]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/scsi/scsi_lib.c
block: introduce blk_rq_is_passthrough
[mirror_ubuntu-artful-kernel.git] / drivers / scsi / scsi_lib.c
CommitLineData
1da177e4 1/*
d285203c
CH
2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
1da177e4
LT
4 *
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
9 */
10
11#include <linux/bio.h>
d3f46f39 12#include <linux/bitops.h>
1da177e4
LT
13#include <linux/blkdev.h>
14#include <linux/completion.h>
15#include <linux/kernel.h>
09703660 16#include <linux/export.h>
1da177e4
LT
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
faead26d 20#include <linux/hardirq.h>
c6132da1 21#include <linux/scatterlist.h>
d285203c 22#include <linux/blk-mq.h>
f1569ff1 23#include <linux/ratelimit.h>
a8aa3978 24#include <asm/unaligned.h>
1da177e4
LT
25
26#include <scsi/scsi.h>
beb40487 27#include <scsi/scsi_cmnd.h>
1da177e4
LT
28#include <scsi/scsi_dbg.h>
29#include <scsi/scsi_device.h>
30#include <scsi/scsi_driver.h>
31#include <scsi/scsi_eh.h>
32#include <scsi/scsi_host.h>
ee14c674 33#include <scsi/scsi_dh.h>
1da177e4 34
3b5382c4
CH
35#include <trace/events/scsi.h>
36
1da177e4
LT
37#include "scsi_priv.h"
38#include "scsi_logging.h"
39
e9c787e6 40static struct kmem_cache *scsi_sdb_cache;
0a6ac4ee
CH
41static struct kmem_cache *scsi_sense_cache;
42static struct kmem_cache *scsi_sense_isadma_cache;
43static DEFINE_MUTEX(scsi_sense_cache_mutex);
44
45static inline struct kmem_cache *
46scsi_select_sense_cache(struct Scsi_Host *shost)
47{
48 return shost->unchecked_isa_dma ?
49 scsi_sense_isadma_cache : scsi_sense_cache;
50}
51
e9c787e6 52static void scsi_free_sense_buffer(struct Scsi_Host *shost,
0a6ac4ee
CH
53 unsigned char *sense_buffer)
54{
55 kmem_cache_free(scsi_select_sense_cache(shost), sense_buffer);
56}
57
e9c787e6
CH
58static unsigned char *scsi_alloc_sense_buffer(struct Scsi_Host *shost,
59 gfp_t gfp_mask, int numa_node)
0a6ac4ee
CH
60{
61 return kmem_cache_alloc_node(scsi_select_sense_cache(shost), gfp_mask,
62 numa_node);
63}
64
65int scsi_init_sense_cache(struct Scsi_Host *shost)
66{
67 struct kmem_cache *cache;
68 int ret = 0;
69
70 cache = scsi_select_sense_cache(shost);
71 if (cache)
72 return 0;
73
74 mutex_lock(&scsi_sense_cache_mutex);
75 if (shost->unchecked_isa_dma) {
76 scsi_sense_isadma_cache =
77 kmem_cache_create("scsi_sense_cache(DMA)",
78 SCSI_SENSE_BUFFERSIZE, 0,
79 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
80 if (!scsi_sense_isadma_cache)
81 ret = -ENOMEM;
82 } else {
83 scsi_sense_cache =
84 kmem_cache_create("scsi_sense_cache",
85 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN, NULL);
86 if (!scsi_sense_cache)
87 ret = -ENOMEM;
88 }
89
90 mutex_unlock(&scsi_sense_cache_mutex);
91 return ret;
92}
6f9a35e2 93
a488e749
JA
94/*
95 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
96 * not change behaviour from the previous unplug mechanism, experimentation
97 * may prove this needs changing.
98 */
99#define SCSI_QUEUE_DELAY 3
100
de3e8bf3
CH
101static void
102scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
1da177e4
LT
103{
104 struct Scsi_Host *host = cmd->device->host;
105 struct scsi_device *device = cmd->device;
f0c0a376 106 struct scsi_target *starget = scsi_target(device);
1da177e4
LT
107
108 /*
d8c37e7b 109 * Set the appropriate busy bit for the device/host.
1da177e4
LT
110 *
111 * If the host/device isn't busy, assume that something actually
112 * completed, and that we should be able to queue a command now.
113 *
114 * Note that the prior mid-layer assumption that any host could
115 * always queue at least one command is now broken. The mid-layer
116 * will implement a user specifiable stall (see
117 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
118 * if a command is requeued with no other commands outstanding
119 * either for the device or for the host.
120 */
f0c0a376
MC
121 switch (reason) {
122 case SCSI_MLQUEUE_HOST_BUSY:
cd9070c9 123 atomic_set(&host->host_blocked, host->max_host_blocked);
f0c0a376
MC
124 break;
125 case SCSI_MLQUEUE_DEVICE_BUSY:
573e5913 126 case SCSI_MLQUEUE_EH_RETRY:
cd9070c9
CH
127 atomic_set(&device->device_blocked,
128 device->max_device_blocked);
f0c0a376
MC
129 break;
130 case SCSI_MLQUEUE_TARGET_BUSY:
cd9070c9
CH
131 atomic_set(&starget->target_blocked,
132 starget->max_target_blocked);
f0c0a376
MC
133 break;
134 }
de3e8bf3
CH
135}
136
d285203c
CH
137static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
138{
139 struct scsi_device *sdev = cmd->device;
d285203c 140
2b053aca 141 blk_mq_requeue_request(cmd->request, true);
d285203c
CH
142 put_device(&sdev->sdev_gendev);
143}
144
de3e8bf3
CH
145/**
146 * __scsi_queue_insert - private queue insertion
147 * @cmd: The SCSI command being requeued
148 * @reason: The reason for the requeue
149 * @unbusy: Whether the queue should be unbusied
150 *
151 * This is a private queue insertion. The public interface
152 * scsi_queue_insert() always assumes the queue should be unbusied
153 * because it's always called before the completion. This function is
154 * for a requeue after completion, which should only occur in this
155 * file.
156 */
157static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
158{
159 struct scsi_device *device = cmd->device;
160 struct request_queue *q = device->request_queue;
161 unsigned long flags;
162
163 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
164 "Inserting command %p into mlqueue\n", cmd));
165
166 scsi_set_blocked(cmd, reason);
1da177e4 167
1da177e4
LT
168 /*
169 * Decrement the counters, since these commands are no longer
170 * active on the host/device.
171 */
4f5299ac
JB
172 if (unbusy)
173 scsi_device_unbusy(device);
1da177e4
LT
174
175 /*
a1bf9d1d 176 * Requeue this command. It will go before all other commands
b485462a
BVA
177 * that are already in the queue. Schedule requeue work under
178 * lock such that the kblockd_schedule_work() call happens
179 * before blk_cleanup_queue() finishes.
a488e749 180 */
644373a4 181 cmd->result = 0;
d285203c
CH
182 if (q->mq_ops) {
183 scsi_mq_requeue_cmd(cmd);
184 return;
185 }
a1bf9d1d 186 spin_lock_irqsave(q->queue_lock, flags);
59897dad 187 blk_requeue_request(q, cmd->request);
59c3d45e 188 kblockd_schedule_work(&device->requeue_work);
b485462a 189 spin_unlock_irqrestore(q->queue_lock, flags);
1da177e4
LT
190}
191
4f5299ac
JB
192/*
193 * Function: scsi_queue_insert()
194 *
195 * Purpose: Insert a command in the midlevel queue.
196 *
197 * Arguments: cmd - command that we are adding to queue.
198 * reason - why we are inserting command to queue.
199 *
200 * Lock status: Assumed that lock is not held upon entry.
201 *
202 * Returns: Nothing.
203 *
204 * Notes: We do this for one of two cases. Either the host is busy
205 * and it cannot accept any more commands for the time being,
206 * or the device returned QUEUE_FULL and can accept no more
207 * commands.
208 * Notes: This could be called either from an interrupt context or a
209 * normal process context.
210 */
84feb166 211void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
4f5299ac 212{
84feb166 213 __scsi_queue_insert(cmd, reason, 1);
4f5299ac 214}
e8064021
CH
215
216static int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
33aa687d 217 int data_direction, void *buffer, unsigned bufflen,
2bfad21e 218 unsigned char *sense, int timeout, int retries, u64 flags,
e8064021 219 req_flags_t rq_flags, int *resid)
39216033
JB
220{
221 struct request *req;
222 int write = (data_direction == DMA_TO_DEVICE);
82ed4db4 223 struct scsi_request *rq;
39216033
JB
224 int ret = DRIVER_ERROR << 24;
225
71baba4b 226 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
a492f075 227 if (IS_ERR(req))
bfe159a5 228 return ret;
82ed4db4
CH
229 rq = scsi_req(req);
230 scsi_req_init(req);
39216033
JB
231
232 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
71baba4b 233 buffer, bufflen, __GFP_RECLAIM))
39216033
JB
234 goto out;
235
82ed4db4
CH
236 rq->cmd_len = COMMAND_SIZE(cmd[0]);
237 memcpy(rq->cmd, cmd, rq->cmd_len);
17e01f21 238 req->retries = retries;
39216033 239 req->timeout = timeout;
e8064021
CH
240 req->cmd_flags |= flags;
241 req->rq_flags |= rq_flags | RQF_QUIET | RQF_PREEMPT;
39216033
JB
242
243 /*
244 * head injection *required* here otherwise quiesce won't work
245 */
246 blk_execute_rq(req->q, NULL, req, 1);
247
bdb2b8ca
AS
248 /*
249 * Some devices (USB mass-storage in particular) may transfer
250 * garbage data together with a residue indicating that the data
251 * is invalid. Prevent the garbage from being misinterpreted
252 * and prevent security leaks by zeroing out the excess data.
253 */
82ed4db4
CH
254 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
255 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
bdb2b8ca 256
f4f4e47e 257 if (resid)
82ed4db4
CH
258 *resid = rq->resid_len;
259 if (sense && rq->sense_len)
260 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
39216033
JB
261 ret = req->errors;
262 out:
263 blk_put_request(req);
264
265 return ret;
266}
e8064021
CH
267
268/**
269 * scsi_execute - insert request and wait for the result
270 * @sdev: scsi device
271 * @cmd: scsi command
272 * @data_direction: data direction
273 * @buffer: data buffer
274 * @bufflen: len of buffer
275 * @sense: optional sense buffer
276 * @timeout: request timeout in seconds
277 * @retries: number of times to retry request
278 * @flags: or into request flags;
279 * @resid: optional residual length
280 *
281 * returns the req->errors value which is the scsi_cmnd result
282 * field.
283 */
284int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
285 int data_direction, void *buffer, unsigned bufflen,
286 unsigned char *sense, int timeout, int retries, u64 flags,
287 int *resid)
288{
289 return __scsi_execute(sdev, cmd, data_direction, buffer, bufflen, sense,
290 timeout, retries, flags, 0, resid);
291}
33aa687d 292EXPORT_SYMBOL(scsi_execute);
39216033 293
9b21493c 294int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
ea73a9f2 295 int data_direction, void *buffer, unsigned bufflen,
f4f4e47e 296 struct scsi_sense_hdr *sshdr, int timeout, int retries,
e8064021 297 int *resid, u64 flags, req_flags_t rq_flags)
ea73a9f2
JB
298{
299 char *sense = NULL;
1ccb48bb 300 int result;
301
ea73a9f2 302 if (sshdr) {
24669f75 303 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
ea73a9f2
JB
304 if (!sense)
305 return DRIVER_ERROR << 24;
ea73a9f2 306 }
e8064021
CH
307 result = __scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
308 sense, timeout, retries, flags, rq_flags, resid);
ea73a9f2 309 if (sshdr)
e514385b 310 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
ea73a9f2
JB
311
312 kfree(sense);
313 return result;
314}
9b21493c 315EXPORT_SYMBOL(scsi_execute_req_flags);
ea73a9f2 316
1da177e4
LT
317/*
318 * Function: scsi_init_cmd_errh()
319 *
320 * Purpose: Initialize cmd fields related to error handling.
321 *
322 * Arguments: cmd - command that is ready to be queued.
323 *
1da177e4
LT
324 * Notes: This function has the job of initializing a number of
325 * fields related to error handling. Typically this will
326 * be called once for each command, as required.
327 */
631c228c 328static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
1da177e4 329{
1da177e4 330 cmd->serial_number = 0;
30b0c37b 331 scsi_set_resid(cmd, 0);
b80ca4f7 332 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1da177e4 333 if (cmd->cmd_len == 0)
db4742dd 334 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1da177e4
LT
335}
336
337void scsi_device_unbusy(struct scsi_device *sdev)
338{
339 struct Scsi_Host *shost = sdev->host;
f0c0a376 340 struct scsi_target *starget = scsi_target(sdev);
1da177e4
LT
341 unsigned long flags;
342
74665016 343 atomic_dec(&shost->host_busy);
2ccbb008
CH
344 if (starget->can_queue > 0)
345 atomic_dec(&starget->target_busy);
74665016 346
939647ee 347 if (unlikely(scsi_host_in_recovery(shost) &&
74665016
CH
348 (shost->host_failed || shost->host_eh_scheduled))) {
349 spin_lock_irqsave(shost->host_lock, flags);
1da177e4 350 scsi_eh_wakeup(shost);
74665016
CH
351 spin_unlock_irqrestore(shost->host_lock, flags);
352 }
353
71e75c97 354 atomic_dec(&sdev->device_busy);
1da177e4
LT
355}
356
d285203c
CH
357static void scsi_kick_queue(struct request_queue *q)
358{
359 if (q->mq_ops)
360 blk_mq_start_hw_queues(q);
361 else
362 blk_run_queue(q);
363}
364
1da177e4
LT
365/*
366 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367 * and call blk_run_queue for all the scsi_devices on the target -
368 * including current_sdev first.
369 *
370 * Called with *no* scsi locks held.
371 */
372static void scsi_single_lun_run(struct scsi_device *current_sdev)
373{
374 struct Scsi_Host *shost = current_sdev->host;
375 struct scsi_device *sdev, *tmp;
376 struct scsi_target *starget = scsi_target(current_sdev);
377 unsigned long flags;
378
379 spin_lock_irqsave(shost->host_lock, flags);
380 starget->starget_sdev_user = NULL;
381 spin_unlock_irqrestore(shost->host_lock, flags);
382
383 /*
384 * Call blk_run_queue for all LUNs on the target, starting with
385 * current_sdev. We race with others (to set starget_sdev_user),
386 * but in most cases, we will be first. Ideally, each LU on the
387 * target would get some limited time or requests on the target.
388 */
d285203c 389 scsi_kick_queue(current_sdev->request_queue);
1da177e4
LT
390
391 spin_lock_irqsave(shost->host_lock, flags);
392 if (starget->starget_sdev_user)
393 goto out;
394 list_for_each_entry_safe(sdev, tmp, &starget->devices,
395 same_target_siblings) {
396 if (sdev == current_sdev)
397 continue;
398 if (scsi_device_get(sdev))
399 continue;
400
401 spin_unlock_irqrestore(shost->host_lock, flags);
d285203c 402 scsi_kick_queue(sdev->request_queue);
1da177e4
LT
403 spin_lock_irqsave(shost->host_lock, flags);
404
405 scsi_device_put(sdev);
406 }
407 out:
408 spin_unlock_irqrestore(shost->host_lock, flags);
409}
410
cd9070c9 411static inline bool scsi_device_is_busy(struct scsi_device *sdev)
9d112517 412{
cd9070c9
CH
413 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
414 return true;
415 if (atomic_read(&sdev->device_blocked) > 0)
416 return true;
417 return false;
9d112517
KU
418}
419
cd9070c9 420static inline bool scsi_target_is_busy(struct scsi_target *starget)
f0c0a376 421{
2ccbb008
CH
422 if (starget->can_queue > 0) {
423 if (atomic_read(&starget->target_busy) >= starget->can_queue)
424 return true;
425 if (atomic_read(&starget->target_blocked) > 0)
426 return true;
427 }
cd9070c9 428 return false;
f0c0a376
MC
429}
430
cd9070c9 431static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
9d112517 432{
cd9070c9
CH
433 if (shost->can_queue > 0 &&
434 atomic_read(&shost->host_busy) >= shost->can_queue)
435 return true;
436 if (atomic_read(&shost->host_blocked) > 0)
437 return true;
438 if (shost->host_self_blocked)
439 return true;
440 return false;
9d112517
KU
441}
442
21a05df5 443static void scsi_starved_list_run(struct Scsi_Host *shost)
1da177e4 444{
2a3a59e5 445 LIST_HEAD(starved_list);
21a05df5 446 struct scsi_device *sdev;
1da177e4
LT
447 unsigned long flags;
448
1da177e4 449 spin_lock_irqsave(shost->host_lock, flags);
2a3a59e5
MC
450 list_splice_init(&shost->starved_list, &starved_list);
451
452 while (!list_empty(&starved_list)) {
e2eb7244
JB
453 struct request_queue *slq;
454
1da177e4
LT
455 /*
456 * As long as shost is accepting commands and we have
457 * starved queues, call blk_run_queue. scsi_request_fn
458 * drops the queue_lock and can add us back to the
459 * starved_list.
460 *
461 * host_lock protects the starved_list and starved_entry.
462 * scsi_request_fn must get the host_lock before checking
463 * or modifying starved_list or starved_entry.
464 */
2a3a59e5 465 if (scsi_host_is_busy(shost))
f0c0a376 466 break;
f0c0a376 467
2a3a59e5
MC
468 sdev = list_entry(starved_list.next,
469 struct scsi_device, starved_entry);
470 list_del_init(&sdev->starved_entry);
f0c0a376
MC
471 if (scsi_target_is_busy(scsi_target(sdev))) {
472 list_move_tail(&sdev->starved_entry,
473 &shost->starved_list);
474 continue;
475 }
476
e2eb7244
JB
477 /*
478 * Once we drop the host lock, a racing scsi_remove_device()
479 * call may remove the sdev from the starved list and destroy
480 * it and the queue. Mitigate by taking a reference to the
481 * queue and never touching the sdev again after we drop the
482 * host lock. Note: if __scsi_remove_device() invokes
483 * blk_cleanup_queue() before the queue is run from this
484 * function then blk_run_queue() will return immediately since
485 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
486 */
487 slq = sdev->request_queue;
488 if (!blk_get_queue(slq))
489 continue;
490 spin_unlock_irqrestore(shost->host_lock, flags);
491
d285203c 492 scsi_kick_queue(slq);
e2eb7244
JB
493 blk_put_queue(slq);
494
495 spin_lock_irqsave(shost->host_lock, flags);
1da177e4 496 }
2a3a59e5
MC
497 /* put any unprocessed entries back */
498 list_splice(&starved_list, &shost->starved_list);
1da177e4 499 spin_unlock_irqrestore(shost->host_lock, flags);
21a05df5
CH
500}
501
502/*
503 * Function: scsi_run_queue()
504 *
505 * Purpose: Select a proper request queue to serve next
506 *
507 * Arguments: q - last request's queue
508 *
509 * Returns: Nothing
510 *
511 * Notes: The previous command was completely finished, start
512 * a new one if possible.
513 */
514static void scsi_run_queue(struct request_queue *q)
515{
516 struct scsi_device *sdev = q->queuedata;
517
518 if (scsi_target(sdev)->single_lun)
519 scsi_single_lun_run(sdev);
520 if (!list_empty(&sdev->host->starved_list))
521 scsi_starved_list_run(sdev->host);
1da177e4 522
d285203c
CH
523 if (q->mq_ops)
524 blk_mq_start_stopped_hw_queues(q, false);
525 else
526 blk_run_queue(q);
1da177e4
LT
527}
528
9937a5e2
JA
529void scsi_requeue_run_queue(struct work_struct *work)
530{
531 struct scsi_device *sdev;
532 struct request_queue *q;
533
534 sdev = container_of(work, struct scsi_device, requeue_work);
535 q = sdev->request_queue;
536 scsi_run_queue(q);
537}
538
1da177e4
LT
539/*
540 * Function: scsi_requeue_command()
541 *
542 * Purpose: Handle post-processing of completed commands.
543 *
544 * Arguments: q - queue to operate on
545 * cmd - command that may need to be requeued.
546 *
547 * Returns: Nothing
548 *
549 * Notes: After command completion, there may be blocks left
550 * over which weren't finished by the previous command
551 * this can be for a number of reasons - the main one is
552 * I/O errors in the middle of the request, in which case
553 * we need to request the blocks that come after the bad
554 * sector.
e91442b6 555 * Notes: Upon return, cmd is a stale pointer.
1da177e4
LT
556 */
557static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
558{
940f5d47 559 struct scsi_device *sdev = cmd->device;
e91442b6 560 struct request *req = cmd->request;
283369cc
TH
561 unsigned long flags;
562
283369cc 563 spin_lock_irqsave(q->queue_lock, flags);
134997a0
CH
564 blk_unprep_request(req);
565 req->special = NULL;
566 scsi_put_command(cmd);
e91442b6 567 blk_requeue_request(q, req);
283369cc 568 spin_unlock_irqrestore(q->queue_lock, flags);
1da177e4
LT
569
570 scsi_run_queue(q);
940f5d47
BVA
571
572 put_device(&sdev->sdev_gendev);
1da177e4
LT
573}
574
1da177e4
LT
575void scsi_run_host_queues(struct Scsi_Host *shost)
576{
577 struct scsi_device *sdev;
578
579 shost_for_each_device(sdev, shost)
580 scsi_run_queue(sdev->request_queue);
581}
582
d285203c
CH
583static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
584{
57292b58 585 if (!blk_rq_is_passthrough(cmd->request)) {
d285203c
CH
586 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
587
588 if (drv->uninit_command)
589 drv->uninit_command(cmd);
590 }
591}
592
593static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
594{
91dbc08d
ML
595 struct scsi_data_buffer *sdb;
596
d285203c 597 if (cmd->sdb.table.nents)
001d63be 598 sg_free_table_chained(&cmd->sdb.table, true);
91dbc08d
ML
599 if (cmd->request->next_rq) {
600 sdb = cmd->request->next_rq->special;
601 if (sdb)
001d63be 602 sg_free_table_chained(&sdb->table, true);
91dbc08d 603 }
d285203c 604 if (scsi_prot_sg_count(cmd))
001d63be 605 sg_free_table_chained(&cmd->prot_sdb->table, true);
d285203c
CH
606}
607
608static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
609{
610 struct scsi_device *sdev = cmd->device;
64bdcbc4 611 struct Scsi_Host *shost = sdev->host;
d285203c
CH
612 unsigned long flags;
613
d285203c
CH
614 scsi_mq_free_sgtables(cmd);
615 scsi_uninit_cmd(cmd);
616
64bdcbc4
KD
617 if (shost->use_cmd_list) {
618 BUG_ON(list_empty(&cmd->list));
619 spin_lock_irqsave(&sdev->list_lock, flags);
620 list_del_init(&cmd->list);
621 spin_unlock_irqrestore(&sdev->list_lock, flags);
622 }
d285203c
CH
623}
624
1da177e4
LT
625/*
626 * Function: scsi_release_buffers()
627 *
c682adf3 628 * Purpose: Free resources allocate for a scsi_command.
1da177e4
LT
629 *
630 * Arguments: cmd - command that we are bailing.
631 *
632 * Lock status: Assumed that no lock is held upon entry.
633 *
634 * Returns: Nothing
635 *
636 * Notes: In the event that an upper level driver rejects a
637 * command, we must release resources allocated during
638 * the __init_io() function. Primarily this would involve
c682adf3 639 * the scatter-gather table.
1da177e4 640 */
f1bea55d 641static void scsi_release_buffers(struct scsi_cmnd *cmd)
1da177e4 642{
c682adf3 643 if (cmd->sdb.table.nents)
001d63be 644 sg_free_table_chained(&cmd->sdb.table, false);
c682adf3
CH
645
646 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
647
648 if (scsi_prot_sg_count(cmd))
001d63be 649 sg_free_table_chained(&cmd->prot_sdb->table, false);
1da177e4
LT
650}
651
c682adf3
CH
652static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
653{
654 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
655
001d63be 656 sg_free_table_chained(&bidi_sdb->table, false);
c682adf3
CH
657 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
658 cmd->request->next_rq->special = NULL;
659}
660
f6d47e74
CH
661static bool scsi_end_request(struct request *req, int error,
662 unsigned int bytes, unsigned int bidi_bytes)
663{
664 struct scsi_cmnd *cmd = req->special;
665 struct scsi_device *sdev = cmd->device;
666 struct request_queue *q = sdev->request_queue;
f6d47e74
CH
667
668 if (blk_update_request(req, error, bytes))
669 return true;
670
671 /* Bidi request must be completed as a whole */
672 if (unlikely(bidi_bytes) &&
673 blk_update_request(req->next_rq, error, bidi_bytes))
674 return true;
675
676 if (blk_queue_add_random(q))
677 add_disk_randomness(req->rq_disk);
678
d285203c
CH
679 if (req->mq_ctx) {
680 /*
c8a446ad 681 * In the MQ case the command gets freed by __blk_mq_end_request,
d285203c
CH
682 * so we have to do all cleanup that depends on it earlier.
683 *
684 * We also can't kick the queues from irq context, so we
685 * will have to defer it to a workqueue.
686 */
687 scsi_mq_uninit_cmd(cmd);
688
c8a446ad 689 __blk_mq_end_request(req, error);
d285203c
CH
690
691 if (scsi_target(sdev)->single_lun ||
692 !list_empty(&sdev->host->starved_list))
693 kblockd_schedule_work(&sdev->requeue_work);
694 else
695 blk_mq_start_stopped_hw_queues(q, true);
d285203c
CH
696 } else {
697 unsigned long flags;
698
f81426a8
DG
699 if (bidi_bytes)
700 scsi_release_bidi_buffers(cmd);
e9c787e6
CH
701 scsi_release_buffers(cmd);
702 scsi_put_command(cmd);
f81426a8 703
d285203c
CH
704 spin_lock_irqsave(q->queue_lock, flags);
705 blk_finish_request(req, error);
706 spin_unlock_irqrestore(q->queue_lock, flags);
707
bb3ec62a 708 scsi_run_queue(q);
d285203c 709 }
f6d47e74 710
bb3ec62a 711 put_device(&sdev->sdev_gendev);
f6d47e74
CH
712 return false;
713}
714
0f7f6234
HR
715/**
716 * __scsi_error_from_host_byte - translate SCSI error code into errno
717 * @cmd: SCSI command (unused)
718 * @result: scsi error code
719 *
720 * Translate SCSI error code into standard UNIX errno.
721 * Return values:
722 * -ENOLINK temporary transport failure
723 * -EREMOTEIO permanent target failure, do not retry
724 * -EBADE permanent nexus failure, retry on other path
a9d6ceb8 725 * -ENOSPC No write space available
7e782af5 726 * -ENODATA Medium error
0f7f6234
HR
727 * -EIO unspecified I/O error
728 */
63583cca
HR
729static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
730{
731 int error = 0;
732
733 switch(host_byte(result)) {
734 case DID_TRANSPORT_FAILFAST:
735 error = -ENOLINK;
736 break;
737 case DID_TARGET_FAILURE:
2082ebc4 738 set_host_byte(cmd, DID_OK);
63583cca
HR
739 error = -EREMOTEIO;
740 break;
741 case DID_NEXUS_FAILURE:
2082ebc4 742 set_host_byte(cmd, DID_OK);
63583cca
HR
743 error = -EBADE;
744 break;
a9d6ceb8
HR
745 case DID_ALLOC_FAILURE:
746 set_host_byte(cmd, DID_OK);
747 error = -ENOSPC;
748 break;
7e782af5
HR
749 case DID_MEDIUM_ERROR:
750 set_host_byte(cmd, DID_OK);
751 error = -ENODATA;
752 break;
63583cca
HR
753 default:
754 error = -EIO;
755 break;
756 }
757
758 return error;
759}
760
1da177e4
LT
761/*
762 * Function: scsi_io_completion()
763 *
764 * Purpose: Completion processing for block device I/O requests.
765 *
766 * Arguments: cmd - command that is finished.
767 *
768 * Lock status: Assumed that no lock is held upon entry.
769 *
770 * Returns: Nothing
771 *
bc85dc50
CH
772 * Notes: We will finish off the specified number of sectors. If we
773 * are done, the command block will be released and the queue
774 * function will be goosed. If we are not done then we have to
b60af5b0 775 * figure out what to do next:
1da177e4 776 *
b60af5b0
AS
777 * a) We can call scsi_requeue_command(). The request
778 * will be unprepared and put back on the queue. Then
779 * a new command will be created for it. This should
780 * be used if we made forward progress, or if we want
781 * to switch from READ(10) to READ(6) for example.
1da177e4 782 *
bc85dc50 783 * b) We can call __scsi_queue_insert(). The request will
b60af5b0
AS
784 * be put back on the queue and retried using the same
785 * command as before, possibly after a delay.
786 *
f6d47e74 787 * c) We can call scsi_end_request() with -EIO to fail
b60af5b0 788 * the remainder of the request.
1da177e4 789 */
03aba2f7 790void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1da177e4
LT
791{
792 int result = cmd->result;
165125e1 793 struct request_queue *q = cmd->device->request_queue;
1da177e4 794 struct request *req = cmd->request;
fa8e36c3 795 int error = 0;
1da177e4 796 struct scsi_sense_hdr sshdr;
4753cbc0 797 bool sense_valid = false;
c11c004b 798 int sense_deferred = 0, level = 0;
b60af5b0
AS
799 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
800 ACTION_DELAYED_RETRY} action;
ee60b2c5 801 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
1da177e4 802
1da177e4
LT
803 if (result) {
804 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
805 if (sense_valid)
806 sense_deferred = scsi_sense_is_deferred(&sshdr);
807 }
631c228c 808
57292b58 809 if (blk_rq_is_passthrough(req)) {
1da177e4 810 if (result) {
82ed4db4 811 if (sense_valid) {
1da177e4
LT
812 /*
813 * SG_IO wants current and deferred errors
814 */
82ed4db4
CH
815 scsi_req(req)->sense_len =
816 min(8 + cmd->sense_buffer[7],
817 SCSI_SENSE_BUFFERSIZE);
1da177e4 818 }
fa8e36c3 819 if (!sense_deferred)
63583cca 820 error = __scsi_error_from_host_byte(cmd, result);
b22f687d 821 }
27c41973
MS
822 /*
823 * __scsi_error_from_host_byte may have reset the host_byte
824 */
825 req->errors = cmd->result;
e6bb7a96 826
82ed4db4 827 scsi_req(req)->resid_len = scsi_get_resid(cmd);
e6bb7a96 828
6f9a35e2 829 if (scsi_bidi_cmnd(cmd)) {
e6bb7a96
FT
830 /*
831 * Bidi commands Must be complete as a whole,
832 * both sides at once.
833 */
82ed4db4 834 scsi_req(req->next_rq)->resid_len = scsi_in(cmd)->resid;
f6d47e74
CH
835 if (scsi_end_request(req, 0, blk_rq_bytes(req),
836 blk_rq_bytes(req->next_rq)))
837 BUG();
6f9a35e2
BH
838 return;
839 }
89fb4cd1
JB
840 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
841 /*
842 * Certain non BLOCK_PC requests are commands that don't
843 * actually transfer anything (FLUSH), so cannot use
844 * good_bytes != blk_rq_bytes(req) as the signal for an error.
845 * This sets the error explicitly for the problem case.
846 */
847 error = __scsi_error_from_host_byte(cmd, result);
1da177e4
LT
848 }
849
57292b58 850 /* no bidi support for !blk_rq_is_passthrough yet */
33659ebb 851 BUG_ON(blk_bidi_rq(req));
30b0c37b 852
1da177e4
LT
853 /*
854 * Next deal with any sectors which we were able to correctly
855 * handle.
856 */
91921e01
HR
857 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
858 "%u sectors total, %d bytes done.\n",
859 blk_rq_sectors(req), good_bytes));
d6b0c537 860
a9bddd74
JB
861 /*
862 * Recovered errors need reporting, but they're always treated
863 * as success, so fiddle the result code here. For BLOCK_PC
864 * we already took a copy of the original into rq->errors which
865 * is what gets returned to the user
866 */
e7efe593
DG
867 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
868 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
869 * print since caller wants ATA registers. Only occurs on
870 * SCSI ATA PASS_THROUGH commands when CK_COND=1
871 */
872 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
873 ;
e8064021 874 else if (!(req->rq_flags & RQF_QUIET))
d811b848 875 scsi_print_sense(cmd);
a9bddd74
JB
876 result = 0;
877 /* BLOCK_PC may have set error */
878 error = 0;
879 }
880
881 /*
a621bac3
JB
882 * special case: failed zero length commands always need to
883 * drop down into the retry code. Otherwise, if we finished
884 * all bytes in the request we are done now.
d6b0c537 885 */
a621bac3
JB
886 if (!(blk_rq_bytes(req) == 0 && error) &&
887 !scsi_end_request(req, error, good_bytes, 0))
f6d47e74 888 return;
bc85dc50
CH
889
890 /*
891 * Kill remainder if no retrys.
892 */
893 if (error && scsi_noretry_cmd(cmd)) {
f6d47e74
CH
894 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
895 BUG();
896 return;
bc85dc50
CH
897 }
898
899 /*
900 * If there had been no error, but we have leftover bytes in the
901 * requeues just queue the command up again.
d6b0c537 902 */
bc85dc50
CH
903 if (result == 0)
904 goto requeue;
03aba2f7 905
63583cca 906 error = __scsi_error_from_host_byte(cmd, result);
3e695f89 907
b60af5b0
AS
908 if (host_byte(result) == DID_RESET) {
909 /* Third party bus reset or reset for error recovery
910 * reasons. Just retry the command and see what
911 * happens.
912 */
913 action = ACTION_RETRY;
914 } else if (sense_valid && !sense_deferred) {
1da177e4
LT
915 switch (sshdr.sense_key) {
916 case UNIT_ATTENTION:
917 if (cmd->device->removable) {
03aba2f7 918 /* Detected disc change. Set a bit
1da177e4
LT
919 * and quietly refuse further access.
920 */
921 cmd->device->changed = 1;
b60af5b0 922 action = ACTION_FAIL;
1da177e4 923 } else {
03aba2f7
LT
924 /* Must have been a power glitch, or a
925 * bus reset. Could not have been a
926 * media change, so we just retry the
b60af5b0 927 * command and see what happens.
03aba2f7 928 */
b60af5b0 929 action = ACTION_RETRY;
1da177e4
LT
930 }
931 break;
932 case ILLEGAL_REQUEST:
03aba2f7
LT
933 /* If we had an ILLEGAL REQUEST returned, then
934 * we may have performed an unsupported
935 * command. The only thing this should be
936 * would be a ten byte read where only a six
937 * byte read was supported. Also, on a system
938 * where READ CAPACITY failed, we may have
939 * read past the end of the disk.
940 */
26a68019
JA
941 if ((cmd->device->use_10_for_rw &&
942 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
1da177e4
LT
943 (cmd->cmnd[0] == READ_10 ||
944 cmd->cmnd[0] == WRITE_10)) {
b60af5b0 945 /* This will issue a new 6-byte command. */
1da177e4 946 cmd->device->use_10_for_rw = 0;
b60af5b0 947 action = ACTION_REPREP;
3e695f89 948 } else if (sshdr.asc == 0x10) /* DIX */ {
3e695f89
MP
949 action = ACTION_FAIL;
950 error = -EILSEQ;
c98a0eb0 951 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
5db44863 952 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
c98a0eb0 953 action = ACTION_FAIL;
66a651aa 954 error = -EREMOTEIO;
b60af5b0
AS
955 } else
956 action = ACTION_FAIL;
957 break;
511e44f4 958 case ABORTED_COMMAND:
126c0982 959 action = ACTION_FAIL;
e6c11dbb 960 if (sshdr.asc == 0x10) /* DIF */
3e695f89 961 error = -EILSEQ;
1da177e4
LT
962 break;
963 case NOT_READY:
03aba2f7 964 /* If the device is in the process of becoming
f3e93f73 965 * ready, or has a temporary blockage, retry.
1da177e4 966 */
f3e93f73
JB
967 if (sshdr.asc == 0x04) {
968 switch (sshdr.ascq) {
969 case 0x01: /* becoming ready */
970 case 0x04: /* format in progress */
971 case 0x05: /* rebuild in progress */
972 case 0x06: /* recalculation in progress */
973 case 0x07: /* operation in progress */
974 case 0x08: /* Long write in progress */
975 case 0x09: /* self test in progress */
d8705f11 976 case 0x14: /* space allocation in progress */
b60af5b0 977 action = ACTION_DELAYED_RETRY;
f3e93f73 978 break;
3dbf6a54 979 default:
3dbf6a54
AS
980 action = ACTION_FAIL;
981 break;
f3e93f73 982 }
e6c11dbb 983 } else
b60af5b0 984 action = ACTION_FAIL;
b60af5b0 985 break;
1da177e4 986 case VOLUME_OVERFLOW:
03aba2f7 987 /* See SSC3rXX or current. */
b60af5b0
AS
988 action = ACTION_FAIL;
989 break;
1da177e4 990 default:
b60af5b0 991 action = ACTION_FAIL;
1da177e4
LT
992 break;
993 }
e6c11dbb 994 } else
b60af5b0 995 action = ACTION_FAIL;
b60af5b0 996
ee60b2c5 997 if (action != ACTION_FAIL &&
e6c11dbb 998 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
ee60b2c5 999 action = ACTION_FAIL;
ee60b2c5 1000
b60af5b0
AS
1001 switch (action) {
1002 case ACTION_FAIL:
1003 /* Give up and fail the remainder of the request */
e8064021 1004 if (!(req->rq_flags & RQF_QUIET)) {
f1569ff1
HR
1005 static DEFINE_RATELIMIT_STATE(_rs,
1006 DEFAULT_RATELIMIT_INTERVAL,
1007 DEFAULT_RATELIMIT_BURST);
1008
1009 if (unlikely(scsi_logging_level))
1010 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1011 SCSI_LOG_MLCOMPLETE_BITS);
1012
1013 /*
1014 * if logging is enabled the failure will be printed
1015 * in scsi_log_completion(), so avoid duplicate messages
1016 */
1017 if (!level && __ratelimit(&_rs)) {
1018 scsi_print_result(cmd, NULL, FAILED);
1019 if (driver_byte(result) & DRIVER_SENSE)
1020 scsi_print_sense(cmd);
1021 scsi_print_command(cmd);
1022 }
3173d8c3 1023 }
f6d47e74
CH
1024 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1025 return;
bc85dc50 1026 /*FALLTHRU*/
b60af5b0 1027 case ACTION_REPREP:
bc85dc50 1028 requeue:
b60af5b0
AS
1029 /* Unprep the request and put it back at the head of the queue.
1030 * A new command will be prepared and issued.
1031 */
d285203c 1032 if (q->mq_ops) {
e8064021 1033 cmd->request->rq_flags &= ~RQF_DONTPREP;
d285203c
CH
1034 scsi_mq_uninit_cmd(cmd);
1035 scsi_mq_requeue_cmd(cmd);
1036 } else {
1037 scsi_release_buffers(cmd);
1038 scsi_requeue_command(q, cmd);
1039 }
b60af5b0
AS
1040 break;
1041 case ACTION_RETRY:
1042 /* Retry the same command immediately */
4f5299ac 1043 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
b60af5b0
AS
1044 break;
1045 case ACTION_DELAYED_RETRY:
1046 /* Retry the same command after a delay */
4f5299ac 1047 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
b60af5b0 1048 break;
1da177e4
LT
1049 }
1050}
1da177e4 1051
3c356bde 1052static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1da177e4 1053{
6f9a35e2 1054 int count;
1da177e4
LT
1055
1056 /*
3b003157 1057 * If sg table allocation fails, requeue request later.
1da177e4 1058 */
f9d03f96
CH
1059 if (unlikely(sg_alloc_table_chained(&sdb->table,
1060 blk_rq_nr_phys_segments(req), sdb->table.sgl)))
1da177e4 1061 return BLKPREP_DEFER;
1da177e4 1062
1da177e4
LT
1063 /*
1064 * Next, walk the list, and fill in the addresses and sizes of
1065 * each segment.
1066 */
30b0c37b
BH
1067 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1068 BUG_ON(count > sdb->table.nents);
1069 sdb->table.nents = count;
fd102b12 1070 sdb->length = blk_rq_payload_bytes(req);
4a03d90e 1071 return BLKPREP_OK;
1da177e4 1072}
6f9a35e2
BH
1073
1074/*
1075 * Function: scsi_init_io()
1076 *
1077 * Purpose: SCSI I/O initialize function.
1078 *
1079 * Arguments: cmd - Command descriptor we wish to initialize
1080 *
1081 * Returns: 0 on success
1082 * BLKPREP_DEFER if the failure is retryable
1083 * BLKPREP_KILL if the failure is fatal
1084 */
3c356bde 1085int scsi_init_io(struct scsi_cmnd *cmd)
6f9a35e2 1086{
5e012aad 1087 struct scsi_device *sdev = cmd->device;
13f05c8d 1088 struct request *rq = cmd->request;
d285203c 1089 bool is_mq = (rq->mq_ctx != NULL);
635d98b1 1090 int error;
13f05c8d 1091
f9d03f96 1092 BUG_ON(!blk_rq_nr_phys_segments(rq));
635d98b1 1093
3c356bde 1094 error = scsi_init_sgtable(rq, &cmd->sdb);
6f9a35e2
BH
1095 if (error)
1096 goto err_exit;
1097
13f05c8d 1098 if (blk_bidi_rq(rq)) {
d285203c
CH
1099 if (!rq->q->mq_ops) {
1100 struct scsi_data_buffer *bidi_sdb =
1101 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1102 if (!bidi_sdb) {
1103 error = BLKPREP_DEFER;
1104 goto err_exit;
1105 }
1106
1107 rq->next_rq->special = bidi_sdb;
6f9a35e2
BH
1108 }
1109
3c356bde 1110 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
6f9a35e2
BH
1111 if (error)
1112 goto err_exit;
1113 }
1114
13f05c8d 1115 if (blk_integrity_rq(rq)) {
7027ad72
MP
1116 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1117 int ivecs, count;
1118
91724c20
EM
1119 if (prot_sdb == NULL) {
1120 /*
1121 * This can happen if someone (e.g. multipath)
1122 * queues a command to a device on an adapter
1123 * that does not support DIX.
1124 */
1125 WARN_ON_ONCE(1);
1126 error = BLKPREP_KILL;
1127 goto err_exit;
1128 }
1129
13f05c8d 1130 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
7027ad72 1131
001d63be 1132 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
22cc3d4c 1133 prot_sdb->table.sgl)) {
7027ad72
MP
1134 error = BLKPREP_DEFER;
1135 goto err_exit;
1136 }
1137
13f05c8d 1138 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
7027ad72
MP
1139 prot_sdb->table.sgl);
1140 BUG_ON(unlikely(count > ivecs));
13f05c8d 1141 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
7027ad72
MP
1142
1143 cmd->prot_sdb = prot_sdb;
1144 cmd->prot_sdb->table.nents = count;
1145 }
1146
d285203c 1147 return BLKPREP_OK;
6f9a35e2 1148err_exit:
d285203c
CH
1149 if (is_mq) {
1150 scsi_mq_free_sgtables(cmd);
1151 } else {
1152 scsi_release_buffers(cmd);
1153 cmd->request->special = NULL;
1154 scsi_put_command(cmd);
1155 put_device(&sdev->sdev_gendev);
1156 }
6f9a35e2
BH
1157 return error;
1158}
bb52d82f 1159EXPORT_SYMBOL(scsi_init_io);
1da177e4 1160
e9c787e6 1161void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
3b003157 1162{
e9c787e6
CH
1163 void *buf = cmd->sense_buffer;
1164 void *prot = cmd->prot_sdb;
1165 unsigned long flags;
3b003157 1166
82ed4db4
CH
1167 /* zero out the cmd, except for the embedded scsi_request */
1168 memset((char *)cmd + sizeof(cmd->req), 0,
1169 sizeof(*cmd) - sizeof(cmd->req));
1170
e9c787e6
CH
1171 cmd->device = dev;
1172 cmd->sense_buffer = buf;
1173 cmd->prot_sdb = prot;
1174 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1175 cmd->jiffies_at_alloc = jiffies;
64a87b24 1176
e9c787e6
CH
1177 spin_lock_irqsave(&dev->list_lock, flags);
1178 list_add_tail(&cmd->list, &dev->cmd_list);
1179 spin_unlock_irqrestore(&dev->list_lock, flags);
3b003157
CH
1180}
1181
4f1e5765 1182static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
7b16318d 1183{
a1b73fc1 1184 struct scsi_cmnd *cmd = req->special;
3b003157
CH
1185
1186 /*
1187 * BLOCK_PC requests may transfer data, in which case they must
1188 * a bio attached to them. Or they might contain a SCSI command
1189 * that does not transfer data, in which case they may optionally
1190 * submit a request without an attached bio.
1191 */
1192 if (req->bio) {
3c356bde 1193 int ret = scsi_init_io(cmd);
3b003157
CH
1194 if (unlikely(ret))
1195 return ret;
1196 } else {
b0790410 1197 BUG_ON(blk_rq_bytes(req));
3b003157 1198
30b0c37b 1199 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
3b003157 1200 }
7b16318d 1201
82ed4db4
CH
1202 cmd->cmd_len = scsi_req(req)->cmd_len;
1203 cmd->cmnd = scsi_req(req)->cmd;
b0790410 1204 cmd->transfersize = blk_rq_bytes(req);
7b16318d 1205 cmd->allowed = req->retries;
3b003157 1206 return BLKPREP_OK;
7b16318d 1207}
7b16318d 1208
3b003157 1209/*
3868cf8e
CH
1210 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1211 * that still need to be translated to SCSI CDBs from the ULD.
3b003157 1212 */
3868cf8e 1213static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1da177e4 1214{
a1b73fc1 1215 struct scsi_cmnd *cmd = req->special;
a6a8d9f8 1216
ee14c674
CH
1217 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1218 int ret = sdev->handler->prep_fn(sdev, req);
a6a8d9f8
CS
1219 if (ret != BLKPREP_OK)
1220 return ret;
1221 }
1222
82ed4db4 1223 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
64a87b24 1224 memset(cmd->cmnd, 0, BLK_MAX_CDB);
3868cf8e 1225 return scsi_cmd_to_driver(cmd)->init_command(cmd);
3b003157
CH
1226}
1227
6af7a4ff
CH
1228static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1229{
1230 struct scsi_cmnd *cmd = req->special;
1231
1232 if (!blk_rq_bytes(req))
1233 cmd->sc_data_direction = DMA_NONE;
1234 else if (rq_data_dir(req) == WRITE)
1235 cmd->sc_data_direction = DMA_TO_DEVICE;
1236 else
1237 cmd->sc_data_direction = DMA_FROM_DEVICE;
1238
1239 switch (req->cmd_type) {
1240 case REQ_TYPE_FS:
1241 return scsi_setup_fs_cmnd(sdev, req);
1242 case REQ_TYPE_BLOCK_PC:
1243 return scsi_setup_blk_pc_cmnd(sdev, req);
1244 default:
1245 return BLKPREP_KILL;
1246 }
1247}
1248
a1b73fc1
CH
1249static int
1250scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
3b003157 1251{
3b003157
CH
1252 int ret = BLKPREP_OK;
1253
1da177e4 1254 /*
3b003157
CH
1255 * If the device is not in running state we will reject some
1256 * or all commands.
1da177e4 1257 */
3b003157
CH
1258 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1259 switch (sdev->sdev_state) {
1260 case SDEV_OFFLINE:
1b8d2620 1261 case SDEV_TRANSPORT_OFFLINE:
3b003157
CH
1262 /*
1263 * If the device is offline we refuse to process any
1264 * commands. The device must be brought online
1265 * before trying any recovery commands.
1266 */
1267 sdev_printk(KERN_ERR, sdev,
1268 "rejecting I/O to offline device\n");
1269 ret = BLKPREP_KILL;
1270 break;
1271 case SDEV_DEL:
1272 /*
1273 * If the device is fully deleted, we refuse to
1274 * process any commands as well.
1275 */
9ccfc756 1276 sdev_printk(KERN_ERR, sdev,
3b003157
CH
1277 "rejecting I/O to dead device\n");
1278 ret = BLKPREP_KILL;
1279 break;
3b003157 1280 case SDEV_BLOCK:
6f4267e3 1281 case SDEV_CREATED_BLOCK:
bba0bdd7
BVA
1282 ret = BLKPREP_DEFER;
1283 break;
1284 case SDEV_QUIESCE:
3b003157
CH
1285 /*
1286 * If the devices is blocked we defer normal commands.
1287 */
e8064021 1288 if (!(req->rq_flags & RQF_PREEMPT))
3b003157
CH
1289 ret = BLKPREP_DEFER;
1290 break;
1291 default:
1292 /*
1293 * For any other not fully online state we only allow
1294 * special commands. In particular any user initiated
1295 * command is not allowed.
1296 */
e8064021 1297 if (!(req->rq_flags & RQF_PREEMPT))
3b003157
CH
1298 ret = BLKPREP_KILL;
1299 break;
1da177e4 1300 }
1da177e4 1301 }
7f9a6bc4
JB
1302 return ret;
1303}
1da177e4 1304
a1b73fc1
CH
1305static int
1306scsi_prep_return(struct request_queue *q, struct request *req, int ret)
7f9a6bc4
JB
1307{
1308 struct scsi_device *sdev = q->queuedata;
1da177e4 1309
3b003157
CH
1310 switch (ret) {
1311 case BLKPREP_KILL:
e1cd3911 1312 case BLKPREP_INVALID:
3b003157 1313 req->errors = DID_NO_CONNECT << 16;
7f9a6bc4
JB
1314 /* release the command and kill it */
1315 if (req->special) {
1316 struct scsi_cmnd *cmd = req->special;
1317 scsi_release_buffers(cmd);
1318 scsi_put_command(cmd);
68c03d91 1319 put_device(&sdev->sdev_gendev);
7f9a6bc4
JB
1320 req->special = NULL;
1321 }
3b003157
CH
1322 break;
1323 case BLKPREP_DEFER:
1da177e4 1324 /*
9934c8c0 1325 * If we defer, the blk_peek_request() returns NULL, but the
a488e749
JA
1326 * queue must be restarted, so we schedule a callback to happen
1327 * shortly.
1da177e4 1328 */
71e75c97 1329 if (atomic_read(&sdev->device_busy) == 0)
a488e749 1330 blk_delay_queue(q, SCSI_QUEUE_DELAY);
3b003157
CH
1331 break;
1332 default:
e8064021 1333 req->rq_flags |= RQF_DONTPREP;
1da177e4
LT
1334 }
1335
3b003157 1336 return ret;
1da177e4 1337}
7f9a6bc4 1338
a1b73fc1 1339static int scsi_prep_fn(struct request_queue *q, struct request *req)
7f9a6bc4
JB
1340{
1341 struct scsi_device *sdev = q->queuedata;
e9c787e6 1342 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
a1b73fc1
CH
1343 int ret;
1344
1345 ret = scsi_prep_state_check(sdev, req);
1346 if (ret != BLKPREP_OK)
1347 goto out;
1348
e9c787e6
CH
1349 if (!req->special) {
1350 /* Bail if we can't get a reference to the device */
1351 if (unlikely(!get_device(&sdev->sdev_gendev))) {
1352 ret = BLKPREP_DEFER;
1353 goto out;
1354 }
1355
1356 scsi_init_command(sdev, cmd);
1357 req->special = cmd;
a1b73fc1 1358 }
7f9a6bc4 1359
e9c787e6
CH
1360 cmd->tag = req->tag;
1361 cmd->request = req;
e9c787e6
CH
1362 cmd->prot_op = SCSI_PROT_NORMAL;
1363
6af7a4ff 1364 ret = scsi_setup_cmnd(sdev, req);
a1b73fc1 1365out:
7f9a6bc4
JB
1366 return scsi_prep_return(q, req, ret);
1367}
a1b73fc1
CH
1368
1369static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1370{
d285203c 1371 scsi_uninit_cmd(req->special);
a1b73fc1 1372}
1da177e4
LT
1373
1374/*
1375 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1376 * return 0.
1377 *
1378 * Called with the queue_lock held.
1379 */
1380static inline int scsi_dev_queue_ready(struct request_queue *q,
1381 struct scsi_device *sdev)
1382{
71e75c97
CH
1383 unsigned int busy;
1384
1385 busy = atomic_inc_return(&sdev->device_busy) - 1;
cd9070c9 1386 if (atomic_read(&sdev->device_blocked)) {
71e75c97
CH
1387 if (busy)
1388 goto out_dec;
1389
1da177e4
LT
1390 /*
1391 * unblock after device_blocked iterates to zero
1392 */
cd9070c9 1393 if (atomic_dec_return(&sdev->device_blocked) > 0) {
d285203c
CH
1394 /*
1395 * For the MQ case we take care of this in the caller.
1396 */
1397 if (!q->mq_ops)
1398 blk_delay_queue(q, SCSI_QUEUE_DELAY);
71e75c97 1399 goto out_dec;
1da177e4 1400 }
71e75c97
CH
1401 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1402 "unblocking device at zero depth\n"));
1da177e4 1403 }
71e75c97
CH
1404
1405 if (busy >= sdev->queue_depth)
1406 goto out_dec;
1da177e4
LT
1407
1408 return 1;
71e75c97
CH
1409out_dec:
1410 atomic_dec(&sdev->device_busy);
1411 return 0;
1da177e4
LT
1412}
1413
f0c0a376
MC
1414/*
1415 * scsi_target_queue_ready: checks if there we can send commands to target
1416 * @sdev: scsi device on starget to check.
f0c0a376
MC
1417 */
1418static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1419 struct scsi_device *sdev)
1420{
1421 struct scsi_target *starget = scsi_target(sdev);
7ae65c0f 1422 unsigned int busy;
f0c0a376
MC
1423
1424 if (starget->single_lun) {
7ae65c0f 1425 spin_lock_irq(shost->host_lock);
f0c0a376 1426 if (starget->starget_sdev_user &&
7ae65c0f
CH
1427 starget->starget_sdev_user != sdev) {
1428 spin_unlock_irq(shost->host_lock);
1429 return 0;
1430 }
f0c0a376 1431 starget->starget_sdev_user = sdev;
7ae65c0f 1432 spin_unlock_irq(shost->host_lock);
f0c0a376
MC
1433 }
1434
2ccbb008
CH
1435 if (starget->can_queue <= 0)
1436 return 1;
1437
7ae65c0f 1438 busy = atomic_inc_return(&starget->target_busy) - 1;
cd9070c9 1439 if (atomic_read(&starget->target_blocked) > 0) {
7ae65c0f
CH
1440 if (busy)
1441 goto starved;
1442
f0c0a376
MC
1443 /*
1444 * unblock after target_blocked iterates to zero
1445 */
cd9070c9 1446 if (atomic_dec_return(&starget->target_blocked) > 0)
7ae65c0f 1447 goto out_dec;
cf68d334
CH
1448
1449 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1450 "unblocking target at zero depth\n"));
f0c0a376
MC
1451 }
1452
2ccbb008 1453 if (busy >= starget->can_queue)
7ae65c0f 1454 goto starved;
f0c0a376 1455
7ae65c0f
CH
1456 return 1;
1457
1458starved:
1459 spin_lock_irq(shost->host_lock);
1460 list_move_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1461 spin_unlock_irq(shost->host_lock);
7ae65c0f 1462out_dec:
2ccbb008
CH
1463 if (starget->can_queue > 0)
1464 atomic_dec(&starget->target_busy);
7ae65c0f 1465 return 0;
f0c0a376
MC
1466}
1467
1da177e4
LT
1468/*
1469 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1470 * return 0. We must end up running the queue again whenever 0 is
1471 * returned, else IO can hang.
1da177e4
LT
1472 */
1473static inline int scsi_host_queue_ready(struct request_queue *q,
1474 struct Scsi_Host *shost,
1475 struct scsi_device *sdev)
1476{
74665016 1477 unsigned int busy;
cf68d334 1478
939647ee 1479 if (scsi_host_in_recovery(shost))
74665016
CH
1480 return 0;
1481
1482 busy = atomic_inc_return(&shost->host_busy) - 1;
cd9070c9 1483 if (atomic_read(&shost->host_blocked) > 0) {
74665016
CH
1484 if (busy)
1485 goto starved;
1486
1da177e4
LT
1487 /*
1488 * unblock after host_blocked iterates to zero
1489 */
cd9070c9 1490 if (atomic_dec_return(&shost->host_blocked) > 0)
74665016 1491 goto out_dec;
cf68d334
CH
1492
1493 SCSI_LOG_MLQUEUE(3,
1494 shost_printk(KERN_INFO, shost,
1495 "unblocking host at zero depth\n"));
1da177e4 1496 }
74665016
CH
1497
1498 if (shost->can_queue > 0 && busy >= shost->can_queue)
1499 goto starved;
1500 if (shost->host_self_blocked)
1501 goto starved;
1da177e4
LT
1502
1503 /* We're OK to process the command, so we can't be starved */
74665016
CH
1504 if (!list_empty(&sdev->starved_entry)) {
1505 spin_lock_irq(shost->host_lock);
1506 if (!list_empty(&sdev->starved_entry))
1507 list_del_init(&sdev->starved_entry);
1508 spin_unlock_irq(shost->host_lock);
1509 }
1da177e4 1510
74665016
CH
1511 return 1;
1512
1513starved:
1514 spin_lock_irq(shost->host_lock);
1515 if (list_empty(&sdev->starved_entry))
1516 list_add_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1517 spin_unlock_irq(shost->host_lock);
74665016
CH
1518out_dec:
1519 atomic_dec(&shost->host_busy);
1520 return 0;
1da177e4
LT
1521}
1522
6c5121b7
KU
1523/*
1524 * Busy state exporting function for request stacking drivers.
1525 *
1526 * For efficiency, no lock is taken to check the busy state of
1527 * shost/starget/sdev, since the returned value is not guaranteed and
1528 * may be changed after request stacking drivers call the function,
1529 * regardless of taking lock or not.
1530 *
67bd9413
BVA
1531 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1532 * needs to return 'not busy'. Otherwise, request stacking drivers
1533 * may hold requests forever.
6c5121b7
KU
1534 */
1535static int scsi_lld_busy(struct request_queue *q)
1536{
1537 struct scsi_device *sdev = q->queuedata;
1538 struct Scsi_Host *shost;
6c5121b7 1539
3f3299d5 1540 if (blk_queue_dying(q))
6c5121b7
KU
1541 return 0;
1542
1543 shost = sdev->host;
6c5121b7 1544
b7e94a16
JN
1545 /*
1546 * Ignore host/starget busy state.
1547 * Since block layer does not have a concept of fairness across
1548 * multiple queues, congestion of host/starget needs to be handled
1549 * in SCSI layer.
1550 */
1551 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
6c5121b7
KU
1552 return 1;
1553
1554 return 0;
1555}
1556
1da177e4 1557/*
e91442b6 1558 * Kill a request for a dead device
1da177e4 1559 */
165125e1 1560static void scsi_kill_request(struct request *req, struct request_queue *q)
1da177e4 1561{
e91442b6 1562 struct scsi_cmnd *cmd = req->special;
03b14708
JS
1563 struct scsi_device *sdev;
1564 struct scsi_target *starget;
1565 struct Scsi_Host *shost;
1da177e4 1566
9934c8c0 1567 blk_start_request(req);
788ce43a 1568
74571813
HR
1569 scmd_printk(KERN_INFO, cmd, "killing request\n");
1570
03b14708
JS
1571 sdev = cmd->device;
1572 starget = scsi_target(sdev);
1573 shost = sdev->host;
e91442b6
JB
1574 scsi_init_cmd_errh(cmd);
1575 cmd->result = DID_NO_CONNECT << 16;
1576 atomic_inc(&cmd->device->iorequest_cnt);
e36e0c80
TH
1577
1578 /*
1579 * SCSI request completion path will do scsi_device_unbusy(),
1580 * bump busy counts. To bump the counters, we need to dance
1581 * with the locks as normal issue path does.
1582 */
71e75c97 1583 atomic_inc(&sdev->device_busy);
74665016 1584 atomic_inc(&shost->host_busy);
2ccbb008
CH
1585 if (starget->can_queue > 0)
1586 atomic_inc(&starget->target_busy);
e36e0c80 1587
242f9dcb 1588 blk_complete_request(req);
1da177e4
LT
1589}
1590
1aea6434
JA
1591static void scsi_softirq_done(struct request *rq)
1592{
242f9dcb
JA
1593 struct scsi_cmnd *cmd = rq->special;
1594 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1aea6434
JA
1595 int disposition;
1596
1597 INIT_LIST_HEAD(&cmd->eh_entry);
1598
242f9dcb
JA
1599 atomic_inc(&cmd->device->iodone_cnt);
1600 if (cmd->result)
1601 atomic_inc(&cmd->device->ioerr_cnt);
1602
1aea6434
JA
1603 disposition = scsi_decide_disposition(cmd);
1604 if (disposition != SUCCESS &&
1605 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1606 sdev_printk(KERN_ERR, cmd->device,
1607 "timing out command, waited %lus\n",
1608 wait_for/HZ);
1609 disposition = SUCCESS;
1610 }
91921e01 1611
1aea6434
JA
1612 scsi_log_completion(cmd, disposition);
1613
1614 switch (disposition) {
1615 case SUCCESS:
1616 scsi_finish_command(cmd);
1617 break;
1618 case NEEDS_RETRY:
596f482a 1619 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1aea6434
JA
1620 break;
1621 case ADD_TO_MLQUEUE:
1622 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1623 break;
1624 default:
1625 if (!scsi_eh_scmd_add(cmd, 0))
1626 scsi_finish_command(cmd);
1627 }
1628}
1629
82042a2c
CH
1630/**
1631 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1632 * @cmd: command block we are dispatching.
1633 *
1634 * Return: nonzero return request was rejected and device's queue needs to be
1635 * plugged.
1636 */
1637static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1638{
1639 struct Scsi_Host *host = cmd->device->host;
1640 int rtn = 0;
1641
1642 atomic_inc(&cmd->device->iorequest_cnt);
1643
1644 /* check if the device is still usable */
1645 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1646 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1647 * returns an immediate error upwards, and signals
1648 * that the device is no longer present */
1649 cmd->result = DID_NO_CONNECT << 16;
1650 goto done;
1651 }
1652
1653 /* Check to see if the scsi lld made this device blocked. */
1654 if (unlikely(scsi_device_blocked(cmd->device))) {
1655 /*
1656 * in blocked state, the command is just put back on
1657 * the device queue. The suspend state has already
1658 * blocked the queue so future requests should not
1659 * occur until the device transitions out of the
1660 * suspend state.
1661 */
1662 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1663 "queuecommand : device blocked\n"));
1664 return SCSI_MLQUEUE_DEVICE_BUSY;
1665 }
1666
1667 /* Store the LUN value in cmnd, if needed. */
1668 if (cmd->device->lun_in_cdb)
1669 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1670 (cmd->device->lun << 5 & 0xe0);
1671
1672 scsi_log_send(cmd);
1673
1674 /*
1675 * Before we queue this command, check if the command
1676 * length exceeds what the host adapter can handle.
1677 */
1678 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1679 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1680 "queuecommand : command too long. "
1681 "cdb_size=%d host->max_cmd_len=%d\n",
1682 cmd->cmd_len, cmd->device->host->max_cmd_len));
1683 cmd->result = (DID_ABORT << 16);
1684 goto done;
1685 }
1686
1687 if (unlikely(host->shost_state == SHOST_DEL)) {
1688 cmd->result = (DID_NO_CONNECT << 16);
1689 goto done;
1690
1691 }
1692
1693 trace_scsi_dispatch_cmd_start(cmd);
1694 rtn = host->hostt->queuecommand(host, cmd);
1695 if (rtn) {
1696 trace_scsi_dispatch_cmd_error(cmd, rtn);
1697 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1698 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1699 rtn = SCSI_MLQUEUE_HOST_BUSY;
1700
1701 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1702 "queuecommand : request rejected\n"));
1703 }
1704
1705 return rtn;
1706 done:
1707 cmd->scsi_done(cmd);
1708 return 0;
1709}
1710
3b5382c4
CH
1711/**
1712 * scsi_done - Invoke completion on finished SCSI command.
1713 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1714 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1715 *
1716 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1717 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1718 * calls blk_complete_request() for further processing.
1719 *
1720 * This function is interrupt context safe.
1721 */
1722static void scsi_done(struct scsi_cmnd *cmd)
1723{
1724 trace_scsi_dispatch_cmd_done(cmd);
1725 blk_complete_request(cmd->request);
1726}
1727
1da177e4
LT
1728/*
1729 * Function: scsi_request_fn()
1730 *
1731 * Purpose: Main strategy routine for SCSI.
1732 *
1733 * Arguments: q - Pointer to actual queue.
1734 *
1735 * Returns: Nothing
1736 *
1737 * Lock status: IO request lock assumed to be held when called.
1738 */
1739static void scsi_request_fn(struct request_queue *q)
613be1f6
BVA
1740 __releases(q->queue_lock)
1741 __acquires(q->queue_lock)
1da177e4
LT
1742{
1743 struct scsi_device *sdev = q->queuedata;
1744 struct Scsi_Host *shost;
1745 struct scsi_cmnd *cmd;
1746 struct request *req;
1747
1da177e4
LT
1748 /*
1749 * To start with, we keep looping until the queue is empty, or until
1750 * the host is no longer able to accept any more requests.
1751 */
1752 shost = sdev->host;
a488e749 1753 for (;;) {
1da177e4
LT
1754 int rtn;
1755 /*
1756 * get next queueable request. We do this early to make sure
91921e01 1757 * that the request is fully prepared even if we cannot
1da177e4
LT
1758 * accept it.
1759 */
9934c8c0 1760 req = blk_peek_request(q);
71e75c97 1761 if (!req)
1da177e4
LT
1762 break;
1763
1764 if (unlikely(!scsi_device_online(sdev))) {
9ccfc756
JB
1765 sdev_printk(KERN_ERR, sdev,
1766 "rejecting I/O to offline device\n");
e91442b6 1767 scsi_kill_request(req, q);
1da177e4
LT
1768 continue;
1769 }
1770
71e75c97
CH
1771 if (!scsi_dev_queue_ready(q, sdev))
1772 break;
1da177e4
LT
1773
1774 /*
1775 * Remove the request from the request list.
1776 */
1777 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
9934c8c0 1778 blk_start_request(req);
1da177e4 1779
cf68d334 1780 spin_unlock_irq(q->queue_lock);
e91442b6
JB
1781 cmd = req->special;
1782 if (unlikely(cmd == NULL)) {
1783 printk(KERN_CRIT "impossible request in %s.\n"
1784 "please mail a stack trace to "
4aff5e23 1785 "linux-scsi@vger.kernel.org\n",
cadbd4a5 1786 __func__);
4aff5e23 1787 blk_dump_rq_flags(req, "foo");
e91442b6
JB
1788 BUG();
1789 }
1da177e4 1790
ecefe8a9
MC
1791 /*
1792 * We hit this when the driver is using a host wide
1793 * tag map. For device level tag maps the queue_depth check
1794 * in the device ready fn would prevent us from trying
1795 * to allocate a tag. Since the map is a shared host resource
1796 * we add the dev to the starved list so it eventually gets
1797 * a run when a tag is freed.
1798 */
e8064021 1799 if (blk_queue_tagged(q) && !(req->rq_flags & RQF_QUEUED)) {
cf68d334 1800 spin_lock_irq(shost->host_lock);
ecefe8a9
MC
1801 if (list_empty(&sdev->starved_entry))
1802 list_add_tail(&sdev->starved_entry,
1803 &shost->starved_list);
cf68d334 1804 spin_unlock_irq(shost->host_lock);
ecefe8a9
MC
1805 goto not_ready;
1806 }
1807
f0c0a376
MC
1808 if (!scsi_target_queue_ready(shost, sdev))
1809 goto not_ready;
1810
1da177e4 1811 if (!scsi_host_queue_ready(q, shost, sdev))
cf68d334 1812 goto host_not_ready;
125c99bc
CH
1813
1814 if (sdev->simple_tags)
1815 cmd->flags |= SCMD_TAGGED;
1816 else
1817 cmd->flags &= ~SCMD_TAGGED;
1da177e4 1818
1da177e4
LT
1819 /*
1820 * Finally, initialize any error handling parameters, and set up
1821 * the timers for timeouts.
1822 */
1823 scsi_init_cmd_errh(cmd);
1824
1825 /*
1826 * Dispatch the command to the low-level driver.
1827 */
3b5382c4 1828 cmd->scsi_done = scsi_done;
1da177e4 1829 rtn = scsi_dispatch_cmd(cmd);
d0d3bbf9
CH
1830 if (rtn) {
1831 scsi_queue_insert(cmd, rtn);
1832 spin_lock_irq(q->queue_lock);
a488e749 1833 goto out_delay;
d0d3bbf9
CH
1834 }
1835 spin_lock_irq(q->queue_lock);
1da177e4
LT
1836 }
1837
613be1f6 1838 return;
1da177e4 1839
cf68d334 1840 host_not_ready:
2ccbb008
CH
1841 if (scsi_target(sdev)->can_queue > 0)
1842 atomic_dec(&scsi_target(sdev)->target_busy);
cf68d334 1843 not_ready:
1da177e4
LT
1844 /*
1845 * lock q, handle tag, requeue req, and decrement device_busy. We
1846 * must return with queue_lock held.
1847 *
1848 * Decrementing device_busy without checking it is OK, as all such
1849 * cases (host limits or settings) should run the queue at some
1850 * later time.
1851 */
1852 spin_lock_irq(q->queue_lock);
1853 blk_requeue_request(q, req);
71e75c97 1854 atomic_dec(&sdev->device_busy);
a488e749 1855out_delay:
480cadc2 1856 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
a488e749 1857 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1da177e4
LT
1858}
1859
d285203c
CH
1860static inline int prep_to_mq(int ret)
1861{
1862 switch (ret) {
1863 case BLKPREP_OK:
2868f13c 1864 return BLK_MQ_RQ_QUEUE_OK;
d285203c
CH
1865 case BLKPREP_DEFER:
1866 return BLK_MQ_RQ_QUEUE_BUSY;
1867 default:
1868 return BLK_MQ_RQ_QUEUE_ERROR;
1869 }
1870}
1871
1872static int scsi_mq_prep_fn(struct request *req)
1873{
1874 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1875 struct scsi_device *sdev = req->q->queuedata;
1876 struct Scsi_Host *shost = sdev->host;
1877 unsigned char *sense_buf = cmd->sense_buffer;
1878 struct scatterlist *sg;
1879
82ed4db4
CH
1880 /* zero out the cmd, except for the embedded scsi_request */
1881 memset((char *)cmd + sizeof(cmd->req), 0,
1882 sizeof(*cmd) - sizeof(cmd->req));
d285203c
CH
1883
1884 req->special = cmd;
1885
1886 cmd->request = req;
1887 cmd->device = sdev;
1888 cmd->sense_buffer = sense_buf;
1889
1890 cmd->tag = req->tag;
1891
d285203c
CH
1892 cmd->prot_op = SCSI_PROT_NORMAL;
1893
1894 INIT_LIST_HEAD(&cmd->list);
1895 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1896 cmd->jiffies_at_alloc = jiffies;
1897
64bdcbc4
KD
1898 if (shost->use_cmd_list) {
1899 spin_lock_irq(&sdev->list_lock);
1900 list_add_tail(&cmd->list, &sdev->cmd_list);
1901 spin_unlock_irq(&sdev->list_lock);
1902 }
d285203c
CH
1903
1904 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1905 cmd->sdb.table.sgl = sg;
1906
1907 if (scsi_host_get_prot(shost)) {
1908 cmd->prot_sdb = (void *)sg +
120bb3e1 1909 min_t(unsigned int,
65e8617f 1910 shost->sg_tablesize, SG_CHUNK_SIZE) *
120bb3e1 1911 sizeof(struct scatterlist);
d285203c
CH
1912 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1913
1914 cmd->prot_sdb->table.sgl =
1915 (struct scatterlist *)(cmd->prot_sdb + 1);
1916 }
1917
1918 if (blk_bidi_rq(req)) {
1919 struct request *next_rq = req->next_rq;
1920 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1921
1922 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1923 bidi_sdb->table.sgl =
1924 (struct scatterlist *)(bidi_sdb + 1);
1925
1926 next_rq->special = bidi_sdb;
1927 }
1928
fe052529
CH
1929 blk_mq_start_request(req);
1930
d285203c
CH
1931 return scsi_setup_cmnd(sdev, req);
1932}
1933
1934static void scsi_mq_done(struct scsi_cmnd *cmd)
1935{
1936 trace_scsi_dispatch_cmd_done(cmd);
f4829a9b 1937 blk_mq_complete_request(cmd->request, cmd->request->errors);
d285203c
CH
1938}
1939
74c45052
JA
1940static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1941 const struct blk_mq_queue_data *bd)
d285203c 1942{
74c45052 1943 struct request *req = bd->rq;
d285203c
CH
1944 struct request_queue *q = req->q;
1945 struct scsi_device *sdev = q->queuedata;
1946 struct Scsi_Host *shost = sdev->host;
1947 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1948 int ret;
1949 int reason;
1950
1951 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
2868f13c 1952 if (ret != BLK_MQ_RQ_QUEUE_OK)
d285203c
CH
1953 goto out;
1954
1955 ret = BLK_MQ_RQ_QUEUE_BUSY;
1956 if (!get_device(&sdev->sdev_gendev))
1957 goto out;
1958
1959 if (!scsi_dev_queue_ready(q, sdev))
1960 goto out_put_device;
1961 if (!scsi_target_queue_ready(shost, sdev))
1962 goto out_dec_device_busy;
1963 if (!scsi_host_queue_ready(q, shost, sdev))
1964 goto out_dec_target_busy;
1965
e8064021 1966 if (!(req->rq_flags & RQF_DONTPREP)) {
d285203c 1967 ret = prep_to_mq(scsi_mq_prep_fn(req));
2868f13c 1968 if (ret != BLK_MQ_RQ_QUEUE_OK)
d285203c 1969 goto out_dec_host_busy;
e8064021 1970 req->rq_flags |= RQF_DONTPREP;
fe052529
CH
1971 } else {
1972 blk_mq_start_request(req);
d285203c
CH
1973 }
1974
125c99bc
CH
1975 if (sdev->simple_tags)
1976 cmd->flags |= SCMD_TAGGED;
b1dd2aac 1977 else
125c99bc 1978 cmd->flags &= ~SCMD_TAGGED;
b1dd2aac 1979
d285203c
CH
1980 scsi_init_cmd_errh(cmd);
1981 cmd->scsi_done = scsi_mq_done;
1982
1983 reason = scsi_dispatch_cmd(cmd);
1984 if (reason) {
1985 scsi_set_blocked(cmd, reason);
1986 ret = BLK_MQ_RQ_QUEUE_BUSY;
1987 goto out_dec_host_busy;
1988 }
1989
1990 return BLK_MQ_RQ_QUEUE_OK;
1991
1992out_dec_host_busy:
1993 atomic_dec(&shost->host_busy);
1994out_dec_target_busy:
1995 if (scsi_target(sdev)->can_queue > 0)
1996 atomic_dec(&scsi_target(sdev)->target_busy);
1997out_dec_device_busy:
1998 atomic_dec(&sdev->device_busy);
1999out_put_device:
2000 put_device(&sdev->sdev_gendev);
2001out:
2002 switch (ret) {
2003 case BLK_MQ_RQ_QUEUE_BUSY:
d285203c
CH
2004 if (atomic_read(&sdev->device_busy) == 0 &&
2005 !scsi_device_blocked(sdev))
2006 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2007 break;
2008 case BLK_MQ_RQ_QUEUE_ERROR:
2009 /*
2010 * Make sure to release all allocated ressources when
2011 * we hit an error, as we will never see this command
2012 * again.
2013 */
e8064021 2014 if (req->rq_flags & RQF_DONTPREP)
d285203c
CH
2015 scsi_mq_uninit_cmd(cmd);
2016 break;
2017 default:
2018 break;
2019 }
2020 return ret;
2021}
2022
0152fb6b
CH
2023static enum blk_eh_timer_return scsi_timeout(struct request *req,
2024 bool reserved)
2025{
2026 if (reserved)
2027 return BLK_EH_RESET_TIMER;
2028 return scsi_times_out(req);
2029}
2030
d285203c
CH
2031static int scsi_init_request(void *data, struct request *rq,
2032 unsigned int hctx_idx, unsigned int request_idx,
2033 unsigned int numa_node)
2034{
0a6ac4ee 2035 struct Scsi_Host *shost = data;
d285203c
CH
2036 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2037
0a6ac4ee
CH
2038 cmd->sense_buffer =
2039 scsi_alloc_sense_buffer(shost, GFP_KERNEL, numa_node);
d285203c
CH
2040 if (!cmd->sense_buffer)
2041 return -ENOMEM;
82ed4db4 2042 cmd->req.sense = cmd->sense_buffer;
d285203c
CH
2043 return 0;
2044}
2045
2046static void scsi_exit_request(void *data, struct request *rq,
2047 unsigned int hctx_idx, unsigned int request_idx)
2048{
0a6ac4ee 2049 struct Scsi_Host *shost = data;
d285203c
CH
2050 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2051
0a6ac4ee 2052 scsi_free_sense_buffer(shost, cmd->sense_buffer);
d285203c
CH
2053}
2054
2d9c5c20
CH
2055static int scsi_map_queues(struct blk_mq_tag_set *set)
2056{
2057 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
2058
2059 if (shost->hostt->map_queues)
2060 return shost->hostt->map_queues(shost);
2061 return blk_mq_map_queues(set);
2062}
2063
f1bea55d 2064static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1da177e4
LT
2065{
2066 struct device *host_dev;
2067 u64 bounce_limit = 0xffffffff;
2068
2069 if (shost->unchecked_isa_dma)
2070 return BLK_BOUNCE_ISA;
2071 /*
2072 * Platforms with virtual-DMA translation
2073 * hardware have no practical limit.
2074 */
2075 if (!PCI_DMA_BUS_IS_PHYS)
2076 return BLK_BOUNCE_ANY;
2077
2078 host_dev = scsi_get_device(shost);
2079 if (host_dev && host_dev->dma_mask)
e83b3664 2080 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1da177e4
LT
2081
2082 return bounce_limit;
2083}
1da177e4 2084
d48777a6 2085void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1da177e4 2086{
6f381fa3 2087 struct device *dev = shost->dma_dev;
1da177e4 2088
a8474ce2
JA
2089 /*
2090 * this limit is imposed by hardware restrictions
2091 */
8a78362c 2092 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
65e8617f 2093 SG_MAX_SEGMENTS));
a8474ce2 2094
13f05c8d
MP
2095 if (scsi_host_prot_dma(shost)) {
2096 shost->sg_prot_tablesize =
2097 min_not_zero(shost->sg_prot_tablesize,
2098 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2099 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2100 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2101 }
2102
086fa5ff 2103 blk_queue_max_hw_sectors(q, shost->max_sectors);
1da177e4
LT
2104 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2105 blk_queue_segment_boundary(q, shost->dma_boundary);
99c84dbd 2106 dma_set_seg_boundary(dev, shost->dma_boundary);
1da177e4 2107
860ac568
FT
2108 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2109
1da177e4 2110 if (!shost->use_clustering)
e692cb66 2111 q->limits.cluster = 0;
465ff318
JB
2112
2113 /*
2114 * set a reasonable default alignment on word boundaries: the
2115 * host and device may alter it using
2116 * blk_queue_update_dma_alignment() later.
2117 */
2118 blk_queue_dma_alignment(q, 0x03);
d285203c 2119}
d48777a6 2120EXPORT_SYMBOL_GPL(__scsi_init_queue);
b58d9154 2121
e9c787e6
CH
2122static int scsi_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp)
2123{
2124 struct Scsi_Host *shost = q->rq_alloc_data;
2125 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2126
2127 memset(cmd, 0, sizeof(*cmd));
2128
2129 cmd->sense_buffer = scsi_alloc_sense_buffer(shost, gfp, NUMA_NO_NODE);
2130 if (!cmd->sense_buffer)
2131 goto fail;
82ed4db4 2132 cmd->req.sense = cmd->sense_buffer;
e9c787e6
CH
2133
2134 if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
2135 cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp);
2136 if (!cmd->prot_sdb)
2137 goto fail_free_sense;
2138 }
2139
2140 return 0;
2141
2142fail_free_sense:
2143 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2144fail:
2145 return -ENOMEM;
2146}
2147
2148static void scsi_exit_rq(struct request_queue *q, struct request *rq)
2149{
2150 struct Scsi_Host *shost = q->rq_alloc_data;
2151 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2152
2153 if (cmd->prot_sdb)
2154 kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
2155 scsi_free_sense_buffer(shost, cmd->sense_buffer);
2156}
2157
b58d9154
FT
2158struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2159{
e9c787e6 2160 struct Scsi_Host *shost = sdev->host;
b58d9154
FT
2161 struct request_queue *q;
2162
e9c787e6 2163 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE);
b58d9154
FT
2164 if (!q)
2165 return NULL;
e9c787e6
CH
2166 q->cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
2167 q->rq_alloc_data = shost;
2168 q->request_fn = scsi_request_fn;
2169 q->init_rq_fn = scsi_init_rq;
2170 q->exit_rq_fn = scsi_exit_rq;
2171
2172 if (blk_init_allocated_queue(q) < 0) {
2173 blk_cleanup_queue(q);
2174 return NULL;
2175 }
b58d9154 2176
e9c787e6 2177 __scsi_init_queue(shost, q);
b58d9154 2178 blk_queue_prep_rq(q, scsi_prep_fn);
a1b73fc1 2179 blk_queue_unprep_rq(q, scsi_unprep_fn);
b58d9154 2180 blk_queue_softirq_done(q, scsi_softirq_done);
242f9dcb 2181 blk_queue_rq_timed_out(q, scsi_times_out);
6c5121b7 2182 blk_queue_lld_busy(q, scsi_lld_busy);
b58d9154
FT
2183 return q;
2184}
1da177e4 2185
d285203c 2186static struct blk_mq_ops scsi_mq_ops = {
d285203c
CH
2187 .queue_rq = scsi_queue_rq,
2188 .complete = scsi_softirq_done,
0152fb6b 2189 .timeout = scsi_timeout,
d285203c
CH
2190 .init_request = scsi_init_request,
2191 .exit_request = scsi_exit_request,
2d9c5c20 2192 .map_queues = scsi_map_queues,
d285203c
CH
2193};
2194
2195struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2196{
2197 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2198 if (IS_ERR(sdev->request_queue))
2199 return NULL;
2200
2201 sdev->request_queue->queuedata = sdev;
2202 __scsi_init_queue(sdev->host, sdev->request_queue);
2203 return sdev->request_queue;
2204}
2205
2206int scsi_mq_setup_tags(struct Scsi_Host *shost)
2207{
2208 unsigned int cmd_size, sgl_size, tbl_size;
2209
2210 tbl_size = shost->sg_tablesize;
65e8617f
ML
2211 if (tbl_size > SG_CHUNK_SIZE)
2212 tbl_size = SG_CHUNK_SIZE;
d285203c
CH
2213 sgl_size = tbl_size * sizeof(struct scatterlist);
2214 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2215 if (scsi_host_get_prot(shost))
2216 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2217
2218 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2219 shost->tag_set.ops = &scsi_mq_ops;
efec4b90 2220 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
d285203c
CH
2221 shost->tag_set.queue_depth = shost->can_queue;
2222 shost->tag_set.cmd_size = cmd_size;
2223 shost->tag_set.numa_node = NUMA_NO_NODE;
2224 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
24391c0d
SL
2225 shost->tag_set.flags |=
2226 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
d285203c
CH
2227 shost->tag_set.driver_data = shost;
2228
2229 return blk_mq_alloc_tag_set(&shost->tag_set);
2230}
2231
2232void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2233{
2234 blk_mq_free_tag_set(&shost->tag_set);
2235}
2236
1da177e4
LT
2237/*
2238 * Function: scsi_block_requests()
2239 *
2240 * Purpose: Utility function used by low-level drivers to prevent further
2241 * commands from being queued to the device.
2242 *
2243 * Arguments: shost - Host in question
2244 *
2245 * Returns: Nothing
2246 *
2247 * Lock status: No locks are assumed held.
2248 *
2249 * Notes: There is no timer nor any other means by which the requests
2250 * get unblocked other than the low-level driver calling
2251 * scsi_unblock_requests().
2252 */
2253void scsi_block_requests(struct Scsi_Host *shost)
2254{
2255 shost->host_self_blocked = 1;
2256}
2257EXPORT_SYMBOL(scsi_block_requests);
2258
2259/*
2260 * Function: scsi_unblock_requests()
2261 *
2262 * Purpose: Utility function used by low-level drivers to allow further
2263 * commands from being queued to the device.
2264 *
2265 * Arguments: shost - Host in question
2266 *
2267 * Returns: Nothing
2268 *
2269 * Lock status: No locks are assumed held.
2270 *
2271 * Notes: There is no timer nor any other means by which the requests
2272 * get unblocked other than the low-level driver calling
2273 * scsi_unblock_requests().
2274 *
2275 * This is done as an API function so that changes to the
2276 * internals of the scsi mid-layer won't require wholesale
2277 * changes to drivers that use this feature.
2278 */
2279void scsi_unblock_requests(struct Scsi_Host *shost)
2280{
2281 shost->host_self_blocked = 0;
2282 scsi_run_host_queues(shost);
2283}
2284EXPORT_SYMBOL(scsi_unblock_requests);
2285
2286int __init scsi_init_queue(void)
2287{
6362abd3
MP
2288 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2289 sizeof(struct scsi_data_buffer),
2290 0, 0, NULL);
2291 if (!scsi_sdb_cache) {
2292 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
f078727b 2293 return -ENOMEM;
6f9a35e2
BH
2294 }
2295
1da177e4
LT
2296 return 0;
2297}
2298
2299void scsi_exit_queue(void)
2300{
0a6ac4ee
CH
2301 kmem_cache_destroy(scsi_sense_cache);
2302 kmem_cache_destroy(scsi_sense_isadma_cache);
6362abd3 2303 kmem_cache_destroy(scsi_sdb_cache);
1da177e4 2304}
5baba830
JB
2305
2306/**
2307 * scsi_mode_select - issue a mode select
2308 * @sdev: SCSI device to be queried
2309 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2310 * @sp: Save page bit (0 == don't save, 1 == save)
2311 * @modepage: mode page being requested
2312 * @buffer: request buffer (may not be smaller than eight bytes)
2313 * @len: length of request buffer.
2314 * @timeout: command timeout
2315 * @retries: number of retries before failing
2316 * @data: returns a structure abstracting the mode header data
eb44820c 2317 * @sshdr: place to put sense data (or NULL if no sense to be collected).
5baba830
JB
2318 * must be SCSI_SENSE_BUFFERSIZE big.
2319 *
2320 * Returns zero if successful; negative error number or scsi
2321 * status on error
2322 *
2323 */
2324int
2325scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2326 unsigned char *buffer, int len, int timeout, int retries,
2327 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2328{
2329 unsigned char cmd[10];
2330 unsigned char *real_buffer;
2331 int ret;
2332
2333 memset(cmd, 0, sizeof(cmd));
2334 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2335
2336 if (sdev->use_10_for_ms) {
2337 if (len > 65535)
2338 return -EINVAL;
2339 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2340 if (!real_buffer)
2341 return -ENOMEM;
2342 memcpy(real_buffer + 8, buffer, len);
2343 len += 8;
2344 real_buffer[0] = 0;
2345 real_buffer[1] = 0;
2346 real_buffer[2] = data->medium_type;
2347 real_buffer[3] = data->device_specific;
2348 real_buffer[4] = data->longlba ? 0x01 : 0;
2349 real_buffer[5] = 0;
2350 real_buffer[6] = data->block_descriptor_length >> 8;
2351 real_buffer[7] = data->block_descriptor_length;
2352
2353 cmd[0] = MODE_SELECT_10;
2354 cmd[7] = len >> 8;
2355 cmd[8] = len;
2356 } else {
2357 if (len > 255 || data->block_descriptor_length > 255 ||
2358 data->longlba)
2359 return -EINVAL;
2360
2361 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2362 if (!real_buffer)
2363 return -ENOMEM;
2364 memcpy(real_buffer + 4, buffer, len);
2365 len += 4;
2366 real_buffer[0] = 0;
2367 real_buffer[1] = data->medium_type;
2368 real_buffer[2] = data->device_specific;
2369 real_buffer[3] = data->block_descriptor_length;
2370
2371
2372 cmd[0] = MODE_SELECT;
2373 cmd[4] = len;
2374 }
2375
2376 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
f4f4e47e 2377 sshdr, timeout, retries, NULL);
5baba830
JB
2378 kfree(real_buffer);
2379 return ret;
2380}
2381EXPORT_SYMBOL_GPL(scsi_mode_select);
2382
1da177e4 2383/**
eb44820c 2384 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1cf72699 2385 * @sdev: SCSI device to be queried
1da177e4
LT
2386 * @dbd: set if mode sense will allow block descriptors to be returned
2387 * @modepage: mode page being requested
2388 * @buffer: request buffer (may not be smaller than eight bytes)
2389 * @len: length of request buffer.
2390 * @timeout: command timeout
2391 * @retries: number of retries before failing
2392 * @data: returns a structure abstracting the mode header data
eb44820c 2393 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1cf72699 2394 * must be SCSI_SENSE_BUFFERSIZE big.
1da177e4
LT
2395 *
2396 * Returns zero if unsuccessful, or the header offset (either 4
2397 * or 8 depending on whether a six or ten byte command was
2398 * issued) if successful.
eb44820c 2399 */
1da177e4 2400int
1cf72699 2401scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1da177e4 2402 unsigned char *buffer, int len, int timeout, int retries,
5baba830
JB
2403 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2404{
1da177e4
LT
2405 unsigned char cmd[12];
2406 int use_10_for_ms;
2407 int header_length;
0ae80ba9 2408 int result, retry_count = retries;
ea73a9f2 2409 struct scsi_sense_hdr my_sshdr;
1da177e4
LT
2410
2411 memset(data, 0, sizeof(*data));
2412 memset(&cmd[0], 0, 12);
2413 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2414 cmd[2] = modepage;
2415
ea73a9f2
JB
2416 /* caller might not be interested in sense, but we need it */
2417 if (!sshdr)
2418 sshdr = &my_sshdr;
2419
1da177e4 2420 retry:
1cf72699 2421 use_10_for_ms = sdev->use_10_for_ms;
1da177e4
LT
2422
2423 if (use_10_for_ms) {
2424 if (len < 8)
2425 len = 8;
2426
2427 cmd[0] = MODE_SENSE_10;
2428 cmd[8] = len;
2429 header_length = 8;
2430 } else {
2431 if (len < 4)
2432 len = 4;
2433
2434 cmd[0] = MODE_SENSE;
2435 cmd[4] = len;
2436 header_length = 4;
2437 }
2438
1da177e4
LT
2439 memset(buffer, 0, len);
2440
1cf72699 2441 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
f4f4e47e 2442 sshdr, timeout, retries, NULL);
1da177e4
LT
2443
2444 /* This code looks awful: what it's doing is making sure an
2445 * ILLEGAL REQUEST sense return identifies the actual command
2446 * byte as the problem. MODE_SENSE commands can return
2447 * ILLEGAL REQUEST if the code page isn't supported */
2448
1cf72699
JB
2449 if (use_10_for_ms && !scsi_status_is_good(result) &&
2450 (driver_byte(result) & DRIVER_SENSE)) {
ea73a9f2
JB
2451 if (scsi_sense_valid(sshdr)) {
2452 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2453 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1da177e4
LT
2454 /*
2455 * Invalid command operation code
2456 */
1cf72699 2457 sdev->use_10_for_ms = 0;
1da177e4
LT
2458 goto retry;
2459 }
2460 }
2461 }
2462
1cf72699 2463 if(scsi_status_is_good(result)) {
6d73c851
AV
2464 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2465 (modepage == 6 || modepage == 8))) {
2466 /* Initio breakage? */
2467 header_length = 0;
2468 data->length = 13;
2469 data->medium_type = 0;
2470 data->device_specific = 0;
2471 data->longlba = 0;
2472 data->block_descriptor_length = 0;
2473 } else if(use_10_for_ms) {
1da177e4
LT
2474 data->length = buffer[0]*256 + buffer[1] + 2;
2475 data->medium_type = buffer[2];
2476 data->device_specific = buffer[3];
2477 data->longlba = buffer[4] & 0x01;
2478 data->block_descriptor_length = buffer[6]*256
2479 + buffer[7];
2480 } else {
2481 data->length = buffer[0] + 1;
2482 data->medium_type = buffer[1];
2483 data->device_specific = buffer[2];
2484 data->block_descriptor_length = buffer[3];
2485 }
6d73c851 2486 data->header_length = header_length;
0ae80ba9
HR
2487 } else if ((status_byte(result) == CHECK_CONDITION) &&
2488 scsi_sense_valid(sshdr) &&
2489 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2490 retry_count--;
2491 goto retry;
1da177e4
LT
2492 }
2493
1cf72699 2494 return result;
1da177e4
LT
2495}
2496EXPORT_SYMBOL(scsi_mode_sense);
2497
001aac25
JB
2498/**
2499 * scsi_test_unit_ready - test if unit is ready
2500 * @sdev: scsi device to change the state of.
2501 * @timeout: command timeout
2502 * @retries: number of retries before failing
2503 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2504 * returning sense. Make sure that this is cleared before passing
2505 * in.
2506 *
2507 * Returns zero if unsuccessful or an error if TUR failed. For
9f8a2c23 2508 * removable media, UNIT_ATTENTION sets ->changed flag.
001aac25 2509 **/
1da177e4 2510int
001aac25
JB
2511scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2512 struct scsi_sense_hdr *sshdr_external)
1da177e4 2513{
1da177e4
LT
2514 char cmd[] = {
2515 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2516 };
001aac25 2517 struct scsi_sense_hdr *sshdr;
1da177e4 2518 int result;
001aac25
JB
2519
2520 if (!sshdr_external)
2521 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2522 else
2523 sshdr = sshdr_external;
2524
2525 /* try to eat the UNIT_ATTENTION if there are enough retries */
2526 do {
2527 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
f4f4e47e 2528 timeout, retries, NULL);
32c356d7
JB
2529 if (sdev->removable && scsi_sense_valid(sshdr) &&
2530 sshdr->sense_key == UNIT_ATTENTION)
2531 sdev->changed = 1;
2532 } while (scsi_sense_valid(sshdr) &&
2533 sshdr->sense_key == UNIT_ATTENTION && --retries);
001aac25 2534
001aac25
JB
2535 if (!sshdr_external)
2536 kfree(sshdr);
1da177e4
LT
2537 return result;
2538}
2539EXPORT_SYMBOL(scsi_test_unit_ready);
2540
2541/**
eb44820c 2542 * scsi_device_set_state - Take the given device through the device state model.
1da177e4
LT
2543 * @sdev: scsi device to change the state of.
2544 * @state: state to change to.
2545 *
2546 * Returns zero if unsuccessful or an error if the requested
2547 * transition is illegal.
eb44820c 2548 */
1da177e4
LT
2549int
2550scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2551{
2552 enum scsi_device_state oldstate = sdev->sdev_state;
2553
2554 if (state == oldstate)
2555 return 0;
2556
2557 switch (state) {
2558 case SDEV_CREATED:
6f4267e3
JB
2559 switch (oldstate) {
2560 case SDEV_CREATED_BLOCK:
2561 break;
2562 default:
2563 goto illegal;
2564 }
2565 break;
1da177e4
LT
2566
2567 case SDEV_RUNNING:
2568 switch (oldstate) {
2569 case SDEV_CREATED:
2570 case SDEV_OFFLINE:
1b8d2620 2571 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2572 case SDEV_QUIESCE:
2573 case SDEV_BLOCK:
2574 break;
2575 default:
2576 goto illegal;
2577 }
2578 break;
2579
2580 case SDEV_QUIESCE:
2581 switch (oldstate) {
2582 case SDEV_RUNNING:
2583 case SDEV_OFFLINE:
1b8d2620 2584 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2585 break;
2586 default:
2587 goto illegal;
2588 }
2589 break;
2590
2591 case SDEV_OFFLINE:
1b8d2620 2592 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2593 switch (oldstate) {
2594 case SDEV_CREATED:
2595 case SDEV_RUNNING:
2596 case SDEV_QUIESCE:
2597 case SDEV_BLOCK:
2598 break;
2599 default:
2600 goto illegal;
2601 }
2602 break;
2603
2604 case SDEV_BLOCK:
2605 switch (oldstate) {
1da177e4 2606 case SDEV_RUNNING:
6f4267e3
JB
2607 case SDEV_CREATED_BLOCK:
2608 break;
2609 default:
2610 goto illegal;
2611 }
2612 break;
2613
2614 case SDEV_CREATED_BLOCK:
2615 switch (oldstate) {
2616 case SDEV_CREATED:
1da177e4
LT
2617 break;
2618 default:
2619 goto illegal;
2620 }
2621 break;
2622
2623 case SDEV_CANCEL:
2624 switch (oldstate) {
2625 case SDEV_CREATED:
2626 case SDEV_RUNNING:
9ea72909 2627 case SDEV_QUIESCE:
1da177e4 2628 case SDEV_OFFLINE:
1b8d2620 2629 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2630 case SDEV_BLOCK:
2631 break;
2632 default:
2633 goto illegal;
2634 }
2635 break;
2636
2637 case SDEV_DEL:
2638 switch (oldstate) {
309bd271
BK
2639 case SDEV_CREATED:
2640 case SDEV_RUNNING:
2641 case SDEV_OFFLINE:
1b8d2620 2642 case SDEV_TRANSPORT_OFFLINE:
1da177e4 2643 case SDEV_CANCEL:
0516c08d 2644 case SDEV_CREATED_BLOCK:
1da177e4
LT
2645 break;
2646 default:
2647 goto illegal;
2648 }
2649 break;
2650
2651 }
2652 sdev->sdev_state = state;
2653 return 0;
2654
2655 illegal:
91921e01 2656 SCSI_LOG_ERROR_RECOVERY(1,
9ccfc756 2657 sdev_printk(KERN_ERR, sdev,
91921e01 2658 "Illegal state transition %s->%s",
9ccfc756
JB
2659 scsi_device_state_name(oldstate),
2660 scsi_device_state_name(state))
1da177e4
LT
2661 );
2662 return -EINVAL;
2663}
2664EXPORT_SYMBOL(scsi_device_set_state);
2665
a341cd0f
JG
2666/**
2667 * sdev_evt_emit - emit a single SCSI device uevent
2668 * @sdev: associated SCSI device
2669 * @evt: event to emit
2670 *
2671 * Send a single uevent (scsi_event) to the associated scsi_device.
2672 */
2673static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2674{
2675 int idx = 0;
2676 char *envp[3];
2677
2678 switch (evt->evt_type) {
2679 case SDEV_EVT_MEDIA_CHANGE:
2680 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2681 break;
279afdfe 2682 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
d3d32891 2683 scsi_rescan_device(&sdev->sdev_gendev);
279afdfe
EM
2684 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2685 break;
2686 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2687 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2688 break;
2689 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2690 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2691 break;
2692 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2693 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2694 break;
2695 case SDEV_EVT_LUN_CHANGE_REPORTED:
2696 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2697 break;
14c3e677
HR
2698 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2699 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2700 break;
a341cd0f
JG
2701 default:
2702 /* do nothing */
2703 break;
2704 }
2705
2706 envp[idx++] = NULL;
2707
2708 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2709}
2710
2711/**
2712 * sdev_evt_thread - send a uevent for each scsi event
2713 * @work: work struct for scsi_device
2714 *
2715 * Dispatch queued events to their associated scsi_device kobjects
2716 * as uevents.
2717 */
2718void scsi_evt_thread(struct work_struct *work)
2719{
2720 struct scsi_device *sdev;
279afdfe 2721 enum scsi_device_event evt_type;
a341cd0f
JG
2722 LIST_HEAD(event_list);
2723
2724 sdev = container_of(work, struct scsi_device, event_work);
2725
279afdfe
EM
2726 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2727 if (test_and_clear_bit(evt_type, sdev->pending_events))
2728 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2729
a341cd0f
JG
2730 while (1) {
2731 struct scsi_event *evt;
2732 struct list_head *this, *tmp;
2733 unsigned long flags;
2734
2735 spin_lock_irqsave(&sdev->list_lock, flags);
2736 list_splice_init(&sdev->event_list, &event_list);
2737 spin_unlock_irqrestore(&sdev->list_lock, flags);
2738
2739 if (list_empty(&event_list))
2740 break;
2741
2742 list_for_each_safe(this, tmp, &event_list) {
2743 evt = list_entry(this, struct scsi_event, node);
2744 list_del(&evt->node);
2745 scsi_evt_emit(sdev, evt);
2746 kfree(evt);
2747 }
2748 }
2749}
2750
2751/**
2752 * sdev_evt_send - send asserted event to uevent thread
2753 * @sdev: scsi_device event occurred on
2754 * @evt: event to send
2755 *
2756 * Assert scsi device event asynchronously.
2757 */
2758void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2759{
2760 unsigned long flags;
2761
4d1566ed
KS
2762#if 0
2763 /* FIXME: currently this check eliminates all media change events
2764 * for polled devices. Need to update to discriminate between AN
2765 * and polled events */
a341cd0f
JG
2766 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2767 kfree(evt);
2768 return;
2769 }
4d1566ed 2770#endif
a341cd0f
JG
2771
2772 spin_lock_irqsave(&sdev->list_lock, flags);
2773 list_add_tail(&evt->node, &sdev->event_list);
2774 schedule_work(&sdev->event_work);
2775 spin_unlock_irqrestore(&sdev->list_lock, flags);
2776}
2777EXPORT_SYMBOL_GPL(sdev_evt_send);
2778
2779/**
2780 * sdev_evt_alloc - allocate a new scsi event
2781 * @evt_type: type of event to allocate
2782 * @gfpflags: GFP flags for allocation
2783 *
2784 * Allocates and returns a new scsi_event.
2785 */
2786struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2787 gfp_t gfpflags)
2788{
2789 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2790 if (!evt)
2791 return NULL;
2792
2793 evt->evt_type = evt_type;
2794 INIT_LIST_HEAD(&evt->node);
2795
2796 /* evt_type-specific initialization, if any */
2797 switch (evt_type) {
2798 case SDEV_EVT_MEDIA_CHANGE:
279afdfe
EM
2799 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2800 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2801 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2802 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2803 case SDEV_EVT_LUN_CHANGE_REPORTED:
14c3e677 2804 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
a341cd0f
JG
2805 default:
2806 /* do nothing */
2807 break;
2808 }
2809
2810 return evt;
2811}
2812EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2813
2814/**
2815 * sdev_evt_send_simple - send asserted event to uevent thread
2816 * @sdev: scsi_device event occurred on
2817 * @evt_type: type of event to send
2818 * @gfpflags: GFP flags for allocation
2819 *
2820 * Assert scsi device event asynchronously, given an event type.
2821 */
2822void sdev_evt_send_simple(struct scsi_device *sdev,
2823 enum scsi_device_event evt_type, gfp_t gfpflags)
2824{
2825 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2826 if (!evt) {
2827 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2828 evt_type);
2829 return;
2830 }
2831
2832 sdev_evt_send(sdev, evt);
2833}
2834EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2835
669f0441
BVA
2836/**
2837 * scsi_request_fn_active() - number of kernel threads inside scsi_request_fn()
2838 * @sdev: SCSI device to count the number of scsi_request_fn() callers for.
2839 */
2840static int scsi_request_fn_active(struct scsi_device *sdev)
2841{
2842 struct request_queue *q = sdev->request_queue;
2843 int request_fn_active;
2844
2845 WARN_ON_ONCE(sdev->host->use_blk_mq);
2846
2847 spin_lock_irq(q->queue_lock);
2848 request_fn_active = q->request_fn_active;
2849 spin_unlock_irq(q->queue_lock);
2850
2851 return request_fn_active;
2852}
2853
2854/**
2855 * scsi_wait_for_queuecommand() - wait for ongoing queuecommand() calls
2856 * @sdev: SCSI device pointer.
2857 *
2858 * Wait until the ongoing shost->hostt->queuecommand() calls that are
2859 * invoked from scsi_request_fn() have finished.
2860 */
2861static void scsi_wait_for_queuecommand(struct scsi_device *sdev)
2862{
2863 WARN_ON_ONCE(sdev->host->use_blk_mq);
2864
2865 while (scsi_request_fn_active(sdev))
2866 msleep(20);
2867}
2868
1da177e4
LT
2869/**
2870 * scsi_device_quiesce - Block user issued commands.
2871 * @sdev: scsi device to quiesce.
2872 *
2873 * This works by trying to transition to the SDEV_QUIESCE state
2874 * (which must be a legal transition). When the device is in this
2875 * state, only special requests will be accepted, all others will
2876 * be deferred. Since special requests may also be requeued requests,
2877 * a successful return doesn't guarantee the device will be
2878 * totally quiescent.
2879 *
2880 * Must be called with user context, may sleep.
2881 *
2882 * Returns zero if unsuccessful or an error if not.
eb44820c 2883 */
1da177e4
LT
2884int
2885scsi_device_quiesce(struct scsi_device *sdev)
2886{
2887 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2888 if (err)
2889 return err;
2890
2891 scsi_run_queue(sdev->request_queue);
71e75c97 2892 while (atomic_read(&sdev->device_busy)) {
1da177e4
LT
2893 msleep_interruptible(200);
2894 scsi_run_queue(sdev->request_queue);
2895 }
2896 return 0;
2897}
2898EXPORT_SYMBOL(scsi_device_quiesce);
2899
2900/**
2901 * scsi_device_resume - Restart user issued commands to a quiesced device.
2902 * @sdev: scsi device to resume.
2903 *
2904 * Moves the device from quiesced back to running and restarts the
2905 * queues.
2906 *
2907 * Must be called with user context, may sleep.
eb44820c 2908 */
a7a20d10 2909void scsi_device_resume(struct scsi_device *sdev)
1da177e4 2910{
a7a20d10
DW
2911 /* check if the device state was mutated prior to resume, and if
2912 * so assume the state is being managed elsewhere (for example
2913 * device deleted during suspend)
2914 */
2915 if (sdev->sdev_state != SDEV_QUIESCE ||
2916 scsi_device_set_state(sdev, SDEV_RUNNING))
1da177e4
LT
2917 return;
2918 scsi_run_queue(sdev->request_queue);
2919}
2920EXPORT_SYMBOL(scsi_device_resume);
2921
2922static void
2923device_quiesce_fn(struct scsi_device *sdev, void *data)
2924{
2925 scsi_device_quiesce(sdev);
2926}
2927
2928void
2929scsi_target_quiesce(struct scsi_target *starget)
2930{
2931 starget_for_each_device(starget, NULL, device_quiesce_fn);
2932}
2933EXPORT_SYMBOL(scsi_target_quiesce);
2934
2935static void
2936device_resume_fn(struct scsi_device *sdev, void *data)
2937{
2938 scsi_device_resume(sdev);
2939}
2940
2941void
2942scsi_target_resume(struct scsi_target *starget)
2943{
2944 starget_for_each_device(starget, NULL, device_resume_fn);
2945}
2946EXPORT_SYMBOL(scsi_target_resume);
2947
2948/**
eb44820c 2949 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
1da177e4
LT
2950 * @sdev: device to block
2951 *
2952 * Block request made by scsi lld's to temporarily stop all
669f0441 2953 * scsi commands on the specified device. May sleep.
1da177e4
LT
2954 *
2955 * Returns zero if successful or error if not
2956 *
2957 * Notes:
2958 * This routine transitions the device to the SDEV_BLOCK state
2959 * (which must be a legal transition). When the device is in this
2960 * state, all commands are deferred until the scsi lld reenables
2961 * the device with scsi_device_unblock or device_block_tmo fires.
669f0441
BVA
2962 *
2963 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2964 * scsi_internal_device_block() has blocked a SCSI device and also
2965 * remove the rport mutex lock and unlock calls from srp_queuecommand().
eb44820c 2966 */
1da177e4
LT
2967int
2968scsi_internal_device_block(struct scsi_device *sdev)
2969{
165125e1 2970 struct request_queue *q = sdev->request_queue;
1da177e4
LT
2971 unsigned long flags;
2972 int err = 0;
2973
2974 err = scsi_device_set_state(sdev, SDEV_BLOCK);
6f4267e3
JB
2975 if (err) {
2976 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2977
2978 if (err)
2979 return err;
2980 }
1da177e4
LT
2981
2982 /*
2983 * The device has transitioned to SDEV_BLOCK. Stop the
2984 * block layer from calling the midlayer with this device's
2985 * request queue.
2986 */
d285203c 2987 if (q->mq_ops) {
7dbbf0fa 2988 blk_mq_quiesce_queue(q);
d285203c
CH
2989 } else {
2990 spin_lock_irqsave(q->queue_lock, flags);
2991 blk_stop_queue(q);
2992 spin_unlock_irqrestore(q->queue_lock, flags);
669f0441 2993 scsi_wait_for_queuecommand(sdev);
d285203c 2994 }
1da177e4
LT
2995
2996 return 0;
2997}
2998EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2999
3000/**
3001 * scsi_internal_device_unblock - resume a device after a block request
3002 * @sdev: device to resume
5d9fb5cc 3003 * @new_state: state to set devices to after unblocking
1da177e4
LT
3004 *
3005 * Called by scsi lld's or the midlayer to restart the device queue
3006 * for the previously suspended scsi device. Called from interrupt or
3007 * normal process context.
3008 *
3009 * Returns zero if successful or error if not.
3010 *
3011 * Notes:
3012 * This routine transitions the device to the SDEV_RUNNING state
5d9fb5cc 3013 * or to one of the offline states (which must be a legal transition)
d075498c 3014 * allowing the midlayer to goose the queue for this device.
eb44820c 3015 */
1da177e4 3016int
5d9fb5cc
MC
3017scsi_internal_device_unblock(struct scsi_device *sdev,
3018 enum scsi_device_state new_state)
1da177e4 3019{
165125e1 3020 struct request_queue *q = sdev->request_queue;
1da177e4 3021 unsigned long flags;
5d9fb5cc
MC
3022
3023 /*
3024 * Try to transition the scsi device to SDEV_RUNNING or one of the
3025 * offlined states and goose the device queue if successful.
1da177e4 3026 */
0e58076b
VC
3027 if ((sdev->sdev_state == SDEV_BLOCK) ||
3028 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
5d9fb5cc
MC
3029 sdev->sdev_state = new_state;
3030 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3031 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3032 new_state == SDEV_OFFLINE)
3033 sdev->sdev_state = new_state;
3034 else
3035 sdev->sdev_state = SDEV_CREATED;
3036 } else if (sdev->sdev_state != SDEV_CANCEL &&
986fe6c7 3037 sdev->sdev_state != SDEV_OFFLINE)
5c10e63c 3038 return -EINVAL;
1da177e4 3039
d285203c
CH
3040 if (q->mq_ops) {
3041 blk_mq_start_stopped_hw_queues(q, false);
3042 } else {
3043 spin_lock_irqsave(q->queue_lock, flags);
3044 blk_start_queue(q);
3045 spin_unlock_irqrestore(q->queue_lock, flags);
3046 }
1da177e4
LT
3047
3048 return 0;
3049}
3050EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3051
3052static void
3053device_block(struct scsi_device *sdev, void *data)
3054{
3055 scsi_internal_device_block(sdev);
3056}
3057
3058static int
3059target_block(struct device *dev, void *data)
3060{
3061 if (scsi_is_target_device(dev))
3062 starget_for_each_device(to_scsi_target(dev), NULL,
3063 device_block);
3064 return 0;
3065}
3066
3067void
3068scsi_target_block(struct device *dev)
3069{
3070 if (scsi_is_target_device(dev))
3071 starget_for_each_device(to_scsi_target(dev), NULL,
3072 device_block);
3073 else
3074 device_for_each_child(dev, NULL, target_block);
3075}
3076EXPORT_SYMBOL_GPL(scsi_target_block);
3077
3078static void
3079device_unblock(struct scsi_device *sdev, void *data)
3080{
5d9fb5cc 3081 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
1da177e4
LT
3082}
3083
3084static int
3085target_unblock(struct device *dev, void *data)
3086{
3087 if (scsi_is_target_device(dev))
5d9fb5cc 3088 starget_for_each_device(to_scsi_target(dev), data,
1da177e4
LT
3089 device_unblock);
3090 return 0;
3091}
3092
3093void
5d9fb5cc 3094scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
1da177e4
LT
3095{
3096 if (scsi_is_target_device(dev))
5d9fb5cc 3097 starget_for_each_device(to_scsi_target(dev), &new_state,
1da177e4
LT
3098 device_unblock);
3099 else
5d9fb5cc 3100 device_for_each_child(dev, &new_state, target_unblock);
1da177e4
LT
3101}
3102EXPORT_SYMBOL_GPL(scsi_target_unblock);
cdb8c2a6
GL
3103
3104/**
3105 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
eb44820c 3106 * @sgl: scatter-gather list
cdb8c2a6
GL
3107 * @sg_count: number of segments in sg
3108 * @offset: offset in bytes into sg, on return offset into the mapped area
3109 * @len: bytes to map, on return number of bytes mapped
3110 *
3111 * Returns virtual address of the start of the mapped page
3112 */
c6132da1 3113void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
cdb8c2a6
GL
3114 size_t *offset, size_t *len)
3115{
3116 int i;
3117 size_t sg_len = 0, len_complete = 0;
c6132da1 3118 struct scatterlist *sg;
cdb8c2a6
GL
3119 struct page *page;
3120
22cfefb5
AM
3121 WARN_ON(!irqs_disabled());
3122
c6132da1 3123 for_each_sg(sgl, sg, sg_count, i) {
cdb8c2a6 3124 len_complete = sg_len; /* Complete sg-entries */
c6132da1 3125 sg_len += sg->length;
cdb8c2a6
GL
3126 if (sg_len > *offset)
3127 break;
3128 }
3129
3130 if (unlikely(i == sg_count)) {
169e1a2a
AM
3131 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3132 "elements %d\n",
cadbd4a5 3133 __func__, sg_len, *offset, sg_count);
cdb8c2a6
GL
3134 WARN_ON(1);
3135 return NULL;
3136 }
3137
3138 /* Offset starting from the beginning of first page in this sg-entry */
c6132da1 3139 *offset = *offset - len_complete + sg->offset;
cdb8c2a6
GL
3140
3141 /* Assumption: contiguous pages can be accessed as "page + i" */
45711f1a 3142 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
cdb8c2a6
GL
3143 *offset &= ~PAGE_MASK;
3144
3145 /* Bytes in this sg-entry from *offset to the end of the page */
3146 sg_len = PAGE_SIZE - *offset;
3147 if (*len > sg_len)
3148 *len = sg_len;
3149
77dfce07 3150 return kmap_atomic(page);
cdb8c2a6
GL
3151}
3152EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3153
3154/**
eb44820c 3155 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
cdb8c2a6
GL
3156 * @virt: virtual address to be unmapped
3157 */
3158void scsi_kunmap_atomic_sg(void *virt)
3159{
77dfce07 3160 kunmap_atomic(virt);
cdb8c2a6
GL
3161}
3162EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
6f4c827e
AL
3163
3164void sdev_disable_disk_events(struct scsi_device *sdev)
3165{
3166 atomic_inc(&sdev->disk_events_disable_depth);
3167}
3168EXPORT_SYMBOL(sdev_disable_disk_events);
3169
3170void sdev_enable_disk_events(struct scsi_device *sdev)
3171{
3172 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3173 return;
3174 atomic_dec(&sdev->disk_events_disable_depth);
3175}
3176EXPORT_SYMBOL(sdev_enable_disk_events);
9983bed3
HR
3177
3178/**
3179 * scsi_vpd_lun_id - return a unique device identification
3180 * @sdev: SCSI device
3181 * @id: buffer for the identification
3182 * @id_len: length of the buffer
3183 *
3184 * Copies a unique device identification into @id based
3185 * on the information in the VPD page 0x83 of the device.
3186 * The string will be formatted as a SCSI name string.
3187 *
3188 * Returns the length of the identification or error on failure.
3189 * If the identifier is longer than the supplied buffer the actual
3190 * identifier length is returned and the buffer is not zero-padded.
3191 */
3192int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3193{
3194 u8 cur_id_type = 0xff;
3195 u8 cur_id_size = 0;
3196 unsigned char *d, *cur_id_str;
3197 unsigned char __rcu *vpd_pg83;
3198 int id_size = -EINVAL;
3199
3200 rcu_read_lock();
3201 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3202 if (!vpd_pg83) {
3203 rcu_read_unlock();
3204 return -ENXIO;
3205 }
3206
3207 /*
3208 * Look for the correct descriptor.
3209 * Order of preference for lun descriptor:
3210 * - SCSI name string
3211 * - NAA IEEE Registered Extended
3212 * - EUI-64 based 16-byte
3213 * - EUI-64 based 12-byte
3214 * - NAA IEEE Registered
3215 * - NAA IEEE Extended
d230823a 3216 * - T10 Vendor ID
9983bed3
HR
3217 * as longer descriptors reduce the likelyhood
3218 * of identification clashes.
3219 */
3220
3221 /* The id string must be at least 20 bytes + terminating NULL byte */
3222 if (id_len < 21) {
3223 rcu_read_unlock();
3224 return -EINVAL;
3225 }
3226
3227 memset(id, 0, id_len);
3228 d = vpd_pg83 + 4;
3229 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3230 /* Skip designators not referring to the LUN */
3231 if ((d[1] & 0x30) != 0x00)
3232 goto next_desig;
3233
3234 switch (d[1] & 0xf) {
d230823a
HR
3235 case 0x1:
3236 /* T10 Vendor ID */
3237 if (cur_id_size > d[3])
3238 break;
3239 /* Prefer anything */
3240 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3241 break;
3242 cur_id_size = d[3];
3243 if (cur_id_size + 4 > id_len)
3244 cur_id_size = id_len - 4;
3245 cur_id_str = d + 4;
3246 cur_id_type = d[1] & 0xf;
3247 id_size = snprintf(id, id_len, "t10.%*pE",
3248 cur_id_size, cur_id_str);
3249 break;
9983bed3
HR
3250 case 0x2:
3251 /* EUI-64 */
3252 if (cur_id_size > d[3])
3253 break;
3254 /* Prefer NAA IEEE Registered Extended */
3255 if (cur_id_type == 0x3 &&
3256 cur_id_size == d[3])
3257 break;
3258 cur_id_size = d[3];
3259 cur_id_str = d + 4;
3260 cur_id_type = d[1] & 0xf;
3261 switch (cur_id_size) {
3262 case 8:
3263 id_size = snprintf(id, id_len,
3264 "eui.%8phN",
3265 cur_id_str);
3266 break;
3267 case 12:
3268 id_size = snprintf(id, id_len,
3269 "eui.%12phN",
3270 cur_id_str);
3271 break;
3272 case 16:
3273 id_size = snprintf(id, id_len,
3274 "eui.%16phN",
3275 cur_id_str);
3276 break;
3277 default:
3278 cur_id_size = 0;
3279 break;
3280 }
3281 break;
3282 case 0x3:
3283 /* NAA */
3284 if (cur_id_size > d[3])
3285 break;
3286 cur_id_size = d[3];
3287 cur_id_str = d + 4;
3288 cur_id_type = d[1] & 0xf;
3289 switch (cur_id_size) {
3290 case 8:
3291 id_size = snprintf(id, id_len,
3292 "naa.%8phN",
3293 cur_id_str);
3294 break;
3295 case 16:
3296 id_size = snprintf(id, id_len,
3297 "naa.%16phN",
3298 cur_id_str);
3299 break;
3300 default:
3301 cur_id_size = 0;
3302 break;
3303 }
3304 break;
3305 case 0x8:
3306 /* SCSI name string */
3307 if (cur_id_size + 4 > d[3])
3308 break;
3309 /* Prefer others for truncated descriptor */
3310 if (cur_id_size && d[3] > id_len)
3311 break;
3312 cur_id_size = id_size = d[3];
3313 cur_id_str = d + 4;
3314 cur_id_type = d[1] & 0xf;
3315 if (cur_id_size >= id_len)
3316 cur_id_size = id_len - 1;
3317 memcpy(id, cur_id_str, cur_id_size);
3318 /* Decrease priority for truncated descriptor */
3319 if (cur_id_size != id_size)
3320 cur_id_size = 6;
3321 break;
3322 default:
3323 break;
3324 }
3325next_desig:
3326 d += d[3] + 4;
3327 }
3328 rcu_read_unlock();
3329
3330 return id_size;
3331}
3332EXPORT_SYMBOL(scsi_vpd_lun_id);
a8aa3978
HR
3333
3334/*
3335 * scsi_vpd_tpg_id - return a target port group identifier
3336 * @sdev: SCSI device
3337 *
3338 * Returns the Target Port Group identifier from the information
3339 * froom VPD page 0x83 of the device.
3340 *
3341 * Returns the identifier or error on failure.
3342 */
3343int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3344{
3345 unsigned char *d;
3346 unsigned char __rcu *vpd_pg83;
3347 int group_id = -EAGAIN, rel_port = -1;
3348
3349 rcu_read_lock();
3350 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3351 if (!vpd_pg83) {
3352 rcu_read_unlock();
3353 return -ENXIO;
3354 }
3355
3356 d = sdev->vpd_pg83 + 4;
3357 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3358 switch (d[1] & 0xf) {
3359 case 0x4:
3360 /* Relative target port */
3361 rel_port = get_unaligned_be16(&d[6]);
3362 break;
3363 case 0x5:
3364 /* Target port group */
3365 group_id = get_unaligned_be16(&d[6]);
3366 break;
3367 default:
3368 break;
3369 }
3370 d += d[3] + 4;
3371 }
3372 rcu_read_unlock();
3373
3374 if (group_id >= 0 && rel_id && rel_port != -1)
3375 *rel_id = rel_port;
3376
3377 return group_id;
3378}
3379EXPORT_SYMBOL(scsi_vpd_tpg_id);