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/ratelimit.h>
25#include <asm/unaligned.h>
26
27#include <scsi/scsi.h>
28#include <scsi/scsi_cmnd.h>
29#include <scsi/scsi_dbg.h>
30#include <scsi/scsi_device.h>
31#include <scsi/scsi_driver.h>
32#include <scsi/scsi_eh.h>
33#include <scsi/scsi_host.h>
34#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35#include <scsi/scsi_dh.h>
36
37#include <trace/events/scsi.h>
38
39#include "scsi_debugfs.h"
40#include "scsi_priv.h"
41#include "scsi_logging.h"
42
43/*
44 * Size of integrity metadata is usually small, 1 inline sg should
45 * cover normal cases.
46 */
47#ifdef CONFIG_ARCH_NO_SG_CHAIN
48#define SCSI_INLINE_PROT_SG_CNT 0
49#define SCSI_INLINE_SG_CNT 0
50#else
51#define SCSI_INLINE_PROT_SG_CNT 1
52#define SCSI_INLINE_SG_CNT 2
53#endif
54
55static struct kmem_cache *scsi_sense_cache;
56static DEFINE_MUTEX(scsi_sense_cache_mutex);
57
58static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
59
60int scsi_init_sense_cache(struct Scsi_Host *shost)
61{
62 int ret = 0;
63
64 mutex_lock(&scsi_sense_cache_mutex);
65 if (!scsi_sense_cache) {
66 scsi_sense_cache =
67 kmem_cache_create_usercopy("scsi_sense_cache",
68 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
69 0, SCSI_SENSE_BUFFERSIZE, NULL);
70 if (!scsi_sense_cache)
71 ret = -ENOMEM;
72 }
73 mutex_unlock(&scsi_sense_cache_mutex);
74 return ret;
75}
76
77/*
78 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
79 * not change behaviour from the previous unplug mechanism, experimentation
80 * may prove this needs changing.
81 */
82#define SCSI_QUEUE_DELAY 3
83
84static void
85scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
86{
87 struct Scsi_Host *host = cmd->device->host;
88 struct scsi_device *device = cmd->device;
89 struct scsi_target *starget = scsi_target(device);
90
91 /*
92 * Set the appropriate busy bit for the device/host.
93 *
94 * If the host/device isn't busy, assume that something actually
95 * completed, and that we should be able to queue a command now.
96 *
97 * Note that the prior mid-layer assumption that any host could
98 * always queue at least one command is now broken. The mid-layer
99 * will implement a user specifiable stall (see
100 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
101 * if a command is requeued with no other commands outstanding
102 * either for the device or for the host.
103 */
104 switch (reason) {
105 case SCSI_MLQUEUE_HOST_BUSY:
106 atomic_set(&host->host_blocked, host->max_host_blocked);
107 break;
108 case SCSI_MLQUEUE_DEVICE_BUSY:
109 case SCSI_MLQUEUE_EH_RETRY:
110 atomic_set(&device->device_blocked,
111 device->max_device_blocked);
112 break;
113 case SCSI_MLQUEUE_TARGET_BUSY:
114 atomic_set(&starget->target_blocked,
115 starget->max_target_blocked);
116 break;
117 }
118}
119
120static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
121{
122 if (cmd->request->rq_flags & RQF_DONTPREP) {
123 cmd->request->rq_flags &= ~RQF_DONTPREP;
124 scsi_mq_uninit_cmd(cmd);
125 } else {
126 WARN_ON_ONCE(true);
127 }
128 blk_mq_requeue_request(cmd->request, true);
129}
130
131/**
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
136 *
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
141 * file.
142 */
143static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
144{
145 struct scsi_device *device = cmd->device;
146
147 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
148 "Inserting command %p into mlqueue\n", cmd));
149
150 scsi_set_blocked(cmd, reason);
151
152 /*
153 * Decrement the counters, since these commands are no longer
154 * active on the host/device.
155 */
156 if (unbusy)
157 scsi_device_unbusy(device, cmd);
158
159 /*
160 * Requeue this command. It will go before all other commands
161 * that are already in the queue. Schedule requeue work under
162 * lock such that the kblockd_schedule_work() call happens
163 * before blk_cleanup_queue() finishes.
164 */
165 cmd->result = 0;
166
167 blk_mq_requeue_request(cmd->request, true);
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
187
188/**
189 * __scsi_execute - insert request and wait for the result
190 * @sdev: scsi device
191 * @cmd: scsi command
192 * @data_direction: data direction
193 * @buffer: data buffer
194 * @bufflen: len of buffer
195 * @sense: optional sense buffer
196 * @sshdr: optional decoded sense header
197 * @timeout: request timeout in HZ
198 * @retries: number of times to retry request
199 * @flags: flags for ->cmd_flags
200 * @rq_flags: flags for ->rq_flags
201 * @resid: optional residual length
202 *
203 * Returns the scsi_cmnd result field if a command was executed, or a negative
204 * Linux error code if we didn't get that far.
205 */
206int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
207 int data_direction, void *buffer, unsigned bufflen,
208 unsigned char *sense, struct scsi_sense_hdr *sshdr,
209 int timeout, int retries, u64 flags, req_flags_t rq_flags,
210 int *resid)
211{
212 struct request *req;
213 struct scsi_request *rq;
214 int ret;
215
216 req = blk_get_request(sdev->request_queue,
217 data_direction == DMA_TO_DEVICE ?
218 REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
219 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
220 if (IS_ERR(req))
221 return PTR_ERR(req);
222
223 rq = scsi_req(req);
224
225 if (bufflen) {
226 ret = blk_rq_map_kern(sdev->request_queue, req,
227 buffer, bufflen, GFP_NOIO);
228 if (ret)
229 goto out;
230 }
231 rq->cmd_len = COMMAND_SIZE(cmd[0]);
232 memcpy(rq->cmd, cmd, rq->cmd_len);
233 rq->retries = retries;
234 req->timeout = timeout;
235 req->cmd_flags |= flags;
236 req->rq_flags |= rq_flags | RQF_QUIET;
237
238 /*
239 * head injection *required* here otherwise quiesce won't work
240 */
241 blk_execute_rq(NULL, req, 1);
242
243 /*
244 * Some devices (USB mass-storage in particular) may transfer
245 * garbage data together with a residue indicating that the data
246 * is invalid. Prevent the garbage from being misinterpreted
247 * and prevent security leaks by zeroing out the excess data.
248 */
249 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
250 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
251
252 if (resid)
253 *resid = rq->resid_len;
254 if (sense && rq->sense_len)
255 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
256 if (sshdr)
257 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
258 ret = rq->result;
259 out:
260 blk_put_request(req);
261
262 return ret;
263}
264EXPORT_SYMBOL(__scsi_execute);
265
266/*
267 * Wake up the error handler if necessary. Avoid as follows that the error
268 * handler is not woken up if host in-flight requests number ==
269 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
270 * with an RCU read lock in this function to ensure that this function in
271 * its entirety either finishes before scsi_eh_scmd_add() increases the
272 * host_failed counter or that it notices the shost state change made by
273 * scsi_eh_scmd_add().
274 */
275static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
276{
277 unsigned long flags;
278
279 rcu_read_lock();
280 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
281 if (unlikely(scsi_host_in_recovery(shost))) {
282 spin_lock_irqsave(shost->host_lock, flags);
283 if (shost->host_failed || shost->host_eh_scheduled)
284 scsi_eh_wakeup(shost);
285 spin_unlock_irqrestore(shost->host_lock, flags);
286 }
287 rcu_read_unlock();
288}
289
290void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
291{
292 struct Scsi_Host *shost = sdev->host;
293 struct scsi_target *starget = scsi_target(sdev);
294
295 scsi_dec_host_busy(shost, cmd);
296
297 if (starget->can_queue > 0)
298 atomic_dec(&starget->target_busy);
299
300 sbitmap_put(&sdev->budget_map, cmd->budget_token);
301 cmd->budget_token = -1;
302}
303
304static void scsi_kick_queue(struct request_queue *q)
305{
306 blk_mq_run_hw_queues(q, false);
307}
308
309/*
310 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
311 * and call blk_run_queue for all the scsi_devices on the target -
312 * including current_sdev first.
313 *
314 * Called with *no* scsi locks held.
315 */
316static void scsi_single_lun_run(struct scsi_device *current_sdev)
317{
318 struct Scsi_Host *shost = current_sdev->host;
319 struct scsi_device *sdev, *tmp;
320 struct scsi_target *starget = scsi_target(current_sdev);
321 unsigned long flags;
322
323 spin_lock_irqsave(shost->host_lock, flags);
324 starget->starget_sdev_user = NULL;
325 spin_unlock_irqrestore(shost->host_lock, flags);
326
327 /*
328 * Call blk_run_queue for all LUNs on the target, starting with
329 * current_sdev. We race with others (to set starget_sdev_user),
330 * but in most cases, we will be first. Ideally, each LU on the
331 * target would get some limited time or requests on the target.
332 */
333 scsi_kick_queue(current_sdev->request_queue);
334
335 spin_lock_irqsave(shost->host_lock, flags);
336 if (starget->starget_sdev_user)
337 goto out;
338 list_for_each_entry_safe(sdev, tmp, &starget->devices,
339 same_target_siblings) {
340 if (sdev == current_sdev)
341 continue;
342 if (scsi_device_get(sdev))
343 continue;
344
345 spin_unlock_irqrestore(shost->host_lock, flags);
346 scsi_kick_queue(sdev->request_queue);
347 spin_lock_irqsave(shost->host_lock, flags);
348
349 scsi_device_put(sdev);
350 }
351 out:
352 spin_unlock_irqrestore(shost->host_lock, flags);
353}
354
355static inline bool scsi_device_is_busy(struct scsi_device *sdev)
356{
357 if (scsi_device_busy(sdev) >= sdev->queue_depth)
358 return true;
359 if (atomic_read(&sdev->device_blocked) > 0)
360 return true;
361 return false;
362}
363
364static inline bool scsi_target_is_busy(struct scsi_target *starget)
365{
366 if (starget->can_queue > 0) {
367 if (atomic_read(&starget->target_busy) >= starget->can_queue)
368 return true;
369 if (atomic_read(&starget->target_blocked) > 0)
370 return true;
371 }
372 return false;
373}
374
375static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
376{
377 if (atomic_read(&shost->host_blocked) > 0)
378 return true;
379 if (shost->host_self_blocked)
380 return true;
381 return false;
382}
383
384static void scsi_starved_list_run(struct Scsi_Host *shost)
385{
386 LIST_HEAD(starved_list);
387 struct scsi_device *sdev;
388 unsigned long flags;
389
390 spin_lock_irqsave(shost->host_lock, flags);
391 list_splice_init(&shost->starved_list, &starved_list);
392
393 while (!list_empty(&starved_list)) {
394 struct request_queue *slq;
395
396 /*
397 * As long as shost is accepting commands and we have
398 * starved queues, call blk_run_queue. scsi_request_fn
399 * drops the queue_lock and can add us back to the
400 * starved_list.
401 *
402 * host_lock protects the starved_list and starved_entry.
403 * scsi_request_fn must get the host_lock before checking
404 * or modifying starved_list or starved_entry.
405 */
406 if (scsi_host_is_busy(shost))
407 break;
408
409 sdev = list_entry(starved_list.next,
410 struct scsi_device, starved_entry);
411 list_del_init(&sdev->starved_entry);
412 if (scsi_target_is_busy(scsi_target(sdev))) {
413 list_move_tail(&sdev->starved_entry,
414 &shost->starved_list);
415 continue;
416 }
417
418 /*
419 * Once we drop the host lock, a racing scsi_remove_device()
420 * call may remove the sdev from the starved list and destroy
421 * it and the queue. Mitigate by taking a reference to the
422 * queue and never touching the sdev again after we drop the
423 * host lock. Note: if __scsi_remove_device() invokes
424 * blk_cleanup_queue() before the queue is run from this
425 * function then blk_run_queue() will return immediately since
426 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
427 */
428 slq = sdev->request_queue;
429 if (!blk_get_queue(slq))
430 continue;
431 spin_unlock_irqrestore(shost->host_lock, flags);
432
433 scsi_kick_queue(slq);
434 blk_put_queue(slq);
435
436 spin_lock_irqsave(shost->host_lock, flags);
437 }
438 /* put any unprocessed entries back */
439 list_splice(&starved_list, &shost->starved_list);
440 spin_unlock_irqrestore(shost->host_lock, flags);
441}
442
443/**
444 * scsi_run_queue - Select a proper request queue to serve next.
445 * @q: last request's queue
446 *
447 * The previous command was completely finished, start a new one if possible.
448 */
449static void scsi_run_queue(struct request_queue *q)
450{
451 struct scsi_device *sdev = q->queuedata;
452
453 if (scsi_target(sdev)->single_lun)
454 scsi_single_lun_run(sdev);
455 if (!list_empty(&sdev->host->starved_list))
456 scsi_starved_list_run(sdev->host);
457
458 blk_mq_run_hw_queues(q, false);
459}
460
461void scsi_requeue_run_queue(struct work_struct *work)
462{
463 struct scsi_device *sdev;
464 struct request_queue *q;
465
466 sdev = container_of(work, struct scsi_device, requeue_work);
467 q = sdev->request_queue;
468 scsi_run_queue(q);
469}
470
471void scsi_run_host_queues(struct Scsi_Host *shost)
472{
473 struct scsi_device *sdev;
474
475 shost_for_each_device(sdev, shost)
476 scsi_run_queue(sdev->request_queue);
477}
478
479static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
480{
481 if (!blk_rq_is_passthrough(cmd->request)) {
482 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
483
484 if (drv->uninit_command)
485 drv->uninit_command(cmd);
486 }
487}
488
489void scsi_free_sgtables(struct scsi_cmnd *cmd)
490{
491 if (cmd->sdb.table.nents)
492 sg_free_table_chained(&cmd->sdb.table,
493 SCSI_INLINE_SG_CNT);
494 if (scsi_prot_sg_count(cmd))
495 sg_free_table_chained(&cmd->prot_sdb->table,
496 SCSI_INLINE_PROT_SG_CNT);
497}
498EXPORT_SYMBOL_GPL(scsi_free_sgtables);
499
500static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
501{
502 scsi_free_sgtables(cmd);
503 scsi_uninit_cmd(cmd);
504}
505
506static void scsi_run_queue_async(struct scsi_device *sdev)
507{
508 if (scsi_target(sdev)->single_lun ||
509 !list_empty(&sdev->host->starved_list)) {
510 kblockd_schedule_work(&sdev->requeue_work);
511 } else {
512 /*
513 * smp_mb() present in sbitmap_queue_clear() or implied in
514 * .end_io is for ordering writing .device_busy in
515 * scsi_device_unbusy() and reading sdev->restarts.
516 */
517 int old = atomic_read(&sdev->restarts);
518
519 /*
520 * ->restarts has to be kept as non-zero if new budget
521 * contention occurs.
522 *
523 * No need to run queue when either another re-run
524 * queue wins in updating ->restarts or a new budget
525 * contention occurs.
526 */
527 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
528 blk_mq_run_hw_queues(sdev->request_queue, true);
529 }
530}
531
532/* Returns false when no more bytes to process, true if there are more */
533static bool scsi_end_request(struct request *req, blk_status_t error,
534 unsigned int bytes)
535{
536 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
537 struct scsi_device *sdev = cmd->device;
538 struct request_queue *q = sdev->request_queue;
539
540 if (blk_update_request(req, error, bytes))
541 return true;
542
543 if (blk_queue_add_random(q))
544 add_disk_randomness(req->rq_disk);
545
546 if (!blk_rq_is_passthrough(req)) {
547 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
548 cmd->flags &= ~SCMD_INITIALIZED;
549 }
550
551 /*
552 * Calling rcu_barrier() is not necessary here because the
553 * SCSI error handler guarantees that the function called by
554 * call_rcu() has been called before scsi_end_request() is
555 * called.
556 */
557 destroy_rcu_head(&cmd->rcu);
558
559 /*
560 * In the MQ case the command gets freed by __blk_mq_end_request,
561 * so we have to do all cleanup that depends on it earlier.
562 *
563 * We also can't kick the queues from irq context, so we
564 * will have to defer it to a workqueue.
565 */
566 scsi_mq_uninit_cmd(cmd);
567
568 /*
569 * queue is still alive, so grab the ref for preventing it
570 * from being cleaned up during running queue.
571 */
572 percpu_ref_get(&q->q_usage_counter);
573
574 __blk_mq_end_request(req, error);
575
576 scsi_run_queue_async(sdev);
577
578 percpu_ref_put(&q->q_usage_counter);
579 return false;
580}
581
582/**
583 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
584 * @cmd: SCSI command
585 * @result: scsi error code
586 *
587 * Translate a SCSI result code into a blk_status_t value. May reset the host
588 * byte of @cmd->result.
589 */
590static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
591{
592 switch (host_byte(result)) {
593 case DID_OK:
594 if (scsi_status_is_good(result))
595 return BLK_STS_OK;
596 return BLK_STS_IOERR;
597 case DID_TRANSPORT_FAILFAST:
598 case DID_TRANSPORT_MARGINAL:
599 return BLK_STS_TRANSPORT;
600 case DID_TARGET_FAILURE:
601 set_host_byte(cmd, DID_OK);
602 return BLK_STS_TARGET;
603 case DID_NEXUS_FAILURE:
604 set_host_byte(cmd, DID_OK);
605 return BLK_STS_NEXUS;
606 case DID_ALLOC_FAILURE:
607 set_host_byte(cmd, DID_OK);
608 return BLK_STS_NOSPC;
609 case DID_MEDIUM_ERROR:
610 set_host_byte(cmd, DID_OK);
611 return BLK_STS_MEDIUM;
612 default:
613 return BLK_STS_IOERR;
614 }
615}
616
617/* Helper for scsi_io_completion() when "reprep" action required. */
618static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
619 struct request_queue *q)
620{
621 /* A new command will be prepared and issued. */
622 scsi_mq_requeue_cmd(cmd);
623}
624
625static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
626{
627 struct request *req = cmd->request;
628 unsigned long wait_for;
629
630 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
631 return false;
632
633 wait_for = (cmd->allowed + 1) * req->timeout;
634 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
635 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
636 wait_for/HZ);
637 return true;
638 }
639 return false;
640}
641
642/* Helper for scsi_io_completion() when special action required. */
643static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
644{
645 struct request_queue *q = cmd->device->request_queue;
646 struct request *req = cmd->request;
647 int level = 0;
648 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
649 ACTION_DELAYED_RETRY} action;
650 struct scsi_sense_hdr sshdr;
651 bool sense_valid;
652 bool sense_current = true; /* false implies "deferred sense" */
653 blk_status_t blk_stat;
654
655 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
656 if (sense_valid)
657 sense_current = !scsi_sense_is_deferred(&sshdr);
658
659 blk_stat = scsi_result_to_blk_status(cmd, result);
660
661 if (host_byte(result) == DID_RESET) {
662 /* Third party bus reset or reset for error recovery
663 * reasons. Just retry the command and see what
664 * happens.
665 */
666 action = ACTION_RETRY;
667 } else if (sense_valid && sense_current) {
668 switch (sshdr.sense_key) {
669 case UNIT_ATTENTION:
670 if (cmd->device->removable) {
671 /* Detected disc change. Set a bit
672 * and quietly refuse further access.
673 */
674 cmd->device->changed = 1;
675 action = ACTION_FAIL;
676 } else {
677 /* Must have been a power glitch, or a
678 * bus reset. Could not have been a
679 * media change, so we just retry the
680 * command and see what happens.
681 */
682 action = ACTION_RETRY;
683 }
684 break;
685 case ILLEGAL_REQUEST:
686 /* If we had an ILLEGAL REQUEST returned, then
687 * we may have performed an unsupported
688 * command. The only thing this should be
689 * would be a ten byte read where only a six
690 * byte read was supported. Also, on a system
691 * where READ CAPACITY failed, we may have
692 * read past the end of the disk.
693 */
694 if ((cmd->device->use_10_for_rw &&
695 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
696 (cmd->cmnd[0] == READ_10 ||
697 cmd->cmnd[0] == WRITE_10)) {
698 /* This will issue a new 6-byte command. */
699 cmd->device->use_10_for_rw = 0;
700 action = ACTION_REPREP;
701 } else if (sshdr.asc == 0x10) /* DIX */ {
702 action = ACTION_FAIL;
703 blk_stat = BLK_STS_PROTECTION;
704 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
705 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
706 action = ACTION_FAIL;
707 blk_stat = BLK_STS_TARGET;
708 } else
709 action = ACTION_FAIL;
710 break;
711 case ABORTED_COMMAND:
712 action = ACTION_FAIL;
713 if (sshdr.asc == 0x10) /* DIF */
714 blk_stat = BLK_STS_PROTECTION;
715 break;
716 case NOT_READY:
717 /* If the device is in the process of becoming
718 * ready, or has a temporary blockage, retry.
719 */
720 if (sshdr.asc == 0x04) {
721 switch (sshdr.ascq) {
722 case 0x01: /* becoming ready */
723 case 0x04: /* format in progress */
724 case 0x05: /* rebuild in progress */
725 case 0x06: /* recalculation in progress */
726 case 0x07: /* operation in progress */
727 case 0x08: /* Long write in progress */
728 case 0x09: /* self test in progress */
729 case 0x11: /* notify (enable spinup) required */
730 case 0x14: /* space allocation in progress */
731 case 0x1a: /* start stop unit in progress */
732 case 0x1b: /* sanitize in progress */
733 case 0x1d: /* configuration in progress */
734 case 0x24: /* depopulation in progress */
735 action = ACTION_DELAYED_RETRY;
736 break;
737 case 0x0a: /* ALUA state transition */
738 blk_stat = BLK_STS_AGAIN;
739 fallthrough;
740 default:
741 action = ACTION_FAIL;
742 break;
743 }
744 } else
745 action = ACTION_FAIL;
746 break;
747 case VOLUME_OVERFLOW:
748 /* See SSC3rXX or current. */
749 action = ACTION_FAIL;
750 break;
751 case DATA_PROTECT:
752 action = ACTION_FAIL;
753 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
754 (sshdr.asc == 0x55 &&
755 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
756 /* Insufficient zone resources */
757 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
758 }
759 break;
760 default:
761 action = ACTION_FAIL;
762 break;
763 }
764 } else
765 action = ACTION_FAIL;
766
767 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
768 action = ACTION_FAIL;
769
770 switch (action) {
771 case ACTION_FAIL:
772 /* Give up and fail the remainder of the request */
773 if (!(req->rq_flags & RQF_QUIET)) {
774 static DEFINE_RATELIMIT_STATE(_rs,
775 DEFAULT_RATELIMIT_INTERVAL,
776 DEFAULT_RATELIMIT_BURST);
777
778 if (unlikely(scsi_logging_level))
779 level =
780 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
781 SCSI_LOG_MLCOMPLETE_BITS);
782
783 /*
784 * if logging is enabled the failure will be printed
785 * in scsi_log_completion(), so avoid duplicate messages
786 */
787 if (!level && __ratelimit(&_rs)) {
788 scsi_print_result(cmd, NULL, FAILED);
789 if (sense_valid)
790 scsi_print_sense(cmd);
791 scsi_print_command(cmd);
792 }
793 }
794 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
795 return;
796 fallthrough;
797 case ACTION_REPREP:
798 scsi_io_completion_reprep(cmd, q);
799 break;
800 case ACTION_RETRY:
801 /* Retry the same command immediately */
802 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
803 break;
804 case ACTION_DELAYED_RETRY:
805 /* Retry the same command after a delay */
806 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
807 break;
808 }
809}
810
811/*
812 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
813 * new result that may suppress further error checking. Also modifies
814 * *blk_statp in some cases.
815 */
816static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
817 blk_status_t *blk_statp)
818{
819 bool sense_valid;
820 bool sense_current = true; /* false implies "deferred sense" */
821 struct request *req = cmd->request;
822 struct scsi_sense_hdr sshdr;
823
824 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
825 if (sense_valid)
826 sense_current = !scsi_sense_is_deferred(&sshdr);
827
828 if (blk_rq_is_passthrough(req)) {
829 if (sense_valid) {
830 /*
831 * SG_IO wants current and deferred errors
832 */
833 scsi_req(req)->sense_len =
834 min(8 + cmd->sense_buffer[7],
835 SCSI_SENSE_BUFFERSIZE);
836 }
837 if (sense_current)
838 *blk_statp = scsi_result_to_blk_status(cmd, result);
839 } else if (blk_rq_bytes(req) == 0 && sense_current) {
840 /*
841 * Flush commands do not transfers any data, and thus cannot use
842 * good_bytes != blk_rq_bytes(req) as the signal for an error.
843 * This sets *blk_statp explicitly for the problem case.
844 */
845 *blk_statp = scsi_result_to_blk_status(cmd, result);
846 }
847 /*
848 * Recovered errors need reporting, but they're always treated as
849 * success, so fiddle the result code here. For passthrough requests
850 * we already took a copy of the original into sreq->result which
851 * is what gets returned to the user
852 */
853 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
854 bool do_print = true;
855 /*
856 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
857 * skip print since caller wants ATA registers. Only occurs
858 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
859 */
860 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
861 do_print = false;
862 else if (req->rq_flags & RQF_QUIET)
863 do_print = false;
864 if (do_print)
865 scsi_print_sense(cmd);
866 result = 0;
867 /* for passthrough, *blk_statp may be set */
868 *blk_statp = BLK_STS_OK;
869 }
870 /*
871 * Another corner case: the SCSI status byte is non-zero but 'good'.
872 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
873 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
874 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
875 * intermediate statuses (both obsolete in SAM-4) as good.
876 */
877 if ((result & 0xff) && scsi_status_is_good(result)) {
878 result = 0;
879 *blk_statp = BLK_STS_OK;
880 }
881 return result;
882}
883
884/**
885 * scsi_io_completion - Completion processing for SCSI commands.
886 * @cmd: command that is finished.
887 * @good_bytes: number of processed bytes.
888 *
889 * We will finish off the specified number of sectors. If we are done, the
890 * command block will be released and the queue function will be goosed. If we
891 * are not done then we have to figure out what to do next:
892 *
893 * a) We can call scsi_io_completion_reprep(). The request will be
894 * unprepared and put back on the queue. Then a new command will
895 * be created for it. This should be used if we made forward
896 * progress, or if we want to switch from READ(10) to READ(6) for
897 * example.
898 *
899 * b) We can call scsi_io_completion_action(). The request will be
900 * put back on the queue and retried using the same command as
901 * before, possibly after a delay.
902 *
903 * c) We can call scsi_end_request() with blk_stat other than
904 * BLK_STS_OK, to fail the remainder of the request.
905 */
906void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
907{
908 int result = cmd->result;
909 struct request_queue *q = cmd->device->request_queue;
910 struct request *req = cmd->request;
911 blk_status_t blk_stat = BLK_STS_OK;
912
913 if (unlikely(result)) /* a nz result may or may not be an error */
914 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
915
916 if (unlikely(blk_rq_is_passthrough(req))) {
917 /*
918 * scsi_result_to_blk_status may have reset the host_byte
919 */
920 scsi_req(req)->result = cmd->result;
921 }
922
923 /*
924 * Next deal with any sectors which we were able to correctly
925 * handle.
926 */
927 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
928 "%u sectors total, %d bytes done.\n",
929 blk_rq_sectors(req), good_bytes));
930
931 /*
932 * Failed, zero length commands always need to drop down
933 * to retry code. Fast path should return in this block.
934 */
935 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
936 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
937 return; /* no bytes remaining */
938 }
939
940 /* Kill remainder if no retries. */
941 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
942 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
943 WARN_ONCE(true,
944 "Bytes remaining after failed, no-retry command");
945 return;
946 }
947
948 /*
949 * If there had been no error, but we have leftover bytes in the
950 * requeues just queue the command up again.
951 */
952 if (likely(result == 0))
953 scsi_io_completion_reprep(cmd, q);
954 else
955 scsi_io_completion_action(cmd, result);
956}
957
958static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
959 struct request *rq)
960{
961 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
962 !op_is_write(req_op(rq)) &&
963 sdev->host->hostt->dma_need_drain(rq);
964}
965
966/**
967 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
968 * @cmd: SCSI command data structure to initialize.
969 *
970 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
971 * for @cmd.
972 *
973 * Returns:
974 * * BLK_STS_OK - on success
975 * * BLK_STS_RESOURCE - if the failure is retryable
976 * * BLK_STS_IOERR - if the failure is fatal
977 */
978blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
979{
980 struct scsi_device *sdev = cmd->device;
981 struct request *rq = cmd->request;
982 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
983 struct scatterlist *last_sg = NULL;
984 blk_status_t ret;
985 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
986 int count;
987
988 if (WARN_ON_ONCE(!nr_segs))
989 return BLK_STS_IOERR;
990
991 /*
992 * Make sure there is space for the drain. The driver must adjust
993 * max_hw_segments to be prepared for this.
994 */
995 if (need_drain)
996 nr_segs++;
997
998 /*
999 * If sg table allocation fails, requeue request later.
1000 */
1001 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1002 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1003 return BLK_STS_RESOURCE;
1004
1005 /*
1006 * Next, walk the list, and fill in the addresses and sizes of
1007 * each segment.
1008 */
1009 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1010
1011 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1012 unsigned int pad_len =
1013 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1014
1015 last_sg->length += pad_len;
1016 cmd->extra_len += pad_len;
1017 }
1018
1019 if (need_drain) {
1020 sg_unmark_end(last_sg);
1021 last_sg = sg_next(last_sg);
1022 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1023 sg_mark_end(last_sg);
1024
1025 cmd->extra_len += sdev->dma_drain_len;
1026 count++;
1027 }
1028
1029 BUG_ON(count > cmd->sdb.table.nents);
1030 cmd->sdb.table.nents = count;
1031 cmd->sdb.length = blk_rq_payload_bytes(rq);
1032
1033 if (blk_integrity_rq(rq)) {
1034 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1035 int ivecs;
1036
1037 if (WARN_ON_ONCE(!prot_sdb)) {
1038 /*
1039 * This can happen if someone (e.g. multipath)
1040 * queues a command to a device on an adapter
1041 * that does not support DIX.
1042 */
1043 ret = BLK_STS_IOERR;
1044 goto out_free_sgtables;
1045 }
1046
1047 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1048
1049 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1050 prot_sdb->table.sgl,
1051 SCSI_INLINE_PROT_SG_CNT)) {
1052 ret = BLK_STS_RESOURCE;
1053 goto out_free_sgtables;
1054 }
1055
1056 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1057 prot_sdb->table.sgl);
1058 BUG_ON(count > ivecs);
1059 BUG_ON(count > queue_max_integrity_segments(rq->q));
1060
1061 cmd->prot_sdb = prot_sdb;
1062 cmd->prot_sdb->table.nents = count;
1063 }
1064
1065 return BLK_STS_OK;
1066out_free_sgtables:
1067 scsi_free_sgtables(cmd);
1068 return ret;
1069}
1070EXPORT_SYMBOL(scsi_alloc_sgtables);
1071
1072/**
1073 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1074 * @rq: Request associated with the SCSI command to be initialized.
1075 *
1076 * This function initializes the members of struct scsi_cmnd that must be
1077 * initialized before request processing starts and that won't be
1078 * reinitialized if a SCSI command is requeued.
1079 *
1080 * Called from inside blk_get_request() for pass-through requests and from
1081 * inside scsi_init_command() for filesystem requests.
1082 */
1083static void scsi_initialize_rq(struct request *rq)
1084{
1085 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1086
1087 scsi_req_init(&cmd->req);
1088 init_rcu_head(&cmd->rcu);
1089 cmd->jiffies_at_alloc = jiffies;
1090 cmd->retries = 0;
1091}
1092
1093/*
1094 * Only called when the request isn't completed by SCSI, and not freed by
1095 * SCSI
1096 */
1097static void scsi_cleanup_rq(struct request *rq)
1098{
1099 if (rq->rq_flags & RQF_DONTPREP) {
1100 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1101 rq->rq_flags &= ~RQF_DONTPREP;
1102 }
1103}
1104
1105/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1106void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1107{
1108 void *buf = cmd->sense_buffer;
1109 void *prot = cmd->prot_sdb;
1110 struct request *rq = blk_mq_rq_from_pdu(cmd);
1111 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1112 unsigned long jiffies_at_alloc;
1113 int retries, to_clear;
1114 bool in_flight;
1115 int budget_token = cmd->budget_token;
1116
1117 if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
1118 flags |= SCMD_INITIALIZED;
1119 scsi_initialize_rq(rq);
1120 }
1121
1122 jiffies_at_alloc = cmd->jiffies_at_alloc;
1123 retries = cmd->retries;
1124 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1125 /*
1126 * Zero out the cmd, except for the embedded scsi_request. Only clear
1127 * the driver-private command data if the LLD does not supply a
1128 * function to initialize that data.
1129 */
1130 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1131 if (!dev->host->hostt->init_cmd_priv)
1132 to_clear += dev->host->hostt->cmd_size;
1133 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1134
1135 cmd->device = dev;
1136 cmd->sense_buffer = buf;
1137 cmd->prot_sdb = prot;
1138 cmd->flags = flags;
1139 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1140 cmd->jiffies_at_alloc = jiffies_at_alloc;
1141 cmd->retries = retries;
1142 if (in_flight)
1143 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1144 cmd->budget_token = budget_token;
1145
1146}
1147
1148static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1149 struct request *req)
1150{
1151 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1152
1153 /*
1154 * Passthrough requests may transfer data, in which case they must
1155 * a bio attached to them. Or they might contain a SCSI command
1156 * that does not transfer data, in which case they may optionally
1157 * submit a request without an attached bio.
1158 */
1159 if (req->bio) {
1160 blk_status_t ret = scsi_alloc_sgtables(cmd);
1161 if (unlikely(ret != BLK_STS_OK))
1162 return ret;
1163 } else {
1164 BUG_ON(blk_rq_bytes(req));
1165
1166 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1167 }
1168
1169 cmd->cmd_len = scsi_req(req)->cmd_len;
1170 if (cmd->cmd_len == 0)
1171 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1172 cmd->cmnd = scsi_req(req)->cmd;
1173 cmd->transfersize = blk_rq_bytes(req);
1174 cmd->allowed = scsi_req(req)->retries;
1175 return BLK_STS_OK;
1176}
1177
1178static blk_status_t
1179scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1180{
1181 switch (sdev->sdev_state) {
1182 case SDEV_CREATED:
1183 return BLK_STS_OK;
1184 case SDEV_OFFLINE:
1185 case SDEV_TRANSPORT_OFFLINE:
1186 /*
1187 * If the device is offline we refuse to process any
1188 * commands. The device must be brought online
1189 * before trying any recovery commands.
1190 */
1191 if (!sdev->offline_already) {
1192 sdev->offline_already = true;
1193 sdev_printk(KERN_ERR, sdev,
1194 "rejecting I/O to offline device\n");
1195 }
1196 return BLK_STS_IOERR;
1197 case SDEV_DEL:
1198 /*
1199 * If the device is fully deleted, we refuse to
1200 * process any commands as well.
1201 */
1202 sdev_printk(KERN_ERR, sdev,
1203 "rejecting I/O to dead device\n");
1204 return BLK_STS_IOERR;
1205 case SDEV_BLOCK:
1206 case SDEV_CREATED_BLOCK:
1207 return BLK_STS_RESOURCE;
1208 case SDEV_QUIESCE:
1209 /*
1210 * If the device is blocked we only accept power management
1211 * commands.
1212 */
1213 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1214 return BLK_STS_RESOURCE;
1215 return BLK_STS_OK;
1216 default:
1217 /*
1218 * For any other not fully online state we only allow
1219 * power management commands.
1220 */
1221 if (req && !(req->rq_flags & RQF_PM))
1222 return BLK_STS_IOERR;
1223 return BLK_STS_OK;
1224 }
1225}
1226
1227/*
1228 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1229 * and return the token else return -1.
1230 */
1231static inline int scsi_dev_queue_ready(struct request_queue *q,
1232 struct scsi_device *sdev)
1233{
1234 int token;
1235
1236 token = sbitmap_get(&sdev->budget_map);
1237 if (atomic_read(&sdev->device_blocked)) {
1238 if (token < 0)
1239 goto out;
1240
1241 if (scsi_device_busy(sdev) > 1)
1242 goto out_dec;
1243
1244 /*
1245 * unblock after device_blocked iterates to zero
1246 */
1247 if (atomic_dec_return(&sdev->device_blocked) > 0)
1248 goto out_dec;
1249 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1250 "unblocking device at zero depth\n"));
1251 }
1252
1253 return token;
1254out_dec:
1255 if (token >= 0)
1256 sbitmap_put(&sdev->budget_map, token);
1257out:
1258 return -1;
1259}
1260
1261/*
1262 * scsi_target_queue_ready: checks if there we can send commands to target
1263 * @sdev: scsi device on starget to check.
1264 */
1265static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1266 struct scsi_device *sdev)
1267{
1268 struct scsi_target *starget = scsi_target(sdev);
1269 unsigned int busy;
1270
1271 if (starget->single_lun) {
1272 spin_lock_irq(shost->host_lock);
1273 if (starget->starget_sdev_user &&
1274 starget->starget_sdev_user != sdev) {
1275 spin_unlock_irq(shost->host_lock);
1276 return 0;
1277 }
1278 starget->starget_sdev_user = sdev;
1279 spin_unlock_irq(shost->host_lock);
1280 }
1281
1282 if (starget->can_queue <= 0)
1283 return 1;
1284
1285 busy = atomic_inc_return(&starget->target_busy) - 1;
1286 if (atomic_read(&starget->target_blocked) > 0) {
1287 if (busy)
1288 goto starved;
1289
1290 /*
1291 * unblock after target_blocked iterates to zero
1292 */
1293 if (atomic_dec_return(&starget->target_blocked) > 0)
1294 goto out_dec;
1295
1296 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1297 "unblocking target at zero depth\n"));
1298 }
1299
1300 if (busy >= starget->can_queue)
1301 goto starved;
1302
1303 return 1;
1304
1305starved:
1306 spin_lock_irq(shost->host_lock);
1307 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1308 spin_unlock_irq(shost->host_lock);
1309out_dec:
1310 if (starget->can_queue > 0)
1311 atomic_dec(&starget->target_busy);
1312 return 0;
1313}
1314
1315/*
1316 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1317 * return 0. We must end up running the queue again whenever 0 is
1318 * returned, else IO can hang.
1319 */
1320static inline int scsi_host_queue_ready(struct request_queue *q,
1321 struct Scsi_Host *shost,
1322 struct scsi_device *sdev,
1323 struct scsi_cmnd *cmd)
1324{
1325 if (scsi_host_in_recovery(shost))
1326 return 0;
1327
1328 if (atomic_read(&shost->host_blocked) > 0) {
1329 if (scsi_host_busy(shost) > 0)
1330 goto starved;
1331
1332 /*
1333 * unblock after host_blocked iterates to zero
1334 */
1335 if (atomic_dec_return(&shost->host_blocked) > 0)
1336 goto out_dec;
1337
1338 SCSI_LOG_MLQUEUE(3,
1339 shost_printk(KERN_INFO, shost,
1340 "unblocking host at zero depth\n"));
1341 }
1342
1343 if (shost->host_self_blocked)
1344 goto starved;
1345
1346 /* We're OK to process the command, so we can't be starved */
1347 if (!list_empty(&sdev->starved_entry)) {
1348 spin_lock_irq(shost->host_lock);
1349 if (!list_empty(&sdev->starved_entry))
1350 list_del_init(&sdev->starved_entry);
1351 spin_unlock_irq(shost->host_lock);
1352 }
1353
1354 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1355
1356 return 1;
1357
1358starved:
1359 spin_lock_irq(shost->host_lock);
1360 if (list_empty(&sdev->starved_entry))
1361 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1362 spin_unlock_irq(shost->host_lock);
1363out_dec:
1364 scsi_dec_host_busy(shost, cmd);
1365 return 0;
1366}
1367
1368/*
1369 * Busy state exporting function for request stacking drivers.
1370 *
1371 * For efficiency, no lock is taken to check the busy state of
1372 * shost/starget/sdev, since the returned value is not guaranteed and
1373 * may be changed after request stacking drivers call the function,
1374 * regardless of taking lock or not.
1375 *
1376 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1377 * needs to return 'not busy'. Otherwise, request stacking drivers
1378 * may hold requests forever.
1379 */
1380static bool scsi_mq_lld_busy(struct request_queue *q)
1381{
1382 struct scsi_device *sdev = q->queuedata;
1383 struct Scsi_Host *shost;
1384
1385 if (blk_queue_dying(q))
1386 return false;
1387
1388 shost = sdev->host;
1389
1390 /*
1391 * Ignore host/starget busy state.
1392 * Since block layer does not have a concept of fairness across
1393 * multiple queues, congestion of host/starget needs to be handled
1394 * in SCSI layer.
1395 */
1396 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1397 return true;
1398
1399 return false;
1400}
1401
1402/*
1403 * Block layer request completion callback. May be called from interrupt
1404 * context.
1405 */
1406static void scsi_complete(struct request *rq)
1407{
1408 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1409 enum scsi_disposition disposition;
1410
1411 INIT_LIST_HEAD(&cmd->eh_entry);
1412
1413 atomic_inc(&cmd->device->iodone_cnt);
1414 if (cmd->result)
1415 atomic_inc(&cmd->device->ioerr_cnt);
1416
1417 disposition = scsi_decide_disposition(cmd);
1418 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1419 disposition = SUCCESS;
1420
1421 scsi_log_completion(cmd, disposition);
1422
1423 switch (disposition) {
1424 case SUCCESS:
1425 scsi_finish_command(cmd);
1426 break;
1427 case NEEDS_RETRY:
1428 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1429 break;
1430 case ADD_TO_MLQUEUE:
1431 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1432 break;
1433 default:
1434 scsi_eh_scmd_add(cmd);
1435 break;
1436 }
1437}
1438
1439/**
1440 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1441 * @cmd: command block we are dispatching.
1442 *
1443 * Return: nonzero return request was rejected and device's queue needs to be
1444 * plugged.
1445 */
1446static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1447{
1448 struct Scsi_Host *host = cmd->device->host;
1449 int rtn = 0;
1450
1451 atomic_inc(&cmd->device->iorequest_cnt);
1452
1453 /* check if the device is still usable */
1454 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1455 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1456 * returns an immediate error upwards, and signals
1457 * that the device is no longer present */
1458 cmd->result = DID_NO_CONNECT << 16;
1459 goto done;
1460 }
1461
1462 /* Check to see if the scsi lld made this device blocked. */
1463 if (unlikely(scsi_device_blocked(cmd->device))) {
1464 /*
1465 * in blocked state, the command is just put back on
1466 * the device queue. The suspend state has already
1467 * blocked the queue so future requests should not
1468 * occur until the device transitions out of the
1469 * suspend state.
1470 */
1471 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1472 "queuecommand : device blocked\n"));
1473 return SCSI_MLQUEUE_DEVICE_BUSY;
1474 }
1475
1476 /* Store the LUN value in cmnd, if needed. */
1477 if (cmd->device->lun_in_cdb)
1478 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1479 (cmd->device->lun << 5 & 0xe0);
1480
1481 scsi_log_send(cmd);
1482
1483 /*
1484 * Before we queue this command, check if the command
1485 * length exceeds what the host adapter can handle.
1486 */
1487 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1488 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1489 "queuecommand : command too long. "
1490 "cdb_size=%d host->max_cmd_len=%d\n",
1491 cmd->cmd_len, cmd->device->host->max_cmd_len));
1492 cmd->result = (DID_ABORT << 16);
1493 goto done;
1494 }
1495
1496 if (unlikely(host->shost_state == SHOST_DEL)) {
1497 cmd->result = (DID_NO_CONNECT << 16);
1498 goto done;
1499
1500 }
1501
1502 trace_scsi_dispatch_cmd_start(cmd);
1503 rtn = host->hostt->queuecommand(host, cmd);
1504 if (rtn) {
1505 trace_scsi_dispatch_cmd_error(cmd, rtn);
1506 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1507 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1508 rtn = SCSI_MLQUEUE_HOST_BUSY;
1509
1510 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1511 "queuecommand : request rejected\n"));
1512 }
1513
1514 return rtn;
1515 done:
1516 cmd->scsi_done(cmd);
1517 return 0;
1518}
1519
1520/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1521static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1522{
1523 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1524 sizeof(struct scatterlist);
1525}
1526
1527static blk_status_t scsi_prepare_cmd(struct request *req)
1528{
1529 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1530 struct scsi_device *sdev = req->q->queuedata;
1531 struct Scsi_Host *shost = sdev->host;
1532 struct scatterlist *sg;
1533
1534 scsi_init_command(sdev, cmd);
1535
1536 cmd->request = req;
1537 cmd->tag = req->tag;
1538 cmd->prot_op = SCSI_PROT_NORMAL;
1539 if (blk_rq_bytes(req))
1540 cmd->sc_data_direction = rq_dma_dir(req);
1541 else
1542 cmd->sc_data_direction = DMA_NONE;
1543
1544 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1545 cmd->sdb.table.sgl = sg;
1546
1547 if (scsi_host_get_prot(shost)) {
1548 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1549
1550 cmd->prot_sdb->table.sgl =
1551 (struct scatterlist *)(cmd->prot_sdb + 1);
1552 }
1553
1554 /*
1555 * Special handling for passthrough commands, which don't go to the ULP
1556 * at all:
1557 */
1558 if (blk_rq_is_passthrough(req))
1559 return scsi_setup_scsi_cmnd(sdev, req);
1560
1561 if (sdev->handler && sdev->handler->prep_fn) {
1562 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1563
1564 if (ret != BLK_STS_OK)
1565 return ret;
1566 }
1567
1568 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1569 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1570 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1571}
1572
1573static void scsi_mq_done(struct scsi_cmnd *cmd)
1574{
1575 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1576 return;
1577 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1578 return;
1579 trace_scsi_dispatch_cmd_done(cmd);
1580 blk_mq_complete_request(cmd->request);
1581}
1582
1583static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1584{
1585 struct scsi_device *sdev = q->queuedata;
1586
1587 sbitmap_put(&sdev->budget_map, budget_token);
1588}
1589
1590static int scsi_mq_get_budget(struct request_queue *q)
1591{
1592 struct scsi_device *sdev = q->queuedata;
1593 int token = scsi_dev_queue_ready(q, sdev);
1594
1595 if (token >= 0)
1596 return token;
1597
1598 atomic_inc(&sdev->restarts);
1599
1600 /*
1601 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1602 * .restarts must be incremented before .device_busy is read because the
1603 * code in scsi_run_queue_async() depends on the order of these operations.
1604 */
1605 smp_mb__after_atomic();
1606
1607 /*
1608 * If all in-flight requests originated from this LUN are completed
1609 * before reading .device_busy, sdev->device_busy will be observed as
1610 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1611 * soon. Otherwise, completion of one of these requests will observe
1612 * the .restarts flag, and the request queue will be run for handling
1613 * this request, see scsi_end_request().
1614 */
1615 if (unlikely(scsi_device_busy(sdev) == 0 &&
1616 !scsi_device_blocked(sdev)))
1617 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1618 return -1;
1619}
1620
1621static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1622{
1623 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1624
1625 cmd->budget_token = token;
1626}
1627
1628static int scsi_mq_get_rq_budget_token(struct request *req)
1629{
1630 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1631
1632 return cmd->budget_token;
1633}
1634
1635static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1636 const struct blk_mq_queue_data *bd)
1637{
1638 struct request *req = bd->rq;
1639 struct request_queue *q = req->q;
1640 struct scsi_device *sdev = q->queuedata;
1641 struct Scsi_Host *shost = sdev->host;
1642 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1643 blk_status_t ret;
1644 int reason;
1645
1646 WARN_ON_ONCE(cmd->budget_token < 0);
1647
1648 /*
1649 * If the device is not in running state we will reject some or all
1650 * commands.
1651 */
1652 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1653 ret = scsi_device_state_check(sdev, req);
1654 if (ret != BLK_STS_OK)
1655 goto out_put_budget;
1656 }
1657
1658 ret = BLK_STS_RESOURCE;
1659 if (!scsi_target_queue_ready(shost, sdev))
1660 goto out_put_budget;
1661 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1662 goto out_dec_target_busy;
1663
1664 if (!(req->rq_flags & RQF_DONTPREP)) {
1665 ret = scsi_prepare_cmd(req);
1666 if (ret != BLK_STS_OK)
1667 goto out_dec_host_busy;
1668 req->rq_flags |= RQF_DONTPREP;
1669 } else {
1670 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1671 }
1672
1673 cmd->flags &= SCMD_PRESERVED_FLAGS;
1674 if (sdev->simple_tags)
1675 cmd->flags |= SCMD_TAGGED;
1676 if (bd->last)
1677 cmd->flags |= SCMD_LAST;
1678
1679 scsi_set_resid(cmd, 0);
1680 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1681 cmd->scsi_done = scsi_mq_done;
1682
1683 blk_mq_start_request(req);
1684 reason = scsi_dispatch_cmd(cmd);
1685 if (reason) {
1686 scsi_set_blocked(cmd, reason);
1687 ret = BLK_STS_RESOURCE;
1688 goto out_dec_host_busy;
1689 }
1690
1691 return BLK_STS_OK;
1692
1693out_dec_host_busy:
1694 scsi_dec_host_busy(shost, cmd);
1695out_dec_target_busy:
1696 if (scsi_target(sdev)->can_queue > 0)
1697 atomic_dec(&scsi_target(sdev)->target_busy);
1698out_put_budget:
1699 scsi_mq_put_budget(q, cmd->budget_token);
1700 cmd->budget_token = -1;
1701 switch (ret) {
1702 case BLK_STS_OK:
1703 break;
1704 case BLK_STS_RESOURCE:
1705 case BLK_STS_ZONE_RESOURCE:
1706 if (scsi_device_blocked(sdev))
1707 ret = BLK_STS_DEV_RESOURCE;
1708 break;
1709 case BLK_STS_AGAIN:
1710 scsi_req(req)->result = DID_BUS_BUSY << 16;
1711 if (req->rq_flags & RQF_DONTPREP)
1712 scsi_mq_uninit_cmd(cmd);
1713 break;
1714 default:
1715 if (unlikely(!scsi_device_online(sdev)))
1716 scsi_req(req)->result = DID_NO_CONNECT << 16;
1717 else
1718 scsi_req(req)->result = DID_ERROR << 16;
1719 /*
1720 * Make sure to release all allocated resources when
1721 * we hit an error, as we will never see this command
1722 * again.
1723 */
1724 if (req->rq_flags & RQF_DONTPREP)
1725 scsi_mq_uninit_cmd(cmd);
1726 scsi_run_queue_async(sdev);
1727 break;
1728 }
1729 return ret;
1730}
1731
1732static enum blk_eh_timer_return scsi_timeout(struct request *req,
1733 bool reserved)
1734{
1735 if (reserved)
1736 return BLK_EH_RESET_TIMER;
1737 return scsi_times_out(req);
1738}
1739
1740static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1741 unsigned int hctx_idx, unsigned int numa_node)
1742{
1743 struct Scsi_Host *shost = set->driver_data;
1744 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1745 struct scatterlist *sg;
1746 int ret = 0;
1747
1748 cmd->sense_buffer =
1749 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1750 if (!cmd->sense_buffer)
1751 return -ENOMEM;
1752 cmd->req.sense = cmd->sense_buffer;
1753
1754 if (scsi_host_get_prot(shost)) {
1755 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1756 shost->hostt->cmd_size;
1757 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1758 }
1759
1760 if (shost->hostt->init_cmd_priv) {
1761 ret = shost->hostt->init_cmd_priv(shost, cmd);
1762 if (ret < 0)
1763 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1764 }
1765
1766 return ret;
1767}
1768
1769static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1770 unsigned int hctx_idx)
1771{
1772 struct Scsi_Host *shost = set->driver_data;
1773 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1774
1775 if (shost->hostt->exit_cmd_priv)
1776 shost->hostt->exit_cmd_priv(shost, cmd);
1777 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1778}
1779
1780
1781static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
1782{
1783 struct Scsi_Host *shost = hctx->driver_data;
1784
1785 if (shost->hostt->mq_poll)
1786 return shost->hostt->mq_poll(shost, hctx->queue_num);
1787
1788 return 0;
1789}
1790
1791static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1792 unsigned int hctx_idx)
1793{
1794 struct Scsi_Host *shost = data;
1795
1796 hctx->driver_data = shost;
1797 return 0;
1798}
1799
1800static int scsi_map_queues(struct blk_mq_tag_set *set)
1801{
1802 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1803
1804 if (shost->hostt->map_queues)
1805 return shost->hostt->map_queues(shost);
1806 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1807}
1808
1809void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1810{
1811 struct device *dev = shost->dma_dev;
1812
1813 /*
1814 * this limit is imposed by hardware restrictions
1815 */
1816 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1817 SG_MAX_SEGMENTS));
1818
1819 if (scsi_host_prot_dma(shost)) {
1820 shost->sg_prot_tablesize =
1821 min_not_zero(shost->sg_prot_tablesize,
1822 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1823 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1824 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1825 }
1826
1827 if (dev->dma_mask) {
1828 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1829 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1830 }
1831 blk_queue_max_hw_sectors(q, shost->max_sectors);
1832 blk_queue_segment_boundary(q, shost->dma_boundary);
1833 dma_set_seg_boundary(dev, shost->dma_boundary);
1834
1835 blk_queue_max_segment_size(q, shost->max_segment_size);
1836 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1837 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1838
1839 /*
1840 * Set a reasonable default alignment: The larger of 32-byte (dword),
1841 * which is a common minimum for HBAs, and the minimum DMA alignment,
1842 * which is set by the platform.
1843 *
1844 * Devices that require a bigger alignment can increase it later.
1845 */
1846 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1847}
1848EXPORT_SYMBOL_GPL(__scsi_init_queue);
1849
1850static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1851 .get_budget = scsi_mq_get_budget,
1852 .put_budget = scsi_mq_put_budget,
1853 .queue_rq = scsi_queue_rq,
1854 .complete = scsi_complete,
1855 .timeout = scsi_timeout,
1856#ifdef CONFIG_BLK_DEBUG_FS
1857 .show_rq = scsi_show_rq,
1858#endif
1859 .init_request = scsi_mq_init_request,
1860 .exit_request = scsi_mq_exit_request,
1861 .initialize_rq_fn = scsi_initialize_rq,
1862 .cleanup_rq = scsi_cleanup_rq,
1863 .busy = scsi_mq_lld_busy,
1864 .map_queues = scsi_map_queues,
1865 .init_hctx = scsi_init_hctx,
1866 .poll = scsi_mq_poll,
1867 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1868 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1869};
1870
1871
1872static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1873{
1874 struct Scsi_Host *shost = hctx->driver_data;
1875
1876 shost->hostt->commit_rqs(shost, hctx->queue_num);
1877}
1878
1879static const struct blk_mq_ops scsi_mq_ops = {
1880 .get_budget = scsi_mq_get_budget,
1881 .put_budget = scsi_mq_put_budget,
1882 .queue_rq = scsi_queue_rq,
1883 .commit_rqs = scsi_commit_rqs,
1884 .complete = scsi_complete,
1885 .timeout = scsi_timeout,
1886#ifdef CONFIG_BLK_DEBUG_FS
1887 .show_rq = scsi_show_rq,
1888#endif
1889 .init_request = scsi_mq_init_request,
1890 .exit_request = scsi_mq_exit_request,
1891 .initialize_rq_fn = scsi_initialize_rq,
1892 .cleanup_rq = scsi_cleanup_rq,
1893 .busy = scsi_mq_lld_busy,
1894 .map_queues = scsi_map_queues,
1895 .init_hctx = scsi_init_hctx,
1896 .poll = scsi_mq_poll,
1897 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
1898 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
1899};
1900
1901int scsi_mq_setup_tags(struct Scsi_Host *shost)
1902{
1903 unsigned int cmd_size, sgl_size;
1904 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1905
1906 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1907 scsi_mq_inline_sgl_size(shost));
1908 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1909 if (scsi_host_get_prot(shost))
1910 cmd_size += sizeof(struct scsi_data_buffer) +
1911 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1912
1913 memset(tag_set, 0, sizeof(*tag_set));
1914 if (shost->hostt->commit_rqs)
1915 tag_set->ops = &scsi_mq_ops;
1916 else
1917 tag_set->ops = &scsi_mq_ops_no_commit;
1918 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1919 tag_set->nr_maps = shost->nr_maps ? : 1;
1920 tag_set->queue_depth = shost->can_queue;
1921 tag_set->cmd_size = cmd_size;
1922 tag_set->numa_node = NUMA_NO_NODE;
1923 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1924 tag_set->flags |=
1925 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1926 tag_set->driver_data = shost;
1927 if (shost->host_tagset)
1928 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1929
1930 return blk_mq_alloc_tag_set(tag_set);
1931}
1932
1933void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1934{
1935 blk_mq_free_tag_set(&shost->tag_set);
1936}
1937
1938/**
1939 * scsi_device_from_queue - return sdev associated with a request_queue
1940 * @q: The request queue to return the sdev from
1941 *
1942 * Return the sdev associated with a request queue or NULL if the
1943 * request_queue does not reference a SCSI device.
1944 */
1945struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1946{
1947 struct scsi_device *sdev = NULL;
1948
1949 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1950 q->mq_ops == &scsi_mq_ops)
1951 sdev = q->queuedata;
1952 if (!sdev || !get_device(&sdev->sdev_gendev))
1953 sdev = NULL;
1954
1955 return sdev;
1956}
1957
1958/**
1959 * scsi_block_requests - Utility function used by low-level drivers to prevent
1960 * further commands from being queued to the device.
1961 * @shost: host in question
1962 *
1963 * There is no timer nor any other means by which the requests get unblocked
1964 * other than the low-level driver calling scsi_unblock_requests().
1965 */
1966void scsi_block_requests(struct Scsi_Host *shost)
1967{
1968 shost->host_self_blocked = 1;
1969}
1970EXPORT_SYMBOL(scsi_block_requests);
1971
1972/**
1973 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1974 * further commands to be queued to the device.
1975 * @shost: host in question
1976 *
1977 * There is no timer nor any other means by which the requests get unblocked
1978 * other than the low-level driver calling scsi_unblock_requests(). This is done
1979 * as an API function so that changes to the internals of the scsi mid-layer
1980 * won't require wholesale changes to drivers that use this feature.
1981 */
1982void scsi_unblock_requests(struct Scsi_Host *shost)
1983{
1984 shost->host_self_blocked = 0;
1985 scsi_run_host_queues(shost);
1986}
1987EXPORT_SYMBOL(scsi_unblock_requests);
1988
1989void scsi_exit_queue(void)
1990{
1991 kmem_cache_destroy(scsi_sense_cache);
1992}
1993
1994/**
1995 * scsi_mode_select - issue a mode select
1996 * @sdev: SCSI device to be queried
1997 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1998 * @sp: Save page bit (0 == don't save, 1 == save)
1999 * @modepage: mode page being requested
2000 * @buffer: request buffer (may not be smaller than eight bytes)
2001 * @len: length of request buffer.
2002 * @timeout: command timeout
2003 * @retries: number of retries before failing
2004 * @data: returns a structure abstracting the mode header data
2005 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2006 * must be SCSI_SENSE_BUFFERSIZE big.
2007 *
2008 * Returns zero if successful; negative error number or scsi
2009 * status on error
2010 *
2011 */
2012int
2013scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2014 unsigned char *buffer, int len, int timeout, int retries,
2015 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2016{
2017 unsigned char cmd[10];
2018 unsigned char *real_buffer;
2019 int ret;
2020
2021 memset(cmd, 0, sizeof(cmd));
2022 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2023
2024 if (sdev->use_10_for_ms) {
2025 if (len > 65535)
2026 return -EINVAL;
2027 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2028 if (!real_buffer)
2029 return -ENOMEM;
2030 memcpy(real_buffer + 8, buffer, len);
2031 len += 8;
2032 real_buffer[0] = 0;
2033 real_buffer[1] = 0;
2034 real_buffer[2] = data->medium_type;
2035 real_buffer[3] = data->device_specific;
2036 real_buffer[4] = data->longlba ? 0x01 : 0;
2037 real_buffer[5] = 0;
2038 real_buffer[6] = data->block_descriptor_length >> 8;
2039 real_buffer[7] = data->block_descriptor_length;
2040
2041 cmd[0] = MODE_SELECT_10;
2042 cmd[7] = len >> 8;
2043 cmd[8] = len;
2044 } else {
2045 if (len > 255 || data->block_descriptor_length > 255 ||
2046 data->longlba)
2047 return -EINVAL;
2048
2049 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2050 if (!real_buffer)
2051 return -ENOMEM;
2052 memcpy(real_buffer + 4, buffer, len);
2053 len += 4;
2054 real_buffer[0] = 0;
2055 real_buffer[1] = data->medium_type;
2056 real_buffer[2] = data->device_specific;
2057 real_buffer[3] = data->block_descriptor_length;
2058
2059 cmd[0] = MODE_SELECT;
2060 cmd[4] = len;
2061 }
2062
2063 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2064 sshdr, timeout, retries, NULL);
2065 kfree(real_buffer);
2066 return ret;
2067}
2068EXPORT_SYMBOL_GPL(scsi_mode_select);
2069
2070/**
2071 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2072 * @sdev: SCSI device to be queried
2073 * @dbd: set if mode sense will allow block descriptors to be returned
2074 * @modepage: mode page being requested
2075 * @buffer: request buffer (may not be smaller than eight bytes)
2076 * @len: length of request buffer.
2077 * @timeout: command timeout
2078 * @retries: number of retries before failing
2079 * @data: returns a structure abstracting the mode header data
2080 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2081 * must be SCSI_SENSE_BUFFERSIZE big.
2082 *
2083 * Returns zero if successful, or a negative error number on failure
2084 */
2085int
2086scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2087 unsigned char *buffer, int len, int timeout, int retries,
2088 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2089{
2090 unsigned char cmd[12];
2091 int use_10_for_ms;
2092 int header_length;
2093 int result, retry_count = retries;
2094 struct scsi_sense_hdr my_sshdr;
2095
2096 memset(data, 0, sizeof(*data));
2097 memset(&cmd[0], 0, 12);
2098
2099 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2100 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2101 cmd[2] = modepage;
2102
2103 /* caller might not be interested in sense, but we need it */
2104 if (!sshdr)
2105 sshdr = &my_sshdr;
2106
2107 retry:
2108 use_10_for_ms = sdev->use_10_for_ms;
2109
2110 if (use_10_for_ms) {
2111 if (len < 8)
2112 len = 8;
2113
2114 cmd[0] = MODE_SENSE_10;
2115 cmd[8] = len;
2116 header_length = 8;
2117 } else {
2118 if (len < 4)
2119 len = 4;
2120
2121 cmd[0] = MODE_SENSE;
2122 cmd[4] = len;
2123 header_length = 4;
2124 }
2125
2126 memset(buffer, 0, len);
2127
2128 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2129 sshdr, timeout, retries, NULL);
2130 if (result < 0)
2131 return result;
2132
2133 /* This code looks awful: what it's doing is making sure an
2134 * ILLEGAL REQUEST sense return identifies the actual command
2135 * byte as the problem. MODE_SENSE commands can return
2136 * ILLEGAL REQUEST if the code page isn't supported */
2137
2138 if (!scsi_status_is_good(result)) {
2139 if (scsi_sense_valid(sshdr)) {
2140 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2141 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2142 /*
2143 * Invalid command operation code
2144 */
2145 if (use_10_for_ms) {
2146 sdev->use_10_for_ms = 0;
2147 goto retry;
2148 }
2149 }
2150 if (scsi_status_is_check_condition(result) &&
2151 sshdr->sense_key == UNIT_ATTENTION &&
2152 retry_count) {
2153 retry_count--;
2154 goto retry;
2155 }
2156 }
2157 return -EIO;
2158 }
2159 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2160 (modepage == 6 || modepage == 8))) {
2161 /* Initio breakage? */
2162 header_length = 0;
2163 data->length = 13;
2164 data->medium_type = 0;
2165 data->device_specific = 0;
2166 data->longlba = 0;
2167 data->block_descriptor_length = 0;
2168 } else if (use_10_for_ms) {
2169 data->length = buffer[0]*256 + buffer[1] + 2;
2170 data->medium_type = buffer[2];
2171 data->device_specific = buffer[3];
2172 data->longlba = buffer[4] & 0x01;
2173 data->block_descriptor_length = buffer[6]*256
2174 + buffer[7];
2175 } else {
2176 data->length = buffer[0] + 1;
2177 data->medium_type = buffer[1];
2178 data->device_specific = buffer[2];
2179 data->block_descriptor_length = buffer[3];
2180 }
2181 data->header_length = header_length;
2182
2183 return 0;
2184}
2185EXPORT_SYMBOL(scsi_mode_sense);
2186
2187/**
2188 * scsi_test_unit_ready - test if unit is ready
2189 * @sdev: scsi device to change the state of.
2190 * @timeout: command timeout
2191 * @retries: number of retries before failing
2192 * @sshdr: outpout pointer for decoded sense information.
2193 *
2194 * Returns zero if unsuccessful or an error if TUR failed. For
2195 * removable media, UNIT_ATTENTION sets ->changed flag.
2196 **/
2197int
2198scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2199 struct scsi_sense_hdr *sshdr)
2200{
2201 char cmd[] = {
2202 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2203 };
2204 int result;
2205
2206 /* try to eat the UNIT_ATTENTION if there are enough retries */
2207 do {
2208 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2209 timeout, 1, NULL);
2210 if (sdev->removable && scsi_sense_valid(sshdr) &&
2211 sshdr->sense_key == UNIT_ATTENTION)
2212 sdev->changed = 1;
2213 } while (scsi_sense_valid(sshdr) &&
2214 sshdr->sense_key == UNIT_ATTENTION && --retries);
2215
2216 return result;
2217}
2218EXPORT_SYMBOL(scsi_test_unit_ready);
2219
2220/**
2221 * scsi_device_set_state - Take the given device through the device state model.
2222 * @sdev: scsi device to change the state of.
2223 * @state: state to change to.
2224 *
2225 * Returns zero if successful or an error if the requested
2226 * transition is illegal.
2227 */
2228int
2229scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2230{
2231 enum scsi_device_state oldstate = sdev->sdev_state;
2232
2233 if (state == oldstate)
2234 return 0;
2235
2236 switch (state) {
2237 case SDEV_CREATED:
2238 switch (oldstate) {
2239 case SDEV_CREATED_BLOCK:
2240 break;
2241 default:
2242 goto illegal;
2243 }
2244 break;
2245
2246 case SDEV_RUNNING:
2247 switch (oldstate) {
2248 case SDEV_CREATED:
2249 case SDEV_OFFLINE:
2250 case SDEV_TRANSPORT_OFFLINE:
2251 case SDEV_QUIESCE:
2252 case SDEV_BLOCK:
2253 break;
2254 default:
2255 goto illegal;
2256 }
2257 break;
2258
2259 case SDEV_QUIESCE:
2260 switch (oldstate) {
2261 case SDEV_RUNNING:
2262 case SDEV_OFFLINE:
2263 case SDEV_TRANSPORT_OFFLINE:
2264 break;
2265 default:
2266 goto illegal;
2267 }
2268 break;
2269
2270 case SDEV_OFFLINE:
2271 case SDEV_TRANSPORT_OFFLINE:
2272 switch (oldstate) {
2273 case SDEV_CREATED:
2274 case SDEV_RUNNING:
2275 case SDEV_QUIESCE:
2276 case SDEV_BLOCK:
2277 break;
2278 default:
2279 goto illegal;
2280 }
2281 break;
2282
2283 case SDEV_BLOCK:
2284 switch (oldstate) {
2285 case SDEV_RUNNING:
2286 case SDEV_CREATED_BLOCK:
2287 case SDEV_QUIESCE:
2288 case SDEV_OFFLINE:
2289 break;
2290 default:
2291 goto illegal;
2292 }
2293 break;
2294
2295 case SDEV_CREATED_BLOCK:
2296 switch (oldstate) {
2297 case SDEV_CREATED:
2298 break;
2299 default:
2300 goto illegal;
2301 }
2302 break;
2303
2304 case SDEV_CANCEL:
2305 switch (oldstate) {
2306 case SDEV_CREATED:
2307 case SDEV_RUNNING:
2308 case SDEV_QUIESCE:
2309 case SDEV_OFFLINE:
2310 case SDEV_TRANSPORT_OFFLINE:
2311 break;
2312 default:
2313 goto illegal;
2314 }
2315 break;
2316
2317 case SDEV_DEL:
2318 switch (oldstate) {
2319 case SDEV_CREATED:
2320 case SDEV_RUNNING:
2321 case SDEV_OFFLINE:
2322 case SDEV_TRANSPORT_OFFLINE:
2323 case SDEV_CANCEL:
2324 case SDEV_BLOCK:
2325 case SDEV_CREATED_BLOCK:
2326 break;
2327 default:
2328 goto illegal;
2329 }
2330 break;
2331
2332 }
2333 sdev->offline_already = false;
2334 sdev->sdev_state = state;
2335 return 0;
2336
2337 illegal:
2338 SCSI_LOG_ERROR_RECOVERY(1,
2339 sdev_printk(KERN_ERR, sdev,
2340 "Illegal state transition %s->%s",
2341 scsi_device_state_name(oldstate),
2342 scsi_device_state_name(state))
2343 );
2344 return -EINVAL;
2345}
2346EXPORT_SYMBOL(scsi_device_set_state);
2347
2348/**
2349 * scsi_evt_emit - emit a single SCSI device uevent
2350 * @sdev: associated SCSI device
2351 * @evt: event to emit
2352 *
2353 * Send a single uevent (scsi_event) to the associated scsi_device.
2354 */
2355static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2356{
2357 int idx = 0;
2358 char *envp[3];
2359
2360 switch (evt->evt_type) {
2361 case SDEV_EVT_MEDIA_CHANGE:
2362 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2363 break;
2364 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2365 scsi_rescan_device(&sdev->sdev_gendev);
2366 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2367 break;
2368 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2369 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2370 break;
2371 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2372 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2373 break;
2374 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2375 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2376 break;
2377 case SDEV_EVT_LUN_CHANGE_REPORTED:
2378 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2379 break;
2380 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2381 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2382 break;
2383 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2384 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2385 break;
2386 default:
2387 /* do nothing */
2388 break;
2389 }
2390
2391 envp[idx++] = NULL;
2392
2393 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2394}
2395
2396/**
2397 * scsi_evt_thread - send a uevent for each scsi event
2398 * @work: work struct for scsi_device
2399 *
2400 * Dispatch queued events to their associated scsi_device kobjects
2401 * as uevents.
2402 */
2403void scsi_evt_thread(struct work_struct *work)
2404{
2405 struct scsi_device *sdev;
2406 enum scsi_device_event evt_type;
2407 LIST_HEAD(event_list);
2408
2409 sdev = container_of(work, struct scsi_device, event_work);
2410
2411 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2412 if (test_and_clear_bit(evt_type, sdev->pending_events))
2413 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2414
2415 while (1) {
2416 struct scsi_event *evt;
2417 struct list_head *this, *tmp;
2418 unsigned long flags;
2419
2420 spin_lock_irqsave(&sdev->list_lock, flags);
2421 list_splice_init(&sdev->event_list, &event_list);
2422 spin_unlock_irqrestore(&sdev->list_lock, flags);
2423
2424 if (list_empty(&event_list))
2425 break;
2426
2427 list_for_each_safe(this, tmp, &event_list) {
2428 evt = list_entry(this, struct scsi_event, node);
2429 list_del(&evt->node);
2430 scsi_evt_emit(sdev, evt);
2431 kfree(evt);
2432 }
2433 }
2434}
2435
2436/**
2437 * sdev_evt_send - send asserted event to uevent thread
2438 * @sdev: scsi_device event occurred on
2439 * @evt: event to send
2440 *
2441 * Assert scsi device event asynchronously.
2442 */
2443void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2444{
2445 unsigned long flags;
2446
2447#if 0
2448 /* FIXME: currently this check eliminates all media change events
2449 * for polled devices. Need to update to discriminate between AN
2450 * and polled events */
2451 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2452 kfree(evt);
2453 return;
2454 }
2455#endif
2456
2457 spin_lock_irqsave(&sdev->list_lock, flags);
2458 list_add_tail(&evt->node, &sdev->event_list);
2459 schedule_work(&sdev->event_work);
2460 spin_unlock_irqrestore(&sdev->list_lock, flags);
2461}
2462EXPORT_SYMBOL_GPL(sdev_evt_send);
2463
2464/**
2465 * sdev_evt_alloc - allocate a new scsi event
2466 * @evt_type: type of event to allocate
2467 * @gfpflags: GFP flags for allocation
2468 *
2469 * Allocates and returns a new scsi_event.
2470 */
2471struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2472 gfp_t gfpflags)
2473{
2474 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2475 if (!evt)
2476 return NULL;
2477
2478 evt->evt_type = evt_type;
2479 INIT_LIST_HEAD(&evt->node);
2480
2481 /* evt_type-specific initialization, if any */
2482 switch (evt_type) {
2483 case SDEV_EVT_MEDIA_CHANGE:
2484 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2485 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2486 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2487 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2488 case SDEV_EVT_LUN_CHANGE_REPORTED:
2489 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2490 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2491 default:
2492 /* do nothing */
2493 break;
2494 }
2495
2496 return evt;
2497}
2498EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2499
2500/**
2501 * sdev_evt_send_simple - send asserted event to uevent thread
2502 * @sdev: scsi_device event occurred on
2503 * @evt_type: type of event to send
2504 * @gfpflags: GFP flags for allocation
2505 *
2506 * Assert scsi device event asynchronously, given an event type.
2507 */
2508void sdev_evt_send_simple(struct scsi_device *sdev,
2509 enum scsi_device_event evt_type, gfp_t gfpflags)
2510{
2511 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2512 if (!evt) {
2513 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2514 evt_type);
2515 return;
2516 }
2517
2518 sdev_evt_send(sdev, evt);
2519}
2520EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2521
2522/**
2523 * scsi_device_quiesce - Block all commands except power management.
2524 * @sdev: scsi device to quiesce.
2525 *
2526 * This works by trying to transition to the SDEV_QUIESCE state
2527 * (which must be a legal transition). When the device is in this
2528 * state, only power management requests will be accepted, all others will
2529 * be deferred.
2530 *
2531 * Must be called with user context, may sleep.
2532 *
2533 * Returns zero if unsuccessful or an error if not.
2534 */
2535int
2536scsi_device_quiesce(struct scsi_device *sdev)
2537{
2538 struct request_queue *q = sdev->request_queue;
2539 int err;
2540
2541 /*
2542 * It is allowed to call scsi_device_quiesce() multiple times from
2543 * the same context but concurrent scsi_device_quiesce() calls are
2544 * not allowed.
2545 */
2546 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2547
2548 if (sdev->quiesced_by == current)
2549 return 0;
2550
2551 blk_set_pm_only(q);
2552
2553 blk_mq_freeze_queue(q);
2554 /*
2555 * Ensure that the effect of blk_set_pm_only() will be visible
2556 * for percpu_ref_tryget() callers that occur after the queue
2557 * unfreeze even if the queue was already frozen before this function
2558 * was called. See also https://lwn.net/Articles/573497/.
2559 */
2560 synchronize_rcu();
2561 blk_mq_unfreeze_queue(q);
2562
2563 mutex_lock(&sdev->state_mutex);
2564 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2565 if (err == 0)
2566 sdev->quiesced_by = current;
2567 else
2568 blk_clear_pm_only(q);
2569 mutex_unlock(&sdev->state_mutex);
2570
2571 return err;
2572}
2573EXPORT_SYMBOL(scsi_device_quiesce);
2574
2575/**
2576 * scsi_device_resume - Restart user issued commands to a quiesced device.
2577 * @sdev: scsi device to resume.
2578 *
2579 * Moves the device from quiesced back to running and restarts the
2580 * queues.
2581 *
2582 * Must be called with user context, may sleep.
2583 */
2584void scsi_device_resume(struct scsi_device *sdev)
2585{
2586 /* check if the device state was mutated prior to resume, and if
2587 * so assume the state is being managed elsewhere (for example
2588 * device deleted during suspend)
2589 */
2590 mutex_lock(&sdev->state_mutex);
2591 if (sdev->sdev_state == SDEV_QUIESCE)
2592 scsi_device_set_state(sdev, SDEV_RUNNING);
2593 if (sdev->quiesced_by) {
2594 sdev->quiesced_by = NULL;
2595 blk_clear_pm_only(sdev->request_queue);
2596 }
2597 mutex_unlock(&sdev->state_mutex);
2598}
2599EXPORT_SYMBOL(scsi_device_resume);
2600
2601static void
2602device_quiesce_fn(struct scsi_device *sdev, void *data)
2603{
2604 scsi_device_quiesce(sdev);
2605}
2606
2607void
2608scsi_target_quiesce(struct scsi_target *starget)
2609{
2610 starget_for_each_device(starget, NULL, device_quiesce_fn);
2611}
2612EXPORT_SYMBOL(scsi_target_quiesce);
2613
2614static void
2615device_resume_fn(struct scsi_device *sdev, void *data)
2616{
2617 scsi_device_resume(sdev);
2618}
2619
2620void
2621scsi_target_resume(struct scsi_target *starget)
2622{
2623 starget_for_each_device(starget, NULL, device_resume_fn);
2624}
2625EXPORT_SYMBOL(scsi_target_resume);
2626
2627/**
2628 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2629 * @sdev: device to block
2630 *
2631 * Pause SCSI command processing on the specified device. Does not sleep.
2632 *
2633 * Returns zero if successful or a negative error code upon failure.
2634 *
2635 * Notes:
2636 * This routine transitions the device to the SDEV_BLOCK state (which must be
2637 * a legal transition). When the device is in this state, command processing
2638 * is paused until the device leaves the SDEV_BLOCK state. See also
2639 * scsi_internal_device_unblock_nowait().
2640 */
2641int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2642{
2643 struct request_queue *q = sdev->request_queue;
2644 int err = 0;
2645
2646 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2647 if (err) {
2648 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2649
2650 if (err)
2651 return err;
2652 }
2653
2654 /*
2655 * The device has transitioned to SDEV_BLOCK. Stop the
2656 * block layer from calling the midlayer with this device's
2657 * request queue.
2658 */
2659 blk_mq_quiesce_queue_nowait(q);
2660 return 0;
2661}
2662EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2663
2664/**
2665 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2666 * @sdev: device to block
2667 *
2668 * Pause SCSI command processing on the specified device and wait until all
2669 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2670 *
2671 * Returns zero if successful or a negative error code upon failure.
2672 *
2673 * Note:
2674 * This routine transitions the device to the SDEV_BLOCK state (which must be
2675 * a legal transition). When the device is in this state, command processing
2676 * is paused until the device leaves the SDEV_BLOCK state. See also
2677 * scsi_internal_device_unblock().
2678 */
2679static int scsi_internal_device_block(struct scsi_device *sdev)
2680{
2681 struct request_queue *q = sdev->request_queue;
2682 int err;
2683
2684 mutex_lock(&sdev->state_mutex);
2685 err = scsi_internal_device_block_nowait(sdev);
2686 if (err == 0)
2687 blk_mq_quiesce_queue(q);
2688 mutex_unlock(&sdev->state_mutex);
2689
2690 return err;
2691}
2692
2693void scsi_start_queue(struct scsi_device *sdev)
2694{
2695 struct request_queue *q = sdev->request_queue;
2696
2697 blk_mq_unquiesce_queue(q);
2698}
2699
2700/**
2701 * scsi_internal_device_unblock_nowait - resume a device after a block request
2702 * @sdev: device to resume
2703 * @new_state: state to set the device to after unblocking
2704 *
2705 * Restart the device queue for a previously suspended SCSI device. Does not
2706 * sleep.
2707 *
2708 * Returns zero if successful or a negative error code upon failure.
2709 *
2710 * Notes:
2711 * This routine transitions the device to the SDEV_RUNNING state or to one of
2712 * the offline states (which must be a legal transition) allowing the midlayer
2713 * to goose the queue for this device.
2714 */
2715int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2716 enum scsi_device_state new_state)
2717{
2718 switch (new_state) {
2719 case SDEV_RUNNING:
2720 case SDEV_TRANSPORT_OFFLINE:
2721 break;
2722 default:
2723 return -EINVAL;
2724 }
2725
2726 /*
2727 * Try to transition the scsi device to SDEV_RUNNING or one of the
2728 * offlined states and goose the device queue if successful.
2729 */
2730 switch (sdev->sdev_state) {
2731 case SDEV_BLOCK:
2732 case SDEV_TRANSPORT_OFFLINE:
2733 sdev->sdev_state = new_state;
2734 break;
2735 case SDEV_CREATED_BLOCK:
2736 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2737 new_state == SDEV_OFFLINE)
2738 sdev->sdev_state = new_state;
2739 else
2740 sdev->sdev_state = SDEV_CREATED;
2741 break;
2742 case SDEV_CANCEL:
2743 case SDEV_OFFLINE:
2744 break;
2745 default:
2746 return -EINVAL;
2747 }
2748 scsi_start_queue(sdev);
2749
2750 return 0;
2751}
2752EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2753
2754/**
2755 * scsi_internal_device_unblock - resume a device after a block request
2756 * @sdev: device to resume
2757 * @new_state: state to set the device to after unblocking
2758 *
2759 * Restart the device queue for a previously suspended SCSI device. May sleep.
2760 *
2761 * Returns zero if successful or a negative error code upon failure.
2762 *
2763 * Notes:
2764 * This routine transitions the device to the SDEV_RUNNING state or to one of
2765 * the offline states (which must be a legal transition) allowing the midlayer
2766 * to goose the queue for this device.
2767 */
2768static int scsi_internal_device_unblock(struct scsi_device *sdev,
2769 enum scsi_device_state new_state)
2770{
2771 int ret;
2772
2773 mutex_lock(&sdev->state_mutex);
2774 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2775 mutex_unlock(&sdev->state_mutex);
2776
2777 return ret;
2778}
2779
2780static void
2781device_block(struct scsi_device *sdev, void *data)
2782{
2783 int ret;
2784
2785 ret = scsi_internal_device_block(sdev);
2786
2787 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2788 dev_name(&sdev->sdev_gendev), ret);
2789}
2790
2791static int
2792target_block(struct device *dev, void *data)
2793{
2794 if (scsi_is_target_device(dev))
2795 starget_for_each_device(to_scsi_target(dev), NULL,
2796 device_block);
2797 return 0;
2798}
2799
2800void
2801scsi_target_block(struct device *dev)
2802{
2803 if (scsi_is_target_device(dev))
2804 starget_for_each_device(to_scsi_target(dev), NULL,
2805 device_block);
2806 else
2807 device_for_each_child(dev, NULL, target_block);
2808}
2809EXPORT_SYMBOL_GPL(scsi_target_block);
2810
2811static void
2812device_unblock(struct scsi_device *sdev, void *data)
2813{
2814 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2815}
2816
2817static int
2818target_unblock(struct device *dev, void *data)
2819{
2820 if (scsi_is_target_device(dev))
2821 starget_for_each_device(to_scsi_target(dev), data,
2822 device_unblock);
2823 return 0;
2824}
2825
2826void
2827scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2828{
2829 if (scsi_is_target_device(dev))
2830 starget_for_each_device(to_scsi_target(dev), &new_state,
2831 device_unblock);
2832 else
2833 device_for_each_child(dev, &new_state, target_unblock);
2834}
2835EXPORT_SYMBOL_GPL(scsi_target_unblock);
2836
2837int
2838scsi_host_block(struct Scsi_Host *shost)
2839{
2840 struct scsi_device *sdev;
2841 int ret = 0;
2842
2843 /*
2844 * Call scsi_internal_device_block_nowait so we can avoid
2845 * calling synchronize_rcu() for each LUN.
2846 */
2847 shost_for_each_device(sdev, shost) {
2848 mutex_lock(&sdev->state_mutex);
2849 ret = scsi_internal_device_block_nowait(sdev);
2850 mutex_unlock(&sdev->state_mutex);
2851 if (ret) {
2852 scsi_device_put(sdev);
2853 break;
2854 }
2855 }
2856
2857 /*
2858 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2859 * calling synchronize_rcu() once is enough.
2860 */
2861 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2862
2863 if (!ret)
2864 synchronize_rcu();
2865
2866 return ret;
2867}
2868EXPORT_SYMBOL_GPL(scsi_host_block);
2869
2870int
2871scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2872{
2873 struct scsi_device *sdev;
2874 int ret = 0;
2875
2876 shost_for_each_device(sdev, shost) {
2877 ret = scsi_internal_device_unblock(sdev, new_state);
2878 if (ret) {
2879 scsi_device_put(sdev);
2880 break;
2881 }
2882 }
2883 return ret;
2884}
2885EXPORT_SYMBOL_GPL(scsi_host_unblock);
2886
2887/**
2888 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2889 * @sgl: scatter-gather list
2890 * @sg_count: number of segments in sg
2891 * @offset: offset in bytes into sg, on return offset into the mapped area
2892 * @len: bytes to map, on return number of bytes mapped
2893 *
2894 * Returns virtual address of the start of the mapped page
2895 */
2896void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2897 size_t *offset, size_t *len)
2898{
2899 int i;
2900 size_t sg_len = 0, len_complete = 0;
2901 struct scatterlist *sg;
2902 struct page *page;
2903
2904 WARN_ON(!irqs_disabled());
2905
2906 for_each_sg(sgl, sg, sg_count, i) {
2907 len_complete = sg_len; /* Complete sg-entries */
2908 sg_len += sg->length;
2909 if (sg_len > *offset)
2910 break;
2911 }
2912
2913 if (unlikely(i == sg_count)) {
2914 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2915 "elements %d\n",
2916 __func__, sg_len, *offset, sg_count);
2917 WARN_ON(1);
2918 return NULL;
2919 }
2920
2921 /* Offset starting from the beginning of first page in this sg-entry */
2922 *offset = *offset - len_complete + sg->offset;
2923
2924 /* Assumption: contiguous pages can be accessed as "page + i" */
2925 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2926 *offset &= ~PAGE_MASK;
2927
2928 /* Bytes in this sg-entry from *offset to the end of the page */
2929 sg_len = PAGE_SIZE - *offset;
2930 if (*len > sg_len)
2931 *len = sg_len;
2932
2933 return kmap_atomic(page);
2934}
2935EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2936
2937/**
2938 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2939 * @virt: virtual address to be unmapped
2940 */
2941void scsi_kunmap_atomic_sg(void *virt)
2942{
2943 kunmap_atomic(virt);
2944}
2945EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2946
2947void sdev_disable_disk_events(struct scsi_device *sdev)
2948{
2949 atomic_inc(&sdev->disk_events_disable_depth);
2950}
2951EXPORT_SYMBOL(sdev_disable_disk_events);
2952
2953void sdev_enable_disk_events(struct scsi_device *sdev)
2954{
2955 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2956 return;
2957 atomic_dec(&sdev->disk_events_disable_depth);
2958}
2959EXPORT_SYMBOL(sdev_enable_disk_events);
2960
2961static unsigned char designator_prio(const unsigned char *d)
2962{
2963 if (d[1] & 0x30)
2964 /* not associated with LUN */
2965 return 0;
2966
2967 if (d[3] == 0)
2968 /* invalid length */
2969 return 0;
2970
2971 /*
2972 * Order of preference for lun descriptor:
2973 * - SCSI name string
2974 * - NAA IEEE Registered Extended
2975 * - EUI-64 based 16-byte
2976 * - EUI-64 based 12-byte
2977 * - NAA IEEE Registered
2978 * - NAA IEEE Extended
2979 * - EUI-64 based 8-byte
2980 * - SCSI name string (truncated)
2981 * - T10 Vendor ID
2982 * as longer descriptors reduce the likelyhood
2983 * of identification clashes.
2984 */
2985
2986 switch (d[1] & 0xf) {
2987 case 8:
2988 /* SCSI name string, variable-length UTF-8 */
2989 return 9;
2990 case 3:
2991 switch (d[4] >> 4) {
2992 case 6:
2993 /* NAA registered extended */
2994 return 8;
2995 case 5:
2996 /* NAA registered */
2997 return 5;
2998 case 4:
2999 /* NAA extended */
3000 return 4;
3001 case 3:
3002 /* NAA locally assigned */
3003 return 1;
3004 default:
3005 break;
3006 }
3007 break;
3008 case 2:
3009 switch (d[3]) {
3010 case 16:
3011 /* EUI64-based, 16 byte */
3012 return 7;
3013 case 12:
3014 /* EUI64-based, 12 byte */
3015 return 6;
3016 case 8:
3017 /* EUI64-based, 8 byte */
3018 return 3;
3019 default:
3020 break;
3021 }
3022 break;
3023 case 1:
3024 /* T10 vendor ID */
3025 return 1;
3026 default:
3027 break;
3028 }
3029
3030 return 0;
3031}
3032
3033/**
3034 * scsi_vpd_lun_id - return a unique device identification
3035 * @sdev: SCSI device
3036 * @id: buffer for the identification
3037 * @id_len: length of the buffer
3038 *
3039 * Copies a unique device identification into @id based
3040 * on the information in the VPD page 0x83 of the device.
3041 * The string will be formatted as a SCSI name string.
3042 *
3043 * Returns the length of the identification or error on failure.
3044 * If the identifier is longer than the supplied buffer the actual
3045 * identifier length is returned and the buffer is not zero-padded.
3046 */
3047int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3048{
3049 u8 cur_id_prio = 0;
3050 u8 cur_id_size = 0;
3051 const unsigned char *d, *cur_id_str;
3052 const struct scsi_vpd *vpd_pg83;
3053 int id_size = -EINVAL;
3054
3055 rcu_read_lock();
3056 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3057 if (!vpd_pg83) {
3058 rcu_read_unlock();
3059 return -ENXIO;
3060 }
3061
3062 /* The id string must be at least 20 bytes + terminating NULL byte */
3063 if (id_len < 21) {
3064 rcu_read_unlock();
3065 return -EINVAL;
3066 }
3067
3068 memset(id, 0, id_len);
3069 for (d = vpd_pg83->data + 4;
3070 d < vpd_pg83->data + vpd_pg83->len;
3071 d += d[3] + 4) {
3072 u8 prio = designator_prio(d);
3073
3074 if (prio == 0 || cur_id_prio > prio)
3075 continue;
3076
3077 switch (d[1] & 0xf) {
3078 case 0x1:
3079 /* T10 Vendor ID */
3080 if (cur_id_size > d[3])
3081 break;
3082 cur_id_prio = prio;
3083 cur_id_size = d[3];
3084 if (cur_id_size + 4 > id_len)
3085 cur_id_size = id_len - 4;
3086 cur_id_str = d + 4;
3087 id_size = snprintf(id, id_len, "t10.%*pE",
3088 cur_id_size, cur_id_str);
3089 break;
3090 case 0x2:
3091 /* EUI-64 */
3092 cur_id_prio = prio;
3093 cur_id_size = d[3];
3094 cur_id_str = d + 4;
3095 switch (cur_id_size) {
3096 case 8:
3097 id_size = snprintf(id, id_len,
3098 "eui.%8phN",
3099 cur_id_str);
3100 break;
3101 case 12:
3102 id_size = snprintf(id, id_len,
3103 "eui.%12phN",
3104 cur_id_str);
3105 break;
3106 case 16:
3107 id_size = snprintf(id, id_len,
3108 "eui.%16phN",
3109 cur_id_str);
3110 break;
3111 default:
3112 break;
3113 }
3114 break;
3115 case 0x3:
3116 /* NAA */
3117 cur_id_prio = prio;
3118 cur_id_size = d[3];
3119 cur_id_str = d + 4;
3120 switch (cur_id_size) {
3121 case 8:
3122 id_size = snprintf(id, id_len,
3123 "naa.%8phN",
3124 cur_id_str);
3125 break;
3126 case 16:
3127 id_size = snprintf(id, id_len,
3128 "naa.%16phN",
3129 cur_id_str);
3130 break;
3131 default:
3132 break;
3133 }
3134 break;
3135 case 0x8:
3136 /* SCSI name string */
3137 if (cur_id_size > d[3])
3138 break;
3139 /* Prefer others for truncated descriptor */
3140 if (d[3] > id_len) {
3141 prio = 2;
3142 if (cur_id_prio > prio)
3143 break;
3144 }
3145 cur_id_prio = prio;
3146 cur_id_size = id_size = d[3];
3147 cur_id_str = d + 4;
3148 if (cur_id_size >= id_len)
3149 cur_id_size = id_len - 1;
3150 memcpy(id, cur_id_str, cur_id_size);
3151 break;
3152 default:
3153 break;
3154 }
3155 }
3156 rcu_read_unlock();
3157
3158 return id_size;
3159}
3160EXPORT_SYMBOL(scsi_vpd_lun_id);
3161
3162/*
3163 * scsi_vpd_tpg_id - return a target port group identifier
3164 * @sdev: SCSI device
3165 *
3166 * Returns the Target Port Group identifier from the information
3167 * froom VPD page 0x83 of the device.
3168 *
3169 * Returns the identifier or error on failure.
3170 */
3171int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3172{
3173 const unsigned char *d;
3174 const struct scsi_vpd *vpd_pg83;
3175 int group_id = -EAGAIN, rel_port = -1;
3176
3177 rcu_read_lock();
3178 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3179 if (!vpd_pg83) {
3180 rcu_read_unlock();
3181 return -ENXIO;
3182 }
3183
3184 d = vpd_pg83->data + 4;
3185 while (d < vpd_pg83->data + vpd_pg83->len) {
3186 switch (d[1] & 0xf) {
3187 case 0x4:
3188 /* Relative target port */
3189 rel_port = get_unaligned_be16(&d[6]);
3190 break;
3191 case 0x5:
3192 /* Target port group */
3193 group_id = get_unaligned_be16(&d[6]);
3194 break;
3195 default:
3196 break;
3197 }
3198 d += d[3] + 4;
3199 }
3200 rcu_read_unlock();
3201
3202 if (group_id >= 0 && rel_id && rel_port != -1)
3203 *rel_id = rel_port;
3204
3205 return group_id;
3206}
3207EXPORT_SYMBOL(scsi_vpd_tpg_id);
3208
3209/**
3210 * scsi_build_sense - build sense data for a command
3211 * @scmd: scsi command for which the sense should be formatted
3212 * @desc: Sense format (non-zero == descriptor format,
3213 * 0 == fixed format)
3214 * @key: Sense key
3215 * @asc: Additional sense code
3216 * @ascq: Additional sense code qualifier
3217 *
3218 **/
3219void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3220{
3221 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3222 scmd->result = SAM_STAT_CHECK_CONDITION;
3223}
3224EXPORT_SYMBOL_GPL(scsi_build_sense);