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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12#include <linux/bio.h>
13#include <linux/bitops.h>
14#include <linux/blkdev.h>
15#include <linux/completion.h>
16#include <linux/kernel.h>
17#include <linux/export.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#include <linux/blk-mq.h>
24#include <linux/blk-integrity.h>
25#include <linux/ratelimit.h>
26#include <linux/unaligned.h>
27
28#include <scsi/scsi.h>
29#include <scsi/scsi_cmnd.h>
30#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_driver.h>
33#include <scsi/scsi_eh.h>
34#include <scsi/scsi_host.h>
35#include <scsi/scsi_transport.h> /* scsi_init_limits() */
36#include <scsi/scsi_dh.h>
37
38#include <trace/events/scsi.h>
39
40#include "scsi_debugfs.h"
41#include "scsi_priv.h"
42#include "scsi_logging.h"
43
44/*
45 * Size of integrity metadata is usually small, 1 inline sg should
46 * cover normal cases.
47 */
48#ifdef CONFIG_ARCH_NO_SG_CHAIN
49#define SCSI_INLINE_PROT_SG_CNT 0
50#define SCSI_INLINE_SG_CNT 0
51#else
52#define SCSI_INLINE_PROT_SG_CNT 1
53#define SCSI_INLINE_SG_CNT 2
54#endif
55
56static struct kmem_cache *scsi_sense_cache;
57static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
61int scsi_init_sense_cache(struct Scsi_Host *shost)
62{
63 int ret = 0;
64
65 mutex_lock(&scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
67 scsi_sense_cache =
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
72 ret = -ENOMEM;
73 }
74 mutex_unlock(&scsi_sense_cache_mutex);
75 return ret;
76}
77
78static void
79scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80{
81 struct Scsi_Host *host = cmd->device->host;
82 struct scsi_device *device = cmd->device;
83 struct scsi_target *starget = scsi_target(device);
84
85 /*
86 * Set the appropriate busy bit for the device/host.
87 *
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
90 *
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
97 */
98 switch (reason) {
99 case SCSI_MLQUEUE_HOST_BUSY:
100 atomic_set(&host->host_blocked, host->max_host_blocked);
101 break;
102 case SCSI_MLQUEUE_DEVICE_BUSY:
103 case SCSI_MLQUEUE_EH_RETRY:
104 atomic_set(&device->device_blocked,
105 device->max_device_blocked);
106 break;
107 case SCSI_MLQUEUE_TARGET_BUSY:
108 atomic_set(&starget->target_blocked,
109 starget->max_target_blocked);
110 break;
111 }
112}
113
114static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115{
116 struct request *rq = scsi_cmd_to_rq(cmd);
117
118 if (rq->rq_flags & RQF_DONTPREP) {
119 rq->rq_flags &= ~RQF_DONTPREP;
120 scsi_mq_uninit_cmd(cmd);
121 } else {
122 WARN_ON_ONCE(true);
123 }
124
125 blk_mq_requeue_request(rq, false);
126 if (!scsi_host_in_recovery(cmd->device->host))
127 blk_mq_delay_kick_requeue_list(rq->q, msecs);
128}
129
130/**
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
135 *
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
140 * file.
141 */
142static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143{
144 struct scsi_device *device = cmd->device;
145
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 "Inserting command %p into mlqueue\n", cmd));
148
149 scsi_set_blocked(cmd, reason);
150
151 /*
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
154 */
155 if (unbusy)
156 scsi_device_unbusy(device, cmd);
157
158 /*
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_mq_destroy_queue() finishes.
163 */
164 cmd->result = 0;
165
166 blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
167 !scsi_host_in_recovery(cmd->device->host));
168}
169
170/**
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
174 *
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
178 *
179 * Context: This could be called either from an interrupt context or a normal
180 * process context.
181 */
182void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183{
184 __scsi_queue_insert(cmd, reason, true);
185}
186
187void scsi_failures_reset_retries(struct scsi_failures *failures)
188{
189 struct scsi_failure *failure;
190
191 failures->total_retries = 0;
192
193 for (failure = failures->failure_definitions; failure->result;
194 failure++)
195 failure->retries = 0;
196}
197EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);
198
199/**
200 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
201 * @scmd: scsi_cmnd to check.
202 * @failures: scsi_failures struct that lists failures to check for.
203 *
204 * Returns -EAGAIN if the caller should retry else 0.
205 */
206static int scsi_check_passthrough(struct scsi_cmnd *scmd,
207 struct scsi_failures *failures)
208{
209 struct scsi_failure *failure;
210 struct scsi_sense_hdr sshdr;
211 enum sam_status status;
212
213 if (!scmd->result)
214 return 0;
215
216 if (!failures)
217 return 0;
218
219 for (failure = failures->failure_definitions; failure->result;
220 failure++) {
221 if (failure->result == SCMD_FAILURE_RESULT_ANY)
222 goto maybe_retry;
223
224 if (host_byte(scmd->result) &&
225 host_byte(scmd->result) == host_byte(failure->result))
226 goto maybe_retry;
227
228 status = status_byte(scmd->result);
229 if (!status)
230 continue;
231
232 if (failure->result == SCMD_FAILURE_STAT_ANY &&
233 !scsi_status_is_good(scmd->result))
234 goto maybe_retry;
235
236 if (status != status_byte(failure->result))
237 continue;
238
239 if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||
240 failure->sense == SCMD_FAILURE_SENSE_ANY)
241 goto maybe_retry;
242
243 if (!scsi_command_normalize_sense(scmd, &sshdr))
244 return 0;
245
246 if (failure->sense != sshdr.sense_key)
247 continue;
248
249 if (failure->asc == SCMD_FAILURE_ASC_ANY)
250 goto maybe_retry;
251
252 if (failure->asc != sshdr.asc)
253 continue;
254
255 if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||
256 failure->ascq == sshdr.ascq)
257 goto maybe_retry;
258 }
259
260 return 0;
261
262maybe_retry:
263 if (failure->allowed) {
264 if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||
265 ++failure->retries <= failure->allowed)
266 return -EAGAIN;
267 } else {
268 if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||
269 ++failures->total_retries <= failures->total_allowed)
270 return -EAGAIN;
271 }
272
273 return 0;
274}
275
276/**
277 * scsi_execute_cmd - insert request and wait for the result
278 * @sdev: scsi_device
279 * @cmd: scsi command
280 * @opf: block layer request cmd_flags
281 * @buffer: data buffer
282 * @bufflen: len of buffer
283 * @timeout: request timeout in HZ
284 * @ml_retries: number of times SCSI midlayer will retry request
285 * @args: Optional args. See struct definition for field descriptions
286 *
287 * Returns the scsi_cmnd result field if a command was executed, or a negative
288 * Linux error code if we didn't get that far.
289 */
290int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
291 blk_opf_t opf, void *buffer, unsigned int bufflen,
292 int timeout, int ml_retries,
293 const struct scsi_exec_args *args)
294{
295 static const struct scsi_exec_args default_args;
296 struct request *req;
297 struct scsi_cmnd *scmd;
298 int ret;
299
300 if (!args)
301 args = &default_args;
302 else if (WARN_ON_ONCE(args->sense &&
303 args->sense_len != SCSI_SENSE_BUFFERSIZE))
304 return -EINVAL;
305
306retry:
307 req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
308 if (IS_ERR(req))
309 return PTR_ERR(req);
310
311 if (bufflen) {
312 ret = blk_rq_map_kern(sdev->request_queue, req,
313 buffer, bufflen, GFP_NOIO);
314 if (ret)
315 goto out;
316 }
317 scmd = blk_mq_rq_to_pdu(req);
318 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
319 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
320 scmd->allowed = ml_retries;
321 scmd->flags |= args->scmd_flags;
322 req->timeout = timeout;
323 req->rq_flags |= RQF_QUIET;
324
325 /*
326 * head injection *required* here otherwise quiesce won't work
327 */
328 blk_execute_rq(req, true);
329
330 if (scsi_check_passthrough(scmd, args->failures) == -EAGAIN) {
331 blk_mq_free_request(req);
332 goto retry;
333 }
334
335 /*
336 * Some devices (USB mass-storage in particular) may transfer
337 * garbage data together with a residue indicating that the data
338 * is invalid. Prevent the garbage from being misinterpreted
339 * and prevent security leaks by zeroing out the excess data.
340 */
341 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
342 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
343
344 if (args->resid)
345 *args->resid = scmd->resid_len;
346 if (args->sense)
347 memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
348 if (args->sshdr)
349 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
350 args->sshdr);
351
352 ret = scmd->result;
353 out:
354 blk_mq_free_request(req);
355
356 return ret;
357}
358EXPORT_SYMBOL(scsi_execute_cmd);
359
360/*
361 * Wake up the error handler if necessary. Avoid as follows that the error
362 * handler is not woken up if host in-flight requests number ==
363 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
364 * with an RCU read lock in this function to ensure that this function in
365 * its entirety either finishes before scsi_eh_scmd_add() increases the
366 * host_failed counter or that it notices the shost state change made by
367 * scsi_eh_scmd_add().
368 */
369static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
370{
371 unsigned long flags;
372
373 rcu_read_lock();
374 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
375 if (unlikely(scsi_host_in_recovery(shost))) {
376 unsigned int busy = scsi_host_busy(shost);
377
378 spin_lock_irqsave(shost->host_lock, flags);
379 if (shost->host_failed || shost->host_eh_scheduled)
380 scsi_eh_wakeup(shost, busy);
381 spin_unlock_irqrestore(shost->host_lock, flags);
382 }
383 rcu_read_unlock();
384}
385
386void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
387{
388 struct Scsi_Host *shost = sdev->host;
389 struct scsi_target *starget = scsi_target(sdev);
390
391 scsi_dec_host_busy(shost, cmd);
392
393 if (starget->can_queue > 0)
394 atomic_dec(&starget->target_busy);
395
396 sbitmap_put(&sdev->budget_map, cmd->budget_token);
397 cmd->budget_token = -1;
398}
399
400/*
401 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
402 * interrupts disabled.
403 */
404static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
405{
406 struct scsi_device *current_sdev = data;
407
408 if (sdev != current_sdev)
409 blk_mq_run_hw_queues(sdev->request_queue, true);
410}
411
412/*
413 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
414 * and call blk_run_queue for all the scsi_devices on the target -
415 * including current_sdev first.
416 *
417 * Called with *no* scsi locks held.
418 */
419static void scsi_single_lun_run(struct scsi_device *current_sdev)
420{
421 struct Scsi_Host *shost = current_sdev->host;
422 struct scsi_target *starget = scsi_target(current_sdev);
423 unsigned long flags;
424
425 spin_lock_irqsave(shost->host_lock, flags);
426 starget->starget_sdev_user = NULL;
427 spin_unlock_irqrestore(shost->host_lock, flags);
428
429 /*
430 * Call blk_run_queue for all LUNs on the target, starting with
431 * current_sdev. We race with others (to set starget_sdev_user),
432 * but in most cases, we will be first. Ideally, each LU on the
433 * target would get some limited time or requests on the target.
434 */
435 blk_mq_run_hw_queues(current_sdev->request_queue,
436 shost->queuecommand_may_block);
437
438 spin_lock_irqsave(shost->host_lock, flags);
439 if (!starget->starget_sdev_user)
440 __starget_for_each_device(starget, current_sdev,
441 scsi_kick_sdev_queue);
442 spin_unlock_irqrestore(shost->host_lock, flags);
443}
444
445static inline bool scsi_device_is_busy(struct scsi_device *sdev)
446{
447 if (scsi_device_busy(sdev) >= sdev->queue_depth)
448 return true;
449 if (atomic_read(&sdev->device_blocked) > 0)
450 return true;
451 return false;
452}
453
454static inline bool scsi_target_is_busy(struct scsi_target *starget)
455{
456 if (starget->can_queue > 0) {
457 if (atomic_read(&starget->target_busy) >= starget->can_queue)
458 return true;
459 if (atomic_read(&starget->target_blocked) > 0)
460 return true;
461 }
462 return false;
463}
464
465static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
466{
467 if (atomic_read(&shost->host_blocked) > 0)
468 return true;
469 if (shost->host_self_blocked)
470 return true;
471 return false;
472}
473
474static void scsi_starved_list_run(struct Scsi_Host *shost)
475{
476 LIST_HEAD(starved_list);
477 struct scsi_device *sdev;
478 unsigned long flags;
479
480 spin_lock_irqsave(shost->host_lock, flags);
481 list_splice_init(&shost->starved_list, &starved_list);
482
483 while (!list_empty(&starved_list)) {
484 struct request_queue *slq;
485
486 /*
487 * As long as shost is accepting commands and we have
488 * starved queues, call blk_run_queue. scsi_request_fn
489 * drops the queue_lock and can add us back to the
490 * starved_list.
491 *
492 * host_lock protects the starved_list and starved_entry.
493 * scsi_request_fn must get the host_lock before checking
494 * or modifying starved_list or starved_entry.
495 */
496 if (scsi_host_is_busy(shost))
497 break;
498
499 sdev = list_entry(starved_list.next,
500 struct scsi_device, starved_entry);
501 list_del_init(&sdev->starved_entry);
502 if (scsi_target_is_busy(scsi_target(sdev))) {
503 list_move_tail(&sdev->starved_entry,
504 &shost->starved_list);
505 continue;
506 }
507
508 /*
509 * Once we drop the host lock, a racing scsi_remove_device()
510 * call may remove the sdev from the starved list and destroy
511 * it and the queue. Mitigate by taking a reference to the
512 * queue and never touching the sdev again after we drop the
513 * host lock. Note: if __scsi_remove_device() invokes
514 * blk_mq_destroy_queue() before the queue is run from this
515 * function then blk_run_queue() will return immediately since
516 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
517 */
518 slq = sdev->request_queue;
519 if (!blk_get_queue(slq))
520 continue;
521 spin_unlock_irqrestore(shost->host_lock, flags);
522
523 blk_mq_run_hw_queues(slq, false);
524 blk_put_queue(slq);
525
526 spin_lock_irqsave(shost->host_lock, flags);
527 }
528 /* put any unprocessed entries back */
529 list_splice(&starved_list, &shost->starved_list);
530 spin_unlock_irqrestore(shost->host_lock, flags);
531}
532
533/**
534 * scsi_run_queue - Select a proper request queue to serve next.
535 * @q: last request's queue
536 *
537 * The previous command was completely finished, start a new one if possible.
538 */
539static void scsi_run_queue(struct request_queue *q)
540{
541 struct scsi_device *sdev = q->queuedata;
542
543 if (scsi_target(sdev)->single_lun)
544 scsi_single_lun_run(sdev);
545 if (!list_empty(&sdev->host->starved_list))
546 scsi_starved_list_run(sdev->host);
547
548 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
549 blk_mq_kick_requeue_list(q);
550}
551
552void scsi_requeue_run_queue(struct work_struct *work)
553{
554 struct scsi_device *sdev;
555 struct request_queue *q;
556
557 sdev = container_of(work, struct scsi_device, requeue_work);
558 q = sdev->request_queue;
559 scsi_run_queue(q);
560}
561
562void scsi_run_host_queues(struct Scsi_Host *shost)
563{
564 struct scsi_device *sdev;
565
566 shost_for_each_device(sdev, shost)
567 scsi_run_queue(sdev->request_queue);
568}
569
570static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
571{
572 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
573 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
574
575 if (drv->uninit_command)
576 drv->uninit_command(cmd);
577 }
578}
579
580void scsi_free_sgtables(struct scsi_cmnd *cmd)
581{
582 if (cmd->sdb.table.nents)
583 sg_free_table_chained(&cmd->sdb.table,
584 SCSI_INLINE_SG_CNT);
585 if (scsi_prot_sg_count(cmd))
586 sg_free_table_chained(&cmd->prot_sdb->table,
587 SCSI_INLINE_PROT_SG_CNT);
588}
589EXPORT_SYMBOL_GPL(scsi_free_sgtables);
590
591static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
592{
593 scsi_free_sgtables(cmd);
594 scsi_uninit_cmd(cmd);
595}
596
597static void scsi_run_queue_async(struct scsi_device *sdev)
598{
599 if (scsi_host_in_recovery(sdev->host))
600 return;
601
602 if (scsi_target(sdev)->single_lun ||
603 !list_empty(&sdev->host->starved_list)) {
604 kblockd_schedule_work(&sdev->requeue_work);
605 } else {
606 /*
607 * smp_mb() present in sbitmap_queue_clear() or implied in
608 * .end_io is for ordering writing .device_busy in
609 * scsi_device_unbusy() and reading sdev->restarts.
610 */
611 int old = atomic_read(&sdev->restarts);
612
613 /*
614 * ->restarts has to be kept as non-zero if new budget
615 * contention occurs.
616 *
617 * No need to run queue when either another re-run
618 * queue wins in updating ->restarts or a new budget
619 * contention occurs.
620 */
621 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
622 blk_mq_run_hw_queues(sdev->request_queue, true);
623 }
624}
625
626/* Returns false when no more bytes to process, true if there are more */
627static bool scsi_end_request(struct request *req, blk_status_t error,
628 unsigned int bytes)
629{
630 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
631 struct scsi_device *sdev = cmd->device;
632 struct request_queue *q = sdev->request_queue;
633
634 if (blk_update_request(req, error, bytes))
635 return true;
636
637 if (q->limits.features & BLK_FEAT_ADD_RANDOM)
638 add_disk_randomness(req->q->disk);
639
640 WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
641 !(cmd->flags & SCMD_INITIALIZED));
642 cmd->flags = 0;
643
644 /*
645 * Calling rcu_barrier() is not necessary here because the
646 * SCSI error handler guarantees that the function called by
647 * call_rcu() has been called before scsi_end_request() is
648 * called.
649 */
650 destroy_rcu_head(&cmd->rcu);
651
652 /*
653 * In the MQ case the command gets freed by __blk_mq_end_request,
654 * so we have to do all cleanup that depends on it earlier.
655 *
656 * We also can't kick the queues from irq context, so we
657 * will have to defer it to a workqueue.
658 */
659 scsi_mq_uninit_cmd(cmd);
660
661 /*
662 * queue is still alive, so grab the ref for preventing it
663 * from being cleaned up during running queue.
664 */
665 percpu_ref_get(&q->q_usage_counter);
666
667 __blk_mq_end_request(req, error);
668
669 scsi_run_queue_async(sdev);
670
671 percpu_ref_put(&q->q_usage_counter);
672 return false;
673}
674
675/**
676 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
677 * @result: scsi error code
678 *
679 * Translate a SCSI result code into a blk_status_t value.
680 */
681static blk_status_t scsi_result_to_blk_status(int result)
682{
683 /*
684 * Check the scsi-ml byte first in case we converted a host or status
685 * byte.
686 */
687 switch (scsi_ml_byte(result)) {
688 case SCSIML_STAT_OK:
689 break;
690 case SCSIML_STAT_RESV_CONFLICT:
691 return BLK_STS_RESV_CONFLICT;
692 case SCSIML_STAT_NOSPC:
693 return BLK_STS_NOSPC;
694 case SCSIML_STAT_MED_ERROR:
695 return BLK_STS_MEDIUM;
696 case SCSIML_STAT_TGT_FAILURE:
697 return BLK_STS_TARGET;
698 case SCSIML_STAT_DL_TIMEOUT:
699 return BLK_STS_DURATION_LIMIT;
700 }
701
702 switch (host_byte(result)) {
703 case DID_OK:
704 if (scsi_status_is_good(result))
705 return BLK_STS_OK;
706 return BLK_STS_IOERR;
707 case DID_TRANSPORT_FAILFAST:
708 case DID_TRANSPORT_MARGINAL:
709 return BLK_STS_TRANSPORT;
710 default:
711 return BLK_STS_IOERR;
712 }
713}
714
715/**
716 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
717 * @rq: request to examine
718 *
719 * Description:
720 * A request could be merge of IOs which require different failure
721 * handling. This function determines the number of bytes which
722 * can be failed from the beginning of the request without
723 * crossing into area which need to be retried further.
724 *
725 * Return:
726 * The number of bytes to fail.
727 */
728static unsigned int scsi_rq_err_bytes(const struct request *rq)
729{
730 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
731 unsigned int bytes = 0;
732 struct bio *bio;
733
734 if (!(rq->rq_flags & RQF_MIXED_MERGE))
735 return blk_rq_bytes(rq);
736
737 /*
738 * Currently the only 'mixing' which can happen is between
739 * different fastfail types. We can safely fail portions
740 * which have all the failfast bits that the first one has -
741 * the ones which are at least as eager to fail as the first
742 * one.
743 */
744 for (bio = rq->bio; bio; bio = bio->bi_next) {
745 if ((bio->bi_opf & ff) != ff)
746 break;
747 bytes += bio->bi_iter.bi_size;
748 }
749
750 /* this could lead to infinite loop */
751 BUG_ON(blk_rq_bytes(rq) && !bytes);
752 return bytes;
753}
754
755static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
756{
757 struct request *req = scsi_cmd_to_rq(cmd);
758 unsigned long wait_for;
759
760 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
761 return false;
762
763 wait_for = (cmd->allowed + 1) * req->timeout;
764 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
765 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
766 wait_for/HZ);
767 return true;
768 }
769 return false;
770}
771
772/*
773 * When ALUA transition state is returned, reprep the cmd to
774 * use the ALUA handler's transition timeout. Delay the reprep
775 * 1 sec to avoid aggressive retries of the target in that
776 * state.
777 */
778#define ALUA_TRANSITION_REPREP_DELAY 1000
779
780/* Helper for scsi_io_completion() when special action required. */
781static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
782{
783 struct request *req = scsi_cmd_to_rq(cmd);
784 int level = 0;
785 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
786 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
787 struct scsi_sense_hdr sshdr;
788 bool sense_valid;
789 bool sense_current = true; /* false implies "deferred sense" */
790 blk_status_t blk_stat;
791
792 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
793 if (sense_valid)
794 sense_current = !scsi_sense_is_deferred(&sshdr);
795
796 blk_stat = scsi_result_to_blk_status(result);
797
798 if (host_byte(result) == DID_RESET) {
799 /* Third party bus reset or reset for error recovery
800 * reasons. Just retry the command and see what
801 * happens.
802 */
803 action = ACTION_RETRY;
804 } else if (sense_valid && sense_current) {
805 switch (sshdr.sense_key) {
806 case UNIT_ATTENTION:
807 if (cmd->device->removable) {
808 /* Detected disc change. Set a bit
809 * and quietly refuse further access.
810 */
811 cmd->device->changed = 1;
812 action = ACTION_FAIL;
813 } else {
814 /* Must have been a power glitch, or a
815 * bus reset. Could not have been a
816 * media change, so we just retry the
817 * command and see what happens.
818 */
819 action = ACTION_RETRY;
820 }
821 break;
822 case ILLEGAL_REQUEST:
823 /* If we had an ILLEGAL REQUEST returned, then
824 * we may have performed an unsupported
825 * command. The only thing this should be
826 * would be a ten byte read where only a six
827 * byte read was supported. Also, on a system
828 * where READ CAPACITY failed, we may have
829 * read past the end of the disk.
830 */
831 if ((cmd->device->use_10_for_rw &&
832 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
833 (cmd->cmnd[0] == READ_10 ||
834 cmd->cmnd[0] == WRITE_10)) {
835 /* This will issue a new 6-byte command. */
836 cmd->device->use_10_for_rw = 0;
837 action = ACTION_REPREP;
838 } else if (sshdr.asc == 0x10) /* DIX */ {
839 action = ACTION_FAIL;
840 blk_stat = BLK_STS_PROTECTION;
841 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
842 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
843 action = ACTION_FAIL;
844 blk_stat = BLK_STS_TARGET;
845 } else
846 action = ACTION_FAIL;
847 break;
848 case ABORTED_COMMAND:
849 action = ACTION_FAIL;
850 if (sshdr.asc == 0x10) /* DIF */
851 blk_stat = BLK_STS_PROTECTION;
852 break;
853 case NOT_READY:
854 /* If the device is in the process of becoming
855 * ready, or has a temporary blockage, retry.
856 */
857 if (sshdr.asc == 0x04) {
858 switch (sshdr.ascq) {
859 case 0x01: /* becoming ready */
860 case 0x04: /* format in progress */
861 case 0x05: /* rebuild in progress */
862 case 0x06: /* recalculation in progress */
863 case 0x07: /* operation in progress */
864 case 0x08: /* Long write in progress */
865 case 0x09: /* self test in progress */
866 case 0x11: /* notify (enable spinup) required */
867 case 0x14: /* space allocation in progress */
868 case 0x1a: /* start stop unit in progress */
869 case 0x1b: /* sanitize in progress */
870 case 0x1d: /* configuration in progress */
871 action = ACTION_DELAYED_RETRY;
872 break;
873 case 0x0a: /* ALUA state transition */
874 action = ACTION_DELAYED_REPREP;
875 break;
876 /*
877 * Depopulation might take many hours,
878 * thus it is not worthwhile to retry.
879 */
880 case 0x24: /* depopulation in progress */
881 case 0x25: /* depopulation restore in progress */
882 fallthrough;
883 default:
884 action = ACTION_FAIL;
885 break;
886 }
887 } else
888 action = ACTION_FAIL;
889 break;
890 case VOLUME_OVERFLOW:
891 /* See SSC3rXX or current. */
892 action = ACTION_FAIL;
893 break;
894 case DATA_PROTECT:
895 action = ACTION_FAIL;
896 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
897 (sshdr.asc == 0x55 &&
898 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
899 /* Insufficient zone resources */
900 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
901 }
902 break;
903 case COMPLETED:
904 fallthrough;
905 default:
906 action = ACTION_FAIL;
907 break;
908 }
909 } else
910 action = ACTION_FAIL;
911
912 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
913 action = ACTION_FAIL;
914
915 switch (action) {
916 case ACTION_FAIL:
917 /* Give up and fail the remainder of the request */
918 if (!(req->rq_flags & RQF_QUIET)) {
919 static DEFINE_RATELIMIT_STATE(_rs,
920 DEFAULT_RATELIMIT_INTERVAL,
921 DEFAULT_RATELIMIT_BURST);
922
923 if (unlikely(scsi_logging_level))
924 level =
925 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
926 SCSI_LOG_MLCOMPLETE_BITS);
927
928 /*
929 * if logging is enabled the failure will be printed
930 * in scsi_log_completion(), so avoid duplicate messages
931 */
932 if (!level && __ratelimit(&_rs)) {
933 scsi_print_result(cmd, NULL, FAILED);
934 if (sense_valid)
935 scsi_print_sense(cmd);
936 scsi_print_command(cmd);
937 }
938 }
939 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
940 return;
941 fallthrough;
942 case ACTION_REPREP:
943 scsi_mq_requeue_cmd(cmd, 0);
944 break;
945 case ACTION_DELAYED_REPREP:
946 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
947 break;
948 case ACTION_RETRY:
949 /* Retry the same command immediately */
950 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
951 break;
952 case ACTION_DELAYED_RETRY:
953 /* Retry the same command after a delay */
954 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
955 break;
956 }
957}
958
959/*
960 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
961 * new result that may suppress further error checking. Also modifies
962 * *blk_statp in some cases.
963 */
964static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
965 blk_status_t *blk_statp)
966{
967 bool sense_valid;
968 bool sense_current = true; /* false implies "deferred sense" */
969 struct request *req = scsi_cmd_to_rq(cmd);
970 struct scsi_sense_hdr sshdr;
971
972 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
973 if (sense_valid)
974 sense_current = !scsi_sense_is_deferred(&sshdr);
975
976 if (blk_rq_is_passthrough(req)) {
977 if (sense_valid) {
978 /*
979 * SG_IO wants current and deferred errors
980 */
981 cmd->sense_len = min(8 + cmd->sense_buffer[7],
982 SCSI_SENSE_BUFFERSIZE);
983 }
984 if (sense_current)
985 *blk_statp = scsi_result_to_blk_status(result);
986 } else if (blk_rq_bytes(req) == 0 && sense_current) {
987 /*
988 * Flush commands do not transfers any data, and thus cannot use
989 * good_bytes != blk_rq_bytes(req) as the signal for an error.
990 * This sets *blk_statp explicitly for the problem case.
991 */
992 *blk_statp = scsi_result_to_blk_status(result);
993 }
994 /*
995 * Recovered errors need reporting, but they're always treated as
996 * success, so fiddle the result code here. For passthrough requests
997 * we already took a copy of the original into sreq->result which
998 * is what gets returned to the user
999 */
1000 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
1001 bool do_print = true;
1002 /*
1003 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
1004 * skip print since caller wants ATA registers. Only occurs
1005 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
1006 */
1007 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1008 do_print = false;
1009 else if (req->rq_flags & RQF_QUIET)
1010 do_print = false;
1011 if (do_print)
1012 scsi_print_sense(cmd);
1013 result = 0;
1014 /* for passthrough, *blk_statp may be set */
1015 *blk_statp = BLK_STS_OK;
1016 }
1017 /*
1018 * Another corner case: the SCSI status byte is non-zero but 'good'.
1019 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1020 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1021 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1022 * intermediate statuses (both obsolete in SAM-4) as good.
1023 */
1024 if ((result & 0xff) && scsi_status_is_good(result)) {
1025 result = 0;
1026 *blk_statp = BLK_STS_OK;
1027 }
1028 return result;
1029}
1030
1031/**
1032 * scsi_io_completion - Completion processing for SCSI commands.
1033 * @cmd: command that is finished.
1034 * @good_bytes: number of processed bytes.
1035 *
1036 * We will finish off the specified number of sectors. If we are done, the
1037 * command block will be released and the queue function will be goosed. If we
1038 * are not done then we have to figure out what to do next:
1039 *
1040 * a) We can call scsi_mq_requeue_cmd(). The request will be
1041 * unprepared and put back on the queue. Then a new command will
1042 * be created for it. This should be used if we made forward
1043 * progress, or if we want to switch from READ(10) to READ(6) for
1044 * example.
1045 *
1046 * b) We can call scsi_io_completion_action(). The request will be
1047 * put back on the queue and retried using the same command as
1048 * before, possibly after a delay.
1049 *
1050 * c) We can call scsi_end_request() with blk_stat other than
1051 * BLK_STS_OK, to fail the remainder of the request.
1052 */
1053void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1054{
1055 int result = cmd->result;
1056 struct request *req = scsi_cmd_to_rq(cmd);
1057 blk_status_t blk_stat = BLK_STS_OK;
1058
1059 if (unlikely(result)) /* a nz result may or may not be an error */
1060 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1061
1062 /*
1063 * Next deal with any sectors which we were able to correctly
1064 * handle.
1065 */
1066 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1067 "%u sectors total, %d bytes done.\n",
1068 blk_rq_sectors(req), good_bytes));
1069
1070 /*
1071 * Failed, zero length commands always need to drop down
1072 * to retry code. Fast path should return in this block.
1073 */
1074 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1075 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1076 return; /* no bytes remaining */
1077 }
1078
1079 /* Kill remainder if no retries. */
1080 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1081 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
1082 WARN_ONCE(true,
1083 "Bytes remaining after failed, no-retry command");
1084 return;
1085 }
1086
1087 /*
1088 * If there had been no error, but we have leftover bytes in the
1089 * request just queue the command up again.
1090 */
1091 if (likely(result == 0))
1092 scsi_mq_requeue_cmd(cmd, 0);
1093 else
1094 scsi_io_completion_action(cmd, result);
1095}
1096
1097static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1098 struct request *rq)
1099{
1100 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1101 !op_is_write(req_op(rq)) &&
1102 sdev->host->hostt->dma_need_drain(rq);
1103}
1104
1105/**
1106 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1107 * @cmd: SCSI command data structure to initialize.
1108 *
1109 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1110 * for @cmd.
1111 *
1112 * Returns:
1113 * * BLK_STS_OK - on success
1114 * * BLK_STS_RESOURCE - if the failure is retryable
1115 * * BLK_STS_IOERR - if the failure is fatal
1116 */
1117blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1118{
1119 struct scsi_device *sdev = cmd->device;
1120 struct request *rq = scsi_cmd_to_rq(cmd);
1121 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1122 struct scatterlist *last_sg = NULL;
1123 blk_status_t ret;
1124 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1125 int count;
1126
1127 if (WARN_ON_ONCE(!nr_segs))
1128 return BLK_STS_IOERR;
1129
1130 /*
1131 * Make sure there is space for the drain. The driver must adjust
1132 * max_hw_segments to be prepared for this.
1133 */
1134 if (need_drain)
1135 nr_segs++;
1136
1137 /*
1138 * If sg table allocation fails, requeue request later.
1139 */
1140 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1141 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1142 return BLK_STS_RESOURCE;
1143
1144 /*
1145 * Next, walk the list, and fill in the addresses and sizes of
1146 * each segment.
1147 */
1148 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1149
1150 if (blk_rq_bytes(rq) & rq->q->limits.dma_pad_mask) {
1151 unsigned int pad_len =
1152 (rq->q->limits.dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1153
1154 last_sg->length += pad_len;
1155 cmd->extra_len += pad_len;
1156 }
1157
1158 if (need_drain) {
1159 sg_unmark_end(last_sg);
1160 last_sg = sg_next(last_sg);
1161 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1162 sg_mark_end(last_sg);
1163
1164 cmd->extra_len += sdev->dma_drain_len;
1165 count++;
1166 }
1167
1168 BUG_ON(count > cmd->sdb.table.nents);
1169 cmd->sdb.table.nents = count;
1170 cmd->sdb.length = blk_rq_payload_bytes(rq);
1171
1172 if (blk_integrity_rq(rq)) {
1173 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1174
1175 if (WARN_ON_ONCE(!prot_sdb)) {
1176 /*
1177 * This can happen if someone (e.g. multipath)
1178 * queues a command to a device on an adapter
1179 * that does not support DIX.
1180 */
1181 ret = BLK_STS_IOERR;
1182 goto out_free_sgtables;
1183 }
1184
1185 if (sg_alloc_table_chained(&prot_sdb->table,
1186 rq->nr_integrity_segments,
1187 prot_sdb->table.sgl,
1188 SCSI_INLINE_PROT_SG_CNT)) {
1189 ret = BLK_STS_RESOURCE;
1190 goto out_free_sgtables;
1191 }
1192
1193 count = blk_rq_map_integrity_sg(rq, prot_sdb->table.sgl);
1194 cmd->prot_sdb = prot_sdb;
1195 cmd->prot_sdb->table.nents = count;
1196 }
1197
1198 return BLK_STS_OK;
1199out_free_sgtables:
1200 scsi_free_sgtables(cmd);
1201 return ret;
1202}
1203EXPORT_SYMBOL(scsi_alloc_sgtables);
1204
1205/**
1206 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1207 * @rq: Request associated with the SCSI command to be initialized.
1208 *
1209 * This function initializes the members of struct scsi_cmnd that must be
1210 * initialized before request processing starts and that won't be
1211 * reinitialized if a SCSI command is requeued.
1212 */
1213static void scsi_initialize_rq(struct request *rq)
1214{
1215 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1216
1217 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1218 cmd->cmd_len = MAX_COMMAND_SIZE;
1219 cmd->sense_len = 0;
1220 init_rcu_head(&cmd->rcu);
1221 cmd->jiffies_at_alloc = jiffies;
1222 cmd->retries = 0;
1223}
1224
1225struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1226 blk_mq_req_flags_t flags)
1227{
1228 struct request *rq;
1229
1230 rq = blk_mq_alloc_request(q, opf, flags);
1231 if (!IS_ERR(rq))
1232 scsi_initialize_rq(rq);
1233 return rq;
1234}
1235EXPORT_SYMBOL_GPL(scsi_alloc_request);
1236
1237/*
1238 * Only called when the request isn't completed by SCSI, and not freed by
1239 * SCSI
1240 */
1241static void scsi_cleanup_rq(struct request *rq)
1242{
1243 if (rq->rq_flags & RQF_DONTPREP) {
1244 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1245 rq->rq_flags &= ~RQF_DONTPREP;
1246 }
1247}
1248
1249/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1250void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1251{
1252 struct request *rq = scsi_cmd_to_rq(cmd);
1253
1254 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1255 cmd->flags |= SCMD_INITIALIZED;
1256 scsi_initialize_rq(rq);
1257 }
1258
1259 cmd->device = dev;
1260 INIT_LIST_HEAD(&cmd->eh_entry);
1261 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1262}
1263
1264static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1265 struct request *req)
1266{
1267 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1268
1269 /*
1270 * Passthrough requests may transfer data, in which case they must
1271 * a bio attached to them. Or they might contain a SCSI command
1272 * that does not transfer data, in which case they may optionally
1273 * submit a request without an attached bio.
1274 */
1275 if (req->bio) {
1276 blk_status_t ret = scsi_alloc_sgtables(cmd);
1277 if (unlikely(ret != BLK_STS_OK))
1278 return ret;
1279 } else {
1280 BUG_ON(blk_rq_bytes(req));
1281
1282 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1283 }
1284
1285 cmd->transfersize = blk_rq_bytes(req);
1286 return BLK_STS_OK;
1287}
1288
1289static blk_status_t
1290scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1291{
1292 switch (sdev->sdev_state) {
1293 case SDEV_CREATED:
1294 return BLK_STS_OK;
1295 case SDEV_OFFLINE:
1296 case SDEV_TRANSPORT_OFFLINE:
1297 /*
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1301 */
1302 if (!sdev->offline_already) {
1303 sdev->offline_already = true;
1304 sdev_printk(KERN_ERR, sdev,
1305 "rejecting I/O to offline device\n");
1306 }
1307 return BLK_STS_IOERR;
1308 case SDEV_DEL:
1309 /*
1310 * If the device is fully deleted, we refuse to
1311 * process any commands as well.
1312 */
1313 sdev_printk(KERN_ERR, sdev,
1314 "rejecting I/O to dead device\n");
1315 return BLK_STS_IOERR;
1316 case SDEV_BLOCK:
1317 case SDEV_CREATED_BLOCK:
1318 return BLK_STS_RESOURCE;
1319 case SDEV_QUIESCE:
1320 /*
1321 * If the device is blocked we only accept power management
1322 * commands.
1323 */
1324 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1325 return BLK_STS_RESOURCE;
1326 return BLK_STS_OK;
1327 default:
1328 /*
1329 * For any other not fully online state we only allow
1330 * power management commands.
1331 */
1332 if (req && !(req->rq_flags & RQF_PM))
1333 return BLK_STS_OFFLINE;
1334 return BLK_STS_OK;
1335 }
1336}
1337
1338/*
1339 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1340 * and return the token else return -1.
1341 */
1342static inline int scsi_dev_queue_ready(struct request_queue *q,
1343 struct scsi_device *sdev)
1344{
1345 int token;
1346
1347 token = sbitmap_get(&sdev->budget_map);
1348 if (token < 0)
1349 return -1;
1350
1351 if (!atomic_read(&sdev->device_blocked))
1352 return token;
1353
1354 /*
1355 * Only unblock if no other commands are pending and
1356 * if device_blocked has decreased to zero
1357 */
1358 if (scsi_device_busy(sdev) > 1 ||
1359 atomic_dec_return(&sdev->device_blocked) > 0) {
1360 sbitmap_put(&sdev->budget_map, token);
1361 return -1;
1362 }
1363
1364 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1365 "unblocking device at zero depth\n"));
1366
1367 return token;
1368}
1369
1370/*
1371 * scsi_target_queue_ready: checks if there we can send commands to target
1372 * @sdev: scsi device on starget to check.
1373 */
1374static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1375 struct scsi_device *sdev)
1376{
1377 struct scsi_target *starget = scsi_target(sdev);
1378 unsigned int busy;
1379
1380 if (starget->single_lun) {
1381 spin_lock_irq(shost->host_lock);
1382 if (starget->starget_sdev_user &&
1383 starget->starget_sdev_user != sdev) {
1384 spin_unlock_irq(shost->host_lock);
1385 return 0;
1386 }
1387 starget->starget_sdev_user = sdev;
1388 spin_unlock_irq(shost->host_lock);
1389 }
1390
1391 if (starget->can_queue <= 0)
1392 return 1;
1393
1394 busy = atomic_inc_return(&starget->target_busy) - 1;
1395 if (atomic_read(&starget->target_blocked) > 0) {
1396 if (busy)
1397 goto starved;
1398
1399 /*
1400 * unblock after target_blocked iterates to zero
1401 */
1402 if (atomic_dec_return(&starget->target_blocked) > 0)
1403 goto out_dec;
1404
1405 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1406 "unblocking target at zero depth\n"));
1407 }
1408
1409 if (busy >= starget->can_queue)
1410 goto starved;
1411
1412 return 1;
1413
1414starved:
1415 spin_lock_irq(shost->host_lock);
1416 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1417 spin_unlock_irq(shost->host_lock);
1418out_dec:
1419 if (starget->can_queue > 0)
1420 atomic_dec(&starget->target_busy);
1421 return 0;
1422}
1423
1424/*
1425 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1426 * return 0. We must end up running the queue again whenever 0 is
1427 * returned, else IO can hang.
1428 */
1429static inline int scsi_host_queue_ready(struct request_queue *q,
1430 struct Scsi_Host *shost,
1431 struct scsi_device *sdev,
1432 struct scsi_cmnd *cmd)
1433{
1434 if (atomic_read(&shost->host_blocked) > 0) {
1435 if (scsi_host_busy(shost) > 0)
1436 goto starved;
1437
1438 /*
1439 * unblock after host_blocked iterates to zero
1440 */
1441 if (atomic_dec_return(&shost->host_blocked) > 0)
1442 goto out_dec;
1443
1444 SCSI_LOG_MLQUEUE(3,
1445 shost_printk(KERN_INFO, shost,
1446 "unblocking host at zero depth\n"));
1447 }
1448
1449 if (shost->host_self_blocked)
1450 goto starved;
1451
1452 /* We're OK to process the command, so we can't be starved */
1453 if (!list_empty(&sdev->starved_entry)) {
1454 spin_lock_irq(shost->host_lock);
1455 if (!list_empty(&sdev->starved_entry))
1456 list_del_init(&sdev->starved_entry);
1457 spin_unlock_irq(shost->host_lock);
1458 }
1459
1460 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1461
1462 return 1;
1463
1464starved:
1465 spin_lock_irq(shost->host_lock);
1466 if (list_empty(&sdev->starved_entry))
1467 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1468 spin_unlock_irq(shost->host_lock);
1469out_dec:
1470 scsi_dec_host_busy(shost, cmd);
1471 return 0;
1472}
1473
1474/*
1475 * Busy state exporting function for request stacking drivers.
1476 *
1477 * For efficiency, no lock is taken to check the busy state of
1478 * shost/starget/sdev, since the returned value is not guaranteed and
1479 * may be changed after request stacking drivers call the function,
1480 * regardless of taking lock or not.
1481 *
1482 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1483 * needs to return 'not busy'. Otherwise, request stacking drivers
1484 * may hold requests forever.
1485 */
1486static bool scsi_mq_lld_busy(struct request_queue *q)
1487{
1488 struct scsi_device *sdev = q->queuedata;
1489 struct Scsi_Host *shost;
1490
1491 if (blk_queue_dying(q))
1492 return false;
1493
1494 shost = sdev->host;
1495
1496 /*
1497 * Ignore host/starget busy state.
1498 * Since block layer does not have a concept of fairness across
1499 * multiple queues, congestion of host/starget needs to be handled
1500 * in SCSI layer.
1501 */
1502 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1503 return true;
1504
1505 return false;
1506}
1507
1508/*
1509 * Block layer request completion callback. May be called from interrupt
1510 * context.
1511 */
1512static void scsi_complete(struct request *rq)
1513{
1514 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1515 enum scsi_disposition disposition;
1516
1517 INIT_LIST_HEAD(&cmd->eh_entry);
1518
1519 atomic_inc(&cmd->device->iodone_cnt);
1520 if (cmd->result)
1521 atomic_inc(&cmd->device->ioerr_cnt);
1522
1523 disposition = scsi_decide_disposition(cmd);
1524 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1525 disposition = SUCCESS;
1526
1527 scsi_log_completion(cmd, disposition);
1528
1529 switch (disposition) {
1530 case SUCCESS:
1531 scsi_finish_command(cmd);
1532 break;
1533 case NEEDS_RETRY:
1534 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1535 break;
1536 case ADD_TO_MLQUEUE:
1537 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1538 break;
1539 default:
1540 scsi_eh_scmd_add(cmd);
1541 break;
1542 }
1543}
1544
1545/**
1546 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1547 * @cmd: command block we are dispatching.
1548 *
1549 * Return: nonzero return request was rejected and device's queue needs to be
1550 * plugged.
1551 */
1552static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1553{
1554 struct Scsi_Host *host = cmd->device->host;
1555 int rtn = 0;
1556
1557 atomic_inc(&cmd->device->iorequest_cnt);
1558
1559 /* check if the device is still usable */
1560 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1561 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1562 * returns an immediate error upwards, and signals
1563 * that the device is no longer present */
1564 cmd->result = DID_NO_CONNECT << 16;
1565 goto done;
1566 }
1567
1568 /* Check to see if the scsi lld made this device blocked. */
1569 if (unlikely(scsi_device_blocked(cmd->device))) {
1570 /*
1571 * in blocked state, the command is just put back on
1572 * the device queue. The suspend state has already
1573 * blocked the queue so future requests should not
1574 * occur until the device transitions out of the
1575 * suspend state.
1576 */
1577 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1578 "queuecommand : device blocked\n"));
1579 atomic_dec(&cmd->device->iorequest_cnt);
1580 return SCSI_MLQUEUE_DEVICE_BUSY;
1581 }
1582
1583 /* Store the LUN value in cmnd, if needed. */
1584 if (cmd->device->lun_in_cdb)
1585 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1586 (cmd->device->lun << 5 & 0xe0);
1587
1588 scsi_log_send(cmd);
1589
1590 /*
1591 * Before we queue this command, check if the command
1592 * length exceeds what the host adapter can handle.
1593 */
1594 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1595 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1596 "queuecommand : command too long. "
1597 "cdb_size=%d host->max_cmd_len=%d\n",
1598 cmd->cmd_len, cmd->device->host->max_cmd_len));
1599 cmd->result = (DID_ABORT << 16);
1600 goto done;
1601 }
1602
1603 if (unlikely(host->shost_state == SHOST_DEL)) {
1604 cmd->result = (DID_NO_CONNECT << 16);
1605 goto done;
1606
1607 }
1608
1609 trace_scsi_dispatch_cmd_start(cmd);
1610 rtn = host->hostt->queuecommand(host, cmd);
1611 if (rtn) {
1612 atomic_dec(&cmd->device->iorequest_cnt);
1613 trace_scsi_dispatch_cmd_error(cmd, rtn);
1614 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1615 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1616 rtn = SCSI_MLQUEUE_HOST_BUSY;
1617
1618 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1619 "queuecommand : request rejected\n"));
1620 }
1621
1622 return rtn;
1623 done:
1624 scsi_done(cmd);
1625 return 0;
1626}
1627
1628/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1629static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1630{
1631 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1632 sizeof(struct scatterlist);
1633}
1634
1635static blk_status_t scsi_prepare_cmd(struct request *req)
1636{
1637 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1638 struct scsi_device *sdev = req->q->queuedata;
1639 struct Scsi_Host *shost = sdev->host;
1640 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1641 struct scatterlist *sg;
1642
1643 scsi_init_command(sdev, cmd);
1644
1645 cmd->eh_eflags = 0;
1646 cmd->prot_type = 0;
1647 cmd->prot_flags = 0;
1648 cmd->submitter = 0;
1649 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1650 cmd->underflow = 0;
1651 cmd->transfersize = 0;
1652 cmd->host_scribble = NULL;
1653 cmd->result = 0;
1654 cmd->extra_len = 0;
1655 cmd->state = 0;
1656 if (in_flight)
1657 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1658
1659 cmd->prot_op = SCSI_PROT_NORMAL;
1660 if (blk_rq_bytes(req))
1661 cmd->sc_data_direction = rq_dma_dir(req);
1662 else
1663 cmd->sc_data_direction = DMA_NONE;
1664
1665 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1666 cmd->sdb.table.sgl = sg;
1667
1668 if (scsi_host_get_prot(shost)) {
1669 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1670
1671 cmd->prot_sdb->table.sgl =
1672 (struct scatterlist *)(cmd->prot_sdb + 1);
1673 }
1674
1675 /*
1676 * Special handling for passthrough commands, which don't go to the ULP
1677 * at all:
1678 */
1679 if (blk_rq_is_passthrough(req))
1680 return scsi_setup_scsi_cmnd(sdev, req);
1681
1682 if (sdev->handler && sdev->handler->prep_fn) {
1683 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1684
1685 if (ret != BLK_STS_OK)
1686 return ret;
1687 }
1688
1689 /* Usually overridden by the ULP */
1690 cmd->allowed = 0;
1691 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1692 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1693}
1694
1695static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1696{
1697 struct request *req = scsi_cmd_to_rq(cmd);
1698
1699 switch (cmd->submitter) {
1700 case SUBMITTED_BY_BLOCK_LAYER:
1701 break;
1702 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1703 return scsi_eh_done(cmd);
1704 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1705 return;
1706 }
1707
1708 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1709 return;
1710 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1711 return;
1712 trace_scsi_dispatch_cmd_done(cmd);
1713
1714 if (complete_directly)
1715 blk_mq_complete_request_direct(req, scsi_complete);
1716 else
1717 blk_mq_complete_request(req);
1718}
1719
1720void scsi_done(struct scsi_cmnd *cmd)
1721{
1722 scsi_done_internal(cmd, false);
1723}
1724EXPORT_SYMBOL(scsi_done);
1725
1726void scsi_done_direct(struct scsi_cmnd *cmd)
1727{
1728 scsi_done_internal(cmd, true);
1729}
1730EXPORT_SYMBOL(scsi_done_direct);
1731
1732static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1733{
1734 struct scsi_device *sdev = q->queuedata;
1735
1736 sbitmap_put(&sdev->budget_map, budget_token);
1737}
1738
1739/*
1740 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1741 * not change behaviour from the previous unplug mechanism, experimentation
1742 * may prove this needs changing.
1743 */
1744#define SCSI_QUEUE_DELAY 3
1745
1746static int scsi_mq_get_budget(struct request_queue *q)
1747{
1748 struct scsi_device *sdev = q->queuedata;
1749 int token = scsi_dev_queue_ready(q, sdev);
1750
1751 if (token >= 0)
1752 return token;
1753
1754 atomic_inc(&sdev->restarts);
1755
1756 /*
1757 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1758 * .restarts must be incremented before .device_busy is read because the
1759 * code in scsi_run_queue_async() depends on the order of these operations.
1760 */
1761 smp_mb__after_atomic();
1762
1763 /*
1764 * If all in-flight requests originated from this LUN are completed
1765 * before reading .device_busy, sdev->device_busy will be observed as
1766 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1767 * soon. Otherwise, completion of one of these requests will observe
1768 * the .restarts flag, and the request queue will be run for handling
1769 * this request, see scsi_end_request().
1770 */
1771 if (unlikely(scsi_device_busy(sdev) == 0 &&
1772 !scsi_device_blocked(sdev)))
1773 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1774 return -1;
1775}
1776
1777static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1778{
1779 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1780
1781 cmd->budget_token = token;
1782}
1783
1784static int scsi_mq_get_rq_budget_token(struct request *req)
1785{
1786 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1787
1788 return cmd->budget_token;
1789}
1790
1791static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1792 const struct blk_mq_queue_data *bd)
1793{
1794 struct request *req = bd->rq;
1795 struct request_queue *q = req->q;
1796 struct scsi_device *sdev = q->queuedata;
1797 struct Scsi_Host *shost = sdev->host;
1798 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1799 blk_status_t ret;
1800 int reason;
1801
1802 WARN_ON_ONCE(cmd->budget_token < 0);
1803
1804 /*
1805 * If the device is not in running state we will reject some or all
1806 * commands.
1807 */
1808 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1809 ret = scsi_device_state_check(sdev, req);
1810 if (ret != BLK_STS_OK)
1811 goto out_put_budget;
1812 }
1813
1814 ret = BLK_STS_RESOURCE;
1815 if (!scsi_target_queue_ready(shost, sdev))
1816 goto out_put_budget;
1817 if (unlikely(scsi_host_in_recovery(shost))) {
1818 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1819 ret = BLK_STS_OFFLINE;
1820 goto out_dec_target_busy;
1821 }
1822 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1823 goto out_dec_target_busy;
1824
1825 /*
1826 * Only clear the driver-private command data if the LLD does not supply
1827 * a function to initialize that data.
1828 */
1829 if (shost->hostt->cmd_size && !shost->hostt->init_cmd_priv)
1830 memset(cmd + 1, 0, shost->hostt->cmd_size);
1831
1832 if (!(req->rq_flags & RQF_DONTPREP)) {
1833 ret = scsi_prepare_cmd(req);
1834 if (ret != BLK_STS_OK)
1835 goto out_dec_host_busy;
1836 req->rq_flags |= RQF_DONTPREP;
1837 } else {
1838 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1839 }
1840
1841 cmd->flags &= SCMD_PRESERVED_FLAGS;
1842 if (sdev->simple_tags)
1843 cmd->flags |= SCMD_TAGGED;
1844 if (bd->last)
1845 cmd->flags |= SCMD_LAST;
1846
1847 scsi_set_resid(cmd, 0);
1848 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1849 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1850
1851 blk_mq_start_request(req);
1852 reason = scsi_dispatch_cmd(cmd);
1853 if (reason) {
1854 scsi_set_blocked(cmd, reason);
1855 ret = BLK_STS_RESOURCE;
1856 goto out_dec_host_busy;
1857 }
1858
1859 return BLK_STS_OK;
1860
1861out_dec_host_busy:
1862 scsi_dec_host_busy(shost, cmd);
1863out_dec_target_busy:
1864 if (scsi_target(sdev)->can_queue > 0)
1865 atomic_dec(&scsi_target(sdev)->target_busy);
1866out_put_budget:
1867 scsi_mq_put_budget(q, cmd->budget_token);
1868 cmd->budget_token = -1;
1869 switch (ret) {
1870 case BLK_STS_OK:
1871 break;
1872 case BLK_STS_RESOURCE:
1873 if (scsi_device_blocked(sdev))
1874 ret = BLK_STS_DEV_RESOURCE;
1875 break;
1876 case BLK_STS_AGAIN:
1877 cmd->result = DID_BUS_BUSY << 16;
1878 if (req->rq_flags & RQF_DONTPREP)
1879 scsi_mq_uninit_cmd(cmd);
1880 break;
1881 default:
1882 if (unlikely(!scsi_device_online(sdev)))
1883 cmd->result = DID_NO_CONNECT << 16;
1884 else
1885 cmd->result = DID_ERROR << 16;
1886 /*
1887 * Make sure to release all allocated resources when
1888 * we hit an error, as we will never see this command
1889 * again.
1890 */
1891 if (req->rq_flags & RQF_DONTPREP)
1892 scsi_mq_uninit_cmd(cmd);
1893 scsi_run_queue_async(sdev);
1894 break;
1895 }
1896 return ret;
1897}
1898
1899static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1900 unsigned int hctx_idx, unsigned int numa_node)
1901{
1902 struct Scsi_Host *shost = set->driver_data;
1903 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1904 struct scatterlist *sg;
1905 int ret = 0;
1906
1907 cmd->sense_buffer =
1908 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1909 if (!cmd->sense_buffer)
1910 return -ENOMEM;
1911
1912 if (scsi_host_get_prot(shost)) {
1913 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1914 shost->hostt->cmd_size;
1915 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1916 }
1917
1918 if (shost->hostt->init_cmd_priv) {
1919 ret = shost->hostt->init_cmd_priv(shost, cmd);
1920 if (ret < 0)
1921 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1922 }
1923
1924 return ret;
1925}
1926
1927static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1928 unsigned int hctx_idx)
1929{
1930 struct Scsi_Host *shost = set->driver_data;
1931 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1932
1933 if (shost->hostt->exit_cmd_priv)
1934 shost->hostt->exit_cmd_priv(shost, cmd);
1935 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1936}
1937
1938
1939static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1940{
1941 struct Scsi_Host *shost = hctx->driver_data;
1942
1943 if (shost->hostt->mq_poll)
1944 return shost->hostt->mq_poll(shost, hctx->queue_num);
1945
1946 return 0;
1947}
1948
1949static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1950 unsigned int hctx_idx)
1951{
1952 struct Scsi_Host *shost = data;
1953
1954 hctx->driver_data = shost;
1955 return 0;
1956}
1957
1958static void scsi_map_queues(struct blk_mq_tag_set *set)
1959{
1960 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1961
1962 if (shost->hostt->map_queues)
1963 return shost->hostt->map_queues(shost);
1964 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1965}
1966
1967void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
1968{
1969 struct device *dev = shost->dma_dev;
1970
1971 memset(lim, 0, sizeof(*lim));
1972 lim->max_segments =
1973 min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);
1974
1975 if (scsi_host_prot_dma(shost)) {
1976 shost->sg_prot_tablesize =
1977 min_not_zero(shost->sg_prot_tablesize,
1978 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1979 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1980 lim->max_integrity_segments = shost->sg_prot_tablesize;
1981 }
1982
1983 lim->max_hw_sectors = shost->max_sectors;
1984 lim->seg_boundary_mask = shost->dma_boundary;
1985 lim->max_segment_size = shost->max_segment_size;
1986 lim->virt_boundary_mask = shost->virt_boundary_mask;
1987 lim->dma_alignment = max_t(unsigned int,
1988 shost->dma_alignment, dma_get_cache_alignment() - 1);
1989
1990 if (shost->no_highmem)
1991 lim->features |= BLK_FEAT_BOUNCE_HIGH;
1992
1993 /*
1994 * Propagate the DMA formation properties to the dma-mapping layer as
1995 * a courtesy service to the LLDDs. This needs to check that the buses
1996 * actually support the DMA API first, though.
1997 */
1998 if (dev->dma_parms) {
1999 dma_set_seg_boundary(dev, shost->dma_boundary);
2000 dma_set_max_seg_size(dev, shost->max_segment_size);
2001 }
2002}
2003EXPORT_SYMBOL_GPL(scsi_init_limits);
2004
2005static const struct blk_mq_ops scsi_mq_ops_no_commit = {
2006 .get_budget = scsi_mq_get_budget,
2007 .put_budget = scsi_mq_put_budget,
2008 .queue_rq = scsi_queue_rq,
2009 .complete = scsi_complete,
2010 .timeout = scsi_timeout,
2011#ifdef CONFIG_BLK_DEBUG_FS
2012 .show_rq = scsi_show_rq,
2013#endif
2014 .init_request = scsi_mq_init_request,
2015 .exit_request = scsi_mq_exit_request,
2016 .cleanup_rq = scsi_cleanup_rq,
2017 .busy = scsi_mq_lld_busy,
2018 .map_queues = scsi_map_queues,
2019 .init_hctx = scsi_init_hctx,
2020 .poll = scsi_mq_poll,
2021 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2022 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2023};
2024
2025
2026static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2027{
2028 struct Scsi_Host *shost = hctx->driver_data;
2029
2030 shost->hostt->commit_rqs(shost, hctx->queue_num);
2031}
2032
2033static const struct blk_mq_ops scsi_mq_ops = {
2034 .get_budget = scsi_mq_get_budget,
2035 .put_budget = scsi_mq_put_budget,
2036 .queue_rq = scsi_queue_rq,
2037 .commit_rqs = scsi_commit_rqs,
2038 .complete = scsi_complete,
2039 .timeout = scsi_timeout,
2040#ifdef CONFIG_BLK_DEBUG_FS
2041 .show_rq = scsi_show_rq,
2042#endif
2043 .init_request = scsi_mq_init_request,
2044 .exit_request = scsi_mq_exit_request,
2045 .cleanup_rq = scsi_cleanup_rq,
2046 .busy = scsi_mq_lld_busy,
2047 .map_queues = scsi_map_queues,
2048 .init_hctx = scsi_init_hctx,
2049 .poll = scsi_mq_poll,
2050 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2051 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2052};
2053
2054int scsi_mq_setup_tags(struct Scsi_Host *shost)
2055{
2056 unsigned int cmd_size, sgl_size;
2057 struct blk_mq_tag_set *tag_set = &shost->tag_set;
2058
2059 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2060 scsi_mq_inline_sgl_size(shost));
2061 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2062 if (scsi_host_get_prot(shost))
2063 cmd_size += sizeof(struct scsi_data_buffer) +
2064 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2065
2066 memset(tag_set, 0, sizeof(*tag_set));
2067 if (shost->hostt->commit_rqs)
2068 tag_set->ops = &scsi_mq_ops;
2069 else
2070 tag_set->ops = &scsi_mq_ops_no_commit;
2071 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2072 tag_set->nr_maps = shost->nr_maps ? : 1;
2073 tag_set->queue_depth = shost->can_queue;
2074 tag_set->cmd_size = cmd_size;
2075 tag_set->numa_node = dev_to_node(shost->dma_dev);
2076 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
2077 tag_set->flags |=
2078 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2079 if (shost->queuecommand_may_block)
2080 tag_set->flags |= BLK_MQ_F_BLOCKING;
2081 tag_set->driver_data = shost;
2082 if (shost->host_tagset)
2083 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2084
2085 return blk_mq_alloc_tag_set(tag_set);
2086}
2087
2088void scsi_mq_free_tags(struct kref *kref)
2089{
2090 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2091 tagset_refcnt);
2092
2093 blk_mq_free_tag_set(&shost->tag_set);
2094 complete(&shost->tagset_freed);
2095}
2096
2097/**
2098 * scsi_device_from_queue - return sdev associated with a request_queue
2099 * @q: The request queue to return the sdev from
2100 *
2101 * Return the sdev associated with a request queue or NULL if the
2102 * request_queue does not reference a SCSI device.
2103 */
2104struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2105{
2106 struct scsi_device *sdev = NULL;
2107
2108 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2109 q->mq_ops == &scsi_mq_ops)
2110 sdev = q->queuedata;
2111 if (!sdev || !get_device(&sdev->sdev_gendev))
2112 sdev = NULL;
2113
2114 return sdev;
2115}
2116/*
2117 * pktcdvd should have been integrated into the SCSI layers, but for historical
2118 * reasons like the old IDE driver it isn't. This export allows it to safely
2119 * probe if a given device is a SCSI one and only attach to that.
2120 */
2121#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2122EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2123#endif
2124
2125/**
2126 * scsi_block_requests - Utility function used by low-level drivers to prevent
2127 * further commands from being queued to the device.
2128 * @shost: host in question
2129 *
2130 * There is no timer nor any other means by which the requests get unblocked
2131 * other than the low-level driver calling scsi_unblock_requests().
2132 */
2133void scsi_block_requests(struct Scsi_Host *shost)
2134{
2135 shost->host_self_blocked = 1;
2136}
2137EXPORT_SYMBOL(scsi_block_requests);
2138
2139/**
2140 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2141 * further commands to be queued to the device.
2142 * @shost: host in question
2143 *
2144 * There is no timer nor any other means by which the requests get unblocked
2145 * other than the low-level driver calling scsi_unblock_requests(). This is done
2146 * as an API function so that changes to the internals of the scsi mid-layer
2147 * won't require wholesale changes to drivers that use this feature.
2148 */
2149void scsi_unblock_requests(struct Scsi_Host *shost)
2150{
2151 shost->host_self_blocked = 0;
2152 scsi_run_host_queues(shost);
2153}
2154EXPORT_SYMBOL(scsi_unblock_requests);
2155
2156void scsi_exit_queue(void)
2157{
2158 kmem_cache_destroy(scsi_sense_cache);
2159}
2160
2161/**
2162 * scsi_mode_select - issue a mode select
2163 * @sdev: SCSI device to be queried
2164 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2165 * @sp: Save page bit (0 == don't save, 1 == save)
2166 * @buffer: request buffer (may not be smaller than eight bytes)
2167 * @len: length of request buffer.
2168 * @timeout: command timeout
2169 * @retries: number of retries before failing
2170 * @data: returns a structure abstracting the mode header data
2171 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2172 * must be SCSI_SENSE_BUFFERSIZE big.
2173 *
2174 * Returns zero if successful; negative error number or scsi
2175 * status on error
2176 *
2177 */
2178int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2179 unsigned char *buffer, int len, int timeout, int retries,
2180 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2181{
2182 unsigned char cmd[10];
2183 unsigned char *real_buffer;
2184 const struct scsi_exec_args exec_args = {
2185 .sshdr = sshdr,
2186 };
2187 int ret;
2188
2189 memset(cmd, 0, sizeof(cmd));
2190 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2191
2192 /*
2193 * Use MODE SELECT(10) if the device asked for it or if the mode page
2194 * and the mode select header cannot fit within the maximumm 255 bytes
2195 * of the MODE SELECT(6) command.
2196 */
2197 if (sdev->use_10_for_ms ||
2198 len + 4 > 255 ||
2199 data->block_descriptor_length > 255) {
2200 if (len > 65535 - 8)
2201 return -EINVAL;
2202 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2203 if (!real_buffer)
2204 return -ENOMEM;
2205 memcpy(real_buffer + 8, buffer, len);
2206 len += 8;
2207 real_buffer[0] = 0;
2208 real_buffer[1] = 0;
2209 real_buffer[2] = data->medium_type;
2210 real_buffer[3] = data->device_specific;
2211 real_buffer[4] = data->longlba ? 0x01 : 0;
2212 real_buffer[5] = 0;
2213 put_unaligned_be16(data->block_descriptor_length,
2214 &real_buffer[6]);
2215
2216 cmd[0] = MODE_SELECT_10;
2217 put_unaligned_be16(len, &cmd[7]);
2218 } else {
2219 if (data->longlba)
2220 return -EINVAL;
2221
2222 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2223 if (!real_buffer)
2224 return -ENOMEM;
2225 memcpy(real_buffer + 4, buffer, len);
2226 len += 4;
2227 real_buffer[0] = 0;
2228 real_buffer[1] = data->medium_type;
2229 real_buffer[2] = data->device_specific;
2230 real_buffer[3] = data->block_descriptor_length;
2231
2232 cmd[0] = MODE_SELECT;
2233 cmd[4] = len;
2234 }
2235
2236 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2237 timeout, retries, &exec_args);
2238 kfree(real_buffer);
2239 return ret;
2240}
2241EXPORT_SYMBOL_GPL(scsi_mode_select);
2242
2243/**
2244 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2245 * @sdev: SCSI device to be queried
2246 * @dbd: set to prevent mode sense from returning block descriptors
2247 * @modepage: mode page being requested
2248 * @subpage: sub-page of the mode page being requested
2249 * @buffer: request buffer (may not be smaller than eight bytes)
2250 * @len: length of request buffer.
2251 * @timeout: command timeout
2252 * @retries: number of retries before failing
2253 * @data: returns a structure abstracting the mode header data
2254 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2255 * must be SCSI_SENSE_BUFFERSIZE big.
2256 *
2257 * Returns zero if successful, or a negative error number on failure
2258 */
2259int
2260scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2261 unsigned char *buffer, int len, int timeout, int retries,
2262 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2263{
2264 unsigned char cmd[12];
2265 int use_10_for_ms;
2266 int header_length;
2267 int result;
2268 struct scsi_sense_hdr my_sshdr;
2269 struct scsi_failure failure_defs[] = {
2270 {
2271 .sense = UNIT_ATTENTION,
2272 .asc = SCMD_FAILURE_ASC_ANY,
2273 .ascq = SCMD_FAILURE_ASCQ_ANY,
2274 .allowed = retries,
2275 .result = SAM_STAT_CHECK_CONDITION,
2276 },
2277 {}
2278 };
2279 struct scsi_failures failures = {
2280 .failure_definitions = failure_defs,
2281 };
2282 const struct scsi_exec_args exec_args = {
2283 /* caller might not be interested in sense, but we need it */
2284 .sshdr = sshdr ? : &my_sshdr,
2285 .failures = &failures,
2286 };
2287
2288 memset(data, 0, sizeof(*data));
2289 memset(&cmd[0], 0, 12);
2290
2291 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2292 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2293 cmd[2] = modepage;
2294 cmd[3] = subpage;
2295
2296 sshdr = exec_args.sshdr;
2297
2298 retry:
2299 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2300
2301 if (use_10_for_ms) {
2302 if (len < 8 || len > 65535)
2303 return -EINVAL;
2304
2305 cmd[0] = MODE_SENSE_10;
2306 put_unaligned_be16(len, &cmd[7]);
2307 header_length = 8;
2308 } else {
2309 if (len < 4)
2310 return -EINVAL;
2311
2312 cmd[0] = MODE_SENSE;
2313 cmd[4] = len;
2314 header_length = 4;
2315 }
2316
2317 memset(buffer, 0, len);
2318
2319 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2320 timeout, retries, &exec_args);
2321 if (result < 0)
2322 return result;
2323
2324 /* This code looks awful: what it's doing is making sure an
2325 * ILLEGAL REQUEST sense return identifies the actual command
2326 * byte as the problem. MODE_SENSE commands can return
2327 * ILLEGAL REQUEST if the code page isn't supported */
2328
2329 if (!scsi_status_is_good(result)) {
2330 if (scsi_sense_valid(sshdr)) {
2331 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2332 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2333 /*
2334 * Invalid command operation code: retry using
2335 * MODE SENSE(6) if this was a MODE SENSE(10)
2336 * request, except if the request mode page is
2337 * too large for MODE SENSE single byte
2338 * allocation length field.
2339 */
2340 if (use_10_for_ms) {
2341 if (len > 255)
2342 return -EIO;
2343 sdev->use_10_for_ms = 0;
2344 goto retry;
2345 }
2346 }
2347 }
2348 return -EIO;
2349 }
2350 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2351 (modepage == 6 || modepage == 8))) {
2352 /* Initio breakage? */
2353 header_length = 0;
2354 data->length = 13;
2355 data->medium_type = 0;
2356 data->device_specific = 0;
2357 data->longlba = 0;
2358 data->block_descriptor_length = 0;
2359 } else if (use_10_for_ms) {
2360 data->length = get_unaligned_be16(&buffer[0]) + 2;
2361 data->medium_type = buffer[2];
2362 data->device_specific = buffer[3];
2363 data->longlba = buffer[4] & 0x01;
2364 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2365 } else {
2366 data->length = buffer[0] + 1;
2367 data->medium_type = buffer[1];
2368 data->device_specific = buffer[2];
2369 data->block_descriptor_length = buffer[3];
2370 }
2371 data->header_length = header_length;
2372
2373 return 0;
2374}
2375EXPORT_SYMBOL(scsi_mode_sense);
2376
2377/**
2378 * scsi_test_unit_ready - test if unit is ready
2379 * @sdev: scsi device to change the state of.
2380 * @timeout: command timeout
2381 * @retries: number of retries before failing
2382 * @sshdr: outpout pointer for decoded sense information.
2383 *
2384 * Returns zero if unsuccessful or an error if TUR failed. For
2385 * removable media, UNIT_ATTENTION sets ->changed flag.
2386 **/
2387int
2388scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2389 struct scsi_sense_hdr *sshdr)
2390{
2391 char cmd[] = {
2392 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2393 };
2394 const struct scsi_exec_args exec_args = {
2395 .sshdr = sshdr,
2396 };
2397 int result;
2398
2399 /* try to eat the UNIT_ATTENTION if there are enough retries */
2400 do {
2401 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2402 timeout, 1, &exec_args);
2403 if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2404 sshdr->sense_key == UNIT_ATTENTION)
2405 sdev->changed = 1;
2406 } while (result > 0 && scsi_sense_valid(sshdr) &&
2407 sshdr->sense_key == UNIT_ATTENTION && --retries);
2408
2409 return result;
2410}
2411EXPORT_SYMBOL(scsi_test_unit_ready);
2412
2413/**
2414 * scsi_device_set_state - Take the given device through the device state model.
2415 * @sdev: scsi device to change the state of.
2416 * @state: state to change to.
2417 *
2418 * Returns zero if successful or an error if the requested
2419 * transition is illegal.
2420 */
2421int
2422scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2423{
2424 enum scsi_device_state oldstate = sdev->sdev_state;
2425
2426 if (state == oldstate)
2427 return 0;
2428
2429 switch (state) {
2430 case SDEV_CREATED:
2431 switch (oldstate) {
2432 case SDEV_CREATED_BLOCK:
2433 break;
2434 default:
2435 goto illegal;
2436 }
2437 break;
2438
2439 case SDEV_RUNNING:
2440 switch (oldstate) {
2441 case SDEV_CREATED:
2442 case SDEV_OFFLINE:
2443 case SDEV_TRANSPORT_OFFLINE:
2444 case SDEV_QUIESCE:
2445 case SDEV_BLOCK:
2446 break;
2447 default:
2448 goto illegal;
2449 }
2450 break;
2451
2452 case SDEV_QUIESCE:
2453 switch (oldstate) {
2454 case SDEV_RUNNING:
2455 case SDEV_OFFLINE:
2456 case SDEV_TRANSPORT_OFFLINE:
2457 break;
2458 default:
2459 goto illegal;
2460 }
2461 break;
2462
2463 case SDEV_OFFLINE:
2464 case SDEV_TRANSPORT_OFFLINE:
2465 switch (oldstate) {
2466 case SDEV_CREATED:
2467 case SDEV_RUNNING:
2468 case SDEV_QUIESCE:
2469 case SDEV_BLOCK:
2470 break;
2471 default:
2472 goto illegal;
2473 }
2474 break;
2475
2476 case SDEV_BLOCK:
2477 switch (oldstate) {
2478 case SDEV_RUNNING:
2479 case SDEV_CREATED_BLOCK:
2480 case SDEV_QUIESCE:
2481 case SDEV_OFFLINE:
2482 break;
2483 default:
2484 goto illegal;
2485 }
2486 break;
2487
2488 case SDEV_CREATED_BLOCK:
2489 switch (oldstate) {
2490 case SDEV_CREATED:
2491 break;
2492 default:
2493 goto illegal;
2494 }
2495 break;
2496
2497 case SDEV_CANCEL:
2498 switch (oldstate) {
2499 case SDEV_CREATED:
2500 case SDEV_RUNNING:
2501 case SDEV_QUIESCE:
2502 case SDEV_OFFLINE:
2503 case SDEV_TRANSPORT_OFFLINE:
2504 break;
2505 default:
2506 goto illegal;
2507 }
2508 break;
2509
2510 case SDEV_DEL:
2511 switch (oldstate) {
2512 case SDEV_CREATED:
2513 case SDEV_RUNNING:
2514 case SDEV_OFFLINE:
2515 case SDEV_TRANSPORT_OFFLINE:
2516 case SDEV_CANCEL:
2517 case SDEV_BLOCK:
2518 case SDEV_CREATED_BLOCK:
2519 break;
2520 default:
2521 goto illegal;
2522 }
2523 break;
2524
2525 }
2526 sdev->offline_already = false;
2527 sdev->sdev_state = state;
2528 return 0;
2529
2530 illegal:
2531 SCSI_LOG_ERROR_RECOVERY(1,
2532 sdev_printk(KERN_ERR, sdev,
2533 "Illegal state transition %s->%s",
2534 scsi_device_state_name(oldstate),
2535 scsi_device_state_name(state))
2536 );
2537 return -EINVAL;
2538}
2539EXPORT_SYMBOL(scsi_device_set_state);
2540
2541/**
2542 * scsi_evt_emit - emit a single SCSI device uevent
2543 * @sdev: associated SCSI device
2544 * @evt: event to emit
2545 *
2546 * Send a single uevent (scsi_event) to the associated scsi_device.
2547 */
2548static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2549{
2550 int idx = 0;
2551 char *envp[3];
2552
2553 switch (evt->evt_type) {
2554 case SDEV_EVT_MEDIA_CHANGE:
2555 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2556 break;
2557 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2558 scsi_rescan_device(sdev);
2559 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2560 break;
2561 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2562 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2563 break;
2564 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2565 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2566 break;
2567 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2568 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2569 break;
2570 case SDEV_EVT_LUN_CHANGE_REPORTED:
2571 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2572 break;
2573 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2574 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2575 break;
2576 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2577 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2578 break;
2579 default:
2580 /* do nothing */
2581 break;
2582 }
2583
2584 envp[idx++] = NULL;
2585
2586 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2587}
2588
2589/**
2590 * scsi_evt_thread - send a uevent for each scsi event
2591 * @work: work struct for scsi_device
2592 *
2593 * Dispatch queued events to their associated scsi_device kobjects
2594 * as uevents.
2595 */
2596void scsi_evt_thread(struct work_struct *work)
2597{
2598 struct scsi_device *sdev;
2599 enum scsi_device_event evt_type;
2600 LIST_HEAD(event_list);
2601
2602 sdev = container_of(work, struct scsi_device, event_work);
2603
2604 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2605 if (test_and_clear_bit(evt_type, sdev->pending_events))
2606 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2607
2608 while (1) {
2609 struct scsi_event *evt;
2610 struct list_head *this, *tmp;
2611 unsigned long flags;
2612
2613 spin_lock_irqsave(&sdev->list_lock, flags);
2614 list_splice_init(&sdev->event_list, &event_list);
2615 spin_unlock_irqrestore(&sdev->list_lock, flags);
2616
2617 if (list_empty(&event_list))
2618 break;
2619
2620 list_for_each_safe(this, tmp, &event_list) {
2621 evt = list_entry(this, struct scsi_event, node);
2622 list_del(&evt->node);
2623 scsi_evt_emit(sdev, evt);
2624 kfree(evt);
2625 }
2626 }
2627}
2628
2629/**
2630 * sdev_evt_send - send asserted event to uevent thread
2631 * @sdev: scsi_device event occurred on
2632 * @evt: event to send
2633 *
2634 * Assert scsi device event asynchronously.
2635 */
2636void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2637{
2638 unsigned long flags;
2639
2640#if 0
2641 /* FIXME: currently this check eliminates all media change events
2642 * for polled devices. Need to update to discriminate between AN
2643 * and polled events */
2644 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2645 kfree(evt);
2646 return;
2647 }
2648#endif
2649
2650 spin_lock_irqsave(&sdev->list_lock, flags);
2651 list_add_tail(&evt->node, &sdev->event_list);
2652 schedule_work(&sdev->event_work);
2653 spin_unlock_irqrestore(&sdev->list_lock, flags);
2654}
2655EXPORT_SYMBOL_GPL(sdev_evt_send);
2656
2657/**
2658 * sdev_evt_alloc - allocate a new scsi event
2659 * @evt_type: type of event to allocate
2660 * @gfpflags: GFP flags for allocation
2661 *
2662 * Allocates and returns a new scsi_event.
2663 */
2664struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2665 gfp_t gfpflags)
2666{
2667 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2668 if (!evt)
2669 return NULL;
2670
2671 evt->evt_type = evt_type;
2672 INIT_LIST_HEAD(&evt->node);
2673
2674 /* evt_type-specific initialization, if any */
2675 switch (evt_type) {
2676 case SDEV_EVT_MEDIA_CHANGE:
2677 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2678 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2679 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2680 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2681 case SDEV_EVT_LUN_CHANGE_REPORTED:
2682 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2683 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2684 default:
2685 /* do nothing */
2686 break;
2687 }
2688
2689 return evt;
2690}
2691EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2692
2693/**
2694 * sdev_evt_send_simple - send asserted event to uevent thread
2695 * @sdev: scsi_device event occurred on
2696 * @evt_type: type of event to send
2697 * @gfpflags: GFP flags for allocation
2698 *
2699 * Assert scsi device event asynchronously, given an event type.
2700 */
2701void sdev_evt_send_simple(struct scsi_device *sdev,
2702 enum scsi_device_event evt_type, gfp_t gfpflags)
2703{
2704 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2705 if (!evt) {
2706 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2707 evt_type);
2708 return;
2709 }
2710
2711 sdev_evt_send(sdev, evt);
2712}
2713EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2714
2715/**
2716 * scsi_device_quiesce - Block all commands except power management.
2717 * @sdev: scsi device to quiesce.
2718 *
2719 * This works by trying to transition to the SDEV_QUIESCE state
2720 * (which must be a legal transition). When the device is in this
2721 * state, only power management requests will be accepted, all others will
2722 * be deferred.
2723 *
2724 * Must be called with user context, may sleep.
2725 *
2726 * Returns zero if unsuccessful or an error if not.
2727 */
2728int
2729scsi_device_quiesce(struct scsi_device *sdev)
2730{
2731 struct request_queue *q = sdev->request_queue;
2732 int err;
2733
2734 /*
2735 * It is allowed to call scsi_device_quiesce() multiple times from
2736 * the same context but concurrent scsi_device_quiesce() calls are
2737 * not allowed.
2738 */
2739 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2740
2741 if (sdev->quiesced_by == current)
2742 return 0;
2743
2744 blk_set_pm_only(q);
2745
2746 blk_mq_freeze_queue(q);
2747 /*
2748 * Ensure that the effect of blk_set_pm_only() will be visible
2749 * for percpu_ref_tryget() callers that occur after the queue
2750 * unfreeze even if the queue was already frozen before this function
2751 * was called. See also https://lwn.net/Articles/573497/.
2752 */
2753 synchronize_rcu();
2754 blk_mq_unfreeze_queue(q);
2755
2756 mutex_lock(&sdev->state_mutex);
2757 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2758 if (err == 0)
2759 sdev->quiesced_by = current;
2760 else
2761 blk_clear_pm_only(q);
2762 mutex_unlock(&sdev->state_mutex);
2763
2764 return err;
2765}
2766EXPORT_SYMBOL(scsi_device_quiesce);
2767
2768/**
2769 * scsi_device_resume - Restart user issued commands to a quiesced device.
2770 * @sdev: scsi device to resume.
2771 *
2772 * Moves the device from quiesced back to running and restarts the
2773 * queues.
2774 *
2775 * Must be called with user context, may sleep.
2776 */
2777void scsi_device_resume(struct scsi_device *sdev)
2778{
2779 /* check if the device state was mutated prior to resume, and if
2780 * so assume the state is being managed elsewhere (for example
2781 * device deleted during suspend)
2782 */
2783 mutex_lock(&sdev->state_mutex);
2784 if (sdev->sdev_state == SDEV_QUIESCE)
2785 scsi_device_set_state(sdev, SDEV_RUNNING);
2786 if (sdev->quiesced_by) {
2787 sdev->quiesced_by = NULL;
2788 blk_clear_pm_only(sdev->request_queue);
2789 }
2790 mutex_unlock(&sdev->state_mutex);
2791}
2792EXPORT_SYMBOL(scsi_device_resume);
2793
2794static void
2795device_quiesce_fn(struct scsi_device *sdev, void *data)
2796{
2797 scsi_device_quiesce(sdev);
2798}
2799
2800void
2801scsi_target_quiesce(struct scsi_target *starget)
2802{
2803 starget_for_each_device(starget, NULL, device_quiesce_fn);
2804}
2805EXPORT_SYMBOL(scsi_target_quiesce);
2806
2807static void
2808device_resume_fn(struct scsi_device *sdev, void *data)
2809{
2810 scsi_device_resume(sdev);
2811}
2812
2813void
2814scsi_target_resume(struct scsi_target *starget)
2815{
2816 starget_for_each_device(starget, NULL, device_resume_fn);
2817}
2818EXPORT_SYMBOL(scsi_target_resume);
2819
2820static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2821{
2822 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2823 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2824
2825 return 0;
2826}
2827
2828void scsi_start_queue(struct scsi_device *sdev)
2829{
2830 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2831 blk_mq_unquiesce_queue(sdev->request_queue);
2832}
2833
2834static void scsi_stop_queue(struct scsi_device *sdev)
2835{
2836 /*
2837 * The atomic variable of ->queue_stopped covers that
2838 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2839 *
2840 * The caller needs to wait until quiesce is done.
2841 */
2842 if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2843 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2844}
2845
2846/**
2847 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2848 * @sdev: device to block
2849 *
2850 * Pause SCSI command processing on the specified device. Does not sleep.
2851 *
2852 * Returns zero if successful or a negative error code upon failure.
2853 *
2854 * Notes:
2855 * This routine transitions the device to the SDEV_BLOCK state (which must be
2856 * a legal transition). When the device is in this state, command processing
2857 * is paused until the device leaves the SDEV_BLOCK state. See also
2858 * scsi_internal_device_unblock_nowait().
2859 */
2860int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2861{
2862 int ret = __scsi_internal_device_block_nowait(sdev);
2863
2864 /*
2865 * The device has transitioned to SDEV_BLOCK. Stop the
2866 * block layer from calling the midlayer with this device's
2867 * request queue.
2868 */
2869 if (!ret)
2870 scsi_stop_queue(sdev);
2871 return ret;
2872}
2873EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2874
2875/**
2876 * scsi_device_block - try to transition to the SDEV_BLOCK state
2877 * @sdev: device to block
2878 * @data: dummy argument, ignored
2879 *
2880 * Pause SCSI command processing on the specified device. Callers must wait
2881 * until all ongoing scsi_queue_rq() calls have finished after this function
2882 * returns.
2883 *
2884 * Note:
2885 * This routine transitions the device to the SDEV_BLOCK state (which must be
2886 * a legal transition). When the device is in this state, command processing
2887 * is paused until the device leaves the SDEV_BLOCK state. See also
2888 * scsi_internal_device_unblock().
2889 */
2890static void scsi_device_block(struct scsi_device *sdev, void *data)
2891{
2892 int err;
2893 enum scsi_device_state state;
2894
2895 mutex_lock(&sdev->state_mutex);
2896 err = __scsi_internal_device_block_nowait(sdev);
2897 state = sdev->sdev_state;
2898 if (err == 0)
2899 /*
2900 * scsi_stop_queue() must be called with the state_mutex
2901 * held. Otherwise a simultaneous scsi_start_queue() call
2902 * might unquiesce the queue before we quiesce it.
2903 */
2904 scsi_stop_queue(sdev);
2905
2906 mutex_unlock(&sdev->state_mutex);
2907
2908 WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2909 __func__, dev_name(&sdev->sdev_gendev), state);
2910}
2911
2912/**
2913 * scsi_internal_device_unblock_nowait - resume a device after a block request
2914 * @sdev: device to resume
2915 * @new_state: state to set the device to after unblocking
2916 *
2917 * Restart the device queue for a previously suspended SCSI device. Does not
2918 * sleep.
2919 *
2920 * Returns zero if successful or a negative error code upon failure.
2921 *
2922 * Notes:
2923 * This routine transitions the device to the SDEV_RUNNING state or to one of
2924 * the offline states (which must be a legal transition) allowing the midlayer
2925 * to goose the queue for this device.
2926 */
2927int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2928 enum scsi_device_state new_state)
2929{
2930 switch (new_state) {
2931 case SDEV_RUNNING:
2932 case SDEV_TRANSPORT_OFFLINE:
2933 break;
2934 default:
2935 return -EINVAL;
2936 }
2937
2938 /*
2939 * Try to transition the scsi device to SDEV_RUNNING or one of the
2940 * offlined states and goose the device queue if successful.
2941 */
2942 switch (sdev->sdev_state) {
2943 case SDEV_BLOCK:
2944 case SDEV_TRANSPORT_OFFLINE:
2945 sdev->sdev_state = new_state;
2946 break;
2947 case SDEV_CREATED_BLOCK:
2948 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2949 new_state == SDEV_OFFLINE)
2950 sdev->sdev_state = new_state;
2951 else
2952 sdev->sdev_state = SDEV_CREATED;
2953 break;
2954 case SDEV_CANCEL:
2955 case SDEV_OFFLINE:
2956 break;
2957 default:
2958 return -EINVAL;
2959 }
2960 scsi_start_queue(sdev);
2961
2962 return 0;
2963}
2964EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2965
2966/**
2967 * scsi_internal_device_unblock - resume a device after a block request
2968 * @sdev: device to resume
2969 * @new_state: state to set the device to after unblocking
2970 *
2971 * Restart the device queue for a previously suspended SCSI device. May sleep.
2972 *
2973 * Returns zero if successful or a negative error code upon failure.
2974 *
2975 * Notes:
2976 * This routine transitions the device to the SDEV_RUNNING state or to one of
2977 * the offline states (which must be a legal transition) allowing the midlayer
2978 * to goose the queue for this device.
2979 */
2980static int scsi_internal_device_unblock(struct scsi_device *sdev,
2981 enum scsi_device_state new_state)
2982{
2983 int ret;
2984
2985 mutex_lock(&sdev->state_mutex);
2986 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2987 mutex_unlock(&sdev->state_mutex);
2988
2989 return ret;
2990}
2991
2992static int
2993target_block(struct device *dev, void *data)
2994{
2995 if (scsi_is_target_device(dev))
2996 starget_for_each_device(to_scsi_target(dev), NULL,
2997 scsi_device_block);
2998 return 0;
2999}
3000
3001/**
3002 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3003 * @dev: a parent device of one or more scsi_target devices
3004 * @shost: the Scsi_Host to which this device belongs
3005 *
3006 * Iterate over all children of @dev, which should be scsi_target devices,
3007 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3008 * ongoing scsi_queue_rq() calls to finish. May sleep.
3009 *
3010 * Note:
3011 * @dev must not itself be a scsi_target device.
3012 */
3013void
3014scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3015{
3016 WARN_ON_ONCE(scsi_is_target_device(dev));
3017 device_for_each_child(dev, NULL, target_block);
3018 blk_mq_wait_quiesce_done(&shost->tag_set);
3019}
3020EXPORT_SYMBOL_GPL(scsi_block_targets);
3021
3022static void
3023device_unblock(struct scsi_device *sdev, void *data)
3024{
3025 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3026}
3027
3028static int
3029target_unblock(struct device *dev, void *data)
3030{
3031 if (scsi_is_target_device(dev))
3032 starget_for_each_device(to_scsi_target(dev), data,
3033 device_unblock);
3034 return 0;
3035}
3036
3037void
3038scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3039{
3040 if (scsi_is_target_device(dev))
3041 starget_for_each_device(to_scsi_target(dev), &new_state,
3042 device_unblock);
3043 else
3044 device_for_each_child(dev, &new_state, target_unblock);
3045}
3046EXPORT_SYMBOL_GPL(scsi_target_unblock);
3047
3048/**
3049 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3050 * @shost: device to block
3051 *
3052 * Pause SCSI command processing for all logical units associated with the SCSI
3053 * host and wait until pending scsi_queue_rq() calls have finished.
3054 *
3055 * Returns zero if successful or a negative error code upon failure.
3056 */
3057int
3058scsi_host_block(struct Scsi_Host *shost)
3059{
3060 struct scsi_device *sdev;
3061 int ret;
3062
3063 /*
3064 * Call scsi_internal_device_block_nowait so we can avoid
3065 * calling synchronize_rcu() for each LUN.
3066 */
3067 shost_for_each_device(sdev, shost) {
3068 mutex_lock(&sdev->state_mutex);
3069 ret = scsi_internal_device_block_nowait(sdev);
3070 mutex_unlock(&sdev->state_mutex);
3071 if (ret) {
3072 scsi_device_put(sdev);
3073 return ret;
3074 }
3075 }
3076
3077 /* Wait for ongoing scsi_queue_rq() calls to finish. */
3078 blk_mq_wait_quiesce_done(&shost->tag_set);
3079
3080 return 0;
3081}
3082EXPORT_SYMBOL_GPL(scsi_host_block);
3083
3084int
3085scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3086{
3087 struct scsi_device *sdev;
3088 int ret = 0;
3089
3090 shost_for_each_device(sdev, shost) {
3091 ret = scsi_internal_device_unblock(sdev, new_state);
3092 if (ret) {
3093 scsi_device_put(sdev);
3094 break;
3095 }
3096 }
3097 return ret;
3098}
3099EXPORT_SYMBOL_GPL(scsi_host_unblock);
3100
3101/**
3102 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3103 * @sgl: scatter-gather list
3104 * @sg_count: number of segments in sg
3105 * @offset: offset in bytes into sg, on return offset into the mapped area
3106 * @len: bytes to map, on return number of bytes mapped
3107 *
3108 * Returns virtual address of the start of the mapped page
3109 */
3110void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3111 size_t *offset, size_t *len)
3112{
3113 int i;
3114 size_t sg_len = 0, len_complete = 0;
3115 struct scatterlist *sg;
3116 struct page *page;
3117
3118 WARN_ON(!irqs_disabled());
3119
3120 for_each_sg(sgl, sg, sg_count, i) {
3121 len_complete = sg_len; /* Complete sg-entries */
3122 sg_len += sg->length;
3123 if (sg_len > *offset)
3124 break;
3125 }
3126
3127 if (unlikely(i == sg_count)) {
3128 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3129 "elements %d\n",
3130 __func__, sg_len, *offset, sg_count);
3131 WARN_ON(1);
3132 return NULL;
3133 }
3134
3135 /* Offset starting from the beginning of first page in this sg-entry */
3136 *offset = *offset - len_complete + sg->offset;
3137
3138 /* Assumption: contiguous pages can be accessed as "page + i" */
3139 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3140 *offset &= ~PAGE_MASK;
3141
3142 /* Bytes in this sg-entry from *offset to the end of the page */
3143 sg_len = PAGE_SIZE - *offset;
3144 if (*len > sg_len)
3145 *len = sg_len;
3146
3147 return kmap_atomic(page);
3148}
3149EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3150
3151/**
3152 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3153 * @virt: virtual address to be unmapped
3154 */
3155void scsi_kunmap_atomic_sg(void *virt)
3156{
3157 kunmap_atomic(virt);
3158}
3159EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3160
3161void sdev_disable_disk_events(struct scsi_device *sdev)
3162{
3163 atomic_inc(&sdev->disk_events_disable_depth);
3164}
3165EXPORT_SYMBOL(sdev_disable_disk_events);
3166
3167void sdev_enable_disk_events(struct scsi_device *sdev)
3168{
3169 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3170 return;
3171 atomic_dec(&sdev->disk_events_disable_depth);
3172}
3173EXPORT_SYMBOL(sdev_enable_disk_events);
3174
3175static unsigned char designator_prio(const unsigned char *d)
3176{
3177 if (d[1] & 0x30)
3178 /* not associated with LUN */
3179 return 0;
3180
3181 if (d[3] == 0)
3182 /* invalid length */
3183 return 0;
3184
3185 /*
3186 * Order of preference for lun descriptor:
3187 * - SCSI name string
3188 * - NAA IEEE Registered Extended
3189 * - EUI-64 based 16-byte
3190 * - EUI-64 based 12-byte
3191 * - NAA IEEE Registered
3192 * - NAA IEEE Extended
3193 * - EUI-64 based 8-byte
3194 * - SCSI name string (truncated)
3195 * - T10 Vendor ID
3196 * as longer descriptors reduce the likelyhood
3197 * of identification clashes.
3198 */
3199
3200 switch (d[1] & 0xf) {
3201 case 8:
3202 /* SCSI name string, variable-length UTF-8 */
3203 return 9;
3204 case 3:
3205 switch (d[4] >> 4) {
3206 case 6:
3207 /* NAA registered extended */
3208 return 8;
3209 case 5:
3210 /* NAA registered */
3211 return 5;
3212 case 4:
3213 /* NAA extended */
3214 return 4;
3215 case 3:
3216 /* NAA locally assigned */
3217 return 1;
3218 default:
3219 break;
3220 }
3221 break;
3222 case 2:
3223 switch (d[3]) {
3224 case 16:
3225 /* EUI64-based, 16 byte */
3226 return 7;
3227 case 12:
3228 /* EUI64-based, 12 byte */
3229 return 6;
3230 case 8:
3231 /* EUI64-based, 8 byte */
3232 return 3;
3233 default:
3234 break;
3235 }
3236 break;
3237 case 1:
3238 /* T10 vendor ID */
3239 return 1;
3240 default:
3241 break;
3242 }
3243
3244 return 0;
3245}
3246
3247/**
3248 * scsi_vpd_lun_id - return a unique device identification
3249 * @sdev: SCSI device
3250 * @id: buffer for the identification
3251 * @id_len: length of the buffer
3252 *
3253 * Copies a unique device identification into @id based
3254 * on the information in the VPD page 0x83 of the device.
3255 * The string will be formatted as a SCSI name string.
3256 *
3257 * Returns the length of the identification or error on failure.
3258 * If the identifier is longer than the supplied buffer the actual
3259 * identifier length is returned and the buffer is not zero-padded.
3260 */
3261int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3262{
3263 u8 cur_id_prio = 0;
3264 u8 cur_id_size = 0;
3265 const unsigned char *d, *cur_id_str;
3266 const struct scsi_vpd *vpd_pg83;
3267 int id_size = -EINVAL;
3268
3269 rcu_read_lock();
3270 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3271 if (!vpd_pg83) {
3272 rcu_read_unlock();
3273 return -ENXIO;
3274 }
3275
3276 /* The id string must be at least 20 bytes + terminating NULL byte */
3277 if (id_len < 21) {
3278 rcu_read_unlock();
3279 return -EINVAL;
3280 }
3281
3282 memset(id, 0, id_len);
3283 for (d = vpd_pg83->data + 4;
3284 d < vpd_pg83->data + vpd_pg83->len;
3285 d += d[3] + 4) {
3286 u8 prio = designator_prio(d);
3287
3288 if (prio == 0 || cur_id_prio > prio)
3289 continue;
3290
3291 switch (d[1] & 0xf) {
3292 case 0x1:
3293 /* T10 Vendor ID */
3294 if (cur_id_size > d[3])
3295 break;
3296 cur_id_prio = prio;
3297 cur_id_size = d[3];
3298 if (cur_id_size + 4 > id_len)
3299 cur_id_size = id_len - 4;
3300 cur_id_str = d + 4;
3301 id_size = snprintf(id, id_len, "t10.%*pE",
3302 cur_id_size, cur_id_str);
3303 break;
3304 case 0x2:
3305 /* EUI-64 */
3306 cur_id_prio = prio;
3307 cur_id_size = d[3];
3308 cur_id_str = d + 4;
3309 switch (cur_id_size) {
3310 case 8:
3311 id_size = snprintf(id, id_len,
3312 "eui.%8phN",
3313 cur_id_str);
3314 break;
3315 case 12:
3316 id_size = snprintf(id, id_len,
3317 "eui.%12phN",
3318 cur_id_str);
3319 break;
3320 case 16:
3321 id_size = snprintf(id, id_len,
3322 "eui.%16phN",
3323 cur_id_str);
3324 break;
3325 default:
3326 break;
3327 }
3328 break;
3329 case 0x3:
3330 /* NAA */
3331 cur_id_prio = prio;
3332 cur_id_size = d[3];
3333 cur_id_str = d + 4;
3334 switch (cur_id_size) {
3335 case 8:
3336 id_size = snprintf(id, id_len,
3337 "naa.%8phN",
3338 cur_id_str);
3339 break;
3340 case 16:
3341 id_size = snprintf(id, id_len,
3342 "naa.%16phN",
3343 cur_id_str);
3344 break;
3345 default:
3346 break;
3347 }
3348 break;
3349 case 0x8:
3350 /* SCSI name string */
3351 if (cur_id_size > d[3])
3352 break;
3353 /* Prefer others for truncated descriptor */
3354 if (d[3] > id_len) {
3355 prio = 2;
3356 if (cur_id_prio > prio)
3357 break;
3358 }
3359 cur_id_prio = prio;
3360 cur_id_size = id_size = d[3];
3361 cur_id_str = d + 4;
3362 if (cur_id_size >= id_len)
3363 cur_id_size = id_len - 1;
3364 memcpy(id, cur_id_str, cur_id_size);
3365 break;
3366 default:
3367 break;
3368 }
3369 }
3370 rcu_read_unlock();
3371
3372 return id_size;
3373}
3374EXPORT_SYMBOL(scsi_vpd_lun_id);
3375
3376/*
3377 * scsi_vpd_tpg_id - return a target port group identifier
3378 * @sdev: SCSI device
3379 *
3380 * Returns the Target Port Group identifier from the information
3381 * froom VPD page 0x83 of the device.
3382 *
3383 * Returns the identifier or error on failure.
3384 */
3385int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3386{
3387 const unsigned char *d;
3388 const struct scsi_vpd *vpd_pg83;
3389 int group_id = -EAGAIN, rel_port = -1;
3390
3391 rcu_read_lock();
3392 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3393 if (!vpd_pg83) {
3394 rcu_read_unlock();
3395 return -ENXIO;
3396 }
3397
3398 d = vpd_pg83->data + 4;
3399 while (d < vpd_pg83->data + vpd_pg83->len) {
3400 switch (d[1] & 0xf) {
3401 case 0x4:
3402 /* Relative target port */
3403 rel_port = get_unaligned_be16(&d[6]);
3404 break;
3405 case 0x5:
3406 /* Target port group */
3407 group_id = get_unaligned_be16(&d[6]);
3408 break;
3409 default:
3410 break;
3411 }
3412 d += d[3] + 4;
3413 }
3414 rcu_read_unlock();
3415
3416 if (group_id >= 0 && rel_id && rel_port != -1)
3417 *rel_id = rel_port;
3418
3419 return group_id;
3420}
3421EXPORT_SYMBOL(scsi_vpd_tpg_id);
3422
3423/**
3424 * scsi_build_sense - build sense data for a command
3425 * @scmd: scsi command for which the sense should be formatted
3426 * @desc: Sense format (non-zero == descriptor format,
3427 * 0 == fixed format)
3428 * @key: Sense key
3429 * @asc: Additional sense code
3430 * @ascq: Additional sense code qualifier
3431 *
3432 **/
3433void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3434{
3435 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3436 scmd->result = SAM_STAT_CHECK_CONDITION;
3437}
3438EXPORT_SYMBOL_GPL(scsi_build_sense);
3439
3440#ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3441#include "scsi_lib_test.c"
3442#endif