Loading...
1/*
2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
3 *
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
8 */
9
10#include <linux/bio.h>
11#include <linux/bitops.h>
12#include <linux/blkdev.h>
13#include <linux/completion.h>
14#include <linux/kernel.h>
15#include <linux/export.h>
16#include <linux/mempool.h>
17#include <linux/slab.h>
18#include <linux/init.h>
19#include <linux/pci.h>
20#include <linux/delay.h>
21#include <linux/hardirq.h>
22#include <linux/scatterlist.h>
23
24#include <scsi/scsi.h>
25#include <scsi/scsi_cmnd.h>
26#include <scsi/scsi_dbg.h>
27#include <scsi/scsi_device.h>
28#include <scsi/scsi_driver.h>
29#include <scsi/scsi_eh.h>
30#include <scsi/scsi_host.h>
31
32#include "scsi_priv.h"
33#include "scsi_logging.h"
34
35
36#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
37#define SG_MEMPOOL_SIZE 2
38
39struct scsi_host_sg_pool {
40 size_t size;
41 char *name;
42 struct kmem_cache *slab;
43 mempool_t *pool;
44};
45
46#define SP(x) { x, "sgpool-" __stringify(x) }
47#if (SCSI_MAX_SG_SEGMENTS < 32)
48#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
49#endif
50static struct scsi_host_sg_pool scsi_sg_pools[] = {
51 SP(8),
52 SP(16),
53#if (SCSI_MAX_SG_SEGMENTS > 32)
54 SP(32),
55#if (SCSI_MAX_SG_SEGMENTS > 64)
56 SP(64),
57#if (SCSI_MAX_SG_SEGMENTS > 128)
58 SP(128),
59#if (SCSI_MAX_SG_SEGMENTS > 256)
60#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
61#endif
62#endif
63#endif
64#endif
65 SP(SCSI_MAX_SG_SEGMENTS)
66};
67#undef SP
68
69struct kmem_cache *scsi_sdb_cache;
70
71/*
72 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
73 * not change behaviour from the previous unplug mechanism, experimentation
74 * may prove this needs changing.
75 */
76#define SCSI_QUEUE_DELAY 3
77
78/*
79 * Function: scsi_unprep_request()
80 *
81 * Purpose: Remove all preparation done for a request, including its
82 * associated scsi_cmnd, so that it can be requeued.
83 *
84 * Arguments: req - request to unprepare
85 *
86 * Lock status: Assumed that no locks are held upon entry.
87 *
88 * Returns: Nothing.
89 */
90static void scsi_unprep_request(struct request *req)
91{
92 struct scsi_cmnd *cmd = req->special;
93
94 blk_unprep_request(req);
95 req->special = NULL;
96
97 scsi_put_command(cmd);
98}
99
100/**
101 * __scsi_queue_insert - private queue insertion
102 * @cmd: The SCSI command being requeued
103 * @reason: The reason for the requeue
104 * @unbusy: Whether the queue should be unbusied
105 *
106 * This is a private queue insertion. The public interface
107 * scsi_queue_insert() always assumes the queue should be unbusied
108 * because it's always called before the completion. This function is
109 * for a requeue after completion, which should only occur in this
110 * file.
111 */
112static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
113{
114 struct Scsi_Host *host = cmd->device->host;
115 struct scsi_device *device = cmd->device;
116 struct scsi_target *starget = scsi_target(device);
117 struct request_queue *q = device->request_queue;
118 unsigned long flags;
119
120 SCSI_LOG_MLQUEUE(1,
121 printk("Inserting command %p into mlqueue\n", cmd));
122
123 /*
124 * Set the appropriate busy bit for the device/host.
125 *
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
128 *
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
135 */
136 switch (reason) {
137 case SCSI_MLQUEUE_HOST_BUSY:
138 host->host_blocked = host->max_host_blocked;
139 break;
140 case SCSI_MLQUEUE_DEVICE_BUSY:
141 case SCSI_MLQUEUE_EH_RETRY:
142 device->device_blocked = device->max_device_blocked;
143 break;
144 case SCSI_MLQUEUE_TARGET_BUSY:
145 starget->target_blocked = starget->max_target_blocked;
146 break;
147 }
148
149 /*
150 * Decrement the counters, since these commands are no longer
151 * active on the host/device.
152 */
153 if (unbusy)
154 scsi_device_unbusy(device);
155
156 /*
157 * Requeue this command. It will go before all other commands
158 * that are already in the queue.
159 */
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
163
164 kblockd_schedule_work(q, &device->requeue_work);
165
166 return 0;
167}
168
169/*
170 * Function: scsi_queue_insert()
171 *
172 * Purpose: Insert a command in the midlevel queue.
173 *
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
176 *
177 * Lock status: Assumed that lock is not held upon entry.
178 *
179 * Returns: Nothing.
180 *
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
184 * commands.
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
187 */
188int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
189{
190 return __scsi_queue_insert(cmd, reason, 1);
191}
192/**
193 * scsi_execute - insert request and wait for the result
194 * @sdev: scsi device
195 * @cmd: scsi command
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
204 *
205 * returns the req->errors value which is the scsi_cmnd result
206 * field.
207 */
208int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209 int data_direction, void *buffer, unsigned bufflen,
210 unsigned char *sense, int timeout, int retries, int flags,
211 int *resid)
212{
213 struct request *req;
214 int write = (data_direction == DMA_TO_DEVICE);
215 int ret = DRIVER_ERROR << 24;
216
217 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
218 if (!req)
219 return ret;
220
221 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
222 buffer, bufflen, __GFP_WAIT))
223 goto out;
224
225 req->cmd_len = COMMAND_SIZE(cmd[0]);
226 memcpy(req->cmd, cmd, req->cmd_len);
227 req->sense = sense;
228 req->sense_len = 0;
229 req->retries = retries;
230 req->timeout = timeout;
231 req->cmd_type = REQ_TYPE_BLOCK_PC;
232 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
233
234 /*
235 * head injection *required* here otherwise quiesce won't work
236 */
237 blk_execute_rq(req->q, NULL, req, 1);
238
239 /*
240 * Some devices (USB mass-storage in particular) may transfer
241 * garbage data together with a residue indicating that the data
242 * is invalid. Prevent the garbage from being misinterpreted
243 * and prevent security leaks by zeroing out the excess data.
244 */
245 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
246 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
247
248 if (resid)
249 *resid = req->resid_len;
250 ret = req->errors;
251 out:
252 blk_put_request(req);
253
254 return ret;
255}
256EXPORT_SYMBOL(scsi_execute);
257
258
259int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
260 int data_direction, void *buffer, unsigned bufflen,
261 struct scsi_sense_hdr *sshdr, int timeout, int retries,
262 int *resid)
263{
264 char *sense = NULL;
265 int result;
266
267 if (sshdr) {
268 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
269 if (!sense)
270 return DRIVER_ERROR << 24;
271 }
272 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
273 sense, timeout, retries, 0, resid);
274 if (sshdr)
275 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
276
277 kfree(sense);
278 return result;
279}
280EXPORT_SYMBOL(scsi_execute_req);
281
282/*
283 * Function: scsi_init_cmd_errh()
284 *
285 * Purpose: Initialize cmd fields related to error handling.
286 *
287 * Arguments: cmd - command that is ready to be queued.
288 *
289 * Notes: This function has the job of initializing a number of
290 * fields related to error handling. Typically this will
291 * be called once for each command, as required.
292 */
293static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
294{
295 cmd->serial_number = 0;
296 scsi_set_resid(cmd, 0);
297 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
298 if (cmd->cmd_len == 0)
299 cmd->cmd_len = scsi_command_size(cmd->cmnd);
300}
301
302void scsi_device_unbusy(struct scsi_device *sdev)
303{
304 struct Scsi_Host *shost = sdev->host;
305 struct scsi_target *starget = scsi_target(sdev);
306 unsigned long flags;
307
308 spin_lock_irqsave(shost->host_lock, flags);
309 shost->host_busy--;
310 starget->target_busy--;
311 if (unlikely(scsi_host_in_recovery(shost) &&
312 (shost->host_failed || shost->host_eh_scheduled)))
313 scsi_eh_wakeup(shost);
314 spin_unlock(shost->host_lock);
315 spin_lock(sdev->request_queue->queue_lock);
316 sdev->device_busy--;
317 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
318}
319
320/*
321 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
322 * and call blk_run_queue for all the scsi_devices on the target -
323 * including current_sdev first.
324 *
325 * Called with *no* scsi locks held.
326 */
327static void scsi_single_lun_run(struct scsi_device *current_sdev)
328{
329 struct Scsi_Host *shost = current_sdev->host;
330 struct scsi_device *sdev, *tmp;
331 struct scsi_target *starget = scsi_target(current_sdev);
332 unsigned long flags;
333
334 spin_lock_irqsave(shost->host_lock, flags);
335 starget->starget_sdev_user = NULL;
336 spin_unlock_irqrestore(shost->host_lock, flags);
337
338 /*
339 * Call blk_run_queue for all LUNs on the target, starting with
340 * current_sdev. We race with others (to set starget_sdev_user),
341 * but in most cases, we will be first. Ideally, each LU on the
342 * target would get some limited time or requests on the target.
343 */
344 blk_run_queue(current_sdev->request_queue);
345
346 spin_lock_irqsave(shost->host_lock, flags);
347 if (starget->starget_sdev_user)
348 goto out;
349 list_for_each_entry_safe(sdev, tmp, &starget->devices,
350 same_target_siblings) {
351 if (sdev == current_sdev)
352 continue;
353 if (scsi_device_get(sdev))
354 continue;
355
356 spin_unlock_irqrestore(shost->host_lock, flags);
357 blk_run_queue(sdev->request_queue);
358 spin_lock_irqsave(shost->host_lock, flags);
359
360 scsi_device_put(sdev);
361 }
362 out:
363 spin_unlock_irqrestore(shost->host_lock, flags);
364}
365
366static inline int scsi_device_is_busy(struct scsi_device *sdev)
367{
368 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
369 return 1;
370
371 return 0;
372}
373
374static inline int scsi_target_is_busy(struct scsi_target *starget)
375{
376 return ((starget->can_queue > 0 &&
377 starget->target_busy >= starget->can_queue) ||
378 starget->target_blocked);
379}
380
381static inline int scsi_host_is_busy(struct Scsi_Host *shost)
382{
383 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
384 shost->host_blocked || shost->host_self_blocked)
385 return 1;
386
387 return 0;
388}
389
390/*
391 * Function: scsi_run_queue()
392 *
393 * Purpose: Select a proper request queue to serve next
394 *
395 * Arguments: q - last request's queue
396 *
397 * Returns: Nothing
398 *
399 * Notes: The previous command was completely finished, start
400 * a new one if possible.
401 */
402static void scsi_run_queue(struct request_queue *q)
403{
404 struct scsi_device *sdev = q->queuedata;
405 struct Scsi_Host *shost;
406 LIST_HEAD(starved_list);
407 unsigned long flags;
408
409 shost = sdev->host;
410 if (scsi_target(sdev)->single_lun)
411 scsi_single_lun_run(sdev);
412
413 spin_lock_irqsave(shost->host_lock, flags);
414 list_splice_init(&shost->starved_list, &starved_list);
415
416 while (!list_empty(&starved_list)) {
417 /*
418 * As long as shost is accepting commands and we have
419 * starved queues, call blk_run_queue. scsi_request_fn
420 * drops the queue_lock and can add us back to the
421 * starved_list.
422 *
423 * host_lock protects the starved_list and starved_entry.
424 * scsi_request_fn must get the host_lock before checking
425 * or modifying starved_list or starved_entry.
426 */
427 if (scsi_host_is_busy(shost))
428 break;
429
430 sdev = list_entry(starved_list.next,
431 struct scsi_device, starved_entry);
432 list_del_init(&sdev->starved_entry);
433 if (scsi_target_is_busy(scsi_target(sdev))) {
434 list_move_tail(&sdev->starved_entry,
435 &shost->starved_list);
436 continue;
437 }
438
439 spin_unlock(shost->host_lock);
440 spin_lock(sdev->request_queue->queue_lock);
441 __blk_run_queue(sdev->request_queue);
442 spin_unlock(sdev->request_queue->queue_lock);
443 spin_lock(shost->host_lock);
444 }
445 /* put any unprocessed entries back */
446 list_splice(&starved_list, &shost->starved_list);
447 spin_unlock_irqrestore(shost->host_lock, flags);
448
449 blk_run_queue(q);
450}
451
452void scsi_requeue_run_queue(struct work_struct *work)
453{
454 struct scsi_device *sdev;
455 struct request_queue *q;
456
457 sdev = container_of(work, struct scsi_device, requeue_work);
458 q = sdev->request_queue;
459 scsi_run_queue(q);
460}
461
462/*
463 * Function: scsi_requeue_command()
464 *
465 * Purpose: Handle post-processing of completed commands.
466 *
467 * Arguments: q - queue to operate on
468 * cmd - command that may need to be requeued.
469 *
470 * Returns: Nothing
471 *
472 * Notes: After command completion, there may be blocks left
473 * over which weren't finished by the previous command
474 * this can be for a number of reasons - the main one is
475 * I/O errors in the middle of the request, in which case
476 * we need to request the blocks that come after the bad
477 * sector.
478 * Notes: Upon return, cmd is a stale pointer.
479 */
480static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
481{
482 struct scsi_device *sdev = cmd->device;
483 struct request *req = cmd->request;
484 unsigned long flags;
485
486 /*
487 * We need to hold a reference on the device to avoid the queue being
488 * killed after the unlock and before scsi_run_queue is invoked which
489 * may happen because scsi_unprep_request() puts the command which
490 * releases its reference on the device.
491 */
492 get_device(&sdev->sdev_gendev);
493
494 spin_lock_irqsave(q->queue_lock, flags);
495 scsi_unprep_request(req);
496 blk_requeue_request(q, req);
497 spin_unlock_irqrestore(q->queue_lock, flags);
498
499 scsi_run_queue(q);
500
501 put_device(&sdev->sdev_gendev);
502}
503
504void scsi_next_command(struct scsi_cmnd *cmd)
505{
506 struct scsi_device *sdev = cmd->device;
507 struct request_queue *q = sdev->request_queue;
508
509 /* need to hold a reference on the device before we let go of the cmd */
510 get_device(&sdev->sdev_gendev);
511
512 scsi_put_command(cmd);
513 scsi_run_queue(q);
514
515 /* ok to remove device now */
516 put_device(&sdev->sdev_gendev);
517}
518
519void scsi_run_host_queues(struct Scsi_Host *shost)
520{
521 struct scsi_device *sdev;
522
523 shost_for_each_device(sdev, shost)
524 scsi_run_queue(sdev->request_queue);
525}
526
527static void __scsi_release_buffers(struct scsi_cmnd *, int);
528
529/*
530 * Function: scsi_end_request()
531 *
532 * Purpose: Post-processing of completed commands (usually invoked at end
533 * of upper level post-processing and scsi_io_completion).
534 *
535 * Arguments: cmd - command that is complete.
536 * error - 0 if I/O indicates success, < 0 for I/O error.
537 * bytes - number of bytes of completed I/O
538 * requeue - indicates whether we should requeue leftovers.
539 *
540 * Lock status: Assumed that lock is not held upon entry.
541 *
542 * Returns: cmd if requeue required, NULL otherwise.
543 *
544 * Notes: This is called for block device requests in order to
545 * mark some number of sectors as complete.
546 *
547 * We are guaranteeing that the request queue will be goosed
548 * at some point during this call.
549 * Notes: If cmd was requeued, upon return it will be a stale pointer.
550 */
551static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
552 int bytes, int requeue)
553{
554 struct request_queue *q = cmd->device->request_queue;
555 struct request *req = cmd->request;
556
557 /*
558 * If there are blocks left over at the end, set up the command
559 * to queue the remainder of them.
560 */
561 if (blk_end_request(req, error, bytes)) {
562 /* kill remainder if no retrys */
563 if (error && scsi_noretry_cmd(cmd))
564 blk_end_request_all(req, error);
565 else {
566 if (requeue) {
567 /*
568 * Bleah. Leftovers again. Stick the
569 * leftovers in the front of the
570 * queue, and goose the queue again.
571 */
572 scsi_release_buffers(cmd);
573 scsi_requeue_command(q, cmd);
574 cmd = NULL;
575 }
576 return cmd;
577 }
578 }
579
580 /*
581 * This will goose the queue request function at the end, so we don't
582 * need to worry about launching another command.
583 */
584 __scsi_release_buffers(cmd, 0);
585 scsi_next_command(cmd);
586 return NULL;
587}
588
589static inline unsigned int scsi_sgtable_index(unsigned short nents)
590{
591 unsigned int index;
592
593 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
594
595 if (nents <= 8)
596 index = 0;
597 else
598 index = get_count_order(nents) - 3;
599
600 return index;
601}
602
603static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
604{
605 struct scsi_host_sg_pool *sgp;
606
607 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
608 mempool_free(sgl, sgp->pool);
609}
610
611static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
612{
613 struct scsi_host_sg_pool *sgp;
614
615 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
616 return mempool_alloc(sgp->pool, gfp_mask);
617}
618
619static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
620 gfp_t gfp_mask)
621{
622 int ret;
623
624 BUG_ON(!nents);
625
626 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
627 gfp_mask, scsi_sg_alloc);
628 if (unlikely(ret))
629 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
630 scsi_sg_free);
631
632 return ret;
633}
634
635static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
636{
637 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
638}
639
640static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
641{
642
643 if (cmd->sdb.table.nents)
644 scsi_free_sgtable(&cmd->sdb);
645
646 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
647
648 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
649 struct scsi_data_buffer *bidi_sdb =
650 cmd->request->next_rq->special;
651 scsi_free_sgtable(bidi_sdb);
652 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
653 cmd->request->next_rq->special = NULL;
654 }
655
656 if (scsi_prot_sg_count(cmd))
657 scsi_free_sgtable(cmd->prot_sdb);
658}
659
660/*
661 * Function: scsi_release_buffers()
662 *
663 * Purpose: Completion processing for block device I/O requests.
664 *
665 * Arguments: cmd - command that we are bailing.
666 *
667 * Lock status: Assumed that no lock is held upon entry.
668 *
669 * Returns: Nothing
670 *
671 * Notes: In the event that an upper level driver rejects a
672 * command, we must release resources allocated during
673 * the __init_io() function. Primarily this would involve
674 * the scatter-gather table, and potentially any bounce
675 * buffers.
676 */
677void scsi_release_buffers(struct scsi_cmnd *cmd)
678{
679 __scsi_release_buffers(cmd, 1);
680}
681EXPORT_SYMBOL(scsi_release_buffers);
682
683static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
684{
685 int error = 0;
686
687 switch(host_byte(result)) {
688 case DID_TRANSPORT_FAILFAST:
689 error = -ENOLINK;
690 break;
691 case DID_TARGET_FAILURE:
692 set_host_byte(cmd, DID_OK);
693 error = -EREMOTEIO;
694 break;
695 case DID_NEXUS_FAILURE:
696 set_host_byte(cmd, DID_OK);
697 error = -EBADE;
698 break;
699 default:
700 error = -EIO;
701 break;
702 }
703
704 return error;
705}
706
707/*
708 * Function: scsi_io_completion()
709 *
710 * Purpose: Completion processing for block device I/O requests.
711 *
712 * Arguments: cmd - command that is finished.
713 *
714 * Lock status: Assumed that no lock is held upon entry.
715 *
716 * Returns: Nothing
717 *
718 * Notes: This function is matched in terms of capabilities to
719 * the function that created the scatter-gather list.
720 * In other words, if there are no bounce buffers
721 * (the normal case for most drivers), we don't need
722 * the logic to deal with cleaning up afterwards.
723 *
724 * We must call scsi_end_request(). This will finish off
725 * the specified number of sectors. If we are done, the
726 * command block will be released and the queue function
727 * will be goosed. If we are not done then we have to
728 * figure out what to do next:
729 *
730 * a) We can call scsi_requeue_command(). The request
731 * will be unprepared and put back on the queue. Then
732 * a new command will be created for it. This should
733 * be used if we made forward progress, or if we want
734 * to switch from READ(10) to READ(6) for example.
735 *
736 * b) We can call scsi_queue_insert(). The request will
737 * be put back on the queue and retried using the same
738 * command as before, possibly after a delay.
739 *
740 * c) We can call blk_end_request() with -EIO to fail
741 * the remainder of the request.
742 */
743void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
744{
745 int result = cmd->result;
746 struct request_queue *q = cmd->device->request_queue;
747 struct request *req = cmd->request;
748 int error = 0;
749 struct scsi_sense_hdr sshdr;
750 int sense_valid = 0;
751 int sense_deferred = 0;
752 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
753 ACTION_DELAYED_RETRY} action;
754 char *description = NULL;
755
756 if (result) {
757 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
758 if (sense_valid)
759 sense_deferred = scsi_sense_is_deferred(&sshdr);
760 }
761
762 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
763 if (result) {
764 if (sense_valid && req->sense) {
765 /*
766 * SG_IO wants current and deferred errors
767 */
768 int len = 8 + cmd->sense_buffer[7];
769
770 if (len > SCSI_SENSE_BUFFERSIZE)
771 len = SCSI_SENSE_BUFFERSIZE;
772 memcpy(req->sense, cmd->sense_buffer, len);
773 req->sense_len = len;
774 }
775 if (!sense_deferred)
776 error = __scsi_error_from_host_byte(cmd, result);
777 }
778 /*
779 * __scsi_error_from_host_byte may have reset the host_byte
780 */
781 req->errors = cmd->result;
782
783 req->resid_len = scsi_get_resid(cmd);
784
785 if (scsi_bidi_cmnd(cmd)) {
786 /*
787 * Bidi commands Must be complete as a whole,
788 * both sides at once.
789 */
790 req->next_rq->resid_len = scsi_in(cmd)->resid;
791
792 scsi_release_buffers(cmd);
793 blk_end_request_all(req, 0);
794
795 scsi_next_command(cmd);
796 return;
797 }
798 }
799
800 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
801 BUG_ON(blk_bidi_rq(req));
802
803 /*
804 * Next deal with any sectors which we were able to correctly
805 * handle.
806 */
807 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
808 "%d bytes done.\n",
809 blk_rq_sectors(req), good_bytes));
810
811 /*
812 * Recovered errors need reporting, but they're always treated
813 * as success, so fiddle the result code here. For BLOCK_PC
814 * we already took a copy of the original into rq->errors which
815 * is what gets returned to the user
816 */
817 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
818 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
819 * print since caller wants ATA registers. Only occurs on
820 * SCSI ATA PASS_THROUGH commands when CK_COND=1
821 */
822 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
823 ;
824 else if (!(req->cmd_flags & REQ_QUIET))
825 scsi_print_sense("", cmd);
826 result = 0;
827 /* BLOCK_PC may have set error */
828 error = 0;
829 }
830
831 /*
832 * A number of bytes were successfully read. If there
833 * are leftovers and there is some kind of error
834 * (result != 0), retry the rest.
835 */
836 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
837 return;
838
839 error = __scsi_error_from_host_byte(cmd, result);
840
841 if (host_byte(result) == DID_RESET) {
842 /* Third party bus reset or reset for error recovery
843 * reasons. Just retry the command and see what
844 * happens.
845 */
846 action = ACTION_RETRY;
847 } else if (sense_valid && !sense_deferred) {
848 switch (sshdr.sense_key) {
849 case UNIT_ATTENTION:
850 if (cmd->device->removable) {
851 /* Detected disc change. Set a bit
852 * and quietly refuse further access.
853 */
854 cmd->device->changed = 1;
855 description = "Media Changed";
856 action = ACTION_FAIL;
857 } else {
858 /* Must have been a power glitch, or a
859 * bus reset. Could not have been a
860 * media change, so we just retry the
861 * command and see what happens.
862 */
863 action = ACTION_RETRY;
864 }
865 break;
866 case ILLEGAL_REQUEST:
867 /* If we had an ILLEGAL REQUEST returned, then
868 * we may have performed an unsupported
869 * command. The only thing this should be
870 * would be a ten byte read where only a six
871 * byte read was supported. Also, on a system
872 * where READ CAPACITY failed, we may have
873 * read past the end of the disk.
874 */
875 if ((cmd->device->use_10_for_rw &&
876 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
877 (cmd->cmnd[0] == READ_10 ||
878 cmd->cmnd[0] == WRITE_10)) {
879 /* This will issue a new 6-byte command. */
880 cmd->device->use_10_for_rw = 0;
881 action = ACTION_REPREP;
882 } else if (sshdr.asc == 0x10) /* DIX */ {
883 description = "Host Data Integrity Failure";
884 action = ACTION_FAIL;
885 error = -EILSEQ;
886 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
887 } else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
888 (cmd->cmnd[0] == UNMAP ||
889 cmd->cmnd[0] == WRITE_SAME_16 ||
890 cmd->cmnd[0] == WRITE_SAME)) {
891 description = "Discard failure";
892 action = ACTION_FAIL;
893 error = -EREMOTEIO;
894 } else
895 action = ACTION_FAIL;
896 break;
897 case ABORTED_COMMAND:
898 action = ACTION_FAIL;
899 if (sshdr.asc == 0x10) { /* DIF */
900 description = "Target Data Integrity Failure";
901 error = -EILSEQ;
902 }
903 break;
904 case NOT_READY:
905 /* If the device is in the process of becoming
906 * ready, or has a temporary blockage, retry.
907 */
908 if (sshdr.asc == 0x04) {
909 switch (sshdr.ascq) {
910 case 0x01: /* becoming ready */
911 case 0x04: /* format in progress */
912 case 0x05: /* rebuild in progress */
913 case 0x06: /* recalculation in progress */
914 case 0x07: /* operation in progress */
915 case 0x08: /* Long write in progress */
916 case 0x09: /* self test in progress */
917 case 0x14: /* space allocation in progress */
918 action = ACTION_DELAYED_RETRY;
919 break;
920 default:
921 description = "Device not ready";
922 action = ACTION_FAIL;
923 break;
924 }
925 } else {
926 description = "Device not ready";
927 action = ACTION_FAIL;
928 }
929 break;
930 case VOLUME_OVERFLOW:
931 /* See SSC3rXX or current. */
932 action = ACTION_FAIL;
933 break;
934 default:
935 description = "Unhandled sense code";
936 action = ACTION_FAIL;
937 break;
938 }
939 } else {
940 description = "Unhandled error code";
941 action = ACTION_FAIL;
942 }
943
944 switch (action) {
945 case ACTION_FAIL:
946 /* Give up and fail the remainder of the request */
947 scsi_release_buffers(cmd);
948 if (!(req->cmd_flags & REQ_QUIET)) {
949 if (description)
950 scmd_printk(KERN_INFO, cmd, "%s\n",
951 description);
952 scsi_print_result(cmd);
953 if (driver_byte(result) & DRIVER_SENSE)
954 scsi_print_sense("", cmd);
955 scsi_print_command(cmd);
956 }
957 if (blk_end_request_err(req, error))
958 scsi_requeue_command(q, cmd);
959 else
960 scsi_next_command(cmd);
961 break;
962 case ACTION_REPREP:
963 /* Unprep the request and put it back at the head of the queue.
964 * A new command will be prepared and issued.
965 */
966 scsi_release_buffers(cmd);
967 scsi_requeue_command(q, cmd);
968 break;
969 case ACTION_RETRY:
970 /* Retry the same command immediately */
971 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
972 break;
973 case ACTION_DELAYED_RETRY:
974 /* Retry the same command after a delay */
975 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
976 break;
977 }
978}
979
980static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
981 gfp_t gfp_mask)
982{
983 int count;
984
985 /*
986 * If sg table allocation fails, requeue request later.
987 */
988 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
989 gfp_mask))) {
990 return BLKPREP_DEFER;
991 }
992
993 req->buffer = NULL;
994
995 /*
996 * Next, walk the list, and fill in the addresses and sizes of
997 * each segment.
998 */
999 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1000 BUG_ON(count > sdb->table.nents);
1001 sdb->table.nents = count;
1002 sdb->length = blk_rq_bytes(req);
1003 return BLKPREP_OK;
1004}
1005
1006/*
1007 * Function: scsi_init_io()
1008 *
1009 * Purpose: SCSI I/O initialize function.
1010 *
1011 * Arguments: cmd - Command descriptor we wish to initialize
1012 *
1013 * Returns: 0 on success
1014 * BLKPREP_DEFER if the failure is retryable
1015 * BLKPREP_KILL if the failure is fatal
1016 */
1017int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1018{
1019 struct request *rq = cmd->request;
1020
1021 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1022 if (error)
1023 goto err_exit;
1024
1025 if (blk_bidi_rq(rq)) {
1026 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1027 scsi_sdb_cache, GFP_ATOMIC);
1028 if (!bidi_sdb) {
1029 error = BLKPREP_DEFER;
1030 goto err_exit;
1031 }
1032
1033 rq->next_rq->special = bidi_sdb;
1034 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1035 if (error)
1036 goto err_exit;
1037 }
1038
1039 if (blk_integrity_rq(rq)) {
1040 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1041 int ivecs, count;
1042
1043 BUG_ON(prot_sdb == NULL);
1044 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1045
1046 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1047 error = BLKPREP_DEFER;
1048 goto err_exit;
1049 }
1050
1051 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1052 prot_sdb->table.sgl);
1053 BUG_ON(unlikely(count > ivecs));
1054 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1055
1056 cmd->prot_sdb = prot_sdb;
1057 cmd->prot_sdb->table.nents = count;
1058 }
1059
1060 return BLKPREP_OK ;
1061
1062err_exit:
1063 scsi_release_buffers(cmd);
1064 cmd->request->special = NULL;
1065 scsi_put_command(cmd);
1066 return error;
1067}
1068EXPORT_SYMBOL(scsi_init_io);
1069
1070static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1071 struct request *req)
1072{
1073 struct scsi_cmnd *cmd;
1074
1075 if (!req->special) {
1076 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1077 if (unlikely(!cmd))
1078 return NULL;
1079 req->special = cmd;
1080 } else {
1081 cmd = req->special;
1082 }
1083
1084 /* pull a tag out of the request if we have one */
1085 cmd->tag = req->tag;
1086 cmd->request = req;
1087
1088 cmd->cmnd = req->cmd;
1089 cmd->prot_op = SCSI_PROT_NORMAL;
1090
1091 return cmd;
1092}
1093
1094int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1095{
1096 struct scsi_cmnd *cmd;
1097 int ret = scsi_prep_state_check(sdev, req);
1098
1099 if (ret != BLKPREP_OK)
1100 return ret;
1101
1102 cmd = scsi_get_cmd_from_req(sdev, req);
1103 if (unlikely(!cmd))
1104 return BLKPREP_DEFER;
1105
1106 /*
1107 * BLOCK_PC requests may transfer data, in which case they must
1108 * a bio attached to them. Or they might contain a SCSI command
1109 * that does not transfer data, in which case they may optionally
1110 * submit a request without an attached bio.
1111 */
1112 if (req->bio) {
1113 int ret;
1114
1115 BUG_ON(!req->nr_phys_segments);
1116
1117 ret = scsi_init_io(cmd, GFP_ATOMIC);
1118 if (unlikely(ret))
1119 return ret;
1120 } else {
1121 BUG_ON(blk_rq_bytes(req));
1122
1123 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1124 req->buffer = NULL;
1125 }
1126
1127 cmd->cmd_len = req->cmd_len;
1128 if (!blk_rq_bytes(req))
1129 cmd->sc_data_direction = DMA_NONE;
1130 else if (rq_data_dir(req) == WRITE)
1131 cmd->sc_data_direction = DMA_TO_DEVICE;
1132 else
1133 cmd->sc_data_direction = DMA_FROM_DEVICE;
1134
1135 cmd->transfersize = blk_rq_bytes(req);
1136 cmd->allowed = req->retries;
1137 return BLKPREP_OK;
1138}
1139EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1140
1141/*
1142 * Setup a REQ_TYPE_FS command. These are simple read/write request
1143 * from filesystems that still need to be translated to SCSI CDBs from
1144 * the ULD.
1145 */
1146int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1147{
1148 struct scsi_cmnd *cmd;
1149 int ret = scsi_prep_state_check(sdev, req);
1150
1151 if (ret != BLKPREP_OK)
1152 return ret;
1153
1154 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1155 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1156 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1157 if (ret != BLKPREP_OK)
1158 return ret;
1159 }
1160
1161 /*
1162 * Filesystem requests must transfer data.
1163 */
1164 BUG_ON(!req->nr_phys_segments);
1165
1166 cmd = scsi_get_cmd_from_req(sdev, req);
1167 if (unlikely(!cmd))
1168 return BLKPREP_DEFER;
1169
1170 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1171 return scsi_init_io(cmd, GFP_ATOMIC);
1172}
1173EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1174
1175int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1176{
1177 int ret = BLKPREP_OK;
1178
1179 /*
1180 * If the device is not in running state we will reject some
1181 * or all commands.
1182 */
1183 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1184 switch (sdev->sdev_state) {
1185 case SDEV_OFFLINE:
1186 /*
1187 * If the device is offline we refuse to process any
1188 * commands. The device must be brought online
1189 * before trying any recovery commands.
1190 */
1191 sdev_printk(KERN_ERR, sdev,
1192 "rejecting I/O to offline device\n");
1193 ret = BLKPREP_KILL;
1194 break;
1195 case SDEV_DEL:
1196 /*
1197 * If the device is fully deleted, we refuse to
1198 * process any commands as well.
1199 */
1200 sdev_printk(KERN_ERR, sdev,
1201 "rejecting I/O to dead device\n");
1202 ret = BLKPREP_KILL;
1203 break;
1204 case SDEV_QUIESCE:
1205 case SDEV_BLOCK:
1206 case SDEV_CREATED_BLOCK:
1207 /*
1208 * If the devices is blocked we defer normal commands.
1209 */
1210 if (!(req->cmd_flags & REQ_PREEMPT))
1211 ret = BLKPREP_DEFER;
1212 break;
1213 default:
1214 /*
1215 * For any other not fully online state we only allow
1216 * special commands. In particular any user initiated
1217 * command is not allowed.
1218 */
1219 if (!(req->cmd_flags & REQ_PREEMPT))
1220 ret = BLKPREP_KILL;
1221 break;
1222 }
1223 }
1224 return ret;
1225}
1226EXPORT_SYMBOL(scsi_prep_state_check);
1227
1228int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1229{
1230 struct scsi_device *sdev = q->queuedata;
1231
1232 switch (ret) {
1233 case BLKPREP_KILL:
1234 req->errors = DID_NO_CONNECT << 16;
1235 /* release the command and kill it */
1236 if (req->special) {
1237 struct scsi_cmnd *cmd = req->special;
1238 scsi_release_buffers(cmd);
1239 scsi_put_command(cmd);
1240 req->special = NULL;
1241 }
1242 break;
1243 case BLKPREP_DEFER:
1244 /*
1245 * If we defer, the blk_peek_request() returns NULL, but the
1246 * queue must be restarted, so we schedule a callback to happen
1247 * shortly.
1248 */
1249 if (sdev->device_busy == 0)
1250 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1251 break;
1252 default:
1253 req->cmd_flags |= REQ_DONTPREP;
1254 }
1255
1256 return ret;
1257}
1258EXPORT_SYMBOL(scsi_prep_return);
1259
1260int scsi_prep_fn(struct request_queue *q, struct request *req)
1261{
1262 struct scsi_device *sdev = q->queuedata;
1263 int ret = BLKPREP_KILL;
1264
1265 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1266 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1267 return scsi_prep_return(q, req, ret);
1268}
1269EXPORT_SYMBOL(scsi_prep_fn);
1270
1271/*
1272 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1273 * return 0.
1274 *
1275 * Called with the queue_lock held.
1276 */
1277static inline int scsi_dev_queue_ready(struct request_queue *q,
1278 struct scsi_device *sdev)
1279{
1280 if (sdev->device_busy == 0 && sdev->device_blocked) {
1281 /*
1282 * unblock after device_blocked iterates to zero
1283 */
1284 if (--sdev->device_blocked == 0) {
1285 SCSI_LOG_MLQUEUE(3,
1286 sdev_printk(KERN_INFO, sdev,
1287 "unblocking device at zero depth\n"));
1288 } else {
1289 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 return 0;
1291 }
1292 }
1293 if (scsi_device_is_busy(sdev))
1294 return 0;
1295
1296 return 1;
1297}
1298
1299
1300/*
1301 * scsi_target_queue_ready: checks if there we can send commands to target
1302 * @sdev: scsi device on starget to check.
1303 *
1304 * Called with the host lock held.
1305 */
1306static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1307 struct scsi_device *sdev)
1308{
1309 struct scsi_target *starget = scsi_target(sdev);
1310
1311 if (starget->single_lun) {
1312 if (starget->starget_sdev_user &&
1313 starget->starget_sdev_user != sdev)
1314 return 0;
1315 starget->starget_sdev_user = sdev;
1316 }
1317
1318 if (starget->target_busy == 0 && starget->target_blocked) {
1319 /*
1320 * unblock after target_blocked iterates to zero
1321 */
1322 if (--starget->target_blocked == 0) {
1323 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1324 "unblocking target at zero depth\n"));
1325 } else
1326 return 0;
1327 }
1328
1329 if (scsi_target_is_busy(starget)) {
1330 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1331 return 0;
1332 }
1333
1334 return 1;
1335}
1336
1337/*
1338 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1339 * return 0. We must end up running the queue again whenever 0 is
1340 * returned, else IO can hang.
1341 *
1342 * Called with host_lock held.
1343 */
1344static inline int scsi_host_queue_ready(struct request_queue *q,
1345 struct Scsi_Host *shost,
1346 struct scsi_device *sdev)
1347{
1348 if (scsi_host_in_recovery(shost))
1349 return 0;
1350 if (shost->host_busy == 0 && shost->host_blocked) {
1351 /*
1352 * unblock after host_blocked iterates to zero
1353 */
1354 if (--shost->host_blocked == 0) {
1355 SCSI_LOG_MLQUEUE(3,
1356 printk("scsi%d unblocking host at zero depth\n",
1357 shost->host_no));
1358 } else {
1359 return 0;
1360 }
1361 }
1362 if (scsi_host_is_busy(shost)) {
1363 if (list_empty(&sdev->starved_entry))
1364 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1365 return 0;
1366 }
1367
1368 /* We're OK to process the command, so we can't be starved */
1369 if (!list_empty(&sdev->starved_entry))
1370 list_del_init(&sdev->starved_entry);
1371
1372 return 1;
1373}
1374
1375/*
1376 * Busy state exporting function for request stacking drivers.
1377 *
1378 * For efficiency, no lock is taken to check the busy state of
1379 * shost/starget/sdev, since the returned value is not guaranteed and
1380 * may be changed after request stacking drivers call the function,
1381 * regardless of taking lock or not.
1382 *
1383 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1384 * needs to return 'not busy'. Otherwise, request stacking drivers
1385 * may hold requests forever.
1386 */
1387static int scsi_lld_busy(struct request_queue *q)
1388{
1389 struct scsi_device *sdev = q->queuedata;
1390 struct Scsi_Host *shost;
1391
1392 if (blk_queue_dead(q))
1393 return 0;
1394
1395 shost = sdev->host;
1396
1397 /*
1398 * Ignore host/starget busy state.
1399 * Since block layer does not have a concept of fairness across
1400 * multiple queues, congestion of host/starget needs to be handled
1401 * in SCSI layer.
1402 */
1403 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1404 return 1;
1405
1406 return 0;
1407}
1408
1409/*
1410 * Kill a request for a dead device
1411 */
1412static void scsi_kill_request(struct request *req, struct request_queue *q)
1413{
1414 struct scsi_cmnd *cmd = req->special;
1415 struct scsi_device *sdev;
1416 struct scsi_target *starget;
1417 struct Scsi_Host *shost;
1418
1419 blk_start_request(req);
1420
1421 scmd_printk(KERN_INFO, cmd, "killing request\n");
1422
1423 sdev = cmd->device;
1424 starget = scsi_target(sdev);
1425 shost = sdev->host;
1426 scsi_init_cmd_errh(cmd);
1427 cmd->result = DID_NO_CONNECT << 16;
1428 atomic_inc(&cmd->device->iorequest_cnt);
1429
1430 /*
1431 * SCSI request completion path will do scsi_device_unbusy(),
1432 * bump busy counts. To bump the counters, we need to dance
1433 * with the locks as normal issue path does.
1434 */
1435 sdev->device_busy++;
1436 spin_unlock(sdev->request_queue->queue_lock);
1437 spin_lock(shost->host_lock);
1438 shost->host_busy++;
1439 starget->target_busy++;
1440 spin_unlock(shost->host_lock);
1441 spin_lock(sdev->request_queue->queue_lock);
1442
1443 blk_complete_request(req);
1444}
1445
1446static void scsi_softirq_done(struct request *rq)
1447{
1448 struct scsi_cmnd *cmd = rq->special;
1449 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1450 int disposition;
1451
1452 INIT_LIST_HEAD(&cmd->eh_entry);
1453
1454 atomic_inc(&cmd->device->iodone_cnt);
1455 if (cmd->result)
1456 atomic_inc(&cmd->device->ioerr_cnt);
1457
1458 disposition = scsi_decide_disposition(cmd);
1459 if (disposition != SUCCESS &&
1460 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1461 sdev_printk(KERN_ERR, cmd->device,
1462 "timing out command, waited %lus\n",
1463 wait_for/HZ);
1464 disposition = SUCCESS;
1465 }
1466
1467 scsi_log_completion(cmd, disposition);
1468
1469 switch (disposition) {
1470 case SUCCESS:
1471 scsi_finish_command(cmd);
1472 break;
1473 case NEEDS_RETRY:
1474 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1475 break;
1476 case ADD_TO_MLQUEUE:
1477 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1478 break;
1479 default:
1480 if (!scsi_eh_scmd_add(cmd, 0))
1481 scsi_finish_command(cmd);
1482 }
1483}
1484
1485/*
1486 * Function: scsi_request_fn()
1487 *
1488 * Purpose: Main strategy routine for SCSI.
1489 *
1490 * Arguments: q - Pointer to actual queue.
1491 *
1492 * Returns: Nothing
1493 *
1494 * Lock status: IO request lock assumed to be held when called.
1495 */
1496static void scsi_request_fn(struct request_queue *q)
1497{
1498 struct scsi_device *sdev = q->queuedata;
1499 struct Scsi_Host *shost;
1500 struct scsi_cmnd *cmd;
1501 struct request *req;
1502
1503 if(!get_device(&sdev->sdev_gendev))
1504 /* We must be tearing the block queue down already */
1505 return;
1506
1507 /*
1508 * To start with, we keep looping until the queue is empty, or until
1509 * the host is no longer able to accept any more requests.
1510 */
1511 shost = sdev->host;
1512 for (;;) {
1513 int rtn;
1514 /*
1515 * get next queueable request. We do this early to make sure
1516 * that the request is fully prepared even if we cannot
1517 * accept it.
1518 */
1519 req = blk_peek_request(q);
1520 if (!req || !scsi_dev_queue_ready(q, sdev))
1521 break;
1522
1523 if (unlikely(!scsi_device_online(sdev))) {
1524 sdev_printk(KERN_ERR, sdev,
1525 "rejecting I/O to offline device\n");
1526 scsi_kill_request(req, q);
1527 continue;
1528 }
1529
1530
1531 /*
1532 * Remove the request from the request list.
1533 */
1534 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1535 blk_start_request(req);
1536 sdev->device_busy++;
1537
1538 spin_unlock(q->queue_lock);
1539 cmd = req->special;
1540 if (unlikely(cmd == NULL)) {
1541 printk(KERN_CRIT "impossible request in %s.\n"
1542 "please mail a stack trace to "
1543 "linux-scsi@vger.kernel.org\n",
1544 __func__);
1545 blk_dump_rq_flags(req, "foo");
1546 BUG();
1547 }
1548 spin_lock(shost->host_lock);
1549
1550 /*
1551 * We hit this when the driver is using a host wide
1552 * tag map. For device level tag maps the queue_depth check
1553 * in the device ready fn would prevent us from trying
1554 * to allocate a tag. Since the map is a shared host resource
1555 * we add the dev to the starved list so it eventually gets
1556 * a run when a tag is freed.
1557 */
1558 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1559 if (list_empty(&sdev->starved_entry))
1560 list_add_tail(&sdev->starved_entry,
1561 &shost->starved_list);
1562 goto not_ready;
1563 }
1564
1565 if (!scsi_target_queue_ready(shost, sdev))
1566 goto not_ready;
1567
1568 if (!scsi_host_queue_ready(q, shost, sdev))
1569 goto not_ready;
1570
1571 scsi_target(sdev)->target_busy++;
1572 shost->host_busy++;
1573
1574 /*
1575 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1576 * take the lock again.
1577 */
1578 spin_unlock_irq(shost->host_lock);
1579
1580 /*
1581 * Finally, initialize any error handling parameters, and set up
1582 * the timers for timeouts.
1583 */
1584 scsi_init_cmd_errh(cmd);
1585
1586 /*
1587 * Dispatch the command to the low-level driver.
1588 */
1589 rtn = scsi_dispatch_cmd(cmd);
1590 spin_lock_irq(q->queue_lock);
1591 if (rtn)
1592 goto out_delay;
1593 }
1594
1595 goto out;
1596
1597 not_ready:
1598 spin_unlock_irq(shost->host_lock);
1599
1600 /*
1601 * lock q, handle tag, requeue req, and decrement device_busy. We
1602 * must return with queue_lock held.
1603 *
1604 * Decrementing device_busy without checking it is OK, as all such
1605 * cases (host limits or settings) should run the queue at some
1606 * later time.
1607 */
1608 spin_lock_irq(q->queue_lock);
1609 blk_requeue_request(q, req);
1610 sdev->device_busy--;
1611out_delay:
1612 if (sdev->device_busy == 0)
1613 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1614out:
1615 /* must be careful here...if we trigger the ->remove() function
1616 * we cannot be holding the q lock */
1617 spin_unlock_irq(q->queue_lock);
1618 put_device(&sdev->sdev_gendev);
1619 spin_lock_irq(q->queue_lock);
1620}
1621
1622u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1623{
1624 struct device *host_dev;
1625 u64 bounce_limit = 0xffffffff;
1626
1627 if (shost->unchecked_isa_dma)
1628 return BLK_BOUNCE_ISA;
1629 /*
1630 * Platforms with virtual-DMA translation
1631 * hardware have no practical limit.
1632 */
1633 if (!PCI_DMA_BUS_IS_PHYS)
1634 return BLK_BOUNCE_ANY;
1635
1636 host_dev = scsi_get_device(shost);
1637 if (host_dev && host_dev->dma_mask)
1638 bounce_limit = *host_dev->dma_mask;
1639
1640 return bounce_limit;
1641}
1642EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1643
1644struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1645 request_fn_proc *request_fn)
1646{
1647 struct request_queue *q;
1648 struct device *dev = shost->dma_dev;
1649
1650 q = blk_init_queue(request_fn, NULL);
1651 if (!q)
1652 return NULL;
1653
1654 /*
1655 * this limit is imposed by hardware restrictions
1656 */
1657 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1658 SCSI_MAX_SG_CHAIN_SEGMENTS));
1659
1660 if (scsi_host_prot_dma(shost)) {
1661 shost->sg_prot_tablesize =
1662 min_not_zero(shost->sg_prot_tablesize,
1663 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1664 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1665 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1666 }
1667
1668 blk_queue_max_hw_sectors(q, shost->max_sectors);
1669 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1670 blk_queue_segment_boundary(q, shost->dma_boundary);
1671 dma_set_seg_boundary(dev, shost->dma_boundary);
1672
1673 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1674
1675 if (!shost->use_clustering)
1676 q->limits.cluster = 0;
1677
1678 /*
1679 * set a reasonable default alignment on word boundaries: the
1680 * host and device may alter it using
1681 * blk_queue_update_dma_alignment() later.
1682 */
1683 blk_queue_dma_alignment(q, 0x03);
1684
1685 return q;
1686}
1687EXPORT_SYMBOL(__scsi_alloc_queue);
1688
1689struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1690{
1691 struct request_queue *q;
1692
1693 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1694 if (!q)
1695 return NULL;
1696
1697 blk_queue_prep_rq(q, scsi_prep_fn);
1698 blk_queue_softirq_done(q, scsi_softirq_done);
1699 blk_queue_rq_timed_out(q, scsi_times_out);
1700 blk_queue_lld_busy(q, scsi_lld_busy);
1701 return q;
1702}
1703
1704/*
1705 * Function: scsi_block_requests()
1706 *
1707 * Purpose: Utility function used by low-level drivers to prevent further
1708 * commands from being queued to the device.
1709 *
1710 * Arguments: shost - Host in question
1711 *
1712 * Returns: Nothing
1713 *
1714 * Lock status: No locks are assumed held.
1715 *
1716 * Notes: There is no timer nor any other means by which the requests
1717 * get unblocked other than the low-level driver calling
1718 * scsi_unblock_requests().
1719 */
1720void scsi_block_requests(struct Scsi_Host *shost)
1721{
1722 shost->host_self_blocked = 1;
1723}
1724EXPORT_SYMBOL(scsi_block_requests);
1725
1726/*
1727 * Function: scsi_unblock_requests()
1728 *
1729 * Purpose: Utility function used by low-level drivers to allow further
1730 * commands from being queued to the device.
1731 *
1732 * Arguments: shost - Host in question
1733 *
1734 * Returns: Nothing
1735 *
1736 * Lock status: No locks are assumed held.
1737 *
1738 * Notes: There is no timer nor any other means by which the requests
1739 * get unblocked other than the low-level driver calling
1740 * scsi_unblock_requests().
1741 *
1742 * This is done as an API function so that changes to the
1743 * internals of the scsi mid-layer won't require wholesale
1744 * changes to drivers that use this feature.
1745 */
1746void scsi_unblock_requests(struct Scsi_Host *shost)
1747{
1748 shost->host_self_blocked = 0;
1749 scsi_run_host_queues(shost);
1750}
1751EXPORT_SYMBOL(scsi_unblock_requests);
1752
1753int __init scsi_init_queue(void)
1754{
1755 int i;
1756
1757 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1758 sizeof(struct scsi_data_buffer),
1759 0, 0, NULL);
1760 if (!scsi_sdb_cache) {
1761 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1762 return -ENOMEM;
1763 }
1764
1765 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1766 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1767 int size = sgp->size * sizeof(struct scatterlist);
1768
1769 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1770 SLAB_HWCACHE_ALIGN, NULL);
1771 if (!sgp->slab) {
1772 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1773 sgp->name);
1774 goto cleanup_sdb;
1775 }
1776
1777 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1778 sgp->slab);
1779 if (!sgp->pool) {
1780 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1781 sgp->name);
1782 goto cleanup_sdb;
1783 }
1784 }
1785
1786 return 0;
1787
1788cleanup_sdb:
1789 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1790 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1791 if (sgp->pool)
1792 mempool_destroy(sgp->pool);
1793 if (sgp->slab)
1794 kmem_cache_destroy(sgp->slab);
1795 }
1796 kmem_cache_destroy(scsi_sdb_cache);
1797
1798 return -ENOMEM;
1799}
1800
1801void scsi_exit_queue(void)
1802{
1803 int i;
1804
1805 kmem_cache_destroy(scsi_sdb_cache);
1806
1807 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1808 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1809 mempool_destroy(sgp->pool);
1810 kmem_cache_destroy(sgp->slab);
1811 }
1812}
1813
1814/**
1815 * scsi_mode_select - issue a mode select
1816 * @sdev: SCSI device to be queried
1817 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1818 * @sp: Save page bit (0 == don't save, 1 == save)
1819 * @modepage: mode page being requested
1820 * @buffer: request buffer (may not be smaller than eight bytes)
1821 * @len: length of request buffer.
1822 * @timeout: command timeout
1823 * @retries: number of retries before failing
1824 * @data: returns a structure abstracting the mode header data
1825 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1826 * must be SCSI_SENSE_BUFFERSIZE big.
1827 *
1828 * Returns zero if successful; negative error number or scsi
1829 * status on error
1830 *
1831 */
1832int
1833scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1834 unsigned char *buffer, int len, int timeout, int retries,
1835 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1836{
1837 unsigned char cmd[10];
1838 unsigned char *real_buffer;
1839 int ret;
1840
1841 memset(cmd, 0, sizeof(cmd));
1842 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1843
1844 if (sdev->use_10_for_ms) {
1845 if (len > 65535)
1846 return -EINVAL;
1847 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1848 if (!real_buffer)
1849 return -ENOMEM;
1850 memcpy(real_buffer + 8, buffer, len);
1851 len += 8;
1852 real_buffer[0] = 0;
1853 real_buffer[1] = 0;
1854 real_buffer[2] = data->medium_type;
1855 real_buffer[3] = data->device_specific;
1856 real_buffer[4] = data->longlba ? 0x01 : 0;
1857 real_buffer[5] = 0;
1858 real_buffer[6] = data->block_descriptor_length >> 8;
1859 real_buffer[7] = data->block_descriptor_length;
1860
1861 cmd[0] = MODE_SELECT_10;
1862 cmd[7] = len >> 8;
1863 cmd[8] = len;
1864 } else {
1865 if (len > 255 || data->block_descriptor_length > 255 ||
1866 data->longlba)
1867 return -EINVAL;
1868
1869 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1870 if (!real_buffer)
1871 return -ENOMEM;
1872 memcpy(real_buffer + 4, buffer, len);
1873 len += 4;
1874 real_buffer[0] = 0;
1875 real_buffer[1] = data->medium_type;
1876 real_buffer[2] = data->device_specific;
1877 real_buffer[3] = data->block_descriptor_length;
1878
1879
1880 cmd[0] = MODE_SELECT;
1881 cmd[4] = len;
1882 }
1883
1884 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1885 sshdr, timeout, retries, NULL);
1886 kfree(real_buffer);
1887 return ret;
1888}
1889EXPORT_SYMBOL_GPL(scsi_mode_select);
1890
1891/**
1892 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1893 * @sdev: SCSI device to be queried
1894 * @dbd: set if mode sense will allow block descriptors to be returned
1895 * @modepage: mode page being requested
1896 * @buffer: request buffer (may not be smaller than eight bytes)
1897 * @len: length of request buffer.
1898 * @timeout: command timeout
1899 * @retries: number of retries before failing
1900 * @data: returns a structure abstracting the mode header data
1901 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1902 * must be SCSI_SENSE_BUFFERSIZE big.
1903 *
1904 * Returns zero if unsuccessful, or the header offset (either 4
1905 * or 8 depending on whether a six or ten byte command was
1906 * issued) if successful.
1907 */
1908int
1909scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1910 unsigned char *buffer, int len, int timeout, int retries,
1911 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1912{
1913 unsigned char cmd[12];
1914 int use_10_for_ms;
1915 int header_length;
1916 int result;
1917 struct scsi_sense_hdr my_sshdr;
1918
1919 memset(data, 0, sizeof(*data));
1920 memset(&cmd[0], 0, 12);
1921 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1922 cmd[2] = modepage;
1923
1924 /* caller might not be interested in sense, but we need it */
1925 if (!sshdr)
1926 sshdr = &my_sshdr;
1927
1928 retry:
1929 use_10_for_ms = sdev->use_10_for_ms;
1930
1931 if (use_10_for_ms) {
1932 if (len < 8)
1933 len = 8;
1934
1935 cmd[0] = MODE_SENSE_10;
1936 cmd[8] = len;
1937 header_length = 8;
1938 } else {
1939 if (len < 4)
1940 len = 4;
1941
1942 cmd[0] = MODE_SENSE;
1943 cmd[4] = len;
1944 header_length = 4;
1945 }
1946
1947 memset(buffer, 0, len);
1948
1949 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1950 sshdr, timeout, retries, NULL);
1951
1952 /* This code looks awful: what it's doing is making sure an
1953 * ILLEGAL REQUEST sense return identifies the actual command
1954 * byte as the problem. MODE_SENSE commands can return
1955 * ILLEGAL REQUEST if the code page isn't supported */
1956
1957 if (use_10_for_ms && !scsi_status_is_good(result) &&
1958 (driver_byte(result) & DRIVER_SENSE)) {
1959 if (scsi_sense_valid(sshdr)) {
1960 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1961 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1962 /*
1963 * Invalid command operation code
1964 */
1965 sdev->use_10_for_ms = 0;
1966 goto retry;
1967 }
1968 }
1969 }
1970
1971 if(scsi_status_is_good(result)) {
1972 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1973 (modepage == 6 || modepage == 8))) {
1974 /* Initio breakage? */
1975 header_length = 0;
1976 data->length = 13;
1977 data->medium_type = 0;
1978 data->device_specific = 0;
1979 data->longlba = 0;
1980 data->block_descriptor_length = 0;
1981 } else if(use_10_for_ms) {
1982 data->length = buffer[0]*256 + buffer[1] + 2;
1983 data->medium_type = buffer[2];
1984 data->device_specific = buffer[3];
1985 data->longlba = buffer[4] & 0x01;
1986 data->block_descriptor_length = buffer[6]*256
1987 + buffer[7];
1988 } else {
1989 data->length = buffer[0] + 1;
1990 data->medium_type = buffer[1];
1991 data->device_specific = buffer[2];
1992 data->block_descriptor_length = buffer[3];
1993 }
1994 data->header_length = header_length;
1995 }
1996
1997 return result;
1998}
1999EXPORT_SYMBOL(scsi_mode_sense);
2000
2001/**
2002 * scsi_test_unit_ready - test if unit is ready
2003 * @sdev: scsi device to change the state of.
2004 * @timeout: command timeout
2005 * @retries: number of retries before failing
2006 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2007 * returning sense. Make sure that this is cleared before passing
2008 * in.
2009 *
2010 * Returns zero if unsuccessful or an error if TUR failed. For
2011 * removable media, UNIT_ATTENTION sets ->changed flag.
2012 **/
2013int
2014scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2015 struct scsi_sense_hdr *sshdr_external)
2016{
2017 char cmd[] = {
2018 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2019 };
2020 struct scsi_sense_hdr *sshdr;
2021 int result;
2022
2023 if (!sshdr_external)
2024 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2025 else
2026 sshdr = sshdr_external;
2027
2028 /* try to eat the UNIT_ATTENTION if there are enough retries */
2029 do {
2030 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2031 timeout, retries, NULL);
2032 if (sdev->removable && scsi_sense_valid(sshdr) &&
2033 sshdr->sense_key == UNIT_ATTENTION)
2034 sdev->changed = 1;
2035 } while (scsi_sense_valid(sshdr) &&
2036 sshdr->sense_key == UNIT_ATTENTION && --retries);
2037
2038 if (!sshdr_external)
2039 kfree(sshdr);
2040 return result;
2041}
2042EXPORT_SYMBOL(scsi_test_unit_ready);
2043
2044/**
2045 * scsi_device_set_state - Take the given device through the device state model.
2046 * @sdev: scsi device to change the state of.
2047 * @state: state to change to.
2048 *
2049 * Returns zero if unsuccessful or an error if the requested
2050 * transition is illegal.
2051 */
2052int
2053scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2054{
2055 enum scsi_device_state oldstate = sdev->sdev_state;
2056
2057 if (state == oldstate)
2058 return 0;
2059
2060 switch (state) {
2061 case SDEV_CREATED:
2062 switch (oldstate) {
2063 case SDEV_CREATED_BLOCK:
2064 break;
2065 default:
2066 goto illegal;
2067 }
2068 break;
2069
2070 case SDEV_RUNNING:
2071 switch (oldstate) {
2072 case SDEV_CREATED:
2073 case SDEV_OFFLINE:
2074 case SDEV_QUIESCE:
2075 case SDEV_BLOCK:
2076 break;
2077 default:
2078 goto illegal;
2079 }
2080 break;
2081
2082 case SDEV_QUIESCE:
2083 switch (oldstate) {
2084 case SDEV_RUNNING:
2085 case SDEV_OFFLINE:
2086 break;
2087 default:
2088 goto illegal;
2089 }
2090 break;
2091
2092 case SDEV_OFFLINE:
2093 switch (oldstate) {
2094 case SDEV_CREATED:
2095 case SDEV_RUNNING:
2096 case SDEV_QUIESCE:
2097 case SDEV_BLOCK:
2098 break;
2099 default:
2100 goto illegal;
2101 }
2102 break;
2103
2104 case SDEV_BLOCK:
2105 switch (oldstate) {
2106 case SDEV_RUNNING:
2107 case SDEV_CREATED_BLOCK:
2108 break;
2109 default:
2110 goto illegal;
2111 }
2112 break;
2113
2114 case SDEV_CREATED_BLOCK:
2115 switch (oldstate) {
2116 case SDEV_CREATED:
2117 break;
2118 default:
2119 goto illegal;
2120 }
2121 break;
2122
2123 case SDEV_CANCEL:
2124 switch (oldstate) {
2125 case SDEV_CREATED:
2126 case SDEV_RUNNING:
2127 case SDEV_QUIESCE:
2128 case SDEV_OFFLINE:
2129 case SDEV_BLOCK:
2130 break;
2131 default:
2132 goto illegal;
2133 }
2134 break;
2135
2136 case SDEV_DEL:
2137 switch (oldstate) {
2138 case SDEV_CREATED:
2139 case SDEV_RUNNING:
2140 case SDEV_OFFLINE:
2141 case SDEV_CANCEL:
2142 break;
2143 default:
2144 goto illegal;
2145 }
2146 break;
2147
2148 }
2149 sdev->sdev_state = state;
2150 return 0;
2151
2152 illegal:
2153 SCSI_LOG_ERROR_RECOVERY(1,
2154 sdev_printk(KERN_ERR, sdev,
2155 "Illegal state transition %s->%s\n",
2156 scsi_device_state_name(oldstate),
2157 scsi_device_state_name(state))
2158 );
2159 return -EINVAL;
2160}
2161EXPORT_SYMBOL(scsi_device_set_state);
2162
2163/**
2164 * sdev_evt_emit - emit a single SCSI device uevent
2165 * @sdev: associated SCSI device
2166 * @evt: event to emit
2167 *
2168 * Send a single uevent (scsi_event) to the associated scsi_device.
2169 */
2170static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2171{
2172 int idx = 0;
2173 char *envp[3];
2174
2175 switch (evt->evt_type) {
2176 case SDEV_EVT_MEDIA_CHANGE:
2177 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2178 break;
2179
2180 default:
2181 /* do nothing */
2182 break;
2183 }
2184
2185 envp[idx++] = NULL;
2186
2187 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2188}
2189
2190/**
2191 * sdev_evt_thread - send a uevent for each scsi event
2192 * @work: work struct for scsi_device
2193 *
2194 * Dispatch queued events to their associated scsi_device kobjects
2195 * as uevents.
2196 */
2197void scsi_evt_thread(struct work_struct *work)
2198{
2199 struct scsi_device *sdev;
2200 LIST_HEAD(event_list);
2201
2202 sdev = container_of(work, struct scsi_device, event_work);
2203
2204 while (1) {
2205 struct scsi_event *evt;
2206 struct list_head *this, *tmp;
2207 unsigned long flags;
2208
2209 spin_lock_irqsave(&sdev->list_lock, flags);
2210 list_splice_init(&sdev->event_list, &event_list);
2211 spin_unlock_irqrestore(&sdev->list_lock, flags);
2212
2213 if (list_empty(&event_list))
2214 break;
2215
2216 list_for_each_safe(this, tmp, &event_list) {
2217 evt = list_entry(this, struct scsi_event, node);
2218 list_del(&evt->node);
2219 scsi_evt_emit(sdev, evt);
2220 kfree(evt);
2221 }
2222 }
2223}
2224
2225/**
2226 * sdev_evt_send - send asserted event to uevent thread
2227 * @sdev: scsi_device event occurred on
2228 * @evt: event to send
2229 *
2230 * Assert scsi device event asynchronously.
2231 */
2232void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2233{
2234 unsigned long flags;
2235
2236#if 0
2237 /* FIXME: currently this check eliminates all media change events
2238 * for polled devices. Need to update to discriminate between AN
2239 * and polled events */
2240 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2241 kfree(evt);
2242 return;
2243 }
2244#endif
2245
2246 spin_lock_irqsave(&sdev->list_lock, flags);
2247 list_add_tail(&evt->node, &sdev->event_list);
2248 schedule_work(&sdev->event_work);
2249 spin_unlock_irqrestore(&sdev->list_lock, flags);
2250}
2251EXPORT_SYMBOL_GPL(sdev_evt_send);
2252
2253/**
2254 * sdev_evt_alloc - allocate a new scsi event
2255 * @evt_type: type of event to allocate
2256 * @gfpflags: GFP flags for allocation
2257 *
2258 * Allocates and returns a new scsi_event.
2259 */
2260struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2261 gfp_t gfpflags)
2262{
2263 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2264 if (!evt)
2265 return NULL;
2266
2267 evt->evt_type = evt_type;
2268 INIT_LIST_HEAD(&evt->node);
2269
2270 /* evt_type-specific initialization, if any */
2271 switch (evt_type) {
2272 case SDEV_EVT_MEDIA_CHANGE:
2273 default:
2274 /* do nothing */
2275 break;
2276 }
2277
2278 return evt;
2279}
2280EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2281
2282/**
2283 * sdev_evt_send_simple - send asserted event to uevent thread
2284 * @sdev: scsi_device event occurred on
2285 * @evt_type: type of event to send
2286 * @gfpflags: GFP flags for allocation
2287 *
2288 * Assert scsi device event asynchronously, given an event type.
2289 */
2290void sdev_evt_send_simple(struct scsi_device *sdev,
2291 enum scsi_device_event evt_type, gfp_t gfpflags)
2292{
2293 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2294 if (!evt) {
2295 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2296 evt_type);
2297 return;
2298 }
2299
2300 sdev_evt_send(sdev, evt);
2301}
2302EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2303
2304/**
2305 * scsi_device_quiesce - Block user issued commands.
2306 * @sdev: scsi device to quiesce.
2307 *
2308 * This works by trying to transition to the SDEV_QUIESCE state
2309 * (which must be a legal transition). When the device is in this
2310 * state, only special requests will be accepted, all others will
2311 * be deferred. Since special requests may also be requeued requests,
2312 * a successful return doesn't guarantee the device will be
2313 * totally quiescent.
2314 *
2315 * Must be called with user context, may sleep.
2316 *
2317 * Returns zero if unsuccessful or an error if not.
2318 */
2319int
2320scsi_device_quiesce(struct scsi_device *sdev)
2321{
2322 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2323 if (err)
2324 return err;
2325
2326 scsi_run_queue(sdev->request_queue);
2327 while (sdev->device_busy) {
2328 msleep_interruptible(200);
2329 scsi_run_queue(sdev->request_queue);
2330 }
2331 return 0;
2332}
2333EXPORT_SYMBOL(scsi_device_quiesce);
2334
2335/**
2336 * scsi_device_resume - Restart user issued commands to a quiesced device.
2337 * @sdev: scsi device to resume.
2338 *
2339 * Moves the device from quiesced back to running and restarts the
2340 * queues.
2341 *
2342 * Must be called with user context, may sleep.
2343 */
2344void scsi_device_resume(struct scsi_device *sdev)
2345{
2346 /* check if the device state was mutated prior to resume, and if
2347 * so assume the state is being managed elsewhere (for example
2348 * device deleted during suspend)
2349 */
2350 if (sdev->sdev_state != SDEV_QUIESCE ||
2351 scsi_device_set_state(sdev, SDEV_RUNNING))
2352 return;
2353 scsi_run_queue(sdev->request_queue);
2354}
2355EXPORT_SYMBOL(scsi_device_resume);
2356
2357static void
2358device_quiesce_fn(struct scsi_device *sdev, void *data)
2359{
2360 scsi_device_quiesce(sdev);
2361}
2362
2363void
2364scsi_target_quiesce(struct scsi_target *starget)
2365{
2366 starget_for_each_device(starget, NULL, device_quiesce_fn);
2367}
2368EXPORT_SYMBOL(scsi_target_quiesce);
2369
2370static void
2371device_resume_fn(struct scsi_device *sdev, void *data)
2372{
2373 scsi_device_resume(sdev);
2374}
2375
2376void
2377scsi_target_resume(struct scsi_target *starget)
2378{
2379 starget_for_each_device(starget, NULL, device_resume_fn);
2380}
2381EXPORT_SYMBOL(scsi_target_resume);
2382
2383/**
2384 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2385 * @sdev: device to block
2386 *
2387 * Block request made by scsi lld's to temporarily stop all
2388 * scsi commands on the specified device. Called from interrupt
2389 * or normal process context.
2390 *
2391 * Returns zero if successful or error if not
2392 *
2393 * Notes:
2394 * This routine transitions the device to the SDEV_BLOCK state
2395 * (which must be a legal transition). When the device is in this
2396 * state, all commands are deferred until the scsi lld reenables
2397 * the device with scsi_device_unblock or device_block_tmo fires.
2398 * This routine assumes the host_lock is held on entry.
2399 */
2400int
2401scsi_internal_device_block(struct scsi_device *sdev)
2402{
2403 struct request_queue *q = sdev->request_queue;
2404 unsigned long flags;
2405 int err = 0;
2406
2407 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2408 if (err) {
2409 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2410
2411 if (err)
2412 return err;
2413 }
2414
2415 /*
2416 * The device has transitioned to SDEV_BLOCK. Stop the
2417 * block layer from calling the midlayer with this device's
2418 * request queue.
2419 */
2420 spin_lock_irqsave(q->queue_lock, flags);
2421 blk_stop_queue(q);
2422 spin_unlock_irqrestore(q->queue_lock, flags);
2423
2424 return 0;
2425}
2426EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2427
2428/**
2429 * scsi_internal_device_unblock - resume a device after a block request
2430 * @sdev: device to resume
2431 *
2432 * Called by scsi lld's or the midlayer to restart the device queue
2433 * for the previously suspended scsi device. Called from interrupt or
2434 * normal process context.
2435 *
2436 * Returns zero if successful or error if not.
2437 *
2438 * Notes:
2439 * This routine transitions the device to the SDEV_RUNNING state
2440 * (which must be a legal transition) allowing the midlayer to
2441 * goose the queue for this device. This routine assumes the
2442 * host_lock is held upon entry.
2443 */
2444int
2445scsi_internal_device_unblock(struct scsi_device *sdev)
2446{
2447 struct request_queue *q = sdev->request_queue;
2448 unsigned long flags;
2449
2450 /*
2451 * Try to transition the scsi device to SDEV_RUNNING
2452 * and goose the device queue if successful.
2453 */
2454 if (sdev->sdev_state == SDEV_BLOCK)
2455 sdev->sdev_state = SDEV_RUNNING;
2456 else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2457 sdev->sdev_state = SDEV_CREATED;
2458 else if (sdev->sdev_state != SDEV_CANCEL &&
2459 sdev->sdev_state != SDEV_OFFLINE)
2460 return -EINVAL;
2461
2462 spin_lock_irqsave(q->queue_lock, flags);
2463 blk_start_queue(q);
2464 spin_unlock_irqrestore(q->queue_lock, flags);
2465
2466 return 0;
2467}
2468EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2469
2470static void
2471device_block(struct scsi_device *sdev, void *data)
2472{
2473 scsi_internal_device_block(sdev);
2474}
2475
2476static int
2477target_block(struct device *dev, void *data)
2478{
2479 if (scsi_is_target_device(dev))
2480 starget_for_each_device(to_scsi_target(dev), NULL,
2481 device_block);
2482 return 0;
2483}
2484
2485void
2486scsi_target_block(struct device *dev)
2487{
2488 if (scsi_is_target_device(dev))
2489 starget_for_each_device(to_scsi_target(dev), NULL,
2490 device_block);
2491 else
2492 device_for_each_child(dev, NULL, target_block);
2493}
2494EXPORT_SYMBOL_GPL(scsi_target_block);
2495
2496static void
2497device_unblock(struct scsi_device *sdev, void *data)
2498{
2499 scsi_internal_device_unblock(sdev);
2500}
2501
2502static int
2503target_unblock(struct device *dev, void *data)
2504{
2505 if (scsi_is_target_device(dev))
2506 starget_for_each_device(to_scsi_target(dev), NULL,
2507 device_unblock);
2508 return 0;
2509}
2510
2511void
2512scsi_target_unblock(struct device *dev)
2513{
2514 if (scsi_is_target_device(dev))
2515 starget_for_each_device(to_scsi_target(dev), NULL,
2516 device_unblock);
2517 else
2518 device_for_each_child(dev, NULL, target_unblock);
2519}
2520EXPORT_SYMBOL_GPL(scsi_target_unblock);
2521
2522/**
2523 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2524 * @sgl: scatter-gather list
2525 * @sg_count: number of segments in sg
2526 * @offset: offset in bytes into sg, on return offset into the mapped area
2527 * @len: bytes to map, on return number of bytes mapped
2528 *
2529 * Returns virtual address of the start of the mapped page
2530 */
2531void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2532 size_t *offset, size_t *len)
2533{
2534 int i;
2535 size_t sg_len = 0, len_complete = 0;
2536 struct scatterlist *sg;
2537 struct page *page;
2538
2539 WARN_ON(!irqs_disabled());
2540
2541 for_each_sg(sgl, sg, sg_count, i) {
2542 len_complete = sg_len; /* Complete sg-entries */
2543 sg_len += sg->length;
2544 if (sg_len > *offset)
2545 break;
2546 }
2547
2548 if (unlikely(i == sg_count)) {
2549 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2550 "elements %d\n",
2551 __func__, sg_len, *offset, sg_count);
2552 WARN_ON(1);
2553 return NULL;
2554 }
2555
2556 /* Offset starting from the beginning of first page in this sg-entry */
2557 *offset = *offset - len_complete + sg->offset;
2558
2559 /* Assumption: contiguous pages can be accessed as "page + i" */
2560 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2561 *offset &= ~PAGE_MASK;
2562
2563 /* Bytes in this sg-entry from *offset to the end of the page */
2564 sg_len = PAGE_SIZE - *offset;
2565 if (*len > sg_len)
2566 *len = sg_len;
2567
2568 return kmap_atomic(page);
2569}
2570EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2571
2572/**
2573 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2574 * @virt: virtual address to be unmapped
2575 */
2576void scsi_kunmap_atomic_sg(void *virt)
2577{
2578 kunmap_atomic(virt);
2579}
2580EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12#include <linux/bio.h>
13#include <linux/bitops.h>
14#include <linux/blkdev.h>
15#include <linux/completion.h>
16#include <linux/kernel.h>
17#include <linux/export.h>
18#include <linux/init.h>
19#include <linux/pci.h>
20#include <linux/delay.h>
21#include <linux/hardirq.h>
22#include <linux/scatterlist.h>
23#include <linux/blk-mq.h>
24#include <linux/blk-integrity.h>
25#include <linux/ratelimit.h>
26#include <linux/unaligned.h>
27
28#include <scsi/scsi.h>
29#include <scsi/scsi_cmnd.h>
30#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_driver.h>
33#include <scsi/scsi_eh.h>
34#include <scsi/scsi_host.h>
35#include <scsi/scsi_transport.h> /* scsi_init_limits() */
36#include <scsi/scsi_dh.h>
37
38#include <trace/events/scsi.h>
39
40#include "scsi_debugfs.h"
41#include "scsi_priv.h"
42#include "scsi_logging.h"
43
44/*
45 * Size of integrity metadata is usually small, 1 inline sg should
46 * cover normal cases.
47 */
48#ifdef CONFIG_ARCH_NO_SG_CHAIN
49#define SCSI_INLINE_PROT_SG_CNT 0
50#define SCSI_INLINE_SG_CNT 0
51#else
52#define SCSI_INLINE_PROT_SG_CNT 1
53#define SCSI_INLINE_SG_CNT 2
54#endif
55
56static struct kmem_cache *scsi_sense_cache;
57static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
61int scsi_init_sense_cache(struct Scsi_Host *shost)
62{
63 int ret = 0;
64
65 mutex_lock(&scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
67 scsi_sense_cache =
68 kmem_cache_create_usercopy("scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
72 ret = -ENOMEM;
73 }
74 mutex_unlock(&scsi_sense_cache_mutex);
75 return ret;
76}
77
78static void
79scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80{
81 struct Scsi_Host *host = cmd->device->host;
82 struct scsi_device *device = cmd->device;
83 struct scsi_target *starget = scsi_target(device);
84
85 /*
86 * Set the appropriate busy bit for the device/host.
87 *
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
90 *
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
97 */
98 switch (reason) {
99 case SCSI_MLQUEUE_HOST_BUSY:
100 atomic_set(&host->host_blocked, host->max_host_blocked);
101 break;
102 case SCSI_MLQUEUE_DEVICE_BUSY:
103 case SCSI_MLQUEUE_EH_RETRY:
104 atomic_set(&device->device_blocked,
105 device->max_device_blocked);
106 break;
107 case SCSI_MLQUEUE_TARGET_BUSY:
108 atomic_set(&starget->target_blocked,
109 starget->max_target_blocked);
110 break;
111 }
112}
113
114static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115{
116 struct request *rq = scsi_cmd_to_rq(cmd);
117
118 if (rq->rq_flags & RQF_DONTPREP) {
119 rq->rq_flags &= ~RQF_DONTPREP;
120 scsi_mq_uninit_cmd(cmd);
121 } else {
122 WARN_ON_ONCE(true);
123 }
124
125 blk_mq_requeue_request(rq, false);
126 if (!scsi_host_in_recovery(cmd->device->host))
127 blk_mq_delay_kick_requeue_list(rq->q, msecs);
128}
129
130/**
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
135 *
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
140 * file.
141 */
142static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143{
144 struct scsi_device *device = cmd->device;
145
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 "Inserting command %p into mlqueue\n", cmd));
148
149 scsi_set_blocked(cmd, reason);
150
151 /*
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
154 */
155 if (unbusy)
156 scsi_device_unbusy(device, cmd);
157
158 /*
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_mq_destroy_queue() finishes.
163 */
164 cmd->result = 0;
165
166 blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
167 !scsi_host_in_recovery(cmd->device->host));
168}
169
170/**
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
174 *
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
178 *
179 * Context: This could be called either from an interrupt context or a normal
180 * process context.
181 */
182void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183{
184 __scsi_queue_insert(cmd, reason, true);
185}
186
187void scsi_failures_reset_retries(struct scsi_failures *failures)
188{
189 struct scsi_failure *failure;
190
191 failures->total_retries = 0;
192
193 for (failure = failures->failure_definitions; failure->result;
194 failure++)
195 failure->retries = 0;
196}
197EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);
198
199/**
200 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
201 * @scmd: scsi_cmnd to check.
202 * @failures: scsi_failures struct that lists failures to check for.
203 *
204 * Returns -EAGAIN if the caller should retry else 0.
205 */
206static int scsi_check_passthrough(struct scsi_cmnd *scmd,
207 struct scsi_failures *failures)
208{
209 struct scsi_failure *failure;
210 struct scsi_sense_hdr sshdr;
211 enum sam_status status;
212
213 if (!scmd->result)
214 return 0;
215
216 if (!failures)
217 return 0;
218
219 for (failure = failures->failure_definitions; failure->result;
220 failure++) {
221 if (failure->result == SCMD_FAILURE_RESULT_ANY)
222 goto maybe_retry;
223
224 if (host_byte(scmd->result) &&
225 host_byte(scmd->result) == host_byte(failure->result))
226 goto maybe_retry;
227
228 status = status_byte(scmd->result);
229 if (!status)
230 continue;
231
232 if (failure->result == SCMD_FAILURE_STAT_ANY &&
233 !scsi_status_is_good(scmd->result))
234 goto maybe_retry;
235
236 if (status != status_byte(failure->result))
237 continue;
238
239 if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||
240 failure->sense == SCMD_FAILURE_SENSE_ANY)
241 goto maybe_retry;
242
243 if (!scsi_command_normalize_sense(scmd, &sshdr))
244 return 0;
245
246 if (failure->sense != sshdr.sense_key)
247 continue;
248
249 if (failure->asc == SCMD_FAILURE_ASC_ANY)
250 goto maybe_retry;
251
252 if (failure->asc != sshdr.asc)
253 continue;
254
255 if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||
256 failure->ascq == sshdr.ascq)
257 goto maybe_retry;
258 }
259
260 return 0;
261
262maybe_retry:
263 if (failure->allowed) {
264 if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||
265 ++failure->retries <= failure->allowed)
266 return -EAGAIN;
267 } else {
268 if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||
269 ++failures->total_retries <= failures->total_allowed)
270 return -EAGAIN;
271 }
272
273 return 0;
274}
275
276/**
277 * scsi_execute_cmd - insert request and wait for the result
278 * @sdev: scsi_device
279 * @cmd: scsi command
280 * @opf: block layer request cmd_flags
281 * @buffer: data buffer
282 * @bufflen: len of buffer
283 * @timeout: request timeout in HZ
284 * @ml_retries: number of times SCSI midlayer will retry request
285 * @args: Optional args. See struct definition for field descriptions
286 *
287 * Returns the scsi_cmnd result field if a command was executed, or a negative
288 * Linux error code if we didn't get that far.
289 */
290int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
291 blk_opf_t opf, void *buffer, unsigned int bufflen,
292 int timeout, int ml_retries,
293 const struct scsi_exec_args *args)
294{
295 static const struct scsi_exec_args default_args;
296 struct request *req;
297 struct scsi_cmnd *scmd;
298 int ret;
299
300 if (!args)
301 args = &default_args;
302 else if (WARN_ON_ONCE(args->sense &&
303 args->sense_len != SCSI_SENSE_BUFFERSIZE))
304 return -EINVAL;
305
306retry:
307 req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
308 if (IS_ERR(req))
309 return PTR_ERR(req);
310
311 if (bufflen) {
312 ret = blk_rq_map_kern(sdev->request_queue, req,
313 buffer, bufflen, GFP_NOIO);
314 if (ret)
315 goto out;
316 }
317 scmd = blk_mq_rq_to_pdu(req);
318 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
319 memcpy(scmd->cmnd, cmd, scmd->cmd_len);
320 scmd->allowed = ml_retries;
321 scmd->flags |= args->scmd_flags;
322 req->timeout = timeout;
323 req->rq_flags |= RQF_QUIET;
324
325 /*
326 * head injection *required* here otherwise quiesce won't work
327 */
328 blk_execute_rq(req, true);
329
330 if (scsi_check_passthrough(scmd, args->failures) == -EAGAIN) {
331 blk_mq_free_request(req);
332 goto retry;
333 }
334
335 /*
336 * Some devices (USB mass-storage in particular) may transfer
337 * garbage data together with a residue indicating that the data
338 * is invalid. Prevent the garbage from being misinterpreted
339 * and prevent security leaks by zeroing out the excess data.
340 */
341 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
342 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
343
344 if (args->resid)
345 *args->resid = scmd->resid_len;
346 if (args->sense)
347 memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
348 if (args->sshdr)
349 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
350 args->sshdr);
351
352 ret = scmd->result;
353 out:
354 blk_mq_free_request(req);
355
356 return ret;
357}
358EXPORT_SYMBOL(scsi_execute_cmd);
359
360/*
361 * Wake up the error handler if necessary. Avoid as follows that the error
362 * handler is not woken up if host in-flight requests number ==
363 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
364 * with an RCU read lock in this function to ensure that this function in
365 * its entirety either finishes before scsi_eh_scmd_add() increases the
366 * host_failed counter or that it notices the shost state change made by
367 * scsi_eh_scmd_add().
368 */
369static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
370{
371 unsigned long flags;
372
373 rcu_read_lock();
374 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
375 if (unlikely(scsi_host_in_recovery(shost))) {
376 unsigned int busy = scsi_host_busy(shost);
377
378 spin_lock_irqsave(shost->host_lock, flags);
379 if (shost->host_failed || shost->host_eh_scheduled)
380 scsi_eh_wakeup(shost, busy);
381 spin_unlock_irqrestore(shost->host_lock, flags);
382 }
383 rcu_read_unlock();
384}
385
386void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
387{
388 struct Scsi_Host *shost = sdev->host;
389 struct scsi_target *starget = scsi_target(sdev);
390
391 scsi_dec_host_busy(shost, cmd);
392
393 if (starget->can_queue > 0)
394 atomic_dec(&starget->target_busy);
395
396 sbitmap_put(&sdev->budget_map, cmd->budget_token);
397 cmd->budget_token = -1;
398}
399
400/*
401 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
402 * interrupts disabled.
403 */
404static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
405{
406 struct scsi_device *current_sdev = data;
407
408 if (sdev != current_sdev)
409 blk_mq_run_hw_queues(sdev->request_queue, true);
410}
411
412/*
413 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
414 * and call blk_run_queue for all the scsi_devices on the target -
415 * including current_sdev first.
416 *
417 * Called with *no* scsi locks held.
418 */
419static void scsi_single_lun_run(struct scsi_device *current_sdev)
420{
421 struct Scsi_Host *shost = current_sdev->host;
422 struct scsi_target *starget = scsi_target(current_sdev);
423 unsigned long flags;
424
425 spin_lock_irqsave(shost->host_lock, flags);
426 starget->starget_sdev_user = NULL;
427 spin_unlock_irqrestore(shost->host_lock, flags);
428
429 /*
430 * Call blk_run_queue for all LUNs on the target, starting with
431 * current_sdev. We race with others (to set starget_sdev_user),
432 * but in most cases, we will be first. Ideally, each LU on the
433 * target would get some limited time or requests on the target.
434 */
435 blk_mq_run_hw_queues(current_sdev->request_queue,
436 shost->queuecommand_may_block);
437
438 spin_lock_irqsave(shost->host_lock, flags);
439 if (!starget->starget_sdev_user)
440 __starget_for_each_device(starget, current_sdev,
441 scsi_kick_sdev_queue);
442 spin_unlock_irqrestore(shost->host_lock, flags);
443}
444
445static inline bool scsi_device_is_busy(struct scsi_device *sdev)
446{
447 if (scsi_device_busy(sdev) >= sdev->queue_depth)
448 return true;
449 if (atomic_read(&sdev->device_blocked) > 0)
450 return true;
451 return false;
452}
453
454static inline bool scsi_target_is_busy(struct scsi_target *starget)
455{
456 if (starget->can_queue > 0) {
457 if (atomic_read(&starget->target_busy) >= starget->can_queue)
458 return true;
459 if (atomic_read(&starget->target_blocked) > 0)
460 return true;
461 }
462 return false;
463}
464
465static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
466{
467 if (atomic_read(&shost->host_blocked) > 0)
468 return true;
469 if (shost->host_self_blocked)
470 return true;
471 return false;
472}
473
474static void scsi_starved_list_run(struct Scsi_Host *shost)
475{
476 LIST_HEAD(starved_list);
477 struct scsi_device *sdev;
478 unsigned long flags;
479
480 spin_lock_irqsave(shost->host_lock, flags);
481 list_splice_init(&shost->starved_list, &starved_list);
482
483 while (!list_empty(&starved_list)) {
484 struct request_queue *slq;
485
486 /*
487 * As long as shost is accepting commands and we have
488 * starved queues, call blk_run_queue. scsi_request_fn
489 * drops the queue_lock and can add us back to the
490 * starved_list.
491 *
492 * host_lock protects the starved_list and starved_entry.
493 * scsi_request_fn must get the host_lock before checking
494 * or modifying starved_list or starved_entry.
495 */
496 if (scsi_host_is_busy(shost))
497 break;
498
499 sdev = list_entry(starved_list.next,
500 struct scsi_device, starved_entry);
501 list_del_init(&sdev->starved_entry);
502 if (scsi_target_is_busy(scsi_target(sdev))) {
503 list_move_tail(&sdev->starved_entry,
504 &shost->starved_list);
505 continue;
506 }
507
508 /*
509 * Once we drop the host lock, a racing scsi_remove_device()
510 * call may remove the sdev from the starved list and destroy
511 * it and the queue. Mitigate by taking a reference to the
512 * queue and never touching the sdev again after we drop the
513 * host lock. Note: if __scsi_remove_device() invokes
514 * blk_mq_destroy_queue() before the queue is run from this
515 * function then blk_run_queue() will return immediately since
516 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
517 */
518 slq = sdev->request_queue;
519 if (!blk_get_queue(slq))
520 continue;
521 spin_unlock_irqrestore(shost->host_lock, flags);
522
523 blk_mq_run_hw_queues(slq, false);
524 blk_put_queue(slq);
525
526 spin_lock_irqsave(shost->host_lock, flags);
527 }
528 /* put any unprocessed entries back */
529 list_splice(&starved_list, &shost->starved_list);
530 spin_unlock_irqrestore(shost->host_lock, flags);
531}
532
533/**
534 * scsi_run_queue - Select a proper request queue to serve next.
535 * @q: last request's queue
536 *
537 * The previous command was completely finished, start a new one if possible.
538 */
539static void scsi_run_queue(struct request_queue *q)
540{
541 struct scsi_device *sdev = q->queuedata;
542
543 if (scsi_target(sdev)->single_lun)
544 scsi_single_lun_run(sdev);
545 if (!list_empty(&sdev->host->starved_list))
546 scsi_starved_list_run(sdev->host);
547
548 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
549 blk_mq_kick_requeue_list(q);
550}
551
552void scsi_requeue_run_queue(struct work_struct *work)
553{
554 struct scsi_device *sdev;
555 struct request_queue *q;
556
557 sdev = container_of(work, struct scsi_device, requeue_work);
558 q = sdev->request_queue;
559 scsi_run_queue(q);
560}
561
562void scsi_run_host_queues(struct Scsi_Host *shost)
563{
564 struct scsi_device *sdev;
565
566 shost_for_each_device(sdev, shost)
567 scsi_run_queue(sdev->request_queue);
568}
569
570static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
571{
572 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
573 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
574
575 if (drv->uninit_command)
576 drv->uninit_command(cmd);
577 }
578}
579
580void scsi_free_sgtables(struct scsi_cmnd *cmd)
581{
582 if (cmd->sdb.table.nents)
583 sg_free_table_chained(&cmd->sdb.table,
584 SCSI_INLINE_SG_CNT);
585 if (scsi_prot_sg_count(cmd))
586 sg_free_table_chained(&cmd->prot_sdb->table,
587 SCSI_INLINE_PROT_SG_CNT);
588}
589EXPORT_SYMBOL_GPL(scsi_free_sgtables);
590
591static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
592{
593 scsi_free_sgtables(cmd);
594 scsi_uninit_cmd(cmd);
595}
596
597static void scsi_run_queue_async(struct scsi_device *sdev)
598{
599 if (scsi_host_in_recovery(sdev->host))
600 return;
601
602 if (scsi_target(sdev)->single_lun ||
603 !list_empty(&sdev->host->starved_list)) {
604 kblockd_schedule_work(&sdev->requeue_work);
605 } else {
606 /*
607 * smp_mb() present in sbitmap_queue_clear() or implied in
608 * .end_io is for ordering writing .device_busy in
609 * scsi_device_unbusy() and reading sdev->restarts.
610 */
611 int old = atomic_read(&sdev->restarts);
612
613 /*
614 * ->restarts has to be kept as non-zero if new budget
615 * contention occurs.
616 *
617 * No need to run queue when either another re-run
618 * queue wins in updating ->restarts or a new budget
619 * contention occurs.
620 */
621 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
622 blk_mq_run_hw_queues(sdev->request_queue, true);
623 }
624}
625
626/* Returns false when no more bytes to process, true if there are more */
627static bool scsi_end_request(struct request *req, blk_status_t error,
628 unsigned int bytes)
629{
630 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
631 struct scsi_device *sdev = cmd->device;
632 struct request_queue *q = sdev->request_queue;
633
634 if (blk_update_request(req, error, bytes))
635 return true;
636
637 if (q->limits.features & BLK_FEAT_ADD_RANDOM)
638 add_disk_randomness(req->q->disk);
639
640 WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
641 !(cmd->flags & SCMD_INITIALIZED));
642 cmd->flags = 0;
643
644 /*
645 * Calling rcu_barrier() is not necessary here because the
646 * SCSI error handler guarantees that the function called by
647 * call_rcu() has been called before scsi_end_request() is
648 * called.
649 */
650 destroy_rcu_head(&cmd->rcu);
651
652 /*
653 * In the MQ case the command gets freed by __blk_mq_end_request,
654 * so we have to do all cleanup that depends on it earlier.
655 *
656 * We also can't kick the queues from irq context, so we
657 * will have to defer it to a workqueue.
658 */
659 scsi_mq_uninit_cmd(cmd);
660
661 /*
662 * queue is still alive, so grab the ref for preventing it
663 * from being cleaned up during running queue.
664 */
665 percpu_ref_get(&q->q_usage_counter);
666
667 __blk_mq_end_request(req, error);
668
669 scsi_run_queue_async(sdev);
670
671 percpu_ref_put(&q->q_usage_counter);
672 return false;
673}
674
675/**
676 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
677 * @result: scsi error code
678 *
679 * Translate a SCSI result code into a blk_status_t value.
680 */
681static blk_status_t scsi_result_to_blk_status(int result)
682{
683 /*
684 * Check the scsi-ml byte first in case we converted a host or status
685 * byte.
686 */
687 switch (scsi_ml_byte(result)) {
688 case SCSIML_STAT_OK:
689 break;
690 case SCSIML_STAT_RESV_CONFLICT:
691 return BLK_STS_RESV_CONFLICT;
692 case SCSIML_STAT_NOSPC:
693 return BLK_STS_NOSPC;
694 case SCSIML_STAT_MED_ERROR:
695 return BLK_STS_MEDIUM;
696 case SCSIML_STAT_TGT_FAILURE:
697 return BLK_STS_TARGET;
698 case SCSIML_STAT_DL_TIMEOUT:
699 return BLK_STS_DURATION_LIMIT;
700 }
701
702 switch (host_byte(result)) {
703 case DID_OK:
704 if (scsi_status_is_good(result))
705 return BLK_STS_OK;
706 return BLK_STS_IOERR;
707 case DID_TRANSPORT_FAILFAST:
708 case DID_TRANSPORT_MARGINAL:
709 return BLK_STS_TRANSPORT;
710 default:
711 return BLK_STS_IOERR;
712 }
713}
714
715/**
716 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
717 * @rq: request to examine
718 *
719 * Description:
720 * A request could be merge of IOs which require different failure
721 * handling. This function determines the number of bytes which
722 * can be failed from the beginning of the request without
723 * crossing into area which need to be retried further.
724 *
725 * Return:
726 * The number of bytes to fail.
727 */
728static unsigned int scsi_rq_err_bytes(const struct request *rq)
729{
730 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
731 unsigned int bytes = 0;
732 struct bio *bio;
733
734 if (!(rq->rq_flags & RQF_MIXED_MERGE))
735 return blk_rq_bytes(rq);
736
737 /*
738 * Currently the only 'mixing' which can happen is between
739 * different fastfail types. We can safely fail portions
740 * which have all the failfast bits that the first one has -
741 * the ones which are at least as eager to fail as the first
742 * one.
743 */
744 for (bio = rq->bio; bio; bio = bio->bi_next) {
745 if ((bio->bi_opf & ff) != ff)
746 break;
747 bytes += bio->bi_iter.bi_size;
748 }
749
750 /* this could lead to infinite loop */
751 BUG_ON(blk_rq_bytes(rq) && !bytes);
752 return bytes;
753}
754
755static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
756{
757 struct request *req = scsi_cmd_to_rq(cmd);
758 unsigned long wait_for;
759
760 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
761 return false;
762
763 wait_for = (cmd->allowed + 1) * req->timeout;
764 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
765 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
766 wait_for/HZ);
767 return true;
768 }
769 return false;
770}
771
772/*
773 * When ALUA transition state is returned, reprep the cmd to
774 * use the ALUA handler's transition timeout. Delay the reprep
775 * 1 sec to avoid aggressive retries of the target in that
776 * state.
777 */
778#define ALUA_TRANSITION_REPREP_DELAY 1000
779
780/* Helper for scsi_io_completion() when special action required. */
781static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
782{
783 struct request *req = scsi_cmd_to_rq(cmd);
784 int level = 0;
785 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
786 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
787 struct scsi_sense_hdr sshdr;
788 bool sense_valid;
789 bool sense_current = true; /* false implies "deferred sense" */
790 blk_status_t blk_stat;
791
792 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
793 if (sense_valid)
794 sense_current = !scsi_sense_is_deferred(&sshdr);
795
796 blk_stat = scsi_result_to_blk_status(result);
797
798 if (host_byte(result) == DID_RESET) {
799 /* Third party bus reset or reset for error recovery
800 * reasons. Just retry the command and see what
801 * happens.
802 */
803 action = ACTION_RETRY;
804 } else if (sense_valid && sense_current) {
805 switch (sshdr.sense_key) {
806 case UNIT_ATTENTION:
807 if (cmd->device->removable) {
808 /* Detected disc change. Set a bit
809 * and quietly refuse further access.
810 */
811 cmd->device->changed = 1;
812 action = ACTION_FAIL;
813 } else {
814 /* Must have been a power glitch, or a
815 * bus reset. Could not have been a
816 * media change, so we just retry the
817 * command and see what happens.
818 */
819 action = ACTION_RETRY;
820 }
821 break;
822 case ILLEGAL_REQUEST:
823 /* If we had an ILLEGAL REQUEST returned, then
824 * we may have performed an unsupported
825 * command. The only thing this should be
826 * would be a ten byte read where only a six
827 * byte read was supported. Also, on a system
828 * where READ CAPACITY failed, we may have
829 * read past the end of the disk.
830 */
831 if ((cmd->device->use_10_for_rw &&
832 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
833 (cmd->cmnd[0] == READ_10 ||
834 cmd->cmnd[0] == WRITE_10)) {
835 /* This will issue a new 6-byte command. */
836 cmd->device->use_10_for_rw = 0;
837 action = ACTION_REPREP;
838 } else if (sshdr.asc == 0x10) /* DIX */ {
839 action = ACTION_FAIL;
840 blk_stat = BLK_STS_PROTECTION;
841 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
842 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
843 action = ACTION_FAIL;
844 blk_stat = BLK_STS_TARGET;
845 } else
846 action = ACTION_FAIL;
847 break;
848 case ABORTED_COMMAND:
849 action = ACTION_FAIL;
850 if (sshdr.asc == 0x10) /* DIF */
851 blk_stat = BLK_STS_PROTECTION;
852 break;
853 case NOT_READY:
854 /* If the device is in the process of becoming
855 * ready, or has a temporary blockage, retry.
856 */
857 if (sshdr.asc == 0x04) {
858 switch (sshdr.ascq) {
859 case 0x01: /* becoming ready */
860 case 0x04: /* format in progress */
861 case 0x05: /* rebuild in progress */
862 case 0x06: /* recalculation in progress */
863 case 0x07: /* operation in progress */
864 case 0x08: /* Long write in progress */
865 case 0x09: /* self test in progress */
866 case 0x11: /* notify (enable spinup) required */
867 case 0x14: /* space allocation in progress */
868 case 0x1a: /* start stop unit in progress */
869 case 0x1b: /* sanitize in progress */
870 case 0x1d: /* configuration in progress */
871 action = ACTION_DELAYED_RETRY;
872 break;
873 case 0x0a: /* ALUA state transition */
874 action = ACTION_DELAYED_REPREP;
875 break;
876 /*
877 * Depopulation might take many hours,
878 * thus it is not worthwhile to retry.
879 */
880 case 0x24: /* depopulation in progress */
881 case 0x25: /* depopulation restore in progress */
882 fallthrough;
883 default:
884 action = ACTION_FAIL;
885 break;
886 }
887 } else
888 action = ACTION_FAIL;
889 break;
890 case VOLUME_OVERFLOW:
891 /* See SSC3rXX or current. */
892 action = ACTION_FAIL;
893 break;
894 case DATA_PROTECT:
895 action = ACTION_FAIL;
896 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
897 (sshdr.asc == 0x55 &&
898 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
899 /* Insufficient zone resources */
900 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
901 }
902 break;
903 case COMPLETED:
904 fallthrough;
905 default:
906 action = ACTION_FAIL;
907 break;
908 }
909 } else
910 action = ACTION_FAIL;
911
912 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
913 action = ACTION_FAIL;
914
915 switch (action) {
916 case ACTION_FAIL:
917 /* Give up and fail the remainder of the request */
918 if (!(req->rq_flags & RQF_QUIET)) {
919 static DEFINE_RATELIMIT_STATE(_rs,
920 DEFAULT_RATELIMIT_INTERVAL,
921 DEFAULT_RATELIMIT_BURST);
922
923 if (unlikely(scsi_logging_level))
924 level =
925 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
926 SCSI_LOG_MLCOMPLETE_BITS);
927
928 /*
929 * if logging is enabled the failure will be printed
930 * in scsi_log_completion(), so avoid duplicate messages
931 */
932 if (!level && __ratelimit(&_rs)) {
933 scsi_print_result(cmd, NULL, FAILED);
934 if (sense_valid)
935 scsi_print_sense(cmd);
936 scsi_print_command(cmd);
937 }
938 }
939 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
940 return;
941 fallthrough;
942 case ACTION_REPREP:
943 scsi_mq_requeue_cmd(cmd, 0);
944 break;
945 case ACTION_DELAYED_REPREP:
946 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
947 break;
948 case ACTION_RETRY:
949 /* Retry the same command immediately */
950 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
951 break;
952 case ACTION_DELAYED_RETRY:
953 /* Retry the same command after a delay */
954 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
955 break;
956 }
957}
958
959/*
960 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
961 * new result that may suppress further error checking. Also modifies
962 * *blk_statp in some cases.
963 */
964static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
965 blk_status_t *blk_statp)
966{
967 bool sense_valid;
968 bool sense_current = true; /* false implies "deferred sense" */
969 struct request *req = scsi_cmd_to_rq(cmd);
970 struct scsi_sense_hdr sshdr;
971
972 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
973 if (sense_valid)
974 sense_current = !scsi_sense_is_deferred(&sshdr);
975
976 if (blk_rq_is_passthrough(req)) {
977 if (sense_valid) {
978 /*
979 * SG_IO wants current and deferred errors
980 */
981 cmd->sense_len = min(8 + cmd->sense_buffer[7],
982 SCSI_SENSE_BUFFERSIZE);
983 }
984 if (sense_current)
985 *blk_statp = scsi_result_to_blk_status(result);
986 } else if (blk_rq_bytes(req) == 0 && sense_current) {
987 /*
988 * Flush commands do not transfers any data, and thus cannot use
989 * good_bytes != blk_rq_bytes(req) as the signal for an error.
990 * This sets *blk_statp explicitly for the problem case.
991 */
992 *blk_statp = scsi_result_to_blk_status(result);
993 }
994 /*
995 * Recovered errors need reporting, but they're always treated as
996 * success, so fiddle the result code here. For passthrough requests
997 * we already took a copy of the original into sreq->result which
998 * is what gets returned to the user
999 */
1000 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
1001 bool do_print = true;
1002 /*
1003 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
1004 * skip print since caller wants ATA registers. Only occurs
1005 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
1006 */
1007 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1008 do_print = false;
1009 else if (req->rq_flags & RQF_QUIET)
1010 do_print = false;
1011 if (do_print)
1012 scsi_print_sense(cmd);
1013 result = 0;
1014 /* for passthrough, *blk_statp may be set */
1015 *blk_statp = BLK_STS_OK;
1016 }
1017 /*
1018 * Another corner case: the SCSI status byte is non-zero but 'good'.
1019 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1020 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1021 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1022 * intermediate statuses (both obsolete in SAM-4) as good.
1023 */
1024 if ((result & 0xff) && scsi_status_is_good(result)) {
1025 result = 0;
1026 *blk_statp = BLK_STS_OK;
1027 }
1028 return result;
1029}
1030
1031/**
1032 * scsi_io_completion - Completion processing for SCSI commands.
1033 * @cmd: command that is finished.
1034 * @good_bytes: number of processed bytes.
1035 *
1036 * We will finish off the specified number of sectors. If we are done, the
1037 * command block will be released and the queue function will be goosed. If we
1038 * are not done then we have to figure out what to do next:
1039 *
1040 * a) We can call scsi_mq_requeue_cmd(). The request will be
1041 * unprepared and put back on the queue. Then a new command will
1042 * be created for it. This should be used if we made forward
1043 * progress, or if we want to switch from READ(10) to READ(6) for
1044 * example.
1045 *
1046 * b) We can call scsi_io_completion_action(). The request will be
1047 * put back on the queue and retried using the same command as
1048 * before, possibly after a delay.
1049 *
1050 * c) We can call scsi_end_request() with blk_stat other than
1051 * BLK_STS_OK, to fail the remainder of the request.
1052 */
1053void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1054{
1055 int result = cmd->result;
1056 struct request *req = scsi_cmd_to_rq(cmd);
1057 blk_status_t blk_stat = BLK_STS_OK;
1058
1059 if (unlikely(result)) /* a nz result may or may not be an error */
1060 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1061
1062 /*
1063 * Next deal with any sectors which we were able to correctly
1064 * handle.
1065 */
1066 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1067 "%u sectors total, %d bytes done.\n",
1068 blk_rq_sectors(req), good_bytes));
1069
1070 /*
1071 * Failed, zero length commands always need to drop down
1072 * to retry code. Fast path should return in this block.
1073 */
1074 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1075 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1076 return; /* no bytes remaining */
1077 }
1078
1079 /* Kill remainder if no retries. */
1080 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1081 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
1082 WARN_ONCE(true,
1083 "Bytes remaining after failed, no-retry command");
1084 return;
1085 }
1086
1087 /*
1088 * If there had been no error, but we have leftover bytes in the
1089 * request just queue the command up again.
1090 */
1091 if (likely(result == 0))
1092 scsi_mq_requeue_cmd(cmd, 0);
1093 else
1094 scsi_io_completion_action(cmd, result);
1095}
1096
1097static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1098 struct request *rq)
1099{
1100 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1101 !op_is_write(req_op(rq)) &&
1102 sdev->host->hostt->dma_need_drain(rq);
1103}
1104
1105/**
1106 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1107 * @cmd: SCSI command data structure to initialize.
1108 *
1109 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1110 * for @cmd.
1111 *
1112 * Returns:
1113 * * BLK_STS_OK - on success
1114 * * BLK_STS_RESOURCE - if the failure is retryable
1115 * * BLK_STS_IOERR - if the failure is fatal
1116 */
1117blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1118{
1119 struct scsi_device *sdev = cmd->device;
1120 struct request *rq = scsi_cmd_to_rq(cmd);
1121 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1122 struct scatterlist *last_sg = NULL;
1123 blk_status_t ret;
1124 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1125 int count;
1126
1127 if (WARN_ON_ONCE(!nr_segs))
1128 return BLK_STS_IOERR;
1129
1130 /*
1131 * Make sure there is space for the drain. The driver must adjust
1132 * max_hw_segments to be prepared for this.
1133 */
1134 if (need_drain)
1135 nr_segs++;
1136
1137 /*
1138 * If sg table allocation fails, requeue request later.
1139 */
1140 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1141 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1142 return BLK_STS_RESOURCE;
1143
1144 /*
1145 * Next, walk the list, and fill in the addresses and sizes of
1146 * each segment.
1147 */
1148 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1149
1150 if (blk_rq_bytes(rq) & rq->q->limits.dma_pad_mask) {
1151 unsigned int pad_len =
1152 (rq->q->limits.dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1153
1154 last_sg->length += pad_len;
1155 cmd->extra_len += pad_len;
1156 }
1157
1158 if (need_drain) {
1159 sg_unmark_end(last_sg);
1160 last_sg = sg_next(last_sg);
1161 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1162 sg_mark_end(last_sg);
1163
1164 cmd->extra_len += sdev->dma_drain_len;
1165 count++;
1166 }
1167
1168 BUG_ON(count > cmd->sdb.table.nents);
1169 cmd->sdb.table.nents = count;
1170 cmd->sdb.length = blk_rq_payload_bytes(rq);
1171
1172 if (blk_integrity_rq(rq)) {
1173 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1174
1175 if (WARN_ON_ONCE(!prot_sdb)) {
1176 /*
1177 * This can happen if someone (e.g. multipath)
1178 * queues a command to a device on an adapter
1179 * that does not support DIX.
1180 */
1181 ret = BLK_STS_IOERR;
1182 goto out_free_sgtables;
1183 }
1184
1185 if (sg_alloc_table_chained(&prot_sdb->table,
1186 rq->nr_integrity_segments,
1187 prot_sdb->table.sgl,
1188 SCSI_INLINE_PROT_SG_CNT)) {
1189 ret = BLK_STS_RESOURCE;
1190 goto out_free_sgtables;
1191 }
1192
1193 count = blk_rq_map_integrity_sg(rq, prot_sdb->table.sgl);
1194 cmd->prot_sdb = prot_sdb;
1195 cmd->prot_sdb->table.nents = count;
1196 }
1197
1198 return BLK_STS_OK;
1199out_free_sgtables:
1200 scsi_free_sgtables(cmd);
1201 return ret;
1202}
1203EXPORT_SYMBOL(scsi_alloc_sgtables);
1204
1205/**
1206 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1207 * @rq: Request associated with the SCSI command to be initialized.
1208 *
1209 * This function initializes the members of struct scsi_cmnd that must be
1210 * initialized before request processing starts and that won't be
1211 * reinitialized if a SCSI command is requeued.
1212 */
1213static void scsi_initialize_rq(struct request *rq)
1214{
1215 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1216
1217 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1218 cmd->cmd_len = MAX_COMMAND_SIZE;
1219 cmd->sense_len = 0;
1220 init_rcu_head(&cmd->rcu);
1221 cmd->jiffies_at_alloc = jiffies;
1222 cmd->retries = 0;
1223}
1224
1225struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1226 blk_mq_req_flags_t flags)
1227{
1228 struct request *rq;
1229
1230 rq = blk_mq_alloc_request(q, opf, flags);
1231 if (!IS_ERR(rq))
1232 scsi_initialize_rq(rq);
1233 return rq;
1234}
1235EXPORT_SYMBOL_GPL(scsi_alloc_request);
1236
1237/*
1238 * Only called when the request isn't completed by SCSI, and not freed by
1239 * SCSI
1240 */
1241static void scsi_cleanup_rq(struct request *rq)
1242{
1243 if (rq->rq_flags & RQF_DONTPREP) {
1244 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1245 rq->rq_flags &= ~RQF_DONTPREP;
1246 }
1247}
1248
1249/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1250void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1251{
1252 struct request *rq = scsi_cmd_to_rq(cmd);
1253
1254 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1255 cmd->flags |= SCMD_INITIALIZED;
1256 scsi_initialize_rq(rq);
1257 }
1258
1259 cmd->device = dev;
1260 INIT_LIST_HEAD(&cmd->eh_entry);
1261 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1262}
1263
1264static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1265 struct request *req)
1266{
1267 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1268
1269 /*
1270 * Passthrough requests may transfer data, in which case they must
1271 * a bio attached to them. Or they might contain a SCSI command
1272 * that does not transfer data, in which case they may optionally
1273 * submit a request without an attached bio.
1274 */
1275 if (req->bio) {
1276 blk_status_t ret = scsi_alloc_sgtables(cmd);
1277 if (unlikely(ret != BLK_STS_OK))
1278 return ret;
1279 } else {
1280 BUG_ON(blk_rq_bytes(req));
1281
1282 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1283 }
1284
1285 cmd->transfersize = blk_rq_bytes(req);
1286 return BLK_STS_OK;
1287}
1288
1289static blk_status_t
1290scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1291{
1292 switch (sdev->sdev_state) {
1293 case SDEV_CREATED:
1294 return BLK_STS_OK;
1295 case SDEV_OFFLINE:
1296 case SDEV_TRANSPORT_OFFLINE:
1297 /*
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1301 */
1302 if (!sdev->offline_already) {
1303 sdev->offline_already = true;
1304 sdev_printk(KERN_ERR, sdev,
1305 "rejecting I/O to offline device\n");
1306 }
1307 return BLK_STS_IOERR;
1308 case SDEV_DEL:
1309 /*
1310 * If the device is fully deleted, we refuse to
1311 * process any commands as well.
1312 */
1313 sdev_printk(KERN_ERR, sdev,
1314 "rejecting I/O to dead device\n");
1315 return BLK_STS_IOERR;
1316 case SDEV_BLOCK:
1317 case SDEV_CREATED_BLOCK:
1318 return BLK_STS_RESOURCE;
1319 case SDEV_QUIESCE:
1320 /*
1321 * If the device is blocked we only accept power management
1322 * commands.
1323 */
1324 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1325 return BLK_STS_RESOURCE;
1326 return BLK_STS_OK;
1327 default:
1328 /*
1329 * For any other not fully online state we only allow
1330 * power management commands.
1331 */
1332 if (req && !(req->rq_flags & RQF_PM))
1333 return BLK_STS_OFFLINE;
1334 return BLK_STS_OK;
1335 }
1336}
1337
1338/*
1339 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1340 * and return the token else return -1.
1341 */
1342static inline int scsi_dev_queue_ready(struct request_queue *q,
1343 struct scsi_device *sdev)
1344{
1345 int token;
1346
1347 token = sbitmap_get(&sdev->budget_map);
1348 if (token < 0)
1349 return -1;
1350
1351 if (!atomic_read(&sdev->device_blocked))
1352 return token;
1353
1354 /*
1355 * Only unblock if no other commands are pending and
1356 * if device_blocked has decreased to zero
1357 */
1358 if (scsi_device_busy(sdev) > 1 ||
1359 atomic_dec_return(&sdev->device_blocked) > 0) {
1360 sbitmap_put(&sdev->budget_map, token);
1361 return -1;
1362 }
1363
1364 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1365 "unblocking device at zero depth\n"));
1366
1367 return token;
1368}
1369
1370/*
1371 * scsi_target_queue_ready: checks if there we can send commands to target
1372 * @sdev: scsi device on starget to check.
1373 */
1374static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1375 struct scsi_device *sdev)
1376{
1377 struct scsi_target *starget = scsi_target(sdev);
1378 unsigned int busy;
1379
1380 if (starget->single_lun) {
1381 spin_lock_irq(shost->host_lock);
1382 if (starget->starget_sdev_user &&
1383 starget->starget_sdev_user != sdev) {
1384 spin_unlock_irq(shost->host_lock);
1385 return 0;
1386 }
1387 starget->starget_sdev_user = sdev;
1388 spin_unlock_irq(shost->host_lock);
1389 }
1390
1391 if (starget->can_queue <= 0)
1392 return 1;
1393
1394 busy = atomic_inc_return(&starget->target_busy) - 1;
1395 if (atomic_read(&starget->target_blocked) > 0) {
1396 if (busy)
1397 goto starved;
1398
1399 /*
1400 * unblock after target_blocked iterates to zero
1401 */
1402 if (atomic_dec_return(&starget->target_blocked) > 0)
1403 goto out_dec;
1404
1405 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1406 "unblocking target at zero depth\n"));
1407 }
1408
1409 if (busy >= starget->can_queue)
1410 goto starved;
1411
1412 return 1;
1413
1414starved:
1415 spin_lock_irq(shost->host_lock);
1416 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1417 spin_unlock_irq(shost->host_lock);
1418out_dec:
1419 if (starget->can_queue > 0)
1420 atomic_dec(&starget->target_busy);
1421 return 0;
1422}
1423
1424/*
1425 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1426 * return 0. We must end up running the queue again whenever 0 is
1427 * returned, else IO can hang.
1428 */
1429static inline int scsi_host_queue_ready(struct request_queue *q,
1430 struct Scsi_Host *shost,
1431 struct scsi_device *sdev,
1432 struct scsi_cmnd *cmd)
1433{
1434 if (atomic_read(&shost->host_blocked) > 0) {
1435 if (scsi_host_busy(shost) > 0)
1436 goto starved;
1437
1438 /*
1439 * unblock after host_blocked iterates to zero
1440 */
1441 if (atomic_dec_return(&shost->host_blocked) > 0)
1442 goto out_dec;
1443
1444 SCSI_LOG_MLQUEUE(3,
1445 shost_printk(KERN_INFO, shost,
1446 "unblocking host at zero depth\n"));
1447 }
1448
1449 if (shost->host_self_blocked)
1450 goto starved;
1451
1452 /* We're OK to process the command, so we can't be starved */
1453 if (!list_empty(&sdev->starved_entry)) {
1454 spin_lock_irq(shost->host_lock);
1455 if (!list_empty(&sdev->starved_entry))
1456 list_del_init(&sdev->starved_entry);
1457 spin_unlock_irq(shost->host_lock);
1458 }
1459
1460 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1461
1462 return 1;
1463
1464starved:
1465 spin_lock_irq(shost->host_lock);
1466 if (list_empty(&sdev->starved_entry))
1467 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1468 spin_unlock_irq(shost->host_lock);
1469out_dec:
1470 scsi_dec_host_busy(shost, cmd);
1471 return 0;
1472}
1473
1474/*
1475 * Busy state exporting function for request stacking drivers.
1476 *
1477 * For efficiency, no lock is taken to check the busy state of
1478 * shost/starget/sdev, since the returned value is not guaranteed and
1479 * may be changed after request stacking drivers call the function,
1480 * regardless of taking lock or not.
1481 *
1482 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1483 * needs to return 'not busy'. Otherwise, request stacking drivers
1484 * may hold requests forever.
1485 */
1486static bool scsi_mq_lld_busy(struct request_queue *q)
1487{
1488 struct scsi_device *sdev = q->queuedata;
1489 struct Scsi_Host *shost;
1490
1491 if (blk_queue_dying(q))
1492 return false;
1493
1494 shost = sdev->host;
1495
1496 /*
1497 * Ignore host/starget busy state.
1498 * Since block layer does not have a concept of fairness across
1499 * multiple queues, congestion of host/starget needs to be handled
1500 * in SCSI layer.
1501 */
1502 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1503 return true;
1504
1505 return false;
1506}
1507
1508/*
1509 * Block layer request completion callback. May be called from interrupt
1510 * context.
1511 */
1512static void scsi_complete(struct request *rq)
1513{
1514 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1515 enum scsi_disposition disposition;
1516
1517 INIT_LIST_HEAD(&cmd->eh_entry);
1518
1519 atomic_inc(&cmd->device->iodone_cnt);
1520 if (cmd->result)
1521 atomic_inc(&cmd->device->ioerr_cnt);
1522
1523 disposition = scsi_decide_disposition(cmd);
1524 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1525 disposition = SUCCESS;
1526
1527 scsi_log_completion(cmd, disposition);
1528
1529 switch (disposition) {
1530 case SUCCESS:
1531 scsi_finish_command(cmd);
1532 break;
1533 case NEEDS_RETRY:
1534 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1535 break;
1536 case ADD_TO_MLQUEUE:
1537 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1538 break;
1539 default:
1540 scsi_eh_scmd_add(cmd);
1541 break;
1542 }
1543}
1544
1545/**
1546 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1547 * @cmd: command block we are dispatching.
1548 *
1549 * Return: nonzero return request was rejected and device's queue needs to be
1550 * plugged.
1551 */
1552static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1553{
1554 struct Scsi_Host *host = cmd->device->host;
1555 int rtn = 0;
1556
1557 atomic_inc(&cmd->device->iorequest_cnt);
1558
1559 /* check if the device is still usable */
1560 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1561 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1562 * returns an immediate error upwards, and signals
1563 * that the device is no longer present */
1564 cmd->result = DID_NO_CONNECT << 16;
1565 goto done;
1566 }
1567
1568 /* Check to see if the scsi lld made this device blocked. */
1569 if (unlikely(scsi_device_blocked(cmd->device))) {
1570 /*
1571 * in blocked state, the command is just put back on
1572 * the device queue. The suspend state has already
1573 * blocked the queue so future requests should not
1574 * occur until the device transitions out of the
1575 * suspend state.
1576 */
1577 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1578 "queuecommand : device blocked\n"));
1579 atomic_dec(&cmd->device->iorequest_cnt);
1580 return SCSI_MLQUEUE_DEVICE_BUSY;
1581 }
1582
1583 /* Store the LUN value in cmnd, if needed. */
1584 if (cmd->device->lun_in_cdb)
1585 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1586 (cmd->device->lun << 5 & 0xe0);
1587
1588 scsi_log_send(cmd);
1589
1590 /*
1591 * Before we queue this command, check if the command
1592 * length exceeds what the host adapter can handle.
1593 */
1594 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1595 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1596 "queuecommand : command too long. "
1597 "cdb_size=%d host->max_cmd_len=%d\n",
1598 cmd->cmd_len, cmd->device->host->max_cmd_len));
1599 cmd->result = (DID_ABORT << 16);
1600 goto done;
1601 }
1602
1603 if (unlikely(host->shost_state == SHOST_DEL)) {
1604 cmd->result = (DID_NO_CONNECT << 16);
1605 goto done;
1606
1607 }
1608
1609 trace_scsi_dispatch_cmd_start(cmd);
1610 rtn = host->hostt->queuecommand(host, cmd);
1611 if (rtn) {
1612 atomic_dec(&cmd->device->iorequest_cnt);
1613 trace_scsi_dispatch_cmd_error(cmd, rtn);
1614 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1615 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1616 rtn = SCSI_MLQUEUE_HOST_BUSY;
1617
1618 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1619 "queuecommand : request rejected\n"));
1620 }
1621
1622 return rtn;
1623 done:
1624 scsi_done(cmd);
1625 return 0;
1626}
1627
1628/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1629static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1630{
1631 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1632 sizeof(struct scatterlist);
1633}
1634
1635static blk_status_t scsi_prepare_cmd(struct request *req)
1636{
1637 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1638 struct scsi_device *sdev = req->q->queuedata;
1639 struct Scsi_Host *shost = sdev->host;
1640 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1641 struct scatterlist *sg;
1642
1643 scsi_init_command(sdev, cmd);
1644
1645 cmd->eh_eflags = 0;
1646 cmd->prot_type = 0;
1647 cmd->prot_flags = 0;
1648 cmd->submitter = 0;
1649 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1650 cmd->underflow = 0;
1651 cmd->transfersize = 0;
1652 cmd->host_scribble = NULL;
1653 cmd->result = 0;
1654 cmd->extra_len = 0;
1655 cmd->state = 0;
1656 if (in_flight)
1657 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1658
1659 cmd->prot_op = SCSI_PROT_NORMAL;
1660 if (blk_rq_bytes(req))
1661 cmd->sc_data_direction = rq_dma_dir(req);
1662 else
1663 cmd->sc_data_direction = DMA_NONE;
1664
1665 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1666 cmd->sdb.table.sgl = sg;
1667
1668 if (scsi_host_get_prot(shost)) {
1669 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1670
1671 cmd->prot_sdb->table.sgl =
1672 (struct scatterlist *)(cmd->prot_sdb + 1);
1673 }
1674
1675 /*
1676 * Special handling for passthrough commands, which don't go to the ULP
1677 * at all:
1678 */
1679 if (blk_rq_is_passthrough(req))
1680 return scsi_setup_scsi_cmnd(sdev, req);
1681
1682 if (sdev->handler && sdev->handler->prep_fn) {
1683 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1684
1685 if (ret != BLK_STS_OK)
1686 return ret;
1687 }
1688
1689 /* Usually overridden by the ULP */
1690 cmd->allowed = 0;
1691 memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1692 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1693}
1694
1695static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1696{
1697 struct request *req = scsi_cmd_to_rq(cmd);
1698
1699 switch (cmd->submitter) {
1700 case SUBMITTED_BY_BLOCK_LAYER:
1701 break;
1702 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1703 return scsi_eh_done(cmd);
1704 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1705 return;
1706 }
1707
1708 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1709 return;
1710 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1711 return;
1712 trace_scsi_dispatch_cmd_done(cmd);
1713
1714 if (complete_directly)
1715 blk_mq_complete_request_direct(req, scsi_complete);
1716 else
1717 blk_mq_complete_request(req);
1718}
1719
1720void scsi_done(struct scsi_cmnd *cmd)
1721{
1722 scsi_done_internal(cmd, false);
1723}
1724EXPORT_SYMBOL(scsi_done);
1725
1726void scsi_done_direct(struct scsi_cmnd *cmd)
1727{
1728 scsi_done_internal(cmd, true);
1729}
1730EXPORT_SYMBOL(scsi_done_direct);
1731
1732static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1733{
1734 struct scsi_device *sdev = q->queuedata;
1735
1736 sbitmap_put(&sdev->budget_map, budget_token);
1737}
1738
1739/*
1740 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1741 * not change behaviour from the previous unplug mechanism, experimentation
1742 * may prove this needs changing.
1743 */
1744#define SCSI_QUEUE_DELAY 3
1745
1746static int scsi_mq_get_budget(struct request_queue *q)
1747{
1748 struct scsi_device *sdev = q->queuedata;
1749 int token = scsi_dev_queue_ready(q, sdev);
1750
1751 if (token >= 0)
1752 return token;
1753
1754 atomic_inc(&sdev->restarts);
1755
1756 /*
1757 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1758 * .restarts must be incremented before .device_busy is read because the
1759 * code in scsi_run_queue_async() depends on the order of these operations.
1760 */
1761 smp_mb__after_atomic();
1762
1763 /*
1764 * If all in-flight requests originated from this LUN are completed
1765 * before reading .device_busy, sdev->device_busy will be observed as
1766 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1767 * soon. Otherwise, completion of one of these requests will observe
1768 * the .restarts flag, and the request queue will be run for handling
1769 * this request, see scsi_end_request().
1770 */
1771 if (unlikely(scsi_device_busy(sdev) == 0 &&
1772 !scsi_device_blocked(sdev)))
1773 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1774 return -1;
1775}
1776
1777static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1778{
1779 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1780
1781 cmd->budget_token = token;
1782}
1783
1784static int scsi_mq_get_rq_budget_token(struct request *req)
1785{
1786 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1787
1788 return cmd->budget_token;
1789}
1790
1791static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1792 const struct blk_mq_queue_data *bd)
1793{
1794 struct request *req = bd->rq;
1795 struct request_queue *q = req->q;
1796 struct scsi_device *sdev = q->queuedata;
1797 struct Scsi_Host *shost = sdev->host;
1798 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1799 blk_status_t ret;
1800 int reason;
1801
1802 WARN_ON_ONCE(cmd->budget_token < 0);
1803
1804 /*
1805 * If the device is not in running state we will reject some or all
1806 * commands.
1807 */
1808 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1809 ret = scsi_device_state_check(sdev, req);
1810 if (ret != BLK_STS_OK)
1811 goto out_put_budget;
1812 }
1813
1814 ret = BLK_STS_RESOURCE;
1815 if (!scsi_target_queue_ready(shost, sdev))
1816 goto out_put_budget;
1817 if (unlikely(scsi_host_in_recovery(shost))) {
1818 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1819 ret = BLK_STS_OFFLINE;
1820 goto out_dec_target_busy;
1821 }
1822 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1823 goto out_dec_target_busy;
1824
1825 /*
1826 * Only clear the driver-private command data if the LLD does not supply
1827 * a function to initialize that data.
1828 */
1829 if (shost->hostt->cmd_size && !shost->hostt->init_cmd_priv)
1830 memset(cmd + 1, 0, shost->hostt->cmd_size);
1831
1832 if (!(req->rq_flags & RQF_DONTPREP)) {
1833 ret = scsi_prepare_cmd(req);
1834 if (ret != BLK_STS_OK)
1835 goto out_dec_host_busy;
1836 req->rq_flags |= RQF_DONTPREP;
1837 } else {
1838 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1839 }
1840
1841 cmd->flags &= SCMD_PRESERVED_FLAGS;
1842 if (sdev->simple_tags)
1843 cmd->flags |= SCMD_TAGGED;
1844 if (bd->last)
1845 cmd->flags |= SCMD_LAST;
1846
1847 scsi_set_resid(cmd, 0);
1848 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1849 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1850
1851 blk_mq_start_request(req);
1852 reason = scsi_dispatch_cmd(cmd);
1853 if (reason) {
1854 scsi_set_blocked(cmd, reason);
1855 ret = BLK_STS_RESOURCE;
1856 goto out_dec_host_busy;
1857 }
1858
1859 return BLK_STS_OK;
1860
1861out_dec_host_busy:
1862 scsi_dec_host_busy(shost, cmd);
1863out_dec_target_busy:
1864 if (scsi_target(sdev)->can_queue > 0)
1865 atomic_dec(&scsi_target(sdev)->target_busy);
1866out_put_budget:
1867 scsi_mq_put_budget(q, cmd->budget_token);
1868 cmd->budget_token = -1;
1869 switch (ret) {
1870 case BLK_STS_OK:
1871 break;
1872 case BLK_STS_RESOURCE:
1873 if (scsi_device_blocked(sdev))
1874 ret = BLK_STS_DEV_RESOURCE;
1875 break;
1876 case BLK_STS_AGAIN:
1877 cmd->result = DID_BUS_BUSY << 16;
1878 if (req->rq_flags & RQF_DONTPREP)
1879 scsi_mq_uninit_cmd(cmd);
1880 break;
1881 default:
1882 if (unlikely(!scsi_device_online(sdev)))
1883 cmd->result = DID_NO_CONNECT << 16;
1884 else
1885 cmd->result = DID_ERROR << 16;
1886 /*
1887 * Make sure to release all allocated resources when
1888 * we hit an error, as we will never see this command
1889 * again.
1890 */
1891 if (req->rq_flags & RQF_DONTPREP)
1892 scsi_mq_uninit_cmd(cmd);
1893 scsi_run_queue_async(sdev);
1894 break;
1895 }
1896 return ret;
1897}
1898
1899static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1900 unsigned int hctx_idx, unsigned int numa_node)
1901{
1902 struct Scsi_Host *shost = set->driver_data;
1903 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1904 struct scatterlist *sg;
1905 int ret = 0;
1906
1907 cmd->sense_buffer =
1908 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1909 if (!cmd->sense_buffer)
1910 return -ENOMEM;
1911
1912 if (scsi_host_get_prot(shost)) {
1913 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1914 shost->hostt->cmd_size;
1915 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1916 }
1917
1918 if (shost->hostt->init_cmd_priv) {
1919 ret = shost->hostt->init_cmd_priv(shost, cmd);
1920 if (ret < 0)
1921 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1922 }
1923
1924 return ret;
1925}
1926
1927static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1928 unsigned int hctx_idx)
1929{
1930 struct Scsi_Host *shost = set->driver_data;
1931 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1932
1933 if (shost->hostt->exit_cmd_priv)
1934 shost->hostt->exit_cmd_priv(shost, cmd);
1935 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1936}
1937
1938
1939static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1940{
1941 struct Scsi_Host *shost = hctx->driver_data;
1942
1943 if (shost->hostt->mq_poll)
1944 return shost->hostt->mq_poll(shost, hctx->queue_num);
1945
1946 return 0;
1947}
1948
1949static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1950 unsigned int hctx_idx)
1951{
1952 struct Scsi_Host *shost = data;
1953
1954 hctx->driver_data = shost;
1955 return 0;
1956}
1957
1958static void scsi_map_queues(struct blk_mq_tag_set *set)
1959{
1960 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1961
1962 if (shost->hostt->map_queues)
1963 return shost->hostt->map_queues(shost);
1964 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1965}
1966
1967void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
1968{
1969 struct device *dev = shost->dma_dev;
1970
1971 memset(lim, 0, sizeof(*lim));
1972 lim->max_segments =
1973 min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);
1974
1975 if (scsi_host_prot_dma(shost)) {
1976 shost->sg_prot_tablesize =
1977 min_not_zero(shost->sg_prot_tablesize,
1978 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1979 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1980 lim->max_integrity_segments = shost->sg_prot_tablesize;
1981 }
1982
1983 lim->max_hw_sectors = shost->max_sectors;
1984 lim->seg_boundary_mask = shost->dma_boundary;
1985 lim->max_segment_size = shost->max_segment_size;
1986 lim->virt_boundary_mask = shost->virt_boundary_mask;
1987 lim->dma_alignment = max_t(unsigned int,
1988 shost->dma_alignment, dma_get_cache_alignment() - 1);
1989
1990 if (shost->no_highmem)
1991 lim->features |= BLK_FEAT_BOUNCE_HIGH;
1992
1993 /*
1994 * Propagate the DMA formation properties to the dma-mapping layer as
1995 * a courtesy service to the LLDDs. This needs to check that the buses
1996 * actually support the DMA API first, though.
1997 */
1998 if (dev->dma_parms) {
1999 dma_set_seg_boundary(dev, shost->dma_boundary);
2000 dma_set_max_seg_size(dev, shost->max_segment_size);
2001 }
2002}
2003EXPORT_SYMBOL_GPL(scsi_init_limits);
2004
2005static const struct blk_mq_ops scsi_mq_ops_no_commit = {
2006 .get_budget = scsi_mq_get_budget,
2007 .put_budget = scsi_mq_put_budget,
2008 .queue_rq = scsi_queue_rq,
2009 .complete = scsi_complete,
2010 .timeout = scsi_timeout,
2011#ifdef CONFIG_BLK_DEBUG_FS
2012 .show_rq = scsi_show_rq,
2013#endif
2014 .init_request = scsi_mq_init_request,
2015 .exit_request = scsi_mq_exit_request,
2016 .cleanup_rq = scsi_cleanup_rq,
2017 .busy = scsi_mq_lld_busy,
2018 .map_queues = scsi_map_queues,
2019 .init_hctx = scsi_init_hctx,
2020 .poll = scsi_mq_poll,
2021 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2022 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2023};
2024
2025
2026static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2027{
2028 struct Scsi_Host *shost = hctx->driver_data;
2029
2030 shost->hostt->commit_rqs(shost, hctx->queue_num);
2031}
2032
2033static const struct blk_mq_ops scsi_mq_ops = {
2034 .get_budget = scsi_mq_get_budget,
2035 .put_budget = scsi_mq_put_budget,
2036 .queue_rq = scsi_queue_rq,
2037 .commit_rqs = scsi_commit_rqs,
2038 .complete = scsi_complete,
2039 .timeout = scsi_timeout,
2040#ifdef CONFIG_BLK_DEBUG_FS
2041 .show_rq = scsi_show_rq,
2042#endif
2043 .init_request = scsi_mq_init_request,
2044 .exit_request = scsi_mq_exit_request,
2045 .cleanup_rq = scsi_cleanup_rq,
2046 .busy = scsi_mq_lld_busy,
2047 .map_queues = scsi_map_queues,
2048 .init_hctx = scsi_init_hctx,
2049 .poll = scsi_mq_poll,
2050 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2051 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2052};
2053
2054int scsi_mq_setup_tags(struct Scsi_Host *shost)
2055{
2056 unsigned int cmd_size, sgl_size;
2057 struct blk_mq_tag_set *tag_set = &shost->tag_set;
2058
2059 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2060 scsi_mq_inline_sgl_size(shost));
2061 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2062 if (scsi_host_get_prot(shost))
2063 cmd_size += sizeof(struct scsi_data_buffer) +
2064 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2065
2066 memset(tag_set, 0, sizeof(*tag_set));
2067 if (shost->hostt->commit_rqs)
2068 tag_set->ops = &scsi_mq_ops;
2069 else
2070 tag_set->ops = &scsi_mq_ops_no_commit;
2071 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2072 tag_set->nr_maps = shost->nr_maps ? : 1;
2073 tag_set->queue_depth = shost->can_queue;
2074 tag_set->cmd_size = cmd_size;
2075 tag_set->numa_node = dev_to_node(shost->dma_dev);
2076 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
2077 tag_set->flags |=
2078 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2079 if (shost->queuecommand_may_block)
2080 tag_set->flags |= BLK_MQ_F_BLOCKING;
2081 tag_set->driver_data = shost;
2082 if (shost->host_tagset)
2083 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2084
2085 return blk_mq_alloc_tag_set(tag_set);
2086}
2087
2088void scsi_mq_free_tags(struct kref *kref)
2089{
2090 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2091 tagset_refcnt);
2092
2093 blk_mq_free_tag_set(&shost->tag_set);
2094 complete(&shost->tagset_freed);
2095}
2096
2097/**
2098 * scsi_device_from_queue - return sdev associated with a request_queue
2099 * @q: The request queue to return the sdev from
2100 *
2101 * Return the sdev associated with a request queue or NULL if the
2102 * request_queue does not reference a SCSI device.
2103 */
2104struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2105{
2106 struct scsi_device *sdev = NULL;
2107
2108 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2109 q->mq_ops == &scsi_mq_ops)
2110 sdev = q->queuedata;
2111 if (!sdev || !get_device(&sdev->sdev_gendev))
2112 sdev = NULL;
2113
2114 return sdev;
2115}
2116/*
2117 * pktcdvd should have been integrated into the SCSI layers, but for historical
2118 * reasons like the old IDE driver it isn't. This export allows it to safely
2119 * probe if a given device is a SCSI one and only attach to that.
2120 */
2121#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2122EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2123#endif
2124
2125/**
2126 * scsi_block_requests - Utility function used by low-level drivers to prevent
2127 * further commands from being queued to the device.
2128 * @shost: host in question
2129 *
2130 * There is no timer nor any other means by which the requests get unblocked
2131 * other than the low-level driver calling scsi_unblock_requests().
2132 */
2133void scsi_block_requests(struct Scsi_Host *shost)
2134{
2135 shost->host_self_blocked = 1;
2136}
2137EXPORT_SYMBOL(scsi_block_requests);
2138
2139/**
2140 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2141 * further commands to be queued to the device.
2142 * @shost: host in question
2143 *
2144 * There is no timer nor any other means by which the requests get unblocked
2145 * other than the low-level driver calling scsi_unblock_requests(). This is done
2146 * as an API function so that changes to the internals of the scsi mid-layer
2147 * won't require wholesale changes to drivers that use this feature.
2148 */
2149void scsi_unblock_requests(struct Scsi_Host *shost)
2150{
2151 shost->host_self_blocked = 0;
2152 scsi_run_host_queues(shost);
2153}
2154EXPORT_SYMBOL(scsi_unblock_requests);
2155
2156void scsi_exit_queue(void)
2157{
2158 kmem_cache_destroy(scsi_sense_cache);
2159}
2160
2161/**
2162 * scsi_mode_select - issue a mode select
2163 * @sdev: SCSI device to be queried
2164 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2165 * @sp: Save page bit (0 == don't save, 1 == save)
2166 * @buffer: request buffer (may not be smaller than eight bytes)
2167 * @len: length of request buffer.
2168 * @timeout: command timeout
2169 * @retries: number of retries before failing
2170 * @data: returns a structure abstracting the mode header data
2171 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2172 * must be SCSI_SENSE_BUFFERSIZE big.
2173 *
2174 * Returns zero if successful; negative error number or scsi
2175 * status on error
2176 *
2177 */
2178int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2179 unsigned char *buffer, int len, int timeout, int retries,
2180 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2181{
2182 unsigned char cmd[10];
2183 unsigned char *real_buffer;
2184 const struct scsi_exec_args exec_args = {
2185 .sshdr = sshdr,
2186 };
2187 int ret;
2188
2189 memset(cmd, 0, sizeof(cmd));
2190 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2191
2192 /*
2193 * Use MODE SELECT(10) if the device asked for it or if the mode page
2194 * and the mode select header cannot fit within the maximumm 255 bytes
2195 * of the MODE SELECT(6) command.
2196 */
2197 if (sdev->use_10_for_ms ||
2198 len + 4 > 255 ||
2199 data->block_descriptor_length > 255) {
2200 if (len > 65535 - 8)
2201 return -EINVAL;
2202 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2203 if (!real_buffer)
2204 return -ENOMEM;
2205 memcpy(real_buffer + 8, buffer, len);
2206 len += 8;
2207 real_buffer[0] = 0;
2208 real_buffer[1] = 0;
2209 real_buffer[2] = data->medium_type;
2210 real_buffer[3] = data->device_specific;
2211 real_buffer[4] = data->longlba ? 0x01 : 0;
2212 real_buffer[5] = 0;
2213 put_unaligned_be16(data->block_descriptor_length,
2214 &real_buffer[6]);
2215
2216 cmd[0] = MODE_SELECT_10;
2217 put_unaligned_be16(len, &cmd[7]);
2218 } else {
2219 if (data->longlba)
2220 return -EINVAL;
2221
2222 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2223 if (!real_buffer)
2224 return -ENOMEM;
2225 memcpy(real_buffer + 4, buffer, len);
2226 len += 4;
2227 real_buffer[0] = 0;
2228 real_buffer[1] = data->medium_type;
2229 real_buffer[2] = data->device_specific;
2230 real_buffer[3] = data->block_descriptor_length;
2231
2232 cmd[0] = MODE_SELECT;
2233 cmd[4] = len;
2234 }
2235
2236 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2237 timeout, retries, &exec_args);
2238 kfree(real_buffer);
2239 return ret;
2240}
2241EXPORT_SYMBOL_GPL(scsi_mode_select);
2242
2243/**
2244 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2245 * @sdev: SCSI device to be queried
2246 * @dbd: set to prevent mode sense from returning block descriptors
2247 * @modepage: mode page being requested
2248 * @subpage: sub-page of the mode page being requested
2249 * @buffer: request buffer (may not be smaller than eight bytes)
2250 * @len: length of request buffer.
2251 * @timeout: command timeout
2252 * @retries: number of retries before failing
2253 * @data: returns a structure abstracting the mode header data
2254 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2255 * must be SCSI_SENSE_BUFFERSIZE big.
2256 *
2257 * Returns zero if successful, or a negative error number on failure
2258 */
2259int
2260scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2261 unsigned char *buffer, int len, int timeout, int retries,
2262 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2263{
2264 unsigned char cmd[12];
2265 int use_10_for_ms;
2266 int header_length;
2267 int result;
2268 struct scsi_sense_hdr my_sshdr;
2269 struct scsi_failure failure_defs[] = {
2270 {
2271 .sense = UNIT_ATTENTION,
2272 .asc = SCMD_FAILURE_ASC_ANY,
2273 .ascq = SCMD_FAILURE_ASCQ_ANY,
2274 .allowed = retries,
2275 .result = SAM_STAT_CHECK_CONDITION,
2276 },
2277 {}
2278 };
2279 struct scsi_failures failures = {
2280 .failure_definitions = failure_defs,
2281 };
2282 const struct scsi_exec_args exec_args = {
2283 /* caller might not be interested in sense, but we need it */
2284 .sshdr = sshdr ? : &my_sshdr,
2285 .failures = &failures,
2286 };
2287
2288 memset(data, 0, sizeof(*data));
2289 memset(&cmd[0], 0, 12);
2290
2291 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2292 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2293 cmd[2] = modepage;
2294 cmd[3] = subpage;
2295
2296 sshdr = exec_args.sshdr;
2297
2298 retry:
2299 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2300
2301 if (use_10_for_ms) {
2302 if (len < 8 || len > 65535)
2303 return -EINVAL;
2304
2305 cmd[0] = MODE_SENSE_10;
2306 put_unaligned_be16(len, &cmd[7]);
2307 header_length = 8;
2308 } else {
2309 if (len < 4)
2310 return -EINVAL;
2311
2312 cmd[0] = MODE_SENSE;
2313 cmd[4] = len;
2314 header_length = 4;
2315 }
2316
2317 memset(buffer, 0, len);
2318
2319 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2320 timeout, retries, &exec_args);
2321 if (result < 0)
2322 return result;
2323
2324 /* This code looks awful: what it's doing is making sure an
2325 * ILLEGAL REQUEST sense return identifies the actual command
2326 * byte as the problem. MODE_SENSE commands can return
2327 * ILLEGAL REQUEST if the code page isn't supported */
2328
2329 if (!scsi_status_is_good(result)) {
2330 if (scsi_sense_valid(sshdr)) {
2331 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2332 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2333 /*
2334 * Invalid command operation code: retry using
2335 * MODE SENSE(6) if this was a MODE SENSE(10)
2336 * request, except if the request mode page is
2337 * too large for MODE SENSE single byte
2338 * allocation length field.
2339 */
2340 if (use_10_for_ms) {
2341 if (len > 255)
2342 return -EIO;
2343 sdev->use_10_for_ms = 0;
2344 goto retry;
2345 }
2346 }
2347 }
2348 return -EIO;
2349 }
2350 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2351 (modepage == 6 || modepage == 8))) {
2352 /* Initio breakage? */
2353 header_length = 0;
2354 data->length = 13;
2355 data->medium_type = 0;
2356 data->device_specific = 0;
2357 data->longlba = 0;
2358 data->block_descriptor_length = 0;
2359 } else if (use_10_for_ms) {
2360 data->length = get_unaligned_be16(&buffer[0]) + 2;
2361 data->medium_type = buffer[2];
2362 data->device_specific = buffer[3];
2363 data->longlba = buffer[4] & 0x01;
2364 data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2365 } else {
2366 data->length = buffer[0] + 1;
2367 data->medium_type = buffer[1];
2368 data->device_specific = buffer[2];
2369 data->block_descriptor_length = buffer[3];
2370 }
2371 data->header_length = header_length;
2372
2373 return 0;
2374}
2375EXPORT_SYMBOL(scsi_mode_sense);
2376
2377/**
2378 * scsi_test_unit_ready - test if unit is ready
2379 * @sdev: scsi device to change the state of.
2380 * @timeout: command timeout
2381 * @retries: number of retries before failing
2382 * @sshdr: outpout pointer for decoded sense information.
2383 *
2384 * Returns zero if unsuccessful or an error if TUR failed. For
2385 * removable media, UNIT_ATTENTION sets ->changed flag.
2386 **/
2387int
2388scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2389 struct scsi_sense_hdr *sshdr)
2390{
2391 char cmd[] = {
2392 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2393 };
2394 const struct scsi_exec_args exec_args = {
2395 .sshdr = sshdr,
2396 };
2397 int result;
2398
2399 /* try to eat the UNIT_ATTENTION if there are enough retries */
2400 do {
2401 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2402 timeout, 1, &exec_args);
2403 if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2404 sshdr->sense_key == UNIT_ATTENTION)
2405 sdev->changed = 1;
2406 } while (result > 0 && scsi_sense_valid(sshdr) &&
2407 sshdr->sense_key == UNIT_ATTENTION && --retries);
2408
2409 return result;
2410}
2411EXPORT_SYMBOL(scsi_test_unit_ready);
2412
2413/**
2414 * scsi_device_set_state - Take the given device through the device state model.
2415 * @sdev: scsi device to change the state of.
2416 * @state: state to change to.
2417 *
2418 * Returns zero if successful or an error if the requested
2419 * transition is illegal.
2420 */
2421int
2422scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2423{
2424 enum scsi_device_state oldstate = sdev->sdev_state;
2425
2426 if (state == oldstate)
2427 return 0;
2428
2429 switch (state) {
2430 case SDEV_CREATED:
2431 switch (oldstate) {
2432 case SDEV_CREATED_BLOCK:
2433 break;
2434 default:
2435 goto illegal;
2436 }
2437 break;
2438
2439 case SDEV_RUNNING:
2440 switch (oldstate) {
2441 case SDEV_CREATED:
2442 case SDEV_OFFLINE:
2443 case SDEV_TRANSPORT_OFFLINE:
2444 case SDEV_QUIESCE:
2445 case SDEV_BLOCK:
2446 break;
2447 default:
2448 goto illegal;
2449 }
2450 break;
2451
2452 case SDEV_QUIESCE:
2453 switch (oldstate) {
2454 case SDEV_RUNNING:
2455 case SDEV_OFFLINE:
2456 case SDEV_TRANSPORT_OFFLINE:
2457 break;
2458 default:
2459 goto illegal;
2460 }
2461 break;
2462
2463 case SDEV_OFFLINE:
2464 case SDEV_TRANSPORT_OFFLINE:
2465 switch (oldstate) {
2466 case SDEV_CREATED:
2467 case SDEV_RUNNING:
2468 case SDEV_QUIESCE:
2469 case SDEV_BLOCK:
2470 break;
2471 default:
2472 goto illegal;
2473 }
2474 break;
2475
2476 case SDEV_BLOCK:
2477 switch (oldstate) {
2478 case SDEV_RUNNING:
2479 case SDEV_CREATED_BLOCK:
2480 case SDEV_QUIESCE:
2481 case SDEV_OFFLINE:
2482 break;
2483 default:
2484 goto illegal;
2485 }
2486 break;
2487
2488 case SDEV_CREATED_BLOCK:
2489 switch (oldstate) {
2490 case SDEV_CREATED:
2491 break;
2492 default:
2493 goto illegal;
2494 }
2495 break;
2496
2497 case SDEV_CANCEL:
2498 switch (oldstate) {
2499 case SDEV_CREATED:
2500 case SDEV_RUNNING:
2501 case SDEV_QUIESCE:
2502 case SDEV_OFFLINE:
2503 case SDEV_TRANSPORT_OFFLINE:
2504 break;
2505 default:
2506 goto illegal;
2507 }
2508 break;
2509
2510 case SDEV_DEL:
2511 switch (oldstate) {
2512 case SDEV_CREATED:
2513 case SDEV_RUNNING:
2514 case SDEV_OFFLINE:
2515 case SDEV_TRANSPORT_OFFLINE:
2516 case SDEV_CANCEL:
2517 case SDEV_BLOCK:
2518 case SDEV_CREATED_BLOCK:
2519 break;
2520 default:
2521 goto illegal;
2522 }
2523 break;
2524
2525 }
2526 sdev->offline_already = false;
2527 sdev->sdev_state = state;
2528 return 0;
2529
2530 illegal:
2531 SCSI_LOG_ERROR_RECOVERY(1,
2532 sdev_printk(KERN_ERR, sdev,
2533 "Illegal state transition %s->%s",
2534 scsi_device_state_name(oldstate),
2535 scsi_device_state_name(state))
2536 );
2537 return -EINVAL;
2538}
2539EXPORT_SYMBOL(scsi_device_set_state);
2540
2541/**
2542 * scsi_evt_emit - emit a single SCSI device uevent
2543 * @sdev: associated SCSI device
2544 * @evt: event to emit
2545 *
2546 * Send a single uevent (scsi_event) to the associated scsi_device.
2547 */
2548static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2549{
2550 int idx = 0;
2551 char *envp[3];
2552
2553 switch (evt->evt_type) {
2554 case SDEV_EVT_MEDIA_CHANGE:
2555 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2556 break;
2557 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2558 scsi_rescan_device(sdev);
2559 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2560 break;
2561 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2562 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2563 break;
2564 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2565 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2566 break;
2567 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2568 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2569 break;
2570 case SDEV_EVT_LUN_CHANGE_REPORTED:
2571 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2572 break;
2573 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2574 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2575 break;
2576 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2577 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2578 break;
2579 default:
2580 /* do nothing */
2581 break;
2582 }
2583
2584 envp[idx++] = NULL;
2585
2586 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2587}
2588
2589/**
2590 * scsi_evt_thread - send a uevent for each scsi event
2591 * @work: work struct for scsi_device
2592 *
2593 * Dispatch queued events to their associated scsi_device kobjects
2594 * as uevents.
2595 */
2596void scsi_evt_thread(struct work_struct *work)
2597{
2598 struct scsi_device *sdev;
2599 enum scsi_device_event evt_type;
2600 LIST_HEAD(event_list);
2601
2602 sdev = container_of(work, struct scsi_device, event_work);
2603
2604 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2605 if (test_and_clear_bit(evt_type, sdev->pending_events))
2606 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2607
2608 while (1) {
2609 struct scsi_event *evt;
2610 struct list_head *this, *tmp;
2611 unsigned long flags;
2612
2613 spin_lock_irqsave(&sdev->list_lock, flags);
2614 list_splice_init(&sdev->event_list, &event_list);
2615 spin_unlock_irqrestore(&sdev->list_lock, flags);
2616
2617 if (list_empty(&event_list))
2618 break;
2619
2620 list_for_each_safe(this, tmp, &event_list) {
2621 evt = list_entry(this, struct scsi_event, node);
2622 list_del(&evt->node);
2623 scsi_evt_emit(sdev, evt);
2624 kfree(evt);
2625 }
2626 }
2627}
2628
2629/**
2630 * sdev_evt_send - send asserted event to uevent thread
2631 * @sdev: scsi_device event occurred on
2632 * @evt: event to send
2633 *
2634 * Assert scsi device event asynchronously.
2635 */
2636void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2637{
2638 unsigned long flags;
2639
2640#if 0
2641 /* FIXME: currently this check eliminates all media change events
2642 * for polled devices. Need to update to discriminate between AN
2643 * and polled events */
2644 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2645 kfree(evt);
2646 return;
2647 }
2648#endif
2649
2650 spin_lock_irqsave(&sdev->list_lock, flags);
2651 list_add_tail(&evt->node, &sdev->event_list);
2652 schedule_work(&sdev->event_work);
2653 spin_unlock_irqrestore(&sdev->list_lock, flags);
2654}
2655EXPORT_SYMBOL_GPL(sdev_evt_send);
2656
2657/**
2658 * sdev_evt_alloc - allocate a new scsi event
2659 * @evt_type: type of event to allocate
2660 * @gfpflags: GFP flags for allocation
2661 *
2662 * Allocates and returns a new scsi_event.
2663 */
2664struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2665 gfp_t gfpflags)
2666{
2667 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2668 if (!evt)
2669 return NULL;
2670
2671 evt->evt_type = evt_type;
2672 INIT_LIST_HEAD(&evt->node);
2673
2674 /* evt_type-specific initialization, if any */
2675 switch (evt_type) {
2676 case SDEV_EVT_MEDIA_CHANGE:
2677 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2678 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2679 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2680 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2681 case SDEV_EVT_LUN_CHANGE_REPORTED:
2682 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2683 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2684 default:
2685 /* do nothing */
2686 break;
2687 }
2688
2689 return evt;
2690}
2691EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2692
2693/**
2694 * sdev_evt_send_simple - send asserted event to uevent thread
2695 * @sdev: scsi_device event occurred on
2696 * @evt_type: type of event to send
2697 * @gfpflags: GFP flags for allocation
2698 *
2699 * Assert scsi device event asynchronously, given an event type.
2700 */
2701void sdev_evt_send_simple(struct scsi_device *sdev,
2702 enum scsi_device_event evt_type, gfp_t gfpflags)
2703{
2704 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2705 if (!evt) {
2706 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2707 evt_type);
2708 return;
2709 }
2710
2711 sdev_evt_send(sdev, evt);
2712}
2713EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2714
2715/**
2716 * scsi_device_quiesce - Block all commands except power management.
2717 * @sdev: scsi device to quiesce.
2718 *
2719 * This works by trying to transition to the SDEV_QUIESCE state
2720 * (which must be a legal transition). When the device is in this
2721 * state, only power management requests will be accepted, all others will
2722 * be deferred.
2723 *
2724 * Must be called with user context, may sleep.
2725 *
2726 * Returns zero if unsuccessful or an error if not.
2727 */
2728int
2729scsi_device_quiesce(struct scsi_device *sdev)
2730{
2731 struct request_queue *q = sdev->request_queue;
2732 int err;
2733
2734 /*
2735 * It is allowed to call scsi_device_quiesce() multiple times from
2736 * the same context but concurrent scsi_device_quiesce() calls are
2737 * not allowed.
2738 */
2739 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2740
2741 if (sdev->quiesced_by == current)
2742 return 0;
2743
2744 blk_set_pm_only(q);
2745
2746 blk_mq_freeze_queue(q);
2747 /*
2748 * Ensure that the effect of blk_set_pm_only() will be visible
2749 * for percpu_ref_tryget() callers that occur after the queue
2750 * unfreeze even if the queue was already frozen before this function
2751 * was called. See also https://lwn.net/Articles/573497/.
2752 */
2753 synchronize_rcu();
2754 blk_mq_unfreeze_queue(q);
2755
2756 mutex_lock(&sdev->state_mutex);
2757 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2758 if (err == 0)
2759 sdev->quiesced_by = current;
2760 else
2761 blk_clear_pm_only(q);
2762 mutex_unlock(&sdev->state_mutex);
2763
2764 return err;
2765}
2766EXPORT_SYMBOL(scsi_device_quiesce);
2767
2768/**
2769 * scsi_device_resume - Restart user issued commands to a quiesced device.
2770 * @sdev: scsi device to resume.
2771 *
2772 * Moves the device from quiesced back to running and restarts the
2773 * queues.
2774 *
2775 * Must be called with user context, may sleep.
2776 */
2777void scsi_device_resume(struct scsi_device *sdev)
2778{
2779 /* check if the device state was mutated prior to resume, and if
2780 * so assume the state is being managed elsewhere (for example
2781 * device deleted during suspend)
2782 */
2783 mutex_lock(&sdev->state_mutex);
2784 if (sdev->sdev_state == SDEV_QUIESCE)
2785 scsi_device_set_state(sdev, SDEV_RUNNING);
2786 if (sdev->quiesced_by) {
2787 sdev->quiesced_by = NULL;
2788 blk_clear_pm_only(sdev->request_queue);
2789 }
2790 mutex_unlock(&sdev->state_mutex);
2791}
2792EXPORT_SYMBOL(scsi_device_resume);
2793
2794static void
2795device_quiesce_fn(struct scsi_device *sdev, void *data)
2796{
2797 scsi_device_quiesce(sdev);
2798}
2799
2800void
2801scsi_target_quiesce(struct scsi_target *starget)
2802{
2803 starget_for_each_device(starget, NULL, device_quiesce_fn);
2804}
2805EXPORT_SYMBOL(scsi_target_quiesce);
2806
2807static void
2808device_resume_fn(struct scsi_device *sdev, void *data)
2809{
2810 scsi_device_resume(sdev);
2811}
2812
2813void
2814scsi_target_resume(struct scsi_target *starget)
2815{
2816 starget_for_each_device(starget, NULL, device_resume_fn);
2817}
2818EXPORT_SYMBOL(scsi_target_resume);
2819
2820static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2821{
2822 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2823 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2824
2825 return 0;
2826}
2827
2828void scsi_start_queue(struct scsi_device *sdev)
2829{
2830 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2831 blk_mq_unquiesce_queue(sdev->request_queue);
2832}
2833
2834static void scsi_stop_queue(struct scsi_device *sdev)
2835{
2836 /*
2837 * The atomic variable of ->queue_stopped covers that
2838 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2839 *
2840 * The caller needs to wait until quiesce is done.
2841 */
2842 if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2843 blk_mq_quiesce_queue_nowait(sdev->request_queue);
2844}
2845
2846/**
2847 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2848 * @sdev: device to block
2849 *
2850 * Pause SCSI command processing on the specified device. Does not sleep.
2851 *
2852 * Returns zero if successful or a negative error code upon failure.
2853 *
2854 * Notes:
2855 * This routine transitions the device to the SDEV_BLOCK state (which must be
2856 * a legal transition). When the device is in this state, command processing
2857 * is paused until the device leaves the SDEV_BLOCK state. See also
2858 * scsi_internal_device_unblock_nowait().
2859 */
2860int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2861{
2862 int ret = __scsi_internal_device_block_nowait(sdev);
2863
2864 /*
2865 * The device has transitioned to SDEV_BLOCK. Stop the
2866 * block layer from calling the midlayer with this device's
2867 * request queue.
2868 */
2869 if (!ret)
2870 scsi_stop_queue(sdev);
2871 return ret;
2872}
2873EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2874
2875/**
2876 * scsi_device_block - try to transition to the SDEV_BLOCK state
2877 * @sdev: device to block
2878 * @data: dummy argument, ignored
2879 *
2880 * Pause SCSI command processing on the specified device. Callers must wait
2881 * until all ongoing scsi_queue_rq() calls have finished after this function
2882 * returns.
2883 *
2884 * Note:
2885 * This routine transitions the device to the SDEV_BLOCK state (which must be
2886 * a legal transition). When the device is in this state, command processing
2887 * is paused until the device leaves the SDEV_BLOCK state. See also
2888 * scsi_internal_device_unblock().
2889 */
2890static void scsi_device_block(struct scsi_device *sdev, void *data)
2891{
2892 int err;
2893 enum scsi_device_state state;
2894
2895 mutex_lock(&sdev->state_mutex);
2896 err = __scsi_internal_device_block_nowait(sdev);
2897 state = sdev->sdev_state;
2898 if (err == 0)
2899 /*
2900 * scsi_stop_queue() must be called with the state_mutex
2901 * held. Otherwise a simultaneous scsi_start_queue() call
2902 * might unquiesce the queue before we quiesce it.
2903 */
2904 scsi_stop_queue(sdev);
2905
2906 mutex_unlock(&sdev->state_mutex);
2907
2908 WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2909 __func__, dev_name(&sdev->sdev_gendev), state);
2910}
2911
2912/**
2913 * scsi_internal_device_unblock_nowait - resume a device after a block request
2914 * @sdev: device to resume
2915 * @new_state: state to set the device to after unblocking
2916 *
2917 * Restart the device queue for a previously suspended SCSI device. Does not
2918 * sleep.
2919 *
2920 * Returns zero if successful or a negative error code upon failure.
2921 *
2922 * Notes:
2923 * This routine transitions the device to the SDEV_RUNNING state or to one of
2924 * the offline states (which must be a legal transition) allowing the midlayer
2925 * to goose the queue for this device.
2926 */
2927int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2928 enum scsi_device_state new_state)
2929{
2930 switch (new_state) {
2931 case SDEV_RUNNING:
2932 case SDEV_TRANSPORT_OFFLINE:
2933 break;
2934 default:
2935 return -EINVAL;
2936 }
2937
2938 /*
2939 * Try to transition the scsi device to SDEV_RUNNING or one of the
2940 * offlined states and goose the device queue if successful.
2941 */
2942 switch (sdev->sdev_state) {
2943 case SDEV_BLOCK:
2944 case SDEV_TRANSPORT_OFFLINE:
2945 sdev->sdev_state = new_state;
2946 break;
2947 case SDEV_CREATED_BLOCK:
2948 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2949 new_state == SDEV_OFFLINE)
2950 sdev->sdev_state = new_state;
2951 else
2952 sdev->sdev_state = SDEV_CREATED;
2953 break;
2954 case SDEV_CANCEL:
2955 case SDEV_OFFLINE:
2956 break;
2957 default:
2958 return -EINVAL;
2959 }
2960 scsi_start_queue(sdev);
2961
2962 return 0;
2963}
2964EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2965
2966/**
2967 * scsi_internal_device_unblock - resume a device after a block request
2968 * @sdev: device to resume
2969 * @new_state: state to set the device to after unblocking
2970 *
2971 * Restart the device queue for a previously suspended SCSI device. May sleep.
2972 *
2973 * Returns zero if successful or a negative error code upon failure.
2974 *
2975 * Notes:
2976 * This routine transitions the device to the SDEV_RUNNING state or to one of
2977 * the offline states (which must be a legal transition) allowing the midlayer
2978 * to goose the queue for this device.
2979 */
2980static int scsi_internal_device_unblock(struct scsi_device *sdev,
2981 enum scsi_device_state new_state)
2982{
2983 int ret;
2984
2985 mutex_lock(&sdev->state_mutex);
2986 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2987 mutex_unlock(&sdev->state_mutex);
2988
2989 return ret;
2990}
2991
2992static int
2993target_block(struct device *dev, void *data)
2994{
2995 if (scsi_is_target_device(dev))
2996 starget_for_each_device(to_scsi_target(dev), NULL,
2997 scsi_device_block);
2998 return 0;
2999}
3000
3001/**
3002 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3003 * @dev: a parent device of one or more scsi_target devices
3004 * @shost: the Scsi_Host to which this device belongs
3005 *
3006 * Iterate over all children of @dev, which should be scsi_target devices,
3007 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3008 * ongoing scsi_queue_rq() calls to finish. May sleep.
3009 *
3010 * Note:
3011 * @dev must not itself be a scsi_target device.
3012 */
3013void
3014scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3015{
3016 WARN_ON_ONCE(scsi_is_target_device(dev));
3017 device_for_each_child(dev, NULL, target_block);
3018 blk_mq_wait_quiesce_done(&shost->tag_set);
3019}
3020EXPORT_SYMBOL_GPL(scsi_block_targets);
3021
3022static void
3023device_unblock(struct scsi_device *sdev, void *data)
3024{
3025 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3026}
3027
3028static int
3029target_unblock(struct device *dev, void *data)
3030{
3031 if (scsi_is_target_device(dev))
3032 starget_for_each_device(to_scsi_target(dev), data,
3033 device_unblock);
3034 return 0;
3035}
3036
3037void
3038scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3039{
3040 if (scsi_is_target_device(dev))
3041 starget_for_each_device(to_scsi_target(dev), &new_state,
3042 device_unblock);
3043 else
3044 device_for_each_child(dev, &new_state, target_unblock);
3045}
3046EXPORT_SYMBOL_GPL(scsi_target_unblock);
3047
3048/**
3049 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3050 * @shost: device to block
3051 *
3052 * Pause SCSI command processing for all logical units associated with the SCSI
3053 * host and wait until pending scsi_queue_rq() calls have finished.
3054 *
3055 * Returns zero if successful or a negative error code upon failure.
3056 */
3057int
3058scsi_host_block(struct Scsi_Host *shost)
3059{
3060 struct scsi_device *sdev;
3061 int ret;
3062
3063 /*
3064 * Call scsi_internal_device_block_nowait so we can avoid
3065 * calling synchronize_rcu() for each LUN.
3066 */
3067 shost_for_each_device(sdev, shost) {
3068 mutex_lock(&sdev->state_mutex);
3069 ret = scsi_internal_device_block_nowait(sdev);
3070 mutex_unlock(&sdev->state_mutex);
3071 if (ret) {
3072 scsi_device_put(sdev);
3073 return ret;
3074 }
3075 }
3076
3077 /* Wait for ongoing scsi_queue_rq() calls to finish. */
3078 blk_mq_wait_quiesce_done(&shost->tag_set);
3079
3080 return 0;
3081}
3082EXPORT_SYMBOL_GPL(scsi_host_block);
3083
3084int
3085scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3086{
3087 struct scsi_device *sdev;
3088 int ret = 0;
3089
3090 shost_for_each_device(sdev, shost) {
3091 ret = scsi_internal_device_unblock(sdev, new_state);
3092 if (ret) {
3093 scsi_device_put(sdev);
3094 break;
3095 }
3096 }
3097 return ret;
3098}
3099EXPORT_SYMBOL_GPL(scsi_host_unblock);
3100
3101/**
3102 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3103 * @sgl: scatter-gather list
3104 * @sg_count: number of segments in sg
3105 * @offset: offset in bytes into sg, on return offset into the mapped area
3106 * @len: bytes to map, on return number of bytes mapped
3107 *
3108 * Returns virtual address of the start of the mapped page
3109 */
3110void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3111 size_t *offset, size_t *len)
3112{
3113 int i;
3114 size_t sg_len = 0, len_complete = 0;
3115 struct scatterlist *sg;
3116 struct page *page;
3117
3118 WARN_ON(!irqs_disabled());
3119
3120 for_each_sg(sgl, sg, sg_count, i) {
3121 len_complete = sg_len; /* Complete sg-entries */
3122 sg_len += sg->length;
3123 if (sg_len > *offset)
3124 break;
3125 }
3126
3127 if (unlikely(i == sg_count)) {
3128 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3129 "elements %d\n",
3130 __func__, sg_len, *offset, sg_count);
3131 WARN_ON(1);
3132 return NULL;
3133 }
3134
3135 /* Offset starting from the beginning of first page in this sg-entry */
3136 *offset = *offset - len_complete + sg->offset;
3137
3138 /* Assumption: contiguous pages can be accessed as "page + i" */
3139 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3140 *offset &= ~PAGE_MASK;
3141
3142 /* Bytes in this sg-entry from *offset to the end of the page */
3143 sg_len = PAGE_SIZE - *offset;
3144 if (*len > sg_len)
3145 *len = sg_len;
3146
3147 return kmap_atomic(page);
3148}
3149EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3150
3151/**
3152 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3153 * @virt: virtual address to be unmapped
3154 */
3155void scsi_kunmap_atomic_sg(void *virt)
3156{
3157 kunmap_atomic(virt);
3158}
3159EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3160
3161void sdev_disable_disk_events(struct scsi_device *sdev)
3162{
3163 atomic_inc(&sdev->disk_events_disable_depth);
3164}
3165EXPORT_SYMBOL(sdev_disable_disk_events);
3166
3167void sdev_enable_disk_events(struct scsi_device *sdev)
3168{
3169 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3170 return;
3171 atomic_dec(&sdev->disk_events_disable_depth);
3172}
3173EXPORT_SYMBOL(sdev_enable_disk_events);
3174
3175static unsigned char designator_prio(const unsigned char *d)
3176{
3177 if (d[1] & 0x30)
3178 /* not associated with LUN */
3179 return 0;
3180
3181 if (d[3] == 0)
3182 /* invalid length */
3183 return 0;
3184
3185 /*
3186 * Order of preference for lun descriptor:
3187 * - SCSI name string
3188 * - NAA IEEE Registered Extended
3189 * - EUI-64 based 16-byte
3190 * - EUI-64 based 12-byte
3191 * - NAA IEEE Registered
3192 * - NAA IEEE Extended
3193 * - EUI-64 based 8-byte
3194 * - SCSI name string (truncated)
3195 * - T10 Vendor ID
3196 * as longer descriptors reduce the likelyhood
3197 * of identification clashes.
3198 */
3199
3200 switch (d[1] & 0xf) {
3201 case 8:
3202 /* SCSI name string, variable-length UTF-8 */
3203 return 9;
3204 case 3:
3205 switch (d[4] >> 4) {
3206 case 6:
3207 /* NAA registered extended */
3208 return 8;
3209 case 5:
3210 /* NAA registered */
3211 return 5;
3212 case 4:
3213 /* NAA extended */
3214 return 4;
3215 case 3:
3216 /* NAA locally assigned */
3217 return 1;
3218 default:
3219 break;
3220 }
3221 break;
3222 case 2:
3223 switch (d[3]) {
3224 case 16:
3225 /* EUI64-based, 16 byte */
3226 return 7;
3227 case 12:
3228 /* EUI64-based, 12 byte */
3229 return 6;
3230 case 8:
3231 /* EUI64-based, 8 byte */
3232 return 3;
3233 default:
3234 break;
3235 }
3236 break;
3237 case 1:
3238 /* T10 vendor ID */
3239 return 1;
3240 default:
3241 break;
3242 }
3243
3244 return 0;
3245}
3246
3247/**
3248 * scsi_vpd_lun_id - return a unique device identification
3249 * @sdev: SCSI device
3250 * @id: buffer for the identification
3251 * @id_len: length of the buffer
3252 *
3253 * Copies a unique device identification into @id based
3254 * on the information in the VPD page 0x83 of the device.
3255 * The string will be formatted as a SCSI name string.
3256 *
3257 * Returns the length of the identification or error on failure.
3258 * If the identifier is longer than the supplied buffer the actual
3259 * identifier length is returned and the buffer is not zero-padded.
3260 */
3261int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3262{
3263 u8 cur_id_prio = 0;
3264 u8 cur_id_size = 0;
3265 const unsigned char *d, *cur_id_str;
3266 const struct scsi_vpd *vpd_pg83;
3267 int id_size = -EINVAL;
3268
3269 rcu_read_lock();
3270 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3271 if (!vpd_pg83) {
3272 rcu_read_unlock();
3273 return -ENXIO;
3274 }
3275
3276 /* The id string must be at least 20 bytes + terminating NULL byte */
3277 if (id_len < 21) {
3278 rcu_read_unlock();
3279 return -EINVAL;
3280 }
3281
3282 memset(id, 0, id_len);
3283 for (d = vpd_pg83->data + 4;
3284 d < vpd_pg83->data + vpd_pg83->len;
3285 d += d[3] + 4) {
3286 u8 prio = designator_prio(d);
3287
3288 if (prio == 0 || cur_id_prio > prio)
3289 continue;
3290
3291 switch (d[1] & 0xf) {
3292 case 0x1:
3293 /* T10 Vendor ID */
3294 if (cur_id_size > d[3])
3295 break;
3296 cur_id_prio = prio;
3297 cur_id_size = d[3];
3298 if (cur_id_size + 4 > id_len)
3299 cur_id_size = id_len - 4;
3300 cur_id_str = d + 4;
3301 id_size = snprintf(id, id_len, "t10.%*pE",
3302 cur_id_size, cur_id_str);
3303 break;
3304 case 0x2:
3305 /* EUI-64 */
3306 cur_id_prio = prio;
3307 cur_id_size = d[3];
3308 cur_id_str = d + 4;
3309 switch (cur_id_size) {
3310 case 8:
3311 id_size = snprintf(id, id_len,
3312 "eui.%8phN",
3313 cur_id_str);
3314 break;
3315 case 12:
3316 id_size = snprintf(id, id_len,
3317 "eui.%12phN",
3318 cur_id_str);
3319 break;
3320 case 16:
3321 id_size = snprintf(id, id_len,
3322 "eui.%16phN",
3323 cur_id_str);
3324 break;
3325 default:
3326 break;
3327 }
3328 break;
3329 case 0x3:
3330 /* NAA */
3331 cur_id_prio = prio;
3332 cur_id_size = d[3];
3333 cur_id_str = d + 4;
3334 switch (cur_id_size) {
3335 case 8:
3336 id_size = snprintf(id, id_len,
3337 "naa.%8phN",
3338 cur_id_str);
3339 break;
3340 case 16:
3341 id_size = snprintf(id, id_len,
3342 "naa.%16phN",
3343 cur_id_str);
3344 break;
3345 default:
3346 break;
3347 }
3348 break;
3349 case 0x8:
3350 /* SCSI name string */
3351 if (cur_id_size > d[3])
3352 break;
3353 /* Prefer others for truncated descriptor */
3354 if (d[3] > id_len) {
3355 prio = 2;
3356 if (cur_id_prio > prio)
3357 break;
3358 }
3359 cur_id_prio = prio;
3360 cur_id_size = id_size = d[3];
3361 cur_id_str = d + 4;
3362 if (cur_id_size >= id_len)
3363 cur_id_size = id_len - 1;
3364 memcpy(id, cur_id_str, cur_id_size);
3365 break;
3366 default:
3367 break;
3368 }
3369 }
3370 rcu_read_unlock();
3371
3372 return id_size;
3373}
3374EXPORT_SYMBOL(scsi_vpd_lun_id);
3375
3376/*
3377 * scsi_vpd_tpg_id - return a target port group identifier
3378 * @sdev: SCSI device
3379 *
3380 * Returns the Target Port Group identifier from the information
3381 * froom VPD page 0x83 of the device.
3382 *
3383 * Returns the identifier or error on failure.
3384 */
3385int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3386{
3387 const unsigned char *d;
3388 const struct scsi_vpd *vpd_pg83;
3389 int group_id = -EAGAIN, rel_port = -1;
3390
3391 rcu_read_lock();
3392 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3393 if (!vpd_pg83) {
3394 rcu_read_unlock();
3395 return -ENXIO;
3396 }
3397
3398 d = vpd_pg83->data + 4;
3399 while (d < vpd_pg83->data + vpd_pg83->len) {
3400 switch (d[1] & 0xf) {
3401 case 0x4:
3402 /* Relative target port */
3403 rel_port = get_unaligned_be16(&d[6]);
3404 break;
3405 case 0x5:
3406 /* Target port group */
3407 group_id = get_unaligned_be16(&d[6]);
3408 break;
3409 default:
3410 break;
3411 }
3412 d += d[3] + 4;
3413 }
3414 rcu_read_unlock();
3415
3416 if (group_id >= 0 && rel_id && rel_port != -1)
3417 *rel_id = rel_port;
3418
3419 return group_id;
3420}
3421EXPORT_SYMBOL(scsi_vpd_tpg_id);
3422
3423/**
3424 * scsi_build_sense - build sense data for a command
3425 * @scmd: scsi command for which the sense should be formatted
3426 * @desc: Sense format (non-zero == descriptor format,
3427 * 0 == fixed format)
3428 * @key: Sense key
3429 * @asc: Additional sense code
3430 * @ascq: Additional sense code qualifier
3431 *
3432 **/
3433void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3434{
3435 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3436 scmd->result = SAM_STAT_CHECK_CONDITION;
3437}
3438EXPORT_SYMBOL_GPL(scsi_build_sense);
3439
3440#ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3441#include "scsi_lib_test.c"
3442#endif