP_OK)
1180 return ret;
1181 }
1182
1183 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1184 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1185}
1186
1187static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1188{
1189 struct scsi_cmnd *cmd = req->special;
1190
1191 if (!blk_rq_bytes(req))
1192 cmd->sc_data_direction = DMA_NONE;
1193 else if (rq_data_dir(req) == WRITE)
1194 cmd->sc_data_direction = DMA_TO_DEVICE;
1195 else
1196 cmd->sc_data_direction = DMA_FROM_DEVICE;
1197
1198 switch (req->cmd_type) {
1199 case REQ_TYPE_FS:
1200 return scsi_setup_fs_cmnd(sdev, req);
1201 case REQ_TYPE_BLOCK_PC:
1202 return scsi_setup_blk_pc_cmnd(sdev, req);
1203 default:
1204 return BLKPREP_KILL;
1205 }
1206}
1207
1208static int
1209scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1210{
1211 int ret = BLKPREP_OK;
1212
1213 /*
1214 * If the device is not in running state we will reject some
1215 * or all commands.
1216 */
1217 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1218 switch (sdev->sdev_state) {
1219 case SDEV_OFFLINE:
1220 case SDEV_TRANSPORT_OFFLINE:
1221 /*
1222 * If the device is offline we refuse to process any
1223 * commands. The device must be brought online
1224 * before trying any recovery commands.
1225 */
1226 sdev_printk(KERN_ERR, sdev,
1227 "rejecting I/O to offline device\n");
1228 ret = BLKPREP_KILL;
1229 break;
1230 case SDEV_DEL:
1231 /*
1232 * If the device is fully deleted, we refuse to
1233 * process any commands as well.
1234 */
1235 sdev_printk(KERN_ERR, sdev,
1236 "rejecting I/O to dead device\n");
1237 ret = BLKPREP_KILL;
1238 break;
1239 case SDEV_BLOCK:
1240 case SDEV_CREATED_BLOCK:
1241 ret = BLKPREP_DEFER;
1242 break;
1243 case SDEV_QUIESCE:
1244 /*
1245 * If the devices is blocked we defer normal commands.
1246 */
1247 if (!(req->rq_flags & RQF_PREEMPT))
1248 ret = BLKPREP_DEFER;
1249 break;
1250 default:
1251 /*
1252 * For any other not fully online state we only allow
1253 * special commands. In particular any user initiated
1254 * command is not allowed.
1255 */
1256 if (!(req->rq_flags & RQF_PREEMPT))
1257 ret = BLKPREP_KILL;
1258 break;
1259 }
1260 }
1261 return ret;
1262}
1263
1264static int
1265scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1266{
1267 struct scsi_device *sdev = q->queuedata;
1268
1269 switch (ret) {
1270 case BLKPREP_KILL:
1271 case BLKPREP_INVALID:
1272 req->errors = DID_NO_CONNECT << 16;
1273 /* release the command and kill it */
1274 if (req->special) {
1275 struct scsi_cmnd *cmd = req->special;
1276 scsi_release_buffers(cmd);
1277 scsi_put_command(cmd);
1278 put_device(&sdev->sdev_gendev);
1279 req->special = NULL;
1280 }
1281 break;
1282 case BLKPREP_DEFER:
1283 /*
1284 * If we defer, the blk_peek_request() returns NULL, but the
1285 * queue must be restarted, so we schedule a callback to happen
1286 * shortly.
1287 */
1288 if (atomic_read(&sdev->device_busy) == 0)
1289 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 break;
1291 default:
1292 req->rq_flags |= RQF_DONTPREP;
1293 }
1294
1295 return ret;
1296}
1297
1298static int scsi_prep_fn(struct request_queue *q, struct request *req)
1299{
1300 struct scsi_device *sdev = q->queuedata;
1301 struct scsi_cmnd *cmd;
1302 int ret;
1303
1304 ret = scsi_prep_state_check(sdev, req);
1305 if (ret != BLKPREP_OK)
1306 goto out;
1307
1308 cmd = scsi_get_cmd_from_req(sdev, req);
1309 if (unlikely(!cmd)) {
1310 ret = BLKPREP_DEFER;
1311 goto out;
1312 }
1313
1314 ret = scsi_setup_cmnd(sdev, req);
1315out:
1316 return scsi_prep_return(q, req, ret);
1317}
1318
1319static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1320{
1321 scsi_uninit_cmd(req->special);
1322}
1323
1324/*
1325 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1326 * return 0.
1327 *
1328 * Called with the queue_lock held.
1329 */
1330static inline int scsi_dev_queue_ready(struct request_queue *q,
1331 struct scsi_device *sdev)
1332{
1333 unsigned int busy;
1334
1335 busy = atomic_inc_return(&sdev->device_busy) - 1;
1336 if (atomic_read(&sdev->device_blocked)) {
1337 if (busy)
1338 goto out_dec;
1339
1340 /*
1341 * unblock after device_blocked iterates to zero
1342 */
1343 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1344 /*
1345 * For the MQ case we take care of this in the caller.
1346 */
1347 if (!q->mq_ops)
1348 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1349 goto out_dec;
1350 }
1351 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1352 "unblocking device at zero depth\n"));
1353 }
1354
1355 if (busy >= sdev->queue_depth)
1356 goto out_dec;
1357
1358 return 1;
1359out_dec:
1360 atomic_dec(&sdev->device_busy);
1361 return 0;
1362}
1363
1364/*
1365 * scsi_target_queue_ready: checks if there we can send commands to target
1366 * @sdev: scsi device on starget to check.
1367 */
1368static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1369 struct scsi_device *sdev)
1370{
1371 struct scsi_target *starget = scsi_target(sdev);
1372 unsigned int busy;
1373
1374 if (starget->single_lun) {
1375 spin_lock_irq(shost->host_lock);
1376 if (starget->starget_sdev_user &&
1377 starget->starget_sdev_user != sdev) {
1378 spin_unlock_irq(shost->host_lock);
1379 return 0;
1380 }
1381 starget->starget_sdev_user = sdev;
1382 spin_unlock_irq(shost->host_lock);
1383 }
1384
1385 if (starget->can_queue <= 0)
1386 return 1;
1387
1388 busy = atomic_inc_return(&starget->target_busy) - 1;
1389 if (atomic_read(&starget->target_blocked) > 0) {
1390 if (busy)
1391 goto starved;
1392
1393 /*
1394 * unblock after target_blocked iterates to zero
1395 */
1396 if (atomic_dec_return(&starget->target_blocked) > 0)
1397 goto out_dec;
1398
1399 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1400 "unblocking target at zero depth\n"));
1401 }
1402
1403 if (busy >= starget->can_queue)
1404 goto starved;
1405
1406 return 1;
1407
1408starved:
1409 spin_lock_irq(shost->host_lock);
1410 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1411 spin_unlock_irq(shost->host_lock);
1412out_dec:
1413 if (starget->can_queue > 0)
1414 atomic_dec(&starget->target_busy);
1415 return 0;
1416}
1417
1418/*
1419 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1420 * return 0. We must end up running the queue again whenever 0 is
1421 * returned, else IO can hang.
1422 */
1423static inline int scsi_host_queue_ready(struct request_queue *q,
1424 struct Scsi_Host *shost,
1425 struct scsi_device *sdev)
1426{
1427 unsigned int busy;
1428
1429 if (scsi_host_in_recovery(shost))
1430 return 0;
1431
1432 busy = atomic_inc_return(&shost->host_busy) - 1;
1433 if (atomic_read(&shost->host_blocked) > 0) {
1434 if (busy)
1435 goto starved;
1436
1437 /*
1438 * unblock after host_blocked iterates to zero
1439 */
1440 if (atomic_dec_return(&shost->host_blocked) > 0)
1441 goto out_dec;
1442
1443 SCSI_LOG_MLQUEUE(3,
1444 shost_printk(KERN_INFO, shost,
1445 "unblocking host at zero depth\n"));
1446 }
1447
1448 if (shost->can_queue > 0 && busy >= shost->can_queue)
1449 goto starved;
1450 if (shost->host_self_blocked)
1451 goto starved;
1452
1453 /* We're OK to process the command, so we can't be starved */
1454 if (!list_empty(&sdev->starved_entry)) {
1455 spin_lock_irq(shost->host_lock);
1456 if (!list_empty(&sdev->starved_entry))
1457 list_del_init(&sdev->starved_entry);
1458 spin_unlock_irq(shost->host_lock);
1459 }
1460
1461 return 1;
1462
1463starved:
1464 spin_lock_irq(shost->host_lock);
1465 if (list_empty(&sdev->starved_entry))
1466 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1467 spin_unlock_irq(shost->host_lock);
1468out_dec:
1469 atomic_dec(&shost->host_busy);
1470 return 0;
1471}
1472
1473/*
1474 * Busy state exporting function for request stacking drivers.
1475 *
1476 * For efficiency, no lock is taken to check the busy state of
1477 * shost/starget/sdev, since the returned value is not guaranteed and
1478 * may be changed after request stacking drivers call the function,
1479 * regardless of taking lock or not.
1480 *
1481 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1482 * needs to return 'not busy'. Otherwise, request stacking drivers
1483 * may hold requests forever.
1484 */
1485static int scsi_lld_busy(struct request_queue *q)
1486{
1487 struct scsi_device *sdev = q->queuedata;
1488 struct Scsi_Host *shost;
1489
1490 if (blk_queue_dying(q))
1491 return 0;
1492
1493 shost = sdev->host;
1494
1495 /*
1496 * Ignore host/starget busy state.
1497 * Since block layer does not have a concept of fairness across
1498 * multiple queues, congestion of host/starget needs to be handled
1499 * in SCSI layer.
1500 */
1501 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1502 return 1;
1503
1504 return 0;
1505}
1506
1507/*
1508 * Kill a request for a dead device
1509 */
1510static void scsi_kill_request(struct request *req, struct request_queue *q)
1511{
1512 struct scsi_cmnd *cmd = req->special;
1513 struct scsi_device *sdev;
1514 struct scsi_target *starget;
1515 struct Scsi_Host *shost;
1516
1517 blk_start_request(req);
1518
1519 scmd_printk(KERN_INFO, cmd, "killing request\n");
1520
1521 sdev = cmd->device;
1522 starget = scsi_target(sdev);
1523 shost = sdev->host;
1524 scsi_init_cmd_errh(cmd);
1525 cmd->result = DID_NO_CONNECT << 16;
1526 atomic_inc(&cmd->device->iorequest_cnt);
1527
1528 /*
1529 * SCSI request completion path will do scsi_device_unbusy(),
1530 * bump busy counts. To bump the counters, we need to dance
1531 * with the locks as normal issue path does.
1532 */
1533 atomic_inc(&sdev->device_busy);
1534 atomic_inc(&shost->host_busy);
1535 if (starget->can_queue > 0)
1536 atomic_inc(&starget->target_busy);
1537
1538 blk_complete_request(req);
1539}
1540
1541static void scsi_softirq_done(struct request *rq)
1542{
1543 struct scsi_cmnd *cmd = rq->special;
1544 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1545 int disposition;
1546
1547 INIT_LIST_HEAD(&cmd->eh_entry);
1548
1549 atomic_inc(&cmd->device->iodone_cnt);
1550 if (cmd->result)
1551 atomic_inc(&cmd->device->ioerr_cnt);
1552
1553 disposition = scsi_decide_disposition(cmd);
1554 if (disposition != SUCCESS &&
1555 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1556 sdev_printk(KERN_ERR, cmd->device,
1557 "timing out command, waited %lus\n",
1558 wait_for/HZ);
1559 disposition = SUCCESS;
1560 }
1561
1562 scsi_log_completion(cmd, disposition);
1563
1564 switch (disposition) {
1565 case SUCCESS:
1566 scsi_finish_command(cmd);
1567 break;
1568 case NEEDS_RETRY:
1569 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1570 break;
1571 case ADD_TO_MLQUEUE:
1572 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1573 break;
1574 default:
1575 if (!scsi_eh_scmd_add(cmd, 0))
1576 scsi_finish_command(cmd);
1577 }
1578}
1579
1580/**
1581 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1582 * @cmd: command block we are dispatching.
1583 *
1584 * Return: nonzero return request was rejected and device's queue needs to be
1585 * plugged.
1586 */
1587static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1588{
1589 struct Scsi_Host *host = cmd->device->host;
1590 int rtn = 0;
1591
1592 atomic_inc(&cmd->device->iorequest_cnt);
1593
1594 /* check if the device is still usable */
1595 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1596 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1597 * returns an immediate error upwards, and signals
1598 * that the device is no longer present */
1599 cmd->result = DID_NO_CONNECT << 16;
1600 goto done;
1601 }
1602
1603 /* Check to see if the scsi lld made this device blocked. */
1604 if (unlikely(scsi_device_blocked(cmd->device))) {
1605 /*
1606 * in blocked state, the command is just put back on
1607 * the device queue. The suspend state has already
1608 * blocked the queue so future requests should not
1609 * occur until the device transitions out of the
1610 * suspend state.
1611 */
1612 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1613 "queuecommand : device blocked\n"));
1614 return SCSI_MLQUEUE_DEVICE_BUSY;
1615 }
1616
1617 /* Store the LUN value in cmnd, if needed. */
1618 if (cmd->device->lun_in_cdb)
1619 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1620 (cmd->device->lun << 5 & 0xe0);
1621
1622 scsi_log_send(cmd);
1623
1624 /*
1625 * Before we queue this command, check if the command
1626 * length exceeds what the host adapter can handle.
1627 */
1628 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1629 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1630 "queuecommand : command too long. "
1631 "cdb_size=%d host->max_cmd_len=%d\n",
1632 cmd->cmd_len, cmd->device->host->max_cmd_len));
1633 cmd->result = (DID_ABORT << 16);
1634 goto done;
1635 }
1636
1637 if (unlikely(host->shost_state == SHOST_DEL)) {
1638 cmd->result = (DID_NO_CONNECT << 16);
1639 goto done;
1640
1641 }
1642
1643 trace_scsi_dispatch_cmd_start(cmd);
1644 rtn = host->hostt->queuecommand(host, cmd);
1645 if (rtn) {
1646 trace_scsi_dispatch_cmd_error(cmd, rtn);
1647 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1648 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1649 rtn = SCSI_MLQUEUE_HOST_BUSY;
1650
1651 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1652 "queuecommand : request rejected\n"));
1653 }
1654
1655 return rtn;
1656 done:
1657 cmd->scsi_done(cmd);
1658 return 0;
1659}
1660
1661/**
1662 * scsi_done - Invoke completion on finished SCSI command.
1663 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1664 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1665 *
1666 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1667 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1668 * calls blk_complete_request() for further processing.
1669 *
1670 * This function is interrupt context safe.
1671 */
1672static void scsi_done(struct scsi_cmnd *cmd)
1673{
1674 trace_scsi_dispatch_cmd_done(cmd);
1675 blk_complete_request(cmd->request);
1676}
1677
1678/*
1679 * Function: scsi_request_fn()
1680 *
1681 * Purpose: Main strategy routine for SCSI.
1682 *
1683 * Arguments: q - Pointer to actual queue.
1684 *
1685 * Returns: Nothing
1686 *
1687 * Lock status: IO request lock assumed to be held when called.
1688 */
1689static void scsi_request_fn(struct request_queue *q)
1690 __releases(q->queue_lock)
1691 __acquires(q->queue_lock)
1692{
1693 struct scsi_device *sdev = q->queuedata;
1694 struct Scsi_Host *shost;
1695 struct scsi_cmnd *cmd;
1696 struct request *req;
1697
1698 /*
1699 * To start with, we keep looping until the queue is empty, or until
1700 * the host is no longer able to accept any more requests.
1701 */
1702 shost = sdev->host;
1703 for (;;) {
1704 int rtn;
1705 /*
1706 * get next queueable request. We do this early to make sure
1707 * that the request is fully prepared even if we cannot
1708 * accept it.
1709 */
1710 req = blk_peek_request(q);
1711 if (!req)
1712 break;
1713
1714 if (unlikely(!scsi_device_online(sdev))) {
1715 sdev_printk(KERN_ERR, sdev,
1716 "rejecting I/O to offline device\n");
1717 scsi_kill_request(req, q);
1718 continue;
1719 }
1720
1721 if (!scsi_dev_queue_ready(q, sdev))
1722 break;
1723
1724 /*
1725 * Remove the request from the request list.
1726 */
1727 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1728 blk_start_request(req);
1729
1730 spin_unlock_irq(q->queue_lock);
1731 cmd = req->special;
1732 if (unlikely(cmd == NULL)) {
1733 printk(KERN_CRIT "impossible request in %s.\n"
1734 "please mail a stack trace to "
1735 "linux-scsi@vger.kernel.org\n",
1736 __func__);
1737 blk_dump_rq_flags(req, "foo");
1738 BUG();
1739 }
1740
1741 /*
1742 * We hit this when the driver is using a host wide
1743 * tag map. For device level tag maps the queue_depth check
1744 * in the device ready fn would prevent us from trying
1745 * to allocate a tag. Since the map is a shared host resource
1746 * we add the dev to the starved list so it eventually gets
1747 * a run when a tag is freed.
1748 */
1749 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
1750 spin_lock_irq(shost->host_lock);
1751 if (list_empty(&sdev->starved_entry))
1752 list_add_tail(&sdev->starved_entry,
1753 &shost->starved_list);
1754 spin_unlock_irq(shost->host_lock);
1755 goto not_ready;
1756 }
1757
1758 if (!scsi_target_queue_ready(shost, sdev))
1759 goto not_ready;
1760
1761 if (!scsi_host_queue_ready(q, shost, sdev))
1762 goto host_not_ready;
1763
1764 if (sdev->simple_tags)
1765 cmd->flags |= SCMD_TAGGED;
1766 else
1767 cmd->flags &= ~SCMD_TAGGED;
1768
1769 /*
1770 * Finally, initialize any error handling parameters, and set up
1771 * the timers for timeouts.
1772 */
1773 scsi_init_cmd_errh(cmd);
1774
1775 /*
1776 * Dispatch the command to the low-level driver.
1777 */
1778 cmd->scsi_done = scsi_done;
1779 rtn = scsi_dispatch_cmd(cmd);
1780 if (rtn) {
1781 scsi_queue_insert(cmd, rtn);
1782 spin_lock_irq(q->queue_lock);
1783 goto out_delay;
1784 }
1785 spin_lock_irq(q->queue_lock);
1786 }
1787
1788 return;
1789
1790 host_not_ready:
1791 if (scsi_target(sdev)->can_queue > 0)
1792 atomic_dec(&scsi_target(sdev)->target_busy);
1793 not_ready:
1794 /*
1795 * lock q, handle tag, requeue req, and decrement device_busy. We
1796 * must return with queue_lock held.
1797 *
1798 * Decrementing device_busy without checking it is OK, as all such
1799 * cases (host limits or settings) should run the queue at some
1800 * later time.
1801 */
1802 spin_lock_irq(q->queue_lock);
1803 blk_requeue_request(q, req);
1804 atomic_dec(&sdev->device_busy);
1805out_delay:
1806 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1807 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1808}
1809
1810static inline int prep_to_mq(int ret)
1811{
1812 switch (ret) {
1813 case BLKPREP_OK:
1814 return BLK_MQ_RQ_QUEUE_OK;
1815 case BLKPREP_DEFER:
1816 return BLK_MQ_RQ_QUEUE_BUSY;
1817 default:
1818 return BLK_MQ_RQ_QUEUE_ERROR;
1819 }
1820}
1821
1822static int scsi_mq_prep_fn(struct request *req)
1823{
1824 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1825 struct scsi_device *sdev = req->q->queuedata;
1826 struct Scsi_Host *shost = sdev->host;
1827 unsigned char *sense_buf = cmd->sense_buffer;
1828 struct scatterlist *sg;
1829
1830 memset(cmd, 0, sizeof(struct scsi_cmnd));
1831
1832 req->special = cmd;
1833
1834 cmd->request = req;
1835 cmd->device = sdev;
1836 cmd->sense_buffer = sense_buf;
1837
1838 cmd->tag = req->tag;
1839
1840 cmd->cmnd = req->cmd;
1841 cmd->prot_op = SCSI_PROT_NORMAL;
1842
1843 INIT_LIST_HEAD(&cmd->list);
1844 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1845 cmd->jiffies_at_alloc = jiffies;
1846
1847 if (shost->use_cmd_list) {
1848 spin_lock_irq(&sdev->list_lock);
1849 list_add_tail(&cmd->list, &sdev->cmd_list);
1850 spin_unlock_irq(&sdev->list_lock);
1851 }
1852
1853 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1854 cmd->sdb.table.sgl = sg;
1855
1856 if (scsi_host_get_prot(shost)) {
1857 cmd->prot_sdb = (void *)sg +
1858 min_t(unsigned int,
1859 shost->sg_tablesize, SG_CHUNK_SIZE) *
1860 sizeof(struct scatterlist);
1861 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1862
1863 cmd->prot_sdb->table.sgl =
1864 (struct scatterlist *)(cmd->prot_sdb + 1);
1865 }
1866
1867 if (blk_bidi_rq(req)) {
1868 struct request *next_rq = req->next_rq;
1869 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1870
1871 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1872 bidi_sdb->table.sgl =
1873 (struct scatterlist *)(bidi_sdb + 1);
1874
1875 next_rq->special = bidi_sdb;
1876 }
1877
1878 blk_mq_start_request(req);
1879
1880 return scsi_setup_cmnd(sdev, req);
1881}
1882
1883static void scsi_mq_done(struct scsi_cmnd *cmd)
1884{
1885 trace_scsi_dispatch_cmd_done(cmd);
1886 blk_mq_complete_request(cmd->request, cmd->request->errors);
1887}
1888
1889static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1890 const struct blk_mq_queue_data *bd)
1891{
1892 struct request *req = bd->rq;
1893 struct request_queue *q = req->q;
1894 struct scsi_device *sdev = q->queuedata;
1895 struct Scsi_Host *shost = sdev->host;
1896 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1897 int ret;
1898 int reason;
1899
1900 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1901 if (ret != BLK_MQ_RQ_QUEUE_OK)
1902 goto out;
1903
1904 ret = BLK_MQ_RQ_QUEUE_BUSY;
1905 if (!get_device(&sdev->sdev_gendev))
1906 goto out;
1907
1908 if (!scsi_dev_queue_ready(q, sdev))
1909 goto out_put_device;
1910 if (!scsi_target_queue_ready(shost, sdev))
1911 goto out_dec_device_busy;
1912 if (!scsi_host_queue_ready(q, shost, sdev))
1913 goto out_dec_target_busy;
1914
1915
1916 if (!(req->rq_flags & RQF_DONTPREP)) {
1917 ret = prep_to_mq(scsi_mq_prep_fn(req));
1918 if (ret != BLK_MQ_RQ_QUEUE_OK)
1919 goto out_dec_host_busy;
1920 req->rq_flags |= RQF_DONTPREP;
1921 } else {
1922 blk_mq_start_request(req);
1923 }
1924
1925 if (sdev->simple_tags)
1926 cmd->flags |= SCMD_TAGGED;
1927 else
1928 cmd->flags &= ~SCMD_TAGGED;
1929
1930 scsi_init_cmd_errh(cmd);
1931 cmd->scsi_done = scsi_mq_done;
1932
1933 reason = scsi_dispatch_cmd(cmd);
1934 if (reason) {
1935 scsi_set_blocked(cmd, reason);
1936 ret = BLK_MQ_RQ_QUEUE_BUSY;
1937 goto out_dec_host_busy;
1938 }
1939
1940 return BLK_MQ_RQ_QUEUE_OK;
1941
1942out_dec_host_busy:
1943 atomic_dec(&shost->host_busy);
1944out_dec_target_busy:
1945 if (scsi_target(sdev)->can_queue > 0)
1946 atomic_dec(&scsi_target(sdev)->target_busy);
1947out_dec_device_busy:
1948 atomic_dec(&sdev->device_busy);
1949out_put_device:
1950 put_device(&sdev->sdev_gendev);
1951out:
1952 switch (ret) {
1953 case BLK_MQ_RQ_QUEUE_BUSY:
1954 if (atomic_read(&sdev->device_busy) == 0 &&
1955 !scsi_device_blocked(sdev))
1956 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
1957 break;
1958 case BLK_MQ_RQ_QUEUE_ERROR:
1959 /*
1960 * Make sure to release all allocated ressources when
1961 * we hit an error, as we will never see this command
1962 * again.
1963 */
1964 if (req->rq_flags & RQF_DONTPREP)
1965 scsi_mq_uninit_cmd(cmd);
1966 break;
1967 default:
1968 break;
1969 }
1970 return ret;
1971}
1972
1973static enum blk_eh_timer_return scsi_timeout(struct request *req,
1974 bool reserved)
1975{
1976 if (reserved)
1977 return BLK_EH_RESET_TIMER;
1978 return scsi_times_out(req);
1979}
1980
1981static int scsi_init_request(void *data, struct request *rq,
1982 unsigned int hctx_idx, unsigned int request_idx,
1983 unsigned int numa_node)
1984{
1985 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1986
1987 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
1988 numa_node);
1989 if (!cmd->sense_buffer)
1990 return -ENOMEM;
1991 return 0;
1992}
1993
1994static void scsi_exit_request(void *data, struct request *rq,
1995 unsigned int hctx_idx, unsigned int request_idx)
1996{
1997 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1998
1999 kfree(cmd->sense_buffer);
2000}
2001
2002static int scsi_map_queues(struct blk_mq_tag_set *set)
2003{
2004 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2005
2006 if (shost->hostt->map_queues)
2007 return shost->hostt->map_queues(shost);
2008 return blk_mq_map_queues(set);
2009}
2010
2011static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2012{
2013 struct device *host_dev;
2014 u64 bounce_limit = 0xffffffff;
2015
2016 if (shost->unchecked_isa_dma)
2017 return BLK_BOUNCE_ISA;
2018 /*
2019 * Platforms with virtual-DMA translation
2020 * hardware have no practical limit.
2021 */
2022 if (!PCI_DMA_BUS_IS_PHYS)
2023 return BLK_BOUNCE_ANY;
2024
2025 host_dev = scsi_get_device(shost);
2026 if (host_dev && host_dev->dma_mask)
2027 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2028
2029 return bounce_limit;
2030}
2031
2032static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2033{
2034 struct device *dev = shost->dma_dev;
2035
2036 /*
2037 * this limit is imposed by hardware restrictions
2038 */
2039 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2040 SG_MAX_SEGMENTS));
2041
2042 if (scsi_host_prot_dma(shost)) {
2043 shost->sg_prot_tablesize =
2044 min_not_zero(shost->sg_prot_tablesize,
2045 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2046 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2047 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2048 }
2049
2050 blk_queue_max_hw_sectors(q, shost->max_sectors);
2051 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2052 blk_queue_segment_boundary(q, shost->dma_boundary);
2053 dma_set_seg_boundary(dev, shost->dma_boundary);
2054
2055 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2056
2057 if (!shost->use_clustering)
2058 q->limits.cluster = 0;
2059
2060 /*
2061 * set a reasonable default alignment on word boundaries: the
2062 * host and device may alter it using
2063 * blk_queue_update_dma_alignment() later.
2064 */
2065 blk_queue_dma_alignment(q, 0x03);
2066}
2067
2068struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2069 request_fn_proc *request_fn)
2070{
2071 struct request_queue *q;
2072
2073 q = blk_init_queue(request_fn, NULL);
2074 if (!q)
2075 return NULL;
2076 __scsi_init_queue(shost, q);
2077 return q;
2078}
2079EXPORT_SYMBOL(__scsi_alloc_queue);
2080
2081struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2082{
2083 struct request_queue *q;
2084
2085 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2086 if (!q)
2087 return NULL;
2088
2089 blk_queue_prep_rq(q, scsi_prep_fn);
2090 blk_queue_unprep_rq(q, scsi_unprep_fn);
2091 blk_queue_softirq_done(q, scsi_softirq_done);
2092 blk_queue_rq_timed_out(q, scsi_times_out);
2093 blk_queue_lld_busy(q, scsi_lld_busy);
2094 return q;
2095}
2096
2097static struct blk_mq_ops scsi_mq_ops = {
2098 .queue_rq = scsi_queue_rq,
2099 .complete = scsi_softirq_done,
2100 .timeout = scsi_timeout,
2101 .init_request = scsi_init_request,
2102 .exit_request = scsi_exit_request,
2103 .map_queues = scsi_map_queues,
2104};
2105
2106struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2107{
2108 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2109 if (IS_ERR(sdev->request_queue))
2110 return NULL;
2111
2112 sdev->request_queue->queuedata = sdev;
2113 __scsi_init_queue(sdev->host, sdev->request_queue);
2114 return sdev->request_queue;
2115}
2116
2117int scsi_mq_setup_tags(struct Scsi_Host *shost)
2118{
2119 unsigned int cmd_size, sgl_size, tbl_size;
2120
2121 tbl_size = shost->sg_tablesize;
2122 if (tbl_size > SG_CHUNK_SIZE)
2123 tbl_size = SG_CHUNK_SIZE;
2124 sgl_size = tbl_size * sizeof(struct scatterlist);
2125 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2126 if (scsi_host_get_prot(shost))
2127 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2128
2129 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2130 shost->tag_set.ops = &scsi_mq_ops;
2131 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2132 shost->tag_set.queue_depth = shost->can_queue;
2133 shost->tag_set.cmd_size = cmd_size;
2134 shost->tag_set.numa_node = NUMA_NO_NODE;
2135 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2136 shost->tag_set.flags |=
2137 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2138 shost->tag_set.driver_data = shost;
2139
2140 return blk_mq_alloc_tag_set(&shost->tag_set);
2141}
2142
2143void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2144{
2145 blk_mq_free_tag_set(&shost->tag_set);
2146}
2147
2148/**
2149 * scsi_device_from_queue - return sdev associated with a request_queue
2150 * @q: The request queue to return the sdev from
2151 *
2152 * Return the sdev associated with a request queue or NULL if the
2153 * request_queue does not reference a SCSI device.
2154 */
2155struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2156{
2157 struct scsi_device *sdev = NULL;
2158
2159 if (q->mq_ops) {
2160 if (q->mq_ops == &scsi_mq_ops)
2161 sdev = q->queuedata;
2162 } else if (q->request_fn == scsi_request_fn)
2163 sdev = q->queuedata;
2164 if (!sdev || !get_device(&sdev->sdev_gendev))
2165 sdev = NULL;
2166
2167 return sdev;
2168}
2169EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2170
2171/*
2172 * Function: scsi_block_requests()
2173 *
2174 * Purpose: Utility function used by low-level drivers to prevent further
2175 * commands from being queued to the device.
2176 *
2177 * Arguments: shost - Host in question
2178 *
2179 * Returns: Nothing
2180 *
2181 * Lock status: No locks are assumed held.
2182 *
2183 * Notes: There is no timer nor any other means by which the requests
2184 * get unblocked other than the low-level driver calling
2185 * scsi_unblock_requests().
2186 */
2187void scsi_block_requests(struct Scsi_Host *shost)
2188{
2189 shost->host_self_blocked = 1;
2190}
2191EXPORT_SYMBOL(scsi_block_requests);
2192
2193/*
2194 * Function: scsi_unblock_requests()
2195 *
2196 * Purpose: Utility function used by low-level drivers to allow further
2197 * commands from being queued to the device.
2198 *
2199 * Arguments: shost - Host in question
2200 *
2201 * Returns: Nothing
2202 *
2203 * Lock status: No locks are assumed held.
2204 *
2205 * Notes: There is no timer nor any other means by which the requests
2206 * get unblocked other than the low-level driver calling
2207 * scsi_unblock_requests().
2208 *
2209 * This is done as an API function so that changes to the
2210 * internals of the scsi mid-layer won't require wholesale
2211 * changes to drivers that use this feature.
2212 */
2213void scsi_unblock_requests(struct Scsi_Host *shost)
2214{
2215 shost->host_self_blocked = 0;
2216 scsi_run_host_queues(shost);
2217}
2218EXPORT_SYMBOL(scsi_unblock_requests);
2219
2220int __init scsi_init_queue(void)
2221{
2222 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2223 sizeof(struct scsi_data_buffer),
2224 0, 0, NULL);
2225 if (!scsi_sdb_cache) {
2226 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2227 return -ENOMEM;
2228 }
2229
2230 return 0;
2231}
2232
2233void scsi_exit_queue(void)
2234{
2235 kmem_cache_destroy(scsi_sdb_cache);
2236}
2237
2238/**
2239 * scsi_mode_select - issue a mode select
2240 * @sdev: SCSI device to be queried
2241 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2242 * @sp: Save page bit (0 == don't save, 1 == save)
2243 * @modepage: mode page being requested
2244 * @buffer: request buffer (may not be smaller than eight bytes)
2245 * @len: length of request buffer.
2246 * @timeout: command timeout
2247 * @retries: number of retries before failing
2248 * @data: returns a structure abstracting the mode header data
2249 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2250 * must be SCSI_SENSE_BUFFERSIZE big.
2251 *
2252 * Returns zero if successful; negative error number or scsi
2253 * status on error
2254 *
2255 */
2256int
2257scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2258 unsigned char *buffer, int len, int timeout, int retries,
2259 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2260{
2261 unsigned char cmd[10];
2262 unsigned char *real_buffer;
2263 int ret;
2264
2265 memset(cmd, 0, sizeof(cmd));
2266 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2267
2268 if (sdev->use_10_for_ms) {
2269 if (len > 65535)
2270 return -EINVAL;
2271 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2272 if (!real_buffer)
2273 return -ENOMEM;
2274 memcpy(real_buffer + 8, buffer, len);
2275 len += 8;
2276 real_buffer[0] = 0;
2277 real_buffer[1] = 0;
2278 real_buffer[2] = data->medium_type;
2279 real_buffer[3] = data->device_specific;
2280 real_buffer[4] = data->longlba ? 0x01 : 0;
2281 real_buffer[5] = 0;
2282 real_buffer[6] = data->block_descriptor_length >> 8;
2283 real_buffer[7] = data->block_descriptor_length;
2284
2285 cmd[0] = MODE_SELECT_10;
2286 cmd[7] = len >> 8;
2287 cmd[8] = len;
2288 } else {
2289 if (len > 255 || data->block_descriptor_length > 255 ||
2290 data->longlba)
2291 return -EINVAL;
2292
2293 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2294 if (!real_buffer)
2295 return -ENOMEM;
2296 memcpy(real_buffer + 4, buffer, len);
2297 len += 4;
2298 real_buffer[0] = 0;
2299 real_buffer[1] = data->medium_type;
2300 real_buffer[2] = data->device_specific;
2301 real_buffer[3] = data->block_descriptor_length;
2302
2303
2304 cmd[0] = MODE_SELECT;
2305 cmd[4] = len;
2306 }
2307
2308 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2309 sshdr, timeout, retries, NULL);
2310 kfree(real_buffer);
2311 return ret;
2312}
2313EXPORT_SYMBOL_GPL(scsi_mode_select);
2314
2315/**
2316 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2317 * @sdev: SCSI device to be queried
2318 * @dbd: set if mode sense will allow block descriptors to be returned
2319 * @modepage: mode page being requested
2320 * @buffer: request buffer (may not be smaller than eight bytes)
2321 * @len: length of request buffer.
2322 * @timeout: command timeout
2323 * @retries: number of retries before failing
2324 * @data: returns a structure abstracting the mode header data
2325 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2326 * must be SCSI_SENSE_BUFFERSIZE big.
2327 *
2328 * Returns zero if unsuccessful, or the header offset (either 4
2329 * or 8 depending on whether a six or ten byte command was
2330 * issued) if successful.
2331 */
2332int
2333scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2334 unsigned char *buffer, int len, int timeout, int retries,
2335 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2336{
2337 unsigned char cmd[12];
2338 int use_10_for_ms;
2339 int header_length;
2340 int result, retry_count = retries;
2341 struct scsi_sense_hdr my_sshdr;
2342
2343 memset(data, 0, sizeof(*data));
2344 memset(&cmd[0], 0, 12);
2345 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2346 cmd[2] = modepage;
2347
2348 /* caller might not be interested in sense, but we need it */
2349 if (!sshdr)
2350 sshdr = &my_sshdr;
2351
2352 retry:
2353 use_10_for_ms = sdev->use_10_for_ms;
2354
2355 if (use_10_for_ms) {
2356 if (len < 8)
2357 len = 8;
2358
2359 cmd[0] = MODE_SENSE_10;
2360 cmd[8] = len;
2361 header_length = 8;
2362 } else {
2363 if (len < 4)
2364 len = 4;
2365
2366 cmd[0] = MODE_SENSE;
2367 cmd[4] = len;
2368 header_length = 4;
2369 }
2370
2371 memset(buffer, 0, len);
2372
2373 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2374 sshdr, timeout, retries, NULL);
2375
2376 /* This code looks awful: what it's doing is making sure an
2377 * ILLEGAL REQUEST sense return identifies the actual command
2378 * byte as the problem. MODE_SENSE commands can return
2379 * ILLEGAL REQUEST if the code page isn't supported */
2380
2381 if (use_10_for_ms && !scsi_status_is_good(result) &&
2382 (driver_byte(result) & DRIVER_SENSE)) {
2383 if (scsi_sense_valid(sshdr)) {
2384 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2385 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2386 /*
2387 * Invalid command operation code
2388 */
2389 sdev->use_10_for_ms = 0;
2390 goto retry;
2391 }
2392 }
2393 }
2394
2395 if(scsi_status_is_good(result)) {
2396 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2397 (modepage == 6 || modepage == 8))) {
2398 /* Initio breakage? */
2399 header_length = 0;
2400 data->length = 13;
2401 data->medium_type = 0;
2402 data->device_specific = 0;
2403 data->longlba = 0;
2404 data->block_descriptor_length = 0;
2405 } else if(use_10_for_ms) {
2406 data->length = buffer[0]*256 + buffer[1] + 2;
2407 data->medium_type = buffer[2];
2408 data->device_specific = buffer[3];
2409 data->longlba = buffer[4] & 0x01;
2410 data->block_descriptor_length = buffer[6]*256
2411 + buffer[7];
2412 } else {
2413 data->length = buffer[0] + 1;
2414 data->medium_type = buffer[1];
2415 data->device_specific = buffer[2];
2416 data->block_descriptor_length = buffer[3];
2417 }
2418 data->header_length = header_length;
2419 } else if ((status_byte(result) == CHECK_CONDITION) &&
2420 scsi_sense_valid(sshdr) &&
2421 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2422 retry_count--;
2423 goto retry;
2424 }
2425
2426 return result;
2427}
2428EXPORT_SYMBOL(scsi_mode_sense);
2429
2430/**
2431 * scsi_test_unit_ready - test if unit is ready
2432 * @sdev: scsi device to change the state of.
2433 * @timeout: command timeout
2434 * @retries: number of retries before failing
2435 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2436 * returning sense. Make sure that this is cleared before passing
2437 * in.
2438 *
2439 * Returns zero if unsuccessful or an error if TUR failed. For
2440 * removable media, UNIT_ATTENTION sets ->changed flag.
2441 **/
2442int
2443scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2444 struct scsi_sense_hdr *sshdr_external)
2445{
2446 char cmd[] = {
2447 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2448 };
2449 struct scsi_sense_hdr *sshdr;
2450 int result;
2451
2452 if (!sshdr_external)
2453 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2454 else
2455 sshdr = sshdr_external;
2456
2457 /* try to eat the UNIT_ATTENTION if there are enough retries */
2458 do {
2459 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2460 timeout, retries, NULL);
2461 if (sdev->removable && scsi_sense_valid(sshdr) &&
2462 sshdr->sense_key == UNIT_ATTENTION)
2463 sdev->changed = 1;
2464 } while (scsi_sense_valid(sshdr) &&
2465 sshdr->sense_key == UNIT_ATTENTION && --retries);
2466
2467 if (!sshdr_external)
2468 kfree(sshdr);
2469 return result;
2470}
2471EXPORT_SYMBOL(scsi_test_unit_ready);
2472
2473/**
2474 * scsi_device_set_state - Take the given device through the device state model.
2475 * @sdev: scsi device to change the state of.
2476 * @state: state to change to.
2477 *
2478 * Returns zero if unsuccessful or an error if the requested
2479 * transition is illegal.
2480 */
2481int
2482scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2483{
2484 enum scsi_device_state oldstate = sdev->sdev_state;
2485
2486 if (state == oldstate)
2487 return 0;
2488
2489 switch (state) {
2490 case SDEV_CREATED:
2491 switch (oldstate) {
2492 case SDEV_CREATED_BLOCK:
2493 break;
2494 default:
2495 goto illegal;
2496 }
2497 break;
2498
2499 case SDEV_RUNNING:
2500 switch (oldstate) {
2501 case SDEV_CREATED:
2502 case SDEV_OFFLINE:
2503 case SDEV_TRANSPORT_OFFLINE:
2504 case SDEV_QUIESCE:
2505 case SDEV_BLOCK:
2506 break;
2507 default:
2508 goto illegal;
2509 }
2510 break;
2511
2512 case SDEV_QUIESCE:
2513 switch (oldstate) {
2514 case SDEV_RUNNING:
2515 case SDEV_OFFLINE:
2516 case SDEV_TRANSPORT_OFFLINE:
2517 break;
2518 default:
2519 goto illegal;
2520 }
2521 break;
2522
2523 case SDEV_OFFLINE:
2524 case SDEV_TRANSPORT_OFFLINE:
2525 switch (oldstate) {
2526 case SDEV_CREATED:
2527 case SDEV_RUNNING:
2528 case SDEV_QUIESCE:
2529 case SDEV_BLOCK:
2530 break;
2531 default:
2532 goto illegal;
2533 }
2534 break;
2535
2536 case SDEV_BLOCK:
2537 switch (oldstate) {
2538 case SDEV_RUNNING:
2539 case SDEV_CREATED_BLOCK:
2540 break;
2541 default:
2542 goto illegal;
2543 }
2544 break;
2545
2546 case SDEV_CREATED_BLOCK:
2547 switch (oldstate) {
2548 case SDEV_CREATED:
2549 break;
2550 default:
2551 goto illegal;
2552 }
2553 break;
2554
2555 case SDEV_CANCEL:
2556 switch (oldstate) {
2557 case SDEV_CREATED:
2558 case SDEV_RUNNING:
2559 case SDEV_QUIESCE:
2560 case SDEV_OFFLINE:
2561 case SDEV_TRANSPORT_OFFLINE:
2562 case SDEV_BLOCK:
2563 break;
2564 default:
2565 goto illegal;
2566 }
2567 break;
2568
2569 case SDEV_DEL:
2570 switch (oldstate) {
2571 case SDEV_CREATED:
2572 case SDEV_RUNNING:
2573 case SDEV_OFFLINE:
2574 case SDEV_TRANSPORT_OFFLINE:
2575 case SDEV_CANCEL:
2576 case SDEV_CREATED_BLOCK:
2577 break;
2578 default:
2579 goto illegal;
2580 }
2581 break;
2582
2583 }
2584 sdev->sdev_state = state;
2585 return 0;
2586
2587 illegal:
2588 SCSI_LOG_ERROR_RECOVERY(1,
2589 sdev_printk(KERN_ERR, sdev,
2590 "Illegal state transition %s->%s",
2591 scsi_device_state_name(oldstate),
2592 scsi_device_state_name(state))
2593 );
2594 return -EINVAL;
2595}
2596EXPORT_SYMBOL(scsi_device_set_state);
2597
2598/**
2599 * sdev_evt_emit - emit a single SCSI device uevent
2600 * @sdev: associated SCSI device
2601 * @evt: event to emit
2602 *
2603 * Send a single uevent (scsi_event) to the associated scsi_device.
2604 */
2605static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2606{
2607 int idx = 0;
2608 char *envp[3];
2609
2610 switch (evt->evt_type) {
2611 case SDEV_EVT_MEDIA_CHANGE:
2612 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2613 break;
2614 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2615 scsi_rescan_device(&sdev->sdev_gendev);
2616 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2617 break;
2618 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2619 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2620 break;
2621 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2622 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2623 break;
2624 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2625 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2626 break;
2627 case SDEV_EVT_LUN_CHANGE_REPORTED:
2628 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2629 break;
2630 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2631 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2632 break;
2633 default:
2634 /* do nothing */
2635 break;
2636 }
2637
2638 envp[idx++] = NULL;
2639
2640 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2641}
2642
2643/**
2644 * sdev_evt_thread - send a uevent for each scsi event
2645 * @work: work struct for scsi_device
2646 *
2647 * Dispatch queued events to their associated scsi_device kobjects
2648 * as uevents.
2649 */
2650void scsi_evt_thread(struct work_struct *work)
2651{
2652 struct scsi_device *sdev;
2653 enum scsi_device_event evt_type;
2654 LIST_HEAD(event_list);
2655
2656 sdev = container_of(work, struct scsi_device, event_work);
2657
2658 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2659 if (test_and_clear_bit(evt_type, sdev->pending_events))
2660 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2661
2662 while (1) {
2663 struct scsi_event *evt;
2664 struct list_head *this, *tmp;
2665 unsigned long flags;
2666
2667 spin_lock_irqsave(&sdev->list_lock, flags);
2668 list_splice_init(&sdev->event_list, &event_list);
2669 spin_unlock_irqrestore(&sdev->list_lock, flags);
2670
2671 if (list_empty(&event_list))
2672 break;
2673
2674 list_for_each_safe(this, tmp, &event_list) {
2675 evt = list_entry(this, struct scsi_event, node);
2676 list_del(&evt->node);
2677 scsi_evt_emit(sdev, evt);
2678 kfree(evt);
2679 }
2680 }
2681}
2682
2683/**
2684 * sdev_evt_send - send asserted event to uevent thread
2685 * @sdev: scsi_device event occurred on
2686 * @evt: event to send
2687 *
2688 * Assert scsi device event asynchronously.
2689 */
2690void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2691{
2692 unsigned long flags;
2693
2694#if 0
2695 /* FIXME: currently this check eliminates all media change events
2696 * for polled devices. Need to update to discriminate between AN
2697 * and polled events */
2698 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2699 kfree(evt);
2700 return;
2701 }
2702#endif
2703
2704 spin_lock_irqsave(&sdev->list_lock, flags);
2705 list_add_tail(&evt->node, &sdev->event_list);
2706 schedule_work(&sdev->event_work);
2707 spin_unlock_irqrestore(&sdev->list_lock, flags);
2708}
2709EXPORT_SYMBOL_GPL(sdev_evt_send);
2710
2711/**
2712 * sdev_evt_alloc - allocate a new scsi event
2713 * @evt_type: type of event to allocate
2714 * @gfpflags: GFP flags for allocation
2715 *
2716 * Allocates and returns a new scsi_event.
2717 */
2718struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2719 gfp_t gfpflags)
2720{
2721 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2722 if (!evt)
2723 return NULL;
2724
2725 evt->evt_type = evt_type;
2726 INIT_LIST_HEAD(&evt->node);
2727
2728 /* evt_type-specific initialization, if any */
2729 switch (evt_type) {
2730 case SDEV_EVT_MEDIA_CHANGE:
2731 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2732 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2733 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2734 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2735 case SDEV_EVT_LUN_CHANGE_REPORTED:
2736 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2737 default:
2738 /* do nothing */
2739 break;
2740 }
2741
2742 return evt;
2743}
2744EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2745
2746/**
2747 * sdev_evt_send_simple - send asserted event to uevent thread
2748 * @sdev: scsi_device event occurred on
2749 * @evt_type: type of event to send
2750 * @gfpflags: GFP flags for allocation
2751 *
2752 * Assert scsi device event asynchronously, given an event type.
2753 */
2754void sdev_evt_send_simple(struct scsi_device *sdev,
2755 enum scsi_device_event evt_type, gfp_t gfpflags)
2756{
2757 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2758 if (!evt) {
2759 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2760 evt_type);
2761 return;
2762 }
2763
2764 sdev_evt_send(sdev, evt);
2765}
2766EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2767
2768/**
2769 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2770 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2771 */
2772static int scsi_request_fn_active(struct scsi_device *sdev)
2773{
2774 struct request_queue *q = sdev->request_queue;
2775 int request_fn_active;
2776
2777 WARN_ON_ONCE(sdev->host->use_blk_mq);
2778
2779 spin_lock_irq(q->queue_lock);
2780 request_fn_active = q->request_fn_active;
2781 spin_unlock_irq(q->queue_lock);
2782
2783 return request_fn_active;
2784}
2785
2786/**
2787 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2788 * @sdev: SCSI device pointer.
2789 *
2790 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2791 * invoked from scsi_request_fn() have finished.
2792 */
2793static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2794{
2795 WARN_ON_ONCE(sdev->host->use_blk_mq);
2796
2797 while (scsi_request_fn_active(sdev))
2798 msleep(20);
2799}
2800
2801/**
2802 * scsi_device_quiesce - Block user issued commands.
2803 * @sdev: scsi device to quiesce.
2804 *
2805 * This works by trying to transition to the SDEV_QUIESCE state
2806 * (which must be a legal transition). When the device is in this
2807 * state, only special requests will be accepted, all others will
2808 * be deferred. Since special requests may also be requeued requests,
2809 * a successful return doesn't guarantee the device will be
2810 * totally quiescent.
2811 *
2812 * Must be called with user context, may sleep.
2813 *
2814 * Returns zero if unsuccessful or an error if not.
2815 */
2816int
2817scsi_device_quiesce(struct scsi_device *sdev)
2818{
2819 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2820 if (err)
2821 return err;
2822
2823 scsi_run_queue(sdev->request_queue);
2824 while (atomic_read(&sdev->device_busy)) {
2825 msleep_interruptible(200);
2826 scsi_run_queue(sdev->request_queue);
2827 }
2828 return 0;
2829}
2830EXPORT_SYMBOL(scsi_device_quiesce);
2831
2832/**
2833 * scsi_device_resume - Restart user issued commands to a quiesced device.
2834 * @sdev: scsi device to resume.
2835 *
2836 * Moves the device from quiesced back to running and restarts the
2837 * queues.
2838 *
2839 * Must be called with user context, may sleep.
2840 */
2841void scsi_device_resume(struct scsi_device *sdev)
2842{
2843 /* check if the device state was mutated prior to resume, and if
2844 * so assume the state is being managed elsewhere (for example
2845 * device deleted during suspend)
2846 */
2847 if (sdev->sdev_state != SDEV_QUIESCE ||
2848 scsi_device_set_state(sdev, SDEV_RUNNING))
2849 return;
2850 scsi_run_queue(sdev->request_queue);
2851}
2852EXPORT_SYMBOL(scsi_device_resume);
2853
2854static void
2855device_quiesce_fn(struct scsi_device *sdev, void *data)
2856{
2857 scsi_device_quiesce(sdev);
2858}
2859
2860void
2861scsi_target_quiesce(struct scsi_target *starget)
2862{
2863 starget_for_each_device(starget, NULL, device_quiesce_fn);
2864}
2865EXPORT_SYMBOL(scsi_target_quiesce);
2866
2867static void
2868device_resume_fn(struct scsi_device *sdev, void *data)
2869{
2870 scsi_device_resume(sdev);
2871}
2872
2873void
2874scsi_target_resume(struct scsi_target *starget)
2875{
2876 starget_for_each_device(starget, NULL, device_resume_fn);
2877}
2878EXPORT_SYMBOL(scsi_target_resume);
2879
2880/**
2881 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2882 * @sdev: device to block
2883 * @wait: Whether or not to wait until ongoing .queuecommand() /
2884 * .queue_rq() calls have finished.
2885 *
2886 * Block request made by scsi lld's to temporarily stop all
2887 * scsi commands on the specified device. May sleep.
2888 *
2889 * Returns zero if successful or error if not
2890 *
2891 * Notes:
2892 * This routine transitions the device to the SDEV_BLOCK state
2893 * (which must be a legal transition). When the device is in this
2894 * state, all commands are deferred until the scsi lld reenables
2895 * the device with scsi_device_unblock or device_block_tmo fires.
2896 *
2897 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2898 * scsi_internal_device_block() has blocked a SCSI device and also
2899 * remove the rport mutex lock and unlock calls from srp_queuecommand().
2900 */
2901int
2902scsi_internal_device_block(struct scsi_device *sdev, bool wait)
2903{
2904 struct request_queue *q = sdev->request_queue;
2905 unsigned long flags;
2906 int err = 0;
2907
2908 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2909 if (err) {
2910 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2911
2912 if (err)
2913 return err;
2914 }
2915
2916 /*
2917 * The device has transitioned to SDEV_BLOCK. Stop the
2918 * block layer from calling the midlayer with this device's
2919 * request queue.
2920 */
2921 if (q->mq_ops) {
2922 if (wait)
2923 blk_mq_quiesce_queue(q);
2924 else
2925 blk_mq_stop_hw_queues(q);
2926 } else {
2927 spin_lock_irqsave(q->queue_lock, flags);
2928 blk_stop_queue(q);
2929 spin_unlock_irqrestore(q->queue_lock, flags);
2930 if (wait)
2931 scsi_wait_for_queuecommand(sdev);
2932 }
2933
2934 return 0;
2935}
2936EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2937
2938/**
2939 * scsi_internal_device_unblock - resume a device after a block request
2940 * @sdev: device to resume
2941 * @new_state: state to set devices to after unblocking
2942 *
2943 * Called by scsi lld's or the midlayer to restart the device queue
2944 * for the previously suspended scsi device. Called from interrupt or
2945 * normal process context.
2946 *
2947 * Returns zero if successful or error if not.
2948 *
2949 * Notes:
2950 * This routine transitions the device to the SDEV_RUNNING state
2951 * or to one of the offline states (which must be a legal transition)
2952 * allowing the midlayer to goose the queue for this device.
2953 */
2954int
2955scsi_internal_device_unblock(struct scsi_device *sdev,
2956 enum scsi_device_state new_state)
2957{
2958 struct request_queue *q = sdev->request_queue;
2959 unsigned long flags;
2960
2961 /*
2962 * Try to transition the scsi device to SDEV_RUNNING or one of the
2963 * offlined states and goose the device queue if successful.
2964 */
2965 if ((sdev->sdev_state == SDEV_BLOCK) ||
2966 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2967 sdev->sdev_state = new_state;
2968 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2969 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2970 new_state == SDEV_OFFLINE)
2971 sdev->sdev_state = new_state;
2972 else
2973 sdev->sdev_state = SDEV_CREATED;
2974 } else if (sdev->sdev_state != SDEV_CANCEL &&
2975 sdev->sdev_state != SDEV_OFFLINE)
2976 return -EINVAL;
2977
2978 if (q->mq_ops) {
2979 blk_mq_start_stopped_hw_queues(q, false);
2980 } else {
2981 spin_lock_irqsave(q->queue_lock, flags);
2982 blk_start_queue(q);
2983 spin_unlock_irqrestore(q->queue_lock, flags);
2984 }
2985
2986 return 0;
2987}
2988EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2989
2990static void
2991device_block(struct scsi_device *sdev, void *data)
2992{
2993 scsi_internal_device_block(sdev, true);
2994}
2995
2996static int
2997target_block(struct device *dev, void *data)
2998{
2999 if (scsi_is_target_device(dev))
3000 starget_for_each_device(to_scsi_target(dev), NULL,
3001 device_block);
3002 return 0;
3003}
3004
3005void
3006scsi_target_block(struct device *dev)
3007{
3008 if (scsi_is_target_device(dev))
3009 starget_for_each_device(to_scsi_target(dev), NULL,
3010 device_block);
3011 else
3012 device_for_each_child(dev, NULL, target_block);
3013}
3014EXPORT_SYMBOL_GPL(scsi_target_block);
3015
3016static void
3017device_unblock(struct scsi_device *sdev, void *data)
3018{
3019 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3020}
3021
3022static int
3023target_unblock(struct device *dev, void *data)
3024{
3025 if (scsi_is_target_device(dev))
3026 starget_for_each_device(to_scsi_target(dev), data,
3027 device_unblock);
3028 return 0;
3029}
3030
3031void
3032scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3033{
3034 if (scsi_is_target_device(dev))
3035 starget_for_each_device(to_scsi_target(dev), &new_state,
3036 device_unblock);
3037 else
3038 device_for_each_child(dev, &new_state, target_unblock);
3039}
3040EXPORT_SYMBOL_GPL(scsi_target_unblock);
3041
3042/**
3043 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3044 * @sgl: scatter-gather list
3045 * @sg_count: number of segments in sg
3046 * @offset: offset in bytes into sg, on return offset into the mapped area
3047 * @len: bytes to map, on return number of bytes mapped
3048 *
3049 * Returns virtual address of the start of the mapped page
3050 */
3051void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3052 size_t *offset, size_t *len)
3053{
3054 int i;
3055 size_t sg_len = 0, len_complete = 0;
3056 struct scatterlist *sg;
3057 struct page *page;
3058
3059 WARN_ON(!irqs_disabled());
3060
3061 for_each_sg(sgl, sg, sg_count, i) {
3062 len_complete = sg_len; /* Complete sg-entries */
3063 sg_len += sg->length;
3064 if (sg_len > *offset)
3065 break;
3066 }
3067
3068 if (unlikely(i == sg_count)) {
3069 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3070 "elements %d\n",
3071 __func__, sg_len, *offset, sg_count);
3072 WARN_ON(1);
3073 return NULL;
3074 }
3075
3076 /* Offset starting from the beginning of first page in this sg-entry */
3077 *offset = *offset - len_complete + sg->offset;
3078
3079 /* Assumption: contiguous pages can be accessed as "page + i" */
3080 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3081 *offset &= ~PAGE_MASK;
3082
3083 /* Bytes in this sg-entry from *offset to the end of the page */
3084 sg_len = PAGE_SIZE - *offset;
3085 if (*len > sg_len)
3086 *len = sg_len;
3087
3088 return kmap_atomic(page);
3089}
3090EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3091
3092/**
3093 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3094 * @virt: virtual address to be unmapped
3095 */
3096void scsi_kunmap_atomic_sg(void *virt)
3097{
3098 kunmap_atomic(virt);
3099}
3100EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3101
3102void sdev_disable_disk_events(struct scsi_device *sdev)
3103{
3104 atomic_inc(&sdev->disk_events_disable_depth);
3105}
3106EXPORT_SYMBOL(sdev_disable_disk_events);
3107
3108void sdev_enable_disk_events(struct scsi_device *sdev)
3109{
3110 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3111 return;
3112 atomic_dec(&sdev->disk_events_disable_depth);
3113}
3114EXPORT_SYMBOL(sdev_enable_disk_events);
3115
3116/**
3117 * scsi_vpd_lun_id - return a unique device identification
3118 * @sdev: SCSI device
3119 * @id: buffer for the identification
3120 * @id_len: length of the buffer
3121 *
3122 * Copies a unique device identification into @id based
3123 * on the information in the VPD page 0x83 of the device.
3124 * The string will be formatted as a SCSI name string.
3125 *
3126 * Returns the length of the identification or error on failure.
3127 * If the identifier is longer than the supplied buffer the actual
3128 * identifier length is returned and the buffer is not zero-padded.
3129 */
3130int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3131{
3132 u8 cur_id_type = 0xff;
3133 u8 cur_id_size = 0;
3134 unsigned char *d, *cur_id_str;
3135 unsigned char __rcu *vpd_pg83;
3136 int id_size = -EINVAL;
3137
3138 rcu_read_lock();
3139 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3140 if (!vpd_pg83) {
3141 rcu_read_unlock();
3142 return -ENXIO;
3143 }
3144
3145 /*
3146 * Look for the correct descriptor.
3147 * Order of preference for lun descriptor:
3148 * - SCSI name string
3149 * - NAA IEEE Registered Extended
3150 * - EUI-64 based 16-byte
3151 * - EUI-64 based 12-byte
3152 * - NAA IEEE Registered
3153 * - NAA IEEE Extended
3154 * - T10 Vendor ID
3155 * as longer descriptors reduce the likelyhood
3156 * of identification clashes.
3157 */
3158
3159 /* The id string must be at least 20 bytes + terminating NULL byte */
3160 if (id_len < 21) {
3161 rcu_read_unlock();
3162 return -EINVAL;
3163 }
3164
3165 memset(id, 0, id_len);
3166 d = vpd_pg83 + 4;
3167 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3168 /* Skip designators not referring to the LUN */
3169 if ((d[1] & 0x30) != 0x00)
3170 goto next_desig;
3171
3172 switch (d[1] & 0xf) {
3173 case 0x1:
3174 /* T10 Vendor ID */
3175 if (cur_id_size > d[3])
3176 break;
3177 /* Prefer anything */
3178 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3179 break;
3180 cur_id_size = d[3];
3181 if (cur_id_size + 4 > id_len)
3182 cur_id_size = id_len - 4;
3183 cur_id_str = d + 4;
3184 cur_id_type = d[1] & 0xf;
3185 id_size = snprintf(id, id_len, "t10.%*pE",
3186 cur_id_size, cur_id_str);
3187 break;
3188 case 0x2:
3189 /* EUI-64 */
3190 if (cur_id_size > d[3])
3191 break;
3192 /* Prefer NAA IEEE Registered Extended */
3193 if (cur_id_type == 0x3 &&
3194 cur_id_size == d[3])
3195 break;
3196 cur_id_size = d[3];
3197 cur_id_str = d + 4;
3198 cur_id_type = d[1] & 0xf;
3199 switch (cur_id_size) {
3200 case 8:
3201 id_size = snprintf(id, id_len,
3202 "eui.%8phN",
3203 cur_id_str);
3204 break;
3205 case 12:
3206 id_size = snprintf(id, id_len,
3207 "eui.%12phN",
3208 cur_id_str);
3209 break;
3210 case 16:
3211 id_size = snprintf(id, id_len,
3212 "eui.%16phN",
3213 cur_id_str);
3214 break;
3215 default:
3216 cur_id_size = 0;
3217 break;
3218 }
3219 break;
3220 case 0x3:
3221 /* NAA */
3222 if (cur_id_size > d[3])
3223 break;
3224 cur_id_size = d[3];
3225 cur_id_str = d + 4;
3226 cur_id_type = d[1] & 0xf;
3227 switch (cur_id_size) {
3228 case 8:
3229 id_size = snprintf(id, id_len,
3230 "naa.%8phN",
3231 cur_id_str);
3232 break;
3233 case 16:
3234 id_size = snprintf(id, id_len,
3235 "naa.%16phN",
3236 cur_id_str);
3237 break;
3238 default:
3239 cur_id_size = 0;
3240 break;
3241 }
3242 break;
3243 case 0x8:
3244 /* SCSI name string */
3245 if (cur_id_size + 4 > d[3])
3246 break;
3247 /* Prefer others for truncated descriptor */
3248 if (cur_id_size && d[3] > id_len)
3249 break;
3250 cur_id_size = id_size = d[3];
3251 cur_id_str = d + 4;
3252 cur_id_type = d[1] & 0xf;
3253 if (cur_id_size >= id_len)
3254 cur_id_size = id_len - 1;
3255 memcpy(id, cur_id_str, cur_id_size);
3256 /* Decrease priority for truncated descriptor */
3257 if (cur_id_size != id_size)
3258 cur_id_size = 6;
3259 break;
3260 default:
3261 break;
3262 }
3263next_desig:
3264 d += d[3] + 4;
3265 }
3266 rcu_read_unlock();
3267
3268 return id_size;
3269}
3270EXPORT_SYMBOL(scsi_vpd_lun_id);
3271
3272/*
3273 * scsi_vpd_tpg_id - return a target port group identifier
3274 * @sdev: SCSI device
3275 *
3276 * Returns the Target Port Group identifier from the information
3277 * froom VPD page 0x83 of the device.
3278 *
3279 * Returns the identifier or error on failure.
3280 */
3281int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3282{
3283 unsigned char *d;
3284 unsigned char __rcu *vpd_pg83;
3285 int group_id = -EAGAIN, rel_port = -1;
3286
3287 rcu_read_lock();
3288 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3289 if (!vpd_pg83) {
3290 rcu_read_unlock();
3291 return -ENXIO;
3292 }
3293
3294 d = sdev->vpd_pg83 + 4;
3295 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3296 switch (d[1] & 0xf) {
3297 case 0x4:
3298 /* Relative target port */
3299 rel_port = get_unaligned_be16(&d[6]);
3300 break;
3301 case 0x5:
3302 /* Target port group */
3303 group_id = get_unaligned_be16(&d[6]);
3304 break;
3305 default:
3306 break;
3307 }
3308 d += d[3] + 4;
3309 }
3310 rcu_read_unlock();
3311
3312 if (group_id >= 0 && rel_id && rel_port != -1)
3313 *rel_id = rel_port;
3314
3315 return group_id;
3316}
3317EXPORT_SYMBOL(scsi_vpd_tpg_id);
1/*
2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
3 *
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.
8 */
9
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>
23
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>
31
32#include "scsi_priv.h"
33#include "scsi_logging.h"
34
35
36#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
37#define SG_MEMPOOL_SIZE 2
38
39struct scsi_host_sg_pool {
40 size_t size;
41 char *name;
42 struct kmem_cache *slab;
43 mempool_t *pool;
44};
45
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)
49#endif
50static struct scsi_host_sg_pool scsi_sg_pools[] = {
51 SP(8),
52 SP(16),
53#if (SCSI_MAX_SG_SEGMENTS > 32)
54 SP(32),
55#if (SCSI_MAX_SG_SEGMENTS > 64)
56 SP(64),
57#if (SCSI_MAX_SG_SEGMENTS > 128)
58 SP(128),
59#if (SCSI_MAX_SG_SEGMENTS > 256)
60#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
61#endif
62#endif
63#endif
64#endif
65 SP(SCSI_MAX_SG_SEGMENTS)
66};
67#undef SP
68
69struct kmem_cache *scsi_sdb_cache;
70
71/*
72 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
73 * not change behaviour from the previous unplug mechanism, experimentation
74 * may prove this needs changing.
75 */
76#define SCSI_QUEUE_DELAY 3
77
78/*
79 * Function: scsi_unprep_request()
80 *
81 * Purpose: Remove all preparation done for a request, including its
82 * associated scsi_cmnd, so that it can be requeued.
83 *
84 * Arguments: req - request to unprepare
85 *
86 * Lock status: Assumed that no locks are held upon entry.
87 *
88 * Returns: Nothing.
89 */
90static void scsi_unprep_request(struct request *req)
91{
92 struct scsi_cmnd *cmd = req->special;
93
94 blk_unprep_request(req);
95 req->special = NULL;
96
97 scsi_put_command(cmd);
98}
99
100/**
101 * __scsi_queue_insert - private queue insertion
102 * @cmd: The SCSI command being requeued
103 * @reason: The reason for the requeue
104 * @unbusy: Whether the queue should be unbusied
105 *
106 * This is a private queue insertion. The public interface
107 * scsi_queue_insert() always assumes the queue should be unbusied
108 * because it's always called before the completion. This function is
109 * for a requeue after completion, which should only occur in this
110 * file.
111 */
112static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
113{
114 struct Scsi_Host *host = cmd->device->host;
115 struct scsi_device *device = cmd->device;
116 struct scsi_target *starget = scsi_target(device);
117 struct request_queue *q = device->request_queue;
118 unsigned long flags;
119
120 SCSI_LOG_MLQUEUE(1,
121 printk("Inserting command %p into mlqueue\n", cmd));
122
123 /*
124 * Set the appropriate busy bit for the device/host.
125 *
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
128 *
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
135 */
136 switch (reason) {
137 case SCSI_MLQUEUE_HOST_BUSY:
138 host->host_blocked = host->max_host_blocked;
139 break;
140 case SCSI_MLQUEUE_DEVICE_BUSY:
141 case SCSI_MLQUEUE_EH_RETRY:
142 device->device_blocked = device->max_device_blocked;
143 break;
144 case SCSI_MLQUEUE_TARGET_BUSY:
145 starget->target_blocked = starget->max_target_blocked;
146 break;
147 }
148
149 /*
150 * Decrement the counters, since these commands are no longer
151 * active on the host/device.
152 */
153 if (unbusy)
154 scsi_device_unbusy(device);
155
156 /*
157 * Requeue this command. It will go before all other commands
158 * that are already in the queue.
159 */
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
163
164 kblockd_schedule_work(q, &device->requeue_work);
165
166 return 0;
167}
168
169/*
170 * Function: scsi_queue_insert()
171 *
172 * Purpose: Insert a command in the midlevel queue.
173 *
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
176 *
177 * Lock status: Assumed that lock is not held upon entry.
178 *
179 * Returns: Nothing.
180 *
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
184 * commands.
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
187 */
188int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
189{
190 return __scsi_queue_insert(cmd, reason, 1);
191}
192/**
193 * scsi_execute - insert request and wait for the result
194 * @sdev: scsi device
195 * @cmd: scsi command
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
204 *
205 * returns the req->errors value which is the scsi_cmnd result
206 * field.
207 */
208int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209 int data_direction, void *buffer, unsigned bufflen,
210 unsigned char *sense, int timeout, int retries, int flags,
211 int *resid)
212{
213 struct request *req;
214 int write = (data_direction == DMA_TO_DEVICE);
215 int ret = DRIVER_ERROR << 24;
216
217 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
218 if (!req)
219 return ret;
220
221 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
222 buffer, bufflen, __GFP_WAIT))
223 goto out;
224
225 req->cmd_len = COMMAND_SIZE(cmd[0]);
226 memcpy(req->cmd, cmd, req->cmd_len);
227 req->sense = sense;
228 req->sense_len = 0;
229 req->retries = retries;
230 req->timeout = timeout;
231 req->cmd_type = REQ_TYPE_BLOCK_PC;
232 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
233
234 /*
235 * head injection *required* here otherwise quiesce won't work
236 */
237 blk_execute_rq(req->q, NULL, req, 1);
238
239 /*
240 * Some devices (USB mass-storage in particular) may transfer
241 * garbage data together with a residue indicating that the data
242 * is invalid. Prevent the garbage from being misinterpreted
243 * and prevent security leaks by zeroing out the excess data.
244 */
245 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
246 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
247
248 if (resid)
249 *resid = req->resid_len;
250 ret = req->errors;
251 out:
252 blk_put_request(req);
253
254 return ret;
255}
256EXPORT_SYMBOL(scsi_execute);
257
258
259int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
260 int data_direction, void *buffer, unsigned bufflen,
261 struct scsi_sense_hdr *sshdr, int timeout, int retries,
262 int *resid)
263{
264 char *sense = NULL;
265 int result;
266
267 if (sshdr) {
268 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
269 if (!sense)
270 return DRIVER_ERROR << 24;
271 }
272 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
273 sense, timeout, retries, 0, resid);
274 if (sshdr)
275 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
276
277 kfree(sense);
278 return result;
279}
280EXPORT_SYMBOL(scsi_execute_req);
281
282/*
283 * Function: scsi_init_cmd_errh()
284 *
285 * Purpose: Initialize cmd fields related to error handling.
286 *
287 * Arguments: cmd - command that is ready to be queued.
288 *
289 * Notes: This function has the job of initializing a number of
290 * fields related to error handling. Typically this will
291 * be called once for each command, as required.
292 */
293static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
294{
295 cmd->serial_number = 0;
296 scsi_set_resid(cmd, 0);
297 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
298 if (cmd->cmd_len == 0)
299 cmd->cmd_len = scsi_command_size(cmd->cmnd);
300}
301
302void scsi_device_unbusy(struct scsi_device *sdev)
303{
304 struct Scsi_Host *shost = sdev->host;
305 struct scsi_target *starget = scsi_target(sdev);
306 unsigned long flags;
307
308 spin_lock_irqsave(shost->host_lock, flags);
309 shost->host_busy--;
310 starget->target_busy--;
311 if (unlikely(scsi_host_in_recovery(shost) &&
312 (shost->host_failed || shost->host_eh_scheduled)))
313 scsi_eh_wakeup(shost);
314 spin_unlock(shost->host_lock);
315 spin_lock(sdev->request_queue->queue_lock);
316 sdev->device_busy--;
317 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
318}
319
320/*
321 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
322 * and call blk_run_queue for all the scsi_devices on the target -
323 * including current_sdev first.
324 *
325 * Called with *no* scsi locks held.
326 */
327static void scsi_single_lun_run(struct scsi_device *current_sdev)
328{
329 struct Scsi_Host *shost = current_sdev->host;
330 struct scsi_device *sdev, *tmp;
331 struct scsi_target *starget = scsi_target(current_sdev);
332 unsigned long flags;
333
334 spin_lock_irqsave(shost->host_lock, flags);
335 starget->starget_sdev_user = NULL;
336 spin_unlock_irqrestore(shost->host_lock, flags);
337
338 /*
339 * Call blk_run_queue for all LUNs on the target, starting with
340 * current_sdev. We race with others (to set starget_sdev_user),
341 * but in most cases, we will be first. Ideally, each LU on the
342 * target would get some limited time or requests on the target.
343 */
344 blk_run_queue(current_sdev->request_queue);
345
346 spin_lock_irqsave(shost->host_lock, flags);
347 if (starget->starget_sdev_user)
348 goto out;
349 list_for_each_entry_safe(sdev, tmp, &starget->devices,
350 same_target_siblings) {
351 if (sdev == current_sdev)
352 continue;
353 if (scsi_device_get(sdev))
354 continue;
355
356 spin_unlock_irqrestore(shost->host_lock, flags);
357 blk_run_queue(sdev->request_queue);
358 spin_lock_irqsave(shost->host_lock, flags);
359
360 scsi_device_put(sdev);
361 }
362 out:
363 spin_unlock_irqrestore(shost->host_lock, flags);
364}
365
366static inline int scsi_device_is_busy(struct scsi_device *sdev)
367{
368 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
369 return 1;
370
371 return 0;
372}
373
374static inline int scsi_target_is_busy(struct scsi_target *starget)
375{
376 return ((starget->can_queue > 0 &&
377 starget->target_busy >= starget->can_queue) ||
378 starget->target_blocked);
379}
380
381static inline int scsi_host_is_busy(struct Scsi_Host *shost)
382{
383 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
384 shost->host_blocked || shost->host_self_blocked)
385 return 1;
386
387 return 0;
388}
389
390/*
391 * Function: scsi_run_queue()
392 *
393 * Purpose: Select a proper request queue to serve next
394 *
395 * Arguments: q - last request's queue
396 *
397 * Returns: Nothing
398 *
399 * Notes: The previous command was completely finished, start
400 * a new one if possible.
401 */
402static void scsi_run_queue(struct request_queue *q)
403{
404 struct scsi_device *sdev = q->queuedata;
405 struct Scsi_Host *shost;
406 LIST_HEAD(starved_list);
407 unsigned long flags;
408
409 shost = sdev->host;
410 if (scsi_target(sdev)->single_lun)
411 scsi_single_lun_run(sdev);
412
413 spin_lock_irqsave(shost->host_lock, flags);
414 list_splice_init(&shost->starved_list, &starved_list);
415
416 while (!list_empty(&starved_list)) {
417 /*
418 * As long as shost is accepting commands and we have
419 * starved queues, call blk_run_queue. scsi_request_fn
420 * drops the queue_lock and can add us back to the
421 * starved_list.
422 *
423 * host_lock protects the starved_list and starved_entry.
424 * scsi_request_fn must get the host_lock before checking
425 * or modifying starved_list or starved_entry.
426 */
427 if (scsi_host_is_busy(shost))
428 break;
429
430 sdev = list_entry(starved_list.next,
431 struct scsi_device, starved_entry);
432 list_del_init(&sdev->starved_entry);
433 if (scsi_target_is_busy(scsi_target(sdev))) {
434 list_move_tail(&sdev->starved_entry,
435 &shost->starved_list);
436 continue;
437 }
438
439 spin_unlock(shost->host_lock);
440 spin_lock(sdev->request_queue->queue_lock);
441 __blk_run_queue(sdev->request_queue);
442 spin_unlock(sdev->request_queue->queue_lock);
443 spin_lock(shost->host_lock);
444 }
445 /* put any unprocessed entries back */
446 list_splice(&starved_list, &shost->starved_list);
447 spin_unlock_irqrestore(shost->host_lock, flags);
448
449 blk_run_queue(q);
450}
451
452void scsi_requeue_run_queue(struct work_struct *work)
453{
454 struct scsi_device *sdev;
455 struct request_queue *q;
456
457 sdev = container_of(work, struct scsi_device, requeue_work);
458 q = sdev->request_queue;
459 scsi_run_queue(q);
460}
461
462/*
463 * Function: scsi_requeue_command()
464 *
465 * Purpose: Handle post-processing of completed commands.
466 *
467 * Arguments: q - queue to operate on
468 * cmd - command that may need to be requeued.
469 *
470 * Returns: Nothing
471 *
472 * Notes: After command completion, there may be blocks left
473 * over which weren't finished by the previous command
474 * this can be for a number of reasons - the main one is
475 * I/O errors in the middle of the request, in which case
476 * we need to request the blocks that come after the bad
477 * sector.
478 * Notes: Upon return, cmd is a stale pointer.
479 */
480static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
481{
482 struct scsi_device *sdev = cmd->device;
483 struct request *req = cmd->request;
484 unsigned long flags;
485
486 /*
487 * We need to hold a reference on the device to avoid the queue being
488 * killed after the unlock and before scsi_run_queue is invoked which
489 * may happen because scsi_unprep_request() puts the command which
490 * releases its reference on the device.
491 */
492 get_device(&sdev->sdev_gendev);
493
494 spin_lock_irqsave(q->queue_lock, flags);
495 scsi_unprep_request(req);
496 blk_requeue_request(q, req);
497 spin_unlock_irqrestore(q->queue_lock, flags);
498
499 scsi_run_queue(q);
500
501 put_device(&sdev->sdev_gendev);
502}
503
504void scsi_next_command(struct scsi_cmnd *cmd)
505{
506 struct scsi_device *sdev = cmd->device;
507 struct request_queue *q = sdev->request_queue;
508
509 /* need to hold a reference on the device before we let go of the cmd */
510 get_device(&sdev->sdev_gendev);
511
512 scsi_put_command(cmd);
513 scsi_run_queue(q);
514
515 /* ok to remove device now */
516 put_device(&sdev->sdev_gendev);
517}
518
519void scsi_run_host_queues(struct Scsi_Host *shost)
520{
521 struct scsi_device *sdev;
522
523 shost_for_each_device(sdev, shost)
524 scsi_run_queue(sdev->request_queue);
525}
526
527static void __scsi_release_buffers(struct scsi_cmnd *, int);
528
529/*
530 * Function: scsi_end_request()
531 *
532 * Purpose: Post-processing of completed commands (usually invoked at end
533 * of upper level post-processing and scsi_io_completion).
534 *
535 * Arguments: cmd - command that is complete.
536 * error - 0 if I/O indicates success, < 0 for I/O error.
537 * bytes - number of bytes of completed I/O
538 * requeue - indicates whether we should requeue leftovers.
539 *
540 * Lock status: Assumed that lock is not held upon entry.
541 *
542 * Returns: cmd if requeue required, NULL otherwise.
543 *
544 * Notes: This is called for block device requests in order to
545 * mark some number of sectors as complete.
546 *
547 * We are guaranteeing that the request queue will be goosed
548 * at some point during this call.
549 * Notes: If cmd was requeued, upon return it will be a stale pointer.
550 */
551static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
552 int bytes, int requeue)
553{
554 struct request_queue *q = cmd->device->request_queue;
555 struct request *req = cmd->request;
556
557 /*
558 * If there are blocks left over at the end, set up the command
559 * to queue the remainder of them.
560 */
561 if (blk_end_request(req, error, bytes)) {
562 /* kill remainder if no retrys */
563 if (error && scsi_noretry_cmd(cmd))
564 blk_end_request_all(req, error);
565 else {
566 if (requeue) {
567 /*
568 * Bleah. Leftovers again. Stick the
569 * leftovers in the front of the
570 * queue, and goose the queue again.
571 */
572 scsi_release_buffers(cmd);
573 scsi_requeue_command(q, cmd);
574 cmd = NULL;
575 }
576 return cmd;
577 }
578 }
579
580 /*
581 * This will goose the queue request function at the end, so we don't
582 * need to worry about launching another command.
583 */
584 __scsi_release_buffers(cmd, 0);
585 scsi_next_command(cmd);
586 return NULL;
587}
588
589static inline unsigned int scsi_sgtable_index(unsigned short nents)
590{
591 unsigned int index;
592
593 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
594
595 if (nents <= 8)
596 index = 0;
597 else
598 index = get_count_order(nents) - 3;
599
600 return index;
601}
602
603static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
604{
605 struct scsi_host_sg_pool *sgp;
606
607 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
608 mempool_free(sgl, sgp->pool);
609}
610
611static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
612{
613 struct scsi_host_sg_pool *sgp;
614
615 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
616 return mempool_alloc(sgp->pool, gfp_mask);
617}
618
619static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
620 gfp_t gfp_mask)
621{
622 int ret;
623
624 BUG_ON(!nents);
625
626 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
627 gfp_mask, scsi_sg_alloc);
628 if (unlikely(ret))
629 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
630 scsi_sg_free);
631
632 return ret;
633}
634
635static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
636{
637 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
638}
639
640static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
641{
642
643 if (cmd->sdb.table.nents)
644 scsi_free_sgtable(&cmd->sdb);
645
646 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
647
648 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
649 struct scsi_data_buffer *bidi_sdb =
650 cmd->request->next_rq->special;
651 scsi_free_sgtable(bidi_sdb);
652 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
653 cmd->request->next_rq->special = NULL;
654 }
655
656 if (scsi_prot_sg_count(cmd))
657 scsi_free_sgtable(cmd->prot_sdb);
658}
659
660/*
661 * Function: scsi_release_buffers()
662 *
663 * Purpose: Completion processing for block device I/O requests.
664 *
665 * Arguments: cmd - command that we are bailing.
666 *
667 * Lock status: Assumed that no lock is held upon entry.
668 *
669 * Returns: Nothing
670 *
671 * Notes: In the event that an upper level driver rejects a
672 * command, we must release resources allocated during
673 * the __init_io() function. Primarily this would involve
674 * the scatter-gather table, and potentially any bounce
675 * buffers.
676 */
677void scsi_release_buffers(struct scsi_cmnd *cmd)
678{
679 __scsi_release_buffers(cmd, 1);
680}
681EXPORT_SYMBOL(scsi_release_buffers);
682
683static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
684{
685 int error = 0;
686
687 switch(host_byte(result)) {
688 case DID_TRANSPORT_FAILFAST:
689 error = -ENOLINK;
690 break;
691 case DID_TARGET_FAILURE:
692 set_host_byte(cmd, DID_OK);
693 error = -EREMOTEIO;
694 break;
695 case DID_NEXUS_FAILURE:
696 set_host_byte(cmd, DID_OK);
697 error = -EBADE;
698 break;
699 default:
700 error = -EIO;
701 break;
702 }
703
704 return error;
705}
706
707/*
708 * Function: scsi_io_completion()
709 *
710 * Purpose: Completion processing for block device I/O requests.
711 *
712 * Arguments: cmd - command that is finished.
713 *
714 * Lock status: Assumed that no lock is held upon entry.
715 *
716 * Returns: Nothing
717 *
718 * Notes: This function is matched in terms of capabilities to
719 * the function that created the scatter-gather list.
720 * In other words, if there are no bounce buffers
721 * (the normal case for most drivers), we don't need
722 * the logic to deal with cleaning up afterwards.
723 *
724 * We must call scsi_end_request(). This will finish off
725 * the specified number of sectors. If we are done, the
726 * command block will be released and the queue function
727 * will be goosed. If we are not done then we have to
728 * figure out what to do next:
729 *
730 * a) We can call scsi_requeue_command(). The request
731 * will be unprepared and put back on the queue. Then
732 * a new command will be created for it. This should
733 * be used if we made forward progress, or if we want
734 * to switch from READ(10) to READ(6) for example.
735 *
736 * b) We can call scsi_queue_insert(). The request will
737 * be put back on the queue and retried using the same
738 * command as before, possibly after a delay.
739 *
740 * c) We can call blk_end_request() with -EIO to fail
741 * the remainder of the request.
742 */
743void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
744{
745 int result = cmd->result;
746 struct request_queue *q = cmd->device->request_queue;
747 struct request *req = cmd->request;
748 int error = 0;
749 struct scsi_sense_hdr sshdr;
750 int sense_valid = 0;
751 int sense_deferred = 0;
752 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
753 ACTION_DELAYED_RETRY} action;
754 char *description = NULL;
755
756 if (result) {
757 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
758 if (sense_valid)
759 sense_deferred = scsi_sense_is_deferred(&sshdr);
760 }
761
762 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
763 if (result) {
764 if (sense_valid && req->sense) {
765 /*
766 * SG_IO wants current and deferred errors
767 */
768 int len = 8 + cmd->sense_buffer[7];
769
770 if (len > SCSI_SENSE_BUFFERSIZE)
771 len = SCSI_SENSE_BUFFERSIZE;
772 memcpy(req->sense, cmd->sense_buffer, len);
773 req->sense_len = len;
774 }
775 if (!sense_deferred)
776 error = __scsi_error_from_host_byte(cmd, result);
777 }
778 /*
779 * __scsi_error_from_host_byte may have reset the host_byte
780 */
781 req->errors = cmd->result;
782
783 req->resid_len = scsi_get_resid(cmd);
784
785 if (scsi_bidi_cmnd(cmd)) {
786 /*
787 * Bidi commands Must be complete as a whole,
788 * both sides at once.
789 */
790 req->next_rq->resid_len = scsi_in(cmd)->resid;
791
792 scsi_release_buffers(cmd);
793 blk_end_request_all(req, 0);
794
795 scsi_next_command(cmd);
796 return;
797 }
798 }
799
800 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
801 BUG_ON(blk_bidi_rq(req));
802
803 /*
804 * Next deal with any sectors which we were able to correctly
805 * handle.
806 */
807 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
808 "%d bytes done.\n",
809 blk_rq_sectors(req), good_bytes));
810
811 /*
812 * Recovered errors need reporting, but they're always treated
813 * as success, so fiddle the result code here. For BLOCK_PC
814 * we already took a copy of the original into rq->errors which
815 * is what gets returned to the user
816 */
817 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
818 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
819 * print since caller wants ATA registers. Only occurs on
820 * SCSI ATA PASS_THROUGH commands when CK_COND=1
821 */
822 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
823 ;
824 else if (!(req->cmd_flags & REQ_QUIET))
825 scsi_print_sense("", cmd);
826 result = 0;
827 /* BLOCK_PC may have set error */
828 error = 0;
829 }
830
831 /*
832 * A number of bytes were successfully read. If there
833 * are leftovers and there is some kind of error
834 * (result != 0), retry the rest.
835 */
836 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
837 return;
838
839 error = __scsi_error_from_host_byte(cmd, result);
840
841 if (host_byte(result) == DID_RESET) {
842 /* Third party bus reset or reset for error recovery
843 * reasons. Just retry the command and see what
844 * happens.
845 */
846 action = ACTION_RETRY;
847 } else if (sense_valid && !sense_deferred) {
848 switch (sshdr.sense_key) {
849 case UNIT_ATTENTION:
850 if (cmd->device->removable) {
851 /* Detected disc change. Set a bit
852 * and quietly refuse further access.
853 */
854 cmd->device->changed = 1;
855 description = "Media Changed";
856 action = ACTION_FAIL;
857 } else {
858 /* Must have been a power glitch, or a
859 * bus reset. Could not have been a
860 * media change, so we just retry the
861 * command and see what happens.
862 */
863 action = ACTION_RETRY;
864 }
865 break;
866 case ILLEGAL_REQUEST:
867 /* If we had an ILLEGAL REQUEST returned, then
868 * we may have performed an unsupported
869 * command. The only thing this should be
870 * would be a ten byte read where only a six
871 * byte read was supported. Also, on a system
872 * where READ CAPACITY failed, we may have
873 * read past the end of the disk.
874 */
875 if ((cmd->device->use_10_for_rw &&
876 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
877 (cmd->cmnd[0] == READ_10 ||
878 cmd->cmnd[0] == WRITE_10)) {
879 /* This will issue a new 6-byte command. */
880 cmd->device->use_10_for_rw = 0;
881 action = ACTION_REPREP;
882 } else if (sshdr.asc == 0x10) /* DIX */ {
883 description = "Host Data Integrity Failure";
884 action = ACTION_FAIL;
885 error = -EILSEQ;
886 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
887 } else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
888 (cmd->cmnd[0] == UNMAP ||
889 cmd->cmnd[0] == WRITE_SAME_16 ||
890 cmd->cmnd[0] == WRITE_SAME)) {
891 description = "Discard failure";
892 action = ACTION_FAIL;
893 error = -EREMOTEIO;
894 } else
895 action = ACTION_FAIL;
896 break;
897 case ABORTED_COMMAND:
898 action = ACTION_FAIL;
899 if (sshdr.asc == 0x10) { /* DIF */
900 description = "Target Data Integrity Failure";
901 error = -EILSEQ;
902 }
903 break;
904 case NOT_READY:
905 /* If the device is in the process of becoming
906 * ready, or has a temporary blockage, retry.
907 */
908 if (sshdr.asc == 0x04) {
909 switch (sshdr.ascq) {
910 case 0x01: /* becoming ready */
911 case 0x04: /* format in progress */
912 case 0x05: /* rebuild in progress */
913 case 0x06: /* recalculation in progress */
914 case 0x07: /* operation in progress */
915 case 0x08: /* Long write in progress */
916 case 0x09: /* self test in progress */
917 case 0x14: /* space allocation in progress */
918 action = ACTION_DELAYED_RETRY;
919 break;
920 default:
921 description = "Device not ready";
922 action = ACTION_FAIL;
923 break;
924 }
925 } else {
926 description = "Device not ready";
927 action = ACTION_FAIL;
928 }
929 break;
930 case VOLUME_OVERFLOW:
931 /* See SSC3rXX or current. */
932 action = ACTION_FAIL;
933 break;
934 default:
935 description = "Unhandled sense code";
936 action = ACTION_FAIL;
937 break;
938 }
939 } else {
940 description = "Unhandled error code";
941 action = ACTION_FAIL;
942 }
943
944 switch (action) {
945 case ACTION_FAIL:
946 /* Give up and fail the remainder of the request */
947 scsi_release_buffers(cmd);
948 if (!(req->cmd_flags & REQ_QUIET)) {
949 if (description)
950 scmd_printk(KERN_INFO, cmd, "%s\n",
951 description);
952 scsi_print_result(cmd);
953 if (driver_byte(result) & DRIVER_SENSE)
954 scsi_print_sense("", cmd);
955 scsi_print_command(cmd);
956 }
957 if (blk_end_request_err(req, error))
958 scsi_requeue_command(q, cmd);
959 else
960 scsi_next_command(cmd);
961 break;
962 case ACTION_REPREP:
963 /* Unprep the request and put it back at the head of the queue.
964 * A new command will be prepared and issued.
965 */
966 scsi_release_buffers(cmd);
967 scsi_requeue_command(q, cmd);
968 break;
969 case ACTION_RETRY:
970 /* Retry the same command immediately */
971 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
972 break;
973 case ACTION_DELAYED_RETRY:
974 /* Retry the same command after a delay */
975 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
976 break;
977 }
978}
979
980static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
981 gfp_t gfp_mask)
982{
983 int count;
984
985 /*
986 * If sg table allocation fails, requeue request later.
987 */
988 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
989 gfp_mask))) {
990 return BLKPREP_DEFER;
991 }
992
993 req->buffer = NULL;
994
995 /*
996 * Next, walk the list, and fill in the addresses and sizes of
997 * each segment.
998 */
999 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1000 BUG_ON(count > sdb->table.nents);
1001 sdb->table.nents = count;
1002 sdb->length = blk_rq_bytes(req);
1003 return BLKPREP_OK;
1004}
1005
1006/*
1007 * Function: scsi_init_io()
1008 *
1009 * Purpose: SCSI I/O initialize function.
1010 *
1011 * Arguments: cmd - Command descriptor we wish to initialize
1012 *
1013 * Returns: 0 on success
1014 * BLKPREP_DEFER if the failure is retryable
1015 * BLKPREP_KILL if the failure is fatal
1016 */
1017int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1018{
1019 struct request *rq = cmd->request;
1020
1021 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1022 if (error)
1023 goto err_exit;
1024
1025 if (blk_bidi_rq(rq)) {
1026 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1027 scsi_sdb_cache, GFP_ATOMIC);
1028 if (!bidi_sdb) {
1029 error = BLKPREP_DEFER;
1030 goto err_exit;
1031 }
1032
1033 rq->next_rq->special = bidi_sdb;
1034 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1035 if (error)
1036 goto err_exit;
1037 }
1038
1039 if (blk_integrity_rq(rq)) {
1040 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1041 int ivecs, count;
1042
1043 BUG_ON(prot_sdb == NULL);
1044 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1045
1046 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1047 error = BLKPREP_DEFER;
1048 goto err_exit;
1049 }
1050
1051 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1052 prot_sdb->table.sgl);
1053 BUG_ON(unlikely(count > ivecs));
1054 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1055
1056 cmd->prot_sdb = prot_sdb;
1057 cmd->prot_sdb->table.nents = count;
1058 }
1059
1060 return BLKPREP_OK ;
1061
1062err_exit:
1063 scsi_release_buffers(cmd);
1064 cmd->request->special = NULL;
1065 scsi_put_command(cmd);
1066 return error;
1067}
1068EXPORT_SYMBOL(scsi_init_io);
1069
1070static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1071 struct request *req)
1072{
1073 struct scsi_cmnd *cmd;
1074
1075 if (!req->special) {
1076 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1077 if (unlikely(!cmd))
1078 return NULL;
1079 req->special = cmd;
1080 } else {
1081 cmd = req->special;
1082 }
1083
1084 /* pull a tag out of the request if we have one */
1085 cmd->tag = req->tag;
1086 cmd->request = req;
1087
1088 cmd->cmnd = req->cmd;
1089 cmd->prot_op = SCSI_PROT_NORMAL;
1090
1091 return cmd;
1092}
1093
1094int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1095{
1096 struct scsi_cmnd *cmd;
1097 int ret = scsi_prep_state_check(sdev, req);
1098
1099 if (ret != BLKPREP_OK)
1100 return ret;
1101
1102 cmd = scsi_get_cmd_from_req(sdev, req);
1103 if (unlikely(!cmd))
1104 return BLKPREP_DEFER;
1105
1106 /*
1107 * BLOCK_PC requests may transfer data, in which case they must
1108 * a bio attached to them. Or they might contain a SCSI command
1109 * that does not transfer data, in which case they may optionally
1110 * submit a request without an attached bio.
1111 */
1112 if (req->bio) {
1113 int ret;
1114
1115 BUG_ON(!req->nr_phys_segments);
1116
1117 ret = scsi_init_io(cmd, GFP_ATOMIC);
1118 if (unlikely(ret))
1119 return ret;
1120 } else {
1121 BUG_ON(blk_rq_bytes(req));
1122
1123 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1124 req->buffer = NULL;
1125 }
1126
1127 cmd->cmd_len = req->cmd_len;
1128 if (!blk_rq_bytes(req))
1129 cmd->sc_data_direction = DMA_NONE;
1130 else if (rq_data_dir(req) == WRITE)
1131 cmd->sc_data_direction = DMA_TO_DEVICE;
1132 else
1133 cmd->sc_data_direction = DMA_FROM_DEVICE;
1134
1135 cmd->transfersize = blk_rq_bytes(req);
1136 cmd->allowed = req->retries;
1137 return BLKPREP_OK;
1138}
1139EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1140
1141/*
1142 * Setup a REQ_TYPE_FS command. These are simple read/write request
1143 * from filesystems that still need to be translated to SCSI CDBs from
1144 * the ULD.
1145 */
1146int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1147{
1148 struct scsi_cmnd *cmd;
1149 int ret = scsi_prep_state_check(sdev, req);
1150
1151 if (ret != BLKPREP_OK)
1152 return ret;
1153
1154 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1155 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1156 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1157 if (ret != BLKPREP_OK)
1158 return ret;
1159 }
1160
1161 /*
1162 * Filesystem requests must transfer data.
1163 */
1164 BUG_ON(!req->nr_phys_segments);
1165
1166 cmd = scsi_get_cmd_from_req(sdev, req);
1167 if (unlikely(!cmd))
1168 return BLKPREP_DEFER;
1169
1170 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1171 return scsi_init_io(cmd, GFP_ATOMIC);
1172}
1173EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1174
1175int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1176{
1177 int ret = BLKPREP_OK;
1178
1179 /*
1180 * If the device is not in running state we will reject some
1181 * or all commands.
1182 */
1183 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1184 switch (sdev->sdev_state) {
1185 case SDEV_OFFLINE:
1186 /*
1187 * If the device is offline we refuse to process any
1188 * commands. The device must be brought online
1189 * before trying any recovery commands.
1190 */
1191 sdev_printk(KERN_ERR, sdev,
1192 "rejecting I/O to offline device\n");
1193 ret = BLKPREP_KILL;
1194 break;
1195 case SDEV_DEL:
1196 /*
1197 * If the device is fully deleted, we refuse to
1198 * process any commands as well.
1199 */
1200 sdev_printk(KERN_ERR, sdev,
1201 "rejecting I/O to dead device\n");
1202 ret = BLKPREP_KILL;
1203 break;
1204 case SDEV_QUIESCE:
1205 case SDEV_BLOCK:
1206 case SDEV_CREATED_BLOCK:
1207 /*
1208 * If the devices is blocked we defer normal commands.
1209 */
1210 if (!(req->cmd_flags & REQ_PREEMPT))
1211 ret = BLKPREP_DEFER;
1212 break;
1213 default:
1214 /*
1215 * For any other not fully online state we only allow
1216 * special commands. In particular any user initiated
1217 * command is not allowed.
1218 */
1219 if (!(req->cmd_flags & REQ_PREEMPT))
1220 ret = BLKPREP_KILL;
1221 break;
1222 }
1223 }
1224 return ret;
1225}
1226EXPORT_SYMBOL(scsi_prep_state_check);
1227
1228int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1229{
1230 struct scsi_device *sdev = q->queuedata;
1231
1232 switch (ret) {
1233 case BLKPREP_KILL:
1234 req->errors = DID_NO_CONNECT << 16;
1235 /* release the command and kill it */
1236 if (req->special) {
1237 struct scsi_cmnd *cmd = req->special;
1238 scsi_release_buffers(cmd);
1239 scsi_put_command(cmd);
1240 req->special = NULL;
1241 }
1242 break;
1243 case BLKPREP_DEFER:
1244 /*
1245 * If we defer, the blk_peek_request() returns NULL, but the
1246 * queue must be restarted, so we schedule a callback to happen
1247 * shortly.
1248 */
1249 if (sdev->device_busy == 0)
1250 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1251 break;
1252 default:
1253 req->cmd_flags |= REQ_DONTPREP;
1254 }
1255
1256 return ret;
1257}
1258EXPORT_SYMBOL(scsi_prep_return);
1259
1260int scsi_prep_fn(struct request_queue *q, struct request *req)
1261{
1262 struct scsi_device *sdev = q->queuedata;
1263 int ret = BLKPREP_KILL;
1264
1265 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1266 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1267 return scsi_prep_return(q, req, ret);
1268}
1269EXPORT_SYMBOL(scsi_prep_fn);
1270
1271/*
1272 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1273 * return 0.
1274 *
1275 * Called with the queue_lock held.
1276 */
1277static inline int scsi_dev_queue_ready(struct request_queue *q,
1278 struct scsi_device *sdev)
1279{
1280 if (sdev->device_busy == 0 && sdev->device_blocked) {
1281 /*
1282 * unblock after device_blocked iterates to zero
1283 */
1284 if (--sdev->device_blocked == 0) {
1285 SCSI_LOG_MLQUEUE(3,
1286 sdev_printk(KERN_INFO, sdev,
1287 "unblocking device at zero depth\n"));
1288 } else {
1289 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 return 0;
1291 }
1292 }
1293 if (scsi_device_is_busy(sdev))
1294 return 0;
1295
1296 return 1;
1297}
1298
1299
1300/*
1301 * scsi_target_queue_ready: checks if there we can send commands to target
1302 * @sdev: scsi device on starget to check.
1303 *
1304 * Called with the host lock held.
1305 */
1306static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1307 struct scsi_device *sdev)
1308{
1309 struct scsi_target *starget = scsi_target(sdev);
1310
1311 if (starget->single_lun) {
1312 if (starget->starget_sdev_user &&
1313 starget->starget_sdev_user != sdev)
1314 return 0;
1315 starget->starget_sdev_user = sdev;
1316 }
1317
1318 if (starget->target_busy == 0 && starget->target_blocked) {
1319 /*
1320 * unblock after target_blocked iterates to zero
1321 */
1322 if (--starget->target_blocked == 0) {
1323 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1324 "unblocking target at zero depth\n"));
1325 } else
1326 return 0;
1327 }
1328
1329 if (scsi_target_is_busy(starget)) {
1330 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1331 return 0;
1332 }
1333
1334 return 1;
1335}
1336
1337/*
1338 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1339 * return 0. We must end up running the queue again whenever 0 is
1340 * returned, else IO can hang.
1341 *
1342 * Called with host_lock held.
1343 */
1344static inline int scsi_host_queue_ready(struct request_queue *q,
1345 struct Scsi_Host *shost,
1346 struct scsi_device *sdev)
1347{
1348 if (scsi_host_in_recovery(shost))
1349 return 0;
1350 if (shost->host_busy == 0 && shost->host_blocked) {
1351 /*
1352 * unblock after host_blocked iterates to zero
1353 */
1354 if (--shost->host_blocked == 0) {
1355 SCSI_LOG_MLQUEUE(3,
1356 printk("scsi%d unblocking host at zero depth\n",
1357 shost->host_no));
1358 } else {
1359 return 0;
1360 }
1361 }
1362 if (scsi_host_is_busy(shost)) {
1363 if (list_empty(&sdev->starved_entry))
1364 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1365 return 0;
1366 }
1367
1368 /* We're OK to process the command, so we can't be starved */
1369 if (!list_empty(&sdev->starved_entry))
1370 list_del_init(&sdev->starved_entry);
1371
1372 return 1;
1373}
1374
1375/*
1376 * Busy state exporting function for request stacking drivers.
1377 *
1378 * For efficiency, no lock is taken to check the busy state of
1379 * shost/starget/sdev, since the returned value is not guaranteed and
1380 * may be changed after request stacking drivers call the function,
1381 * regardless of taking lock or not.
1382 *
1383 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1384 * needs to return 'not busy'. Otherwise, request stacking drivers
1385 * may hold requests forever.
1386 */
1387static int scsi_lld_busy(struct request_queue *q)
1388{
1389 struct scsi_device *sdev = q->queuedata;
1390 struct Scsi_Host *shost;
1391
1392 if (blk_queue_dead(q))
1393 return 0;
1394
1395 shost = sdev->host;
1396
1397 /*
1398 * Ignore host/starget busy state.
1399 * Since block layer does not have a concept of fairness across
1400 * multiple queues, congestion of host/starget needs to be handled
1401 * in SCSI layer.
1402 */
1403 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1404 return 1;
1405
1406 return 0;
1407}
1408
1409/*
1410 * Kill a request for a dead device
1411 */
1412static void scsi_kill_request(struct request *req, struct request_queue *q)
1413{
1414 struct scsi_cmnd *cmd = req->special;
1415 struct scsi_device *sdev;
1416 struct scsi_target *starget;
1417 struct Scsi_Host *shost;
1418
1419 blk_start_request(req);
1420
1421 scmd_printk(KERN_INFO, cmd, "killing request\n");
1422
1423 sdev = cmd->device;
1424 starget = scsi_target(sdev);
1425 shost = sdev->host;
1426 scsi_init_cmd_errh(cmd);
1427 cmd->result = DID_NO_CONNECT << 16;
1428 atomic_inc(&cmd->device->iorequest_cnt);
1429
1430 /*
1431 * SCSI request completion path will do scsi_device_unbusy(),
1432 * bump busy counts. To bump the counters, we need to dance
1433 * with the locks as normal issue path does.
1434 */
1435 sdev->device_busy++;
1436 spin_unlock(sdev->request_queue->queue_lock);
1437 spin_lock(shost->host_lock);
1438 shost->host_busy++;
1439 starget->target_busy++;
1440 spin_unlock(shost->host_lock);
1441 spin_lock(sdev->request_queue->queue_lock);
1442
1443 blk_complete_request(req);
1444}
1445
1446static void scsi_softirq_done(struct request *rq)
1447{
1448 struct scsi_cmnd *cmd = rq->special;
1449 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1450 int disposition;
1451
1452 INIT_LIST_HEAD(&cmd->eh_entry);
1453
1454 atomic_inc(&cmd->device->iodone_cnt);
1455 if (cmd->result)
1456 atomic_inc(&cmd->device->ioerr_cnt);
1457
1458 disposition = scsi_decide_disposition(cmd);
1459 if (disposition != SUCCESS &&
1460 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1461 sdev_printk(KERN_ERR, cmd->device,
1462 "timing out command, waited %lus\n",
1463 wait_for/HZ);
1464 disposition = SUCCESS;
1465 }
1466
1467 scsi_log_completion(cmd, disposition);
1468
1469 switch (disposition) {
1470 case SUCCESS:
1471 scsi_finish_command(cmd);
1472 break;
1473 case NEEDS_RETRY:
1474 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1475 break;
1476 case ADD_TO_MLQUEUE:
1477 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1478 break;
1479 default:
1480 if (!scsi_eh_scmd_add(cmd, 0))
1481 scsi_finish_command(cmd);
1482 }
1483}
1484
1485/*
1486 * Function: scsi_request_fn()
1487 *
1488 * Purpose: Main strategy routine for SCSI.
1489 *
1490 * Arguments: q - Pointer to actual queue.
1491 *
1492 * Returns: Nothing
1493 *
1494 * Lock status: IO request lock assumed to be held when called.
1495 */
1496static void scsi_request_fn(struct request_queue *q)
1497{
1498 struct scsi_device *sdev = q->queuedata;
1499 struct Scsi_Host *shost;
1500 struct scsi_cmnd *cmd;
1501 struct request *req;
1502
1503 if(!get_device(&sdev->sdev_gendev))
1504 /* We must be tearing the block queue down already */
1505 return;
1506
1507 /*
1508 * To start with, we keep looping until the queue is empty, or until
1509 * the host is no longer able to accept any more requests.
1510 */
1511 shost = sdev->host;
1512 for (;;) {
1513 int rtn;
1514 /*
1515 * get next queueable request. We do this early to make sure
1516 * that the request is fully prepared even if we cannot
1517 * accept it.
1518 */
1519 req = blk_peek_request(q);
1520 if (!req || !scsi_dev_queue_ready(q, sdev))
1521 break;
1522
1523 if (unlikely(!scsi_device_online(sdev))) {
1524 sdev_printk(KERN_ERR, sdev,
1525 "rejecting I/O to offline device\n");
1526 scsi_kill_request(req, q);
1527 continue;
1528 }
1529
1530
1531 /*
1532 * Remove the request from the request list.
1533 */
1534 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1535 blk_start_request(req);
1536 sdev->device_busy++;
1537
1538 spin_unlock(q->queue_lock);
1539 cmd = req->special;
1540 if (unlikely(cmd == NULL)) {
1541 printk(KERN_CRIT "impossible request in %s.\n"
1542 "please mail a stack trace to "
1543 "linux-scsi@vger.kernel.org\n",
1544 __func__);
1545 blk_dump_rq_flags(req, "foo");
1546 BUG();
1547 }
1548 spin_lock(shost->host_lock);
1549
1550 /*
1551 * We hit this when the driver is using a host wide
1552 * tag map. For device level tag maps the queue_depth check
1553 * in the device ready fn would prevent us from trying
1554 * to allocate a tag. Since the map is a shared host resource
1555 * we add the dev to the starved list so it eventually gets
1556 * a run when a tag is freed.
1557 */
1558 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1559 if (list_empty(&sdev->starved_entry))
1560 list_add_tail(&sdev->starved_entry,
1561 &shost->starved_list);
1562 goto not_ready;
1563 }
1564
1565 if (!scsi_target_queue_ready(shost, sdev))
1566 goto not_ready;
1567
1568 if (!scsi_host_queue_ready(q, shost, sdev))
1569 goto not_ready;
1570
1571 scsi_target(sdev)->target_busy++;
1572 shost->host_busy++;
1573
1574 /*
1575 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1576 * take the lock again.
1577 */
1578 spin_unlock_irq(shost->host_lock);
1579
1580 /*
1581 * Finally, initialize any error handling parameters, and set up
1582 * the timers for timeouts.
1583 */
1584 scsi_init_cmd_errh(cmd);
1585
1586 /*
1587 * Dispatch the command to the low-level driver.
1588 */
1589 rtn = scsi_dispatch_cmd(cmd);
1590 spin_lock_irq(q->queue_lock);
1591 if (rtn)
1592 goto out_delay;
1593 }
1594
1595 goto out;
1596
1597 not_ready:
1598 spin_unlock_irq(shost->host_lock);
1599
1600 /*
1601 * lock q, handle tag, requeue req, and decrement device_busy. We
1602 * must return with queue_lock held.
1603 *
1604 * Decrementing device_busy without checking it is OK, as all such
1605 * cases (host limits or settings) should run the queue at some
1606 * later time.
1607 */
1608 spin_lock_irq(q->queue_lock);
1609 blk_requeue_request(q, req);
1610 sdev->device_busy--;
1611out_delay:
1612 if (sdev->device_busy == 0)
1613 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1614out:
1615 /* must be careful here...if we trigger the ->remove() function
1616 * we cannot be holding the q lock */
1617 spin_unlock_irq(q->queue_lock);
1618 put_device(&sdev->sdev_gendev);
1619 spin_lock_irq(q->queue_lock);
1620}
1621
1622u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1623{
1624 struct device *host_dev;
1625 u64 bounce_limit = 0xffffffff;
1626
1627 if (shost->unchecked_isa_dma)
1628 return BLK_BOUNCE_ISA;
1629 /*
1630 * Platforms with virtual-DMA translation
1631 * hardware have no practical limit.
1632 */
1633 if (!PCI_DMA_BUS_IS_PHYS)
1634 return BLK_BOUNCE_ANY;
1635
1636 host_dev = scsi_get_device(shost);
1637 if (host_dev && host_dev->dma_mask)
1638 bounce_limit = *host_dev->dma_mask;
1639
1640 return bounce_limit;
1641}
1642EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1643
1644struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1645 request_fn_proc *request_fn)
1646{
1647 struct request_queue *q;
1648 struct device *dev = shost->dma_dev;
1649
1650 q = blk_init_queue(request_fn, NULL);
1651 if (!q)
1652 return NULL;
1653
1654 /*
1655 * this limit is imposed by hardware restrictions
1656 */
1657 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1658 SCSI_MAX_SG_CHAIN_SEGMENTS));
1659
1660 if (scsi_host_prot_dma(shost)) {
1661 shost->sg_prot_tablesize =
1662 min_not_zero(shost->sg_prot_tablesize,
1663 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1664 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1665 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1666 }
1667
1668 blk_queue_max_hw_sectors(q, shost->max_sectors);
1669 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1670 blk_queue_segment_boundary(q, shost->dma_boundary);
1671 dma_set_seg_boundary(dev, shost->dma_boundary);
1672
1673 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1674
1675 if (!shost->use_clustering)
1676 q->limits.cluster = 0;
1677
1678 /*
1679 * set a reasonable default alignment on word boundaries: the
1680 * host and device may alter it using
1681 * blk_queue_update_dma_alignment() later.
1682 */
1683 blk_queue_dma_alignment(q, 0x03);
1684
1685 return q;
1686}
1687EXPORT_SYMBOL(__scsi_alloc_queue);
1688
1689struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1690{
1691 struct request_queue *q;
1692
1693 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1694 if (!q)
1695 return NULL;
1696
1697 blk_queue_prep_rq(q, scsi_prep_fn);
1698 blk_queue_softirq_done(q, scsi_softirq_done);
1699 blk_queue_rq_timed_out(q, scsi_times_out);
1700 blk_queue_lld_busy(q, scsi_lld_busy);
1701 return q;
1702}
1703
1704/*
1705 * Function: scsi_block_requests()
1706 *
1707 * Purpose: Utility function used by low-level drivers to prevent further
1708 * commands from being queued to the device.
1709 *
1710 * Arguments: shost - Host in question
1711 *
1712 * Returns: Nothing
1713 *
1714 * Lock status: No locks are assumed held.
1715 *
1716 * Notes: There is no timer nor any other means by which the requests
1717 * get unblocked other than the low-level driver calling
1718 * scsi_unblock_requests().
1719 */
1720void scsi_block_requests(struct Scsi_Host *shost)
1721{
1722 shost->host_self_blocked = 1;
1723}
1724EXPORT_SYMBOL(scsi_block_requests);
1725
1726/*
1727 * Function: scsi_unblock_requests()
1728 *
1729 * Purpose: Utility function used by low-level drivers to allow further
1730 * commands from being queued to the device.
1731 *
1732 * Arguments: shost - Host in question
1733 *
1734 * Returns: Nothing
1735 *
1736 * Lock status: No locks are assumed held.
1737 *
1738 * Notes: There is no timer nor any other means by which the requests
1739 * get unblocked other than the low-level driver calling
1740 * scsi_unblock_requests().
1741 *
1742 * This is done as an API function so that changes to the
1743 * internals of the scsi mid-layer won't require wholesale
1744 * changes to drivers that use this feature.
1745 */
1746void scsi_unblock_requests(struct Scsi_Host *shost)
1747{
1748 shost->host_self_blocked = 0;
1749 scsi_run_host_queues(shost);
1750}
1751EXPORT_SYMBOL(scsi_unblock_requests);
1752
1753int __init scsi_init_queue(void)
1754{
1755 int i;
1756
1757 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1758 sizeof(struct scsi_data_buffer),
1759 0, 0, NULL);
1760 if (!scsi_sdb_cache) {
1761 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1762 return -ENOMEM;
1763 }
1764
1765 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1766 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1767 int size = sgp->size * sizeof(struct scatterlist);
1768
1769 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1770 SLAB_HWCACHE_ALIGN, NULL);
1771 if (!sgp->slab) {
1772 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1773 sgp->name);
1774 goto cleanup_sdb;
1775 }
1776
1777 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1778 sgp->slab);
1779 if (!sgp->pool) {
1780 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1781 sgp->name);
1782 goto cleanup_sdb;
1783 }
1784 }
1785
1786 return 0;
1787
1788cleanup_sdb:
1789 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1790 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1791 if (sgp->pool)
1792 mempool_destroy(sgp->pool);
1793 if (sgp->slab)
1794 kmem_cache_destroy(sgp->slab);
1795 }
1796 kmem_cache_destroy(scsi_sdb_cache);
1797
1798 return -ENOMEM;
1799}
1800
1801void scsi_exit_queue(void)
1802{
1803 int i;
1804
1805 kmem_cache_destroy(scsi_sdb_cache);
1806
1807 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1808 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1809 mempool_destroy(sgp->pool);
1810 kmem_cache_destroy(sgp->slab);
1811 }
1812}
1813
1814/**
1815 * scsi_mode_select - issue a mode select
1816 * @sdev: SCSI device to be queried
1817 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1818 * @sp: Save page bit (0 == don't save, 1 == save)
1819 * @modepage: mode page being requested
1820 * @buffer: request buffer (may not be smaller than eight bytes)
1821 * @len: length of request buffer.
1822 * @timeout: command timeout
1823 * @retries: number of retries before failing
1824 * @data: returns a structure abstracting the mode header data
1825 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1826 * must be SCSI_SENSE_BUFFERSIZE big.
1827 *
1828 * Returns zero if successful; negative error number or scsi
1829 * status on error
1830 *
1831 */
1832int
1833scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1834 unsigned char *buffer, int len, int timeout, int retries,
1835 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1836{
1837 unsigned char cmd[10];
1838 unsigned char *real_buffer;
1839 int ret;
1840
1841 memset(cmd, 0, sizeof(cmd));
1842 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1843
1844 if (sdev->use_10_for_ms) {
1845 if (len > 65535)
1846 return -EINVAL;
1847 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1848 if (!real_buffer)
1849 return -ENOMEM;
1850 memcpy(real_buffer + 8, buffer, len);
1851 len += 8;
1852 real_buffer[0] = 0;
1853 real_buffer[1] = 0;
1854 real_buffer[2] = data->medium_type;
1855 real_buffer[3] = data->device_specific;
1856 real_buffer[4] = data->longlba ? 0x01 : 0;
1857 real_buffer[5] = 0;
1858 real_buffer[6] = data->block_descriptor_length >> 8;
1859 real_buffer[7] = data->block_descriptor_length;
1860
1861 cmd[0] = MODE_SELECT_10;
1862 cmd[7] = len >> 8;
1863 cmd[8] = len;
1864 } else {
1865 if (len > 255 || data->block_descriptor_length > 255 ||
1866 data->longlba)
1867 return -EINVAL;
1868
1869 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1870 if (!real_buffer)
1871 return -ENOMEM;
1872 memcpy(real_buffer + 4, buffer, len);
1873 len += 4;
1874 real_buffer[0] = 0;
1875 real_buffer[1] = data->medium_type;
1876 real_buffer[2] = data->device_specific;
1877 real_buffer[3] = data->block_descriptor_length;
1878
1879
1880 cmd[0] = MODE_SELECT;
1881 cmd[4] = len;
1882 }
1883
1884 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1885 sshdr, timeout, retries, NULL);
1886 kfree(real_buffer);
1887 return ret;
1888}
1889EXPORT_SYMBOL_GPL(scsi_mode_select);
1890
1891/**
1892 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1893 * @sdev: SCSI device to be queried
1894 * @dbd: set if mode sense will allow block descriptors to be returned
1895 * @modepage: mode page being requested
1896 * @buffer: request buffer (may not be smaller than eight bytes)
1897 * @len: length of request buffer.
1898 * @timeout: command timeout
1899 * @retries: number of retries before failing
1900 * @data: returns a structure abstracting the mode header data
1901 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1902 * must be SCSI_SENSE_BUFFERSIZE big.
1903 *
1904 * Returns zero if unsuccessful, or the header offset (either 4
1905 * or 8 depending on whether a six or ten byte command was
1906 * issued) if successful.
1907 */
1908int
1909scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1910 unsigned char *buffer, int len, int timeout, int retries,
1911 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1912{
1913 unsigned char cmd[12];
1914 int use_10_for_ms;
1915 int header_length;
1916 int result;
1917 struct scsi_sense_hdr my_sshdr;
1918
1919 memset(data, 0, sizeof(*data));
1920 memset(&cmd[0], 0, 12);
1921 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1922 cmd[2] = modepage;
1923
1924 /* caller might not be interested in sense, but we need it */
1925 if (!sshdr)
1926 sshdr = &my_sshdr;
1927
1928 retry:
1929 use_10_for_ms = sdev->use_10_for_ms;
1930
1931 if (use_10_for_ms) {
1932 if (len < 8)
1933 len = 8;
1934
1935 cmd[0] = MODE_SENSE_10;
1936 cmd[8] = len;
1937 header_length = 8;
1938 } else {
1939 if (len < 4)
1940 len = 4;
1941
1942 cmd[0] = MODE_SENSE;
1943 cmd[4] = len;
1944 header_length = 4;
1945 }
1946
1947 memset(buffer, 0, len);
1948
1949 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1950 sshdr, timeout, retries, NULL);
1951
1952 /* This code looks awful: what it's doing is making sure an
1953 * ILLEGAL REQUEST sense return identifies the actual command
1954 * byte as the problem. MODE_SENSE commands can return
1955 * ILLEGAL REQUEST if the code page isn't supported */
1956
1957 if (use_10_for_ms && !scsi_status_is_good(result) &&
1958 (driver_byte(result) & DRIVER_SENSE)) {
1959 if (scsi_sense_valid(sshdr)) {
1960 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1961 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1962 /*
1963 * Invalid command operation code
1964 */
1965 sdev->use_10_for_ms = 0;
1966 goto retry;
1967 }
1968 }
1969 }
1970
1971 if(scsi_status_is_good(result)) {
1972 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1973 (modepage == 6 || modepage == 8))) {
1974 /* Initio breakage? */
1975 header_length = 0;
1976 data->length = 13;
1977 data->medium_type = 0;
1978 data->device_specific = 0;
1979 data->longlba = 0;
1980 data->block_descriptor_length = 0;
1981 } else if(use_10_for_ms) {
1982 data->length = buffer[0]*256 + buffer[1] + 2;
1983 data->medium_type = buffer[2];
1984 data->device_specific = buffer[3];
1985 data->longlba = buffer[4] & 0x01;
1986 data->block_descriptor_length = buffer[6]*256
1987 + buffer[7];
1988 } else {
1989 data->length = buffer[0] + 1;
1990 data->medium_type = buffer[1];
1991 data->device_specific = buffer[2];
1992 data->block_descriptor_length = buffer[3];
1993 }
1994 data->header_length = header_length;
1995 }
1996
1997 return result;
1998}
1999EXPORT_SYMBOL(scsi_mode_sense);
2000
2001/**
2002 * scsi_test_unit_ready - test if unit is ready
2003 * @sdev: scsi device to change the state of.
2004 * @timeout: command timeout
2005 * @retries: number of retries before failing
2006 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2007 * returning sense. Make sure that this is cleared before passing
2008 * in.
2009 *
2010 * Returns zero if unsuccessful or an error if TUR failed. For
2011 * removable media, UNIT_ATTENTION sets ->changed flag.
2012 **/
2013int
2014scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2015 struct scsi_sense_hdr *sshdr_external)
2016{
2017 char cmd[] = {
2018 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2019 };
2020 struct scsi_sense_hdr *sshdr;
2021 int result;
2022
2023 if (!sshdr_external)
2024 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2025 else
2026 sshdr = sshdr_external;
2027
2028 /* try to eat the UNIT_ATTENTION if there are enough retries */
2029 do {
2030 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2031 timeout, retries, NULL);
2032 if (sdev->removable && scsi_sense_valid(sshdr) &&
2033 sshdr->sense_key == UNIT_ATTENTION)
2034 sdev->changed = 1;
2035 } while (scsi_sense_valid(sshdr) &&
2036 sshdr->sense_key == UNIT_ATTENTION && --retries);
2037
2038 if (!sshdr_external)
2039 kfree(sshdr);
2040 return result;
2041}
2042EXPORT_SYMBOL(scsi_test_unit_ready);
2043
2044/**
2045 * scsi_device_set_state - Take the given device through the device state model.
2046 * @sdev: scsi device to change the state of.
2047 * @state: state to change to.
2048 *
2049 * Returns zero if unsuccessful or an error if the requested
2050 * transition is illegal.
2051 */
2052int
2053scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2054{
2055 enum scsi_device_state oldstate = sdev->sdev_state;
2056
2057 if (state == oldstate)
2058 return 0;
2059
2060 switch (state) {
2061 case SDEV_CREATED:
2062 switch (oldstate) {
2063 case SDEV_CREATED_BLOCK:
2064 break;
2065 default:
2066 goto illegal;
2067 }
2068 break;
2069
2070 case SDEV_RUNNING:
2071 switch (oldstate) {
2072 case SDEV_CREATED:
2073 case SDEV_OFFLINE:
2074 case SDEV_QUIESCE:
2075 case SDEV_BLOCK:
2076 break;
2077 default:
2078 goto illegal;
2079 }
2080 break;
2081
2082 case SDEV_QUIESCE:
2083 switch (oldstate) {
2084 case SDEV_RUNNING:
2085 case SDEV_OFFLINE:
2086 break;
2087 default:
2088 goto illegal;
2089 }
2090 break;
2091
2092 case SDEV_OFFLINE:
2093 switch (oldstate) {
2094 case SDEV_CREATED:
2095 case SDEV_RUNNING:
2096 case SDEV_QUIESCE:
2097 case SDEV_BLOCK:
2098 break;
2099 default:
2100 goto illegal;
2101 }
2102 break;
2103
2104 case SDEV_BLOCK:
2105 switch (oldstate) {
2106 case SDEV_RUNNING:
2107 case SDEV_CREATED_BLOCK:
2108 break;
2109 default:
2110 goto illegal;
2111 }
2112 break;
2113
2114 case SDEV_CREATED_BLOCK:
2115 switch (oldstate) {
2116 case SDEV_CREATED:
2117 break;
2118 default:
2119 goto illegal;
2120 }
2121 break;
2122
2123 case SDEV_CANCEL:
2124 switch (oldstate) {
2125 case SDEV_CREATED:
2126 case SDEV_RUNNING:
2127 case SDEV_QUIESCE:
2128 case SDEV_OFFLINE:
2129 case SDEV_BLOCK:
2130 break;
2131 default:
2132 goto illegal;
2133 }
2134 break;
2135
2136 case SDEV_DEL:
2137 switch (oldstate) {
2138 case SDEV_CREATED:
2139 case SDEV_RUNNING:
2140 case SDEV_OFFLINE:
2141 case SDEV_CANCEL:
2142 break;
2143 default:
2144 goto illegal;
2145 }
2146 break;
2147
2148 }
2149 sdev->sdev_state = state;
2150 return 0;
2151
2152 illegal:
2153 SCSI_LOG_ERROR_RECOVERY(1,
2154 sdev_printk(KERN_ERR, sdev,
2155 "Illegal state transition %s->%s\n",
2156 scsi_device_state_name(oldstate),
2157 scsi_device_state_name(state))
2158 );
2159 return -EINVAL;
2160}
2161EXPORT_SYMBOL(scsi_device_set_state);
2162
2163/**
2164 * sdev_evt_emit - emit a single SCSI device uevent
2165 * @sdev: associated SCSI device
2166 * @evt: event to emit
2167 *
2168 * Send a single uevent (scsi_event) to the associated scsi_device.
2169 */
2170static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2171{
2172 int idx = 0;
2173 char *envp[3];
2174
2175 switch (evt->evt_type) {
2176 case SDEV_EVT_MEDIA_CHANGE:
2177 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2178 break;
2179
2180 default:
2181 /* do nothing */
2182 break;
2183 }
2184
2185 envp[idx++] = NULL;
2186
2187 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2188}
2189
2190/**
2191 * sdev_evt_thread - send a uevent for each scsi event
2192 * @work: work struct for scsi_device
2193 *
2194 * Dispatch queued events to their associated scsi_device kobjects
2195 * as uevents.
2196 */
2197void scsi_evt_thread(struct work_struct *work)
2198{
2199 struct scsi_device *sdev;
2200 LIST_HEAD(event_list);
2201
2202 sdev = container_of(work, struct scsi_device, event_work);
2203
2204 while (1) {
2205 struct scsi_event *evt;
2206 struct list_head *this, *tmp;
2207 unsigned long flags;
2208
2209 spin_lock_irqsave(&sdev->list_lock, flags);
2210 list_splice_init(&sdev->event_list, &event_list);
2211 spin_unlock_irqrestore(&sdev->list_lock, flags);
2212
2213 if (list_empty(&event_list))
2214 break;
2215
2216 list_for_each_safe(this, tmp, &event_list) {
2217 evt = list_entry(this, struct scsi_event, node);
2218 list_del(&evt->node);
2219 scsi_evt_emit(sdev, evt);
2220 kfree(evt);
2221 }
2222 }
2223}
2224
2225/**
2226 * sdev_evt_send - send asserted event to uevent thread
2227 * @sdev: scsi_device event occurred on
2228 * @evt: event to send
2229 *
2230 * Assert scsi device event asynchronously.
2231 */
2232void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2233{
2234 unsigned long flags;
2235
2236#if 0
2237 /* FIXME: currently this check eliminates all media change events
2238 * for polled devices. Need to update to discriminate between AN
2239 * and polled events */
2240 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2241 kfree(evt);
2242 return;
2243 }
2244#endif
2245
2246 spin_lock_irqsave(&sdev->list_lock, flags);
2247 list_add_tail(&evt->node, &sdev->event_list);
2248 schedule_work(&sdev->event_work);
2249 spin_unlock_irqrestore(&sdev->list_lock, flags);
2250}
2251EXPORT_SYMBOL_GPL(sdev_evt_send);
2252
2253/**
2254 * sdev_evt_alloc - allocate a new scsi event
2255 * @evt_type: type of event to allocate
2256 * @gfpflags: GFP flags for allocation
2257 *
2258 * Allocates and returns a new scsi_event.
2259 */
2260struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2261 gfp_t gfpflags)
2262{
2263 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2264 if (!evt)
2265 return NULL;
2266
2267 evt->evt_type = evt_type;
2268 INIT_LIST_HEAD(&evt->node);
2269
2270 /* evt_type-specific initialization, if any */
2271 switch (evt_type) {
2272 case SDEV_EVT_MEDIA_CHANGE:
2273 default:
2274 /* do nothing */
2275 break;
2276 }
2277
2278 return evt;
2279}
2280EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2281
2282/**
2283 * sdev_evt_send_simple - send asserted event to uevent thread
2284 * @sdev: scsi_device event occurred on
2285 * @evt_type: type of event to send
2286 * @gfpflags: GFP flags for allocation
2287 *
2288 * Assert scsi device event asynchronously, given an event type.
2289 */
2290void sdev_evt_send_simple(struct scsi_device *sdev,
2291 enum scsi_device_event evt_type, gfp_t gfpflags)
2292{
2293 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2294 if (!evt) {
2295 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2296 evt_type);
2297 return;
2298 }
2299
2300 sdev_evt_send(sdev, evt);
2301}
2302EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2303
2304/**
2305 * scsi_device_quiesce - Block user issued commands.
2306 * @sdev: scsi device to quiesce.
2307 *
2308 * This works by trying to transition to the SDEV_QUIESCE state
2309 * (which must be a legal transition). When the device is in this
2310 * state, only special requests will be accepted, all others will
2311 * be deferred. Since special requests may also be requeued requests,
2312 * a successful return doesn't guarantee the device will be
2313 * totally quiescent.
2314 *
2315 * Must be called with user context, may sleep.
2316 *
2317 * Returns zero if unsuccessful or an error if not.
2318 */
2319int
2320scsi_device_quiesce(struct scsi_device *sdev)
2321{
2322 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2323 if (err)
2324 return err;
2325
2326 scsi_run_queue(sdev->request_queue);
2327 while (sdev->device_busy) {
2328 msleep_interruptible(200);
2329 scsi_run_queue(sdev->request_queue);
2330 }
2331 return 0;
2332}
2333EXPORT_SYMBOL(scsi_device_quiesce);
2334
2335/**
2336 * scsi_device_resume - Restart user issued commands to a quiesced device.
2337 * @sdev: scsi device to resume.
2338 *
2339 * Moves the device from quiesced back to running and restarts the
2340 * queues.
2341 *
2342 * Must be called with user context, may sleep.
2343 */
2344void scsi_device_resume(struct scsi_device *sdev)
2345{
2346 /* check if the device state was mutated prior to resume, and if
2347 * so assume the state is being managed elsewhere (for example
2348 * device deleted during suspend)
2349 */
2350 if (sdev->sdev_state != SDEV_QUIESCE ||
2351 scsi_device_set_state(sdev, SDEV_RUNNING))
2352 return;
2353 scsi_run_queue(sdev->request_queue);
2354}
2355EXPORT_SYMBOL(scsi_device_resume);
2356
2357static void
2358device_quiesce_fn(struct scsi_device *sdev, void *data)
2359{
2360 scsi_device_quiesce(sdev);
2361}
2362
2363void
2364scsi_target_quiesce(struct scsi_target *starget)
2365{
2366 starget_for_each_device(starget, NULL, device_quiesce_fn);
2367}
2368EXPORT_SYMBOL(scsi_target_quiesce);
2369
2370static void
2371device_resume_fn(struct scsi_device *sdev, void *data)
2372{
2373 scsi_device_resume(sdev);
2374}
2375
2376void
2377scsi_target_resume(struct scsi_target *starget)
2378{
2379 starget_for_each_device(starget, NULL, device_resume_fn);
2380}
2381EXPORT_SYMBOL(scsi_target_resume);
2382
2383/**
2384 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2385 * @sdev: device to block
2386 *
2387 * Block request made by scsi lld's to temporarily stop all
2388 * scsi commands on the specified device. Called from interrupt
2389 * or normal process context.
2390 *
2391 * Returns zero if successful or error if not
2392 *
2393 * Notes:
2394 * This routine transitions the device to the SDEV_BLOCK state
2395 * (which must be a legal transition). When the device is in this
2396 * state, all commands are deferred until the scsi lld reenables
2397 * the device with scsi_device_unblock or device_block_tmo fires.
2398 * This routine assumes the host_lock is held on entry.
2399 */
2400int
2401scsi_internal_device_block(struct scsi_device *sdev)
2402{
2403 struct request_queue *q = sdev->request_queue;
2404 unsigned long flags;
2405 int err = 0;
2406
2407 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2408 if (err) {
2409 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2410
2411 if (err)
2412 return err;
2413 }
2414
2415 /*
2416 * The device has transitioned to SDEV_BLOCK. Stop the
2417 * block layer from calling the midlayer with this device's
2418 * request queue.
2419 */
2420 spin_lock_irqsave(q->queue_lock, flags);
2421 blk_stop_queue(q);
2422 spin_unlock_irqrestore(q->queue_lock, flags);
2423
2424 return 0;
2425}
2426EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2427
2428/**
2429 * scsi_internal_device_unblock - resume a device after a block request
2430 * @sdev: device to resume
2431 *
2432 * Called by scsi lld's or the midlayer to restart the device queue
2433 * for the previously suspended scsi device. Called from interrupt or
2434 * normal process context.
2435 *
2436 * Returns zero if successful or error if not.
2437 *
2438 * Notes:
2439 * This routine transitions the device to the SDEV_RUNNING state
2440 * (which must be a legal transition) allowing the midlayer to
2441 * goose the queue for this device. This routine assumes the
2442 * host_lock is held upon entry.
2443 */
2444int
2445scsi_internal_device_unblock(struct scsi_device *sdev)
2446{
2447 struct request_queue *q = sdev->request_queue;
2448 unsigned long flags;
2449
2450 /*
2451 * Try to transition the scsi device to SDEV_RUNNING
2452 * and goose the device queue if successful.
2453 */
2454 if (sdev->sdev_state == SDEV_BLOCK)
2455 sdev->sdev_state = SDEV_RUNNING;
2456 else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2457 sdev->sdev_state = SDEV_CREATED;
2458 else if (sdev->sdev_state != SDEV_CANCEL &&
2459 sdev->sdev_state != SDEV_OFFLINE)
2460 return -EINVAL;
2461
2462 spin_lock_irqsave(q->queue_lock, flags);
2463 blk_start_queue(q);
2464 spin_unlock_irqrestore(q->queue_lock, flags);
2465
2466 return 0;
2467}
2468EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2469
2470static void
2471device_block(struct scsi_device *sdev, void *data)
2472{
2473 scsi_internal_device_block(sdev);
2474}
2475
2476static int
2477target_block(struct device *dev, void *data)
2478{
2479 if (scsi_is_target_device(dev))
2480 starget_for_each_device(to_scsi_target(dev), NULL,
2481 device_block);
2482 return 0;
2483}
2484
2485void
2486scsi_target_block(struct device *dev)
2487{
2488 if (scsi_is_target_device(dev))
2489 starget_for_each_device(to_scsi_target(dev), NULL,
2490 device_block);
2491 else
2492 device_for_each_child(dev, NULL, target_block);
2493}
2494EXPORT_SYMBOL_GPL(scsi_target_block);
2495
2496static void
2497device_unblock(struct scsi_device *sdev, void *data)
2498{
2499 scsi_internal_device_unblock(sdev);
2500}
2501
2502static int
2503target_unblock(struct device *dev, void *data)
2504{
2505 if (scsi_is_target_device(dev))
2506 starget_for_each_device(to_scsi_target(dev), NULL,
2507 device_unblock);
2508 return 0;
2509}
2510
2511void
2512scsi_target_unblock(struct device *dev)
2513{
2514 if (scsi_is_target_device(dev))
2515 starget_for_each_device(to_scsi_target(dev), NULL,
2516 device_unblock);
2517 else
2518 device_for_each_child(dev, NULL, target_unblock);
2519}
2520EXPORT_SYMBOL_GPL(scsi_target_unblock);
2521
2522/**
2523 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2524 * @sgl: scatter-gather list
2525 * @sg_count: number of segments in sg
2526 * @offset: offset in bytes into sg, on return offset into the mapped area
2527 * @len: bytes to map, on return number of bytes mapped
2528 *
2529 * Returns virtual address of the start of the mapped page
2530 */
2531void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2532 size_t *offset, size_t *len)
2533{
2534 int i;
2535 size_t sg_len = 0, len_complete = 0;
2536 struct scatterlist *sg;
2537 struct page *page;
2538
2539 WARN_ON(!irqs_disabled());
2540
2541 for_each_sg(sgl, sg, sg_count, i) {
2542 len_complete = sg_len; /* Complete sg-entries */
2543 sg_len += sg->length;
2544 if (sg_len > *offset)
2545 break;
2546 }
2547
2548 if (unlikely(i == sg_count)) {
2549 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2550 "elements %d\n",
2551 __func__, sg_len, *offset, sg_count);
2552 WARN_ON(1);
2553 return NULL;
2554 }
2555
2556 /* Offset starting from the beginning of first page in this sg-entry */
2557 *offset = *offset - len_complete + sg->offset;
2558
2559 /* Assumption: contiguous pages can be accessed as "page + i" */
2560 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2561 *offset &= ~PAGE_MASK;
2562
2563 /* Bytes in this sg-entry from *offset to the end of the page */
2564 sg_len = PAGE_SIZE - *offset;
2565 if (*len > sg_len)
2566 *len = sg_len;
2567
2568 return kmap_atomic(page);
2569}
2570EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2571
2572/**
2573 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2574 * @virt: virtual address to be unmapped
2575 */
2576void scsi_kunmap_atomic_sg(void *virt)
2577{
2578 kunmap_atomic(virt);
2579}
2580EXPORT_SYMBOL(scsi_kunmap_atomic_sg);