block: provide a blk_rq_map_sg variant that returns the last element
[linux-block.git] / drivers / scsi / scsi_lib.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
1da177e4 2/*
d285203c
CH
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
1da177e4
LT
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12#include <linux/bio.h>
d3f46f39 13#include <linux/bitops.h>
1da177e4
LT
14#include <linux/blkdev.h>
15#include <linux/completion.h>
16#include <linux/kernel.h>
09703660 17#include <linux/export.h>
1da177e4
LT
18#include <linux/init.h>
19#include <linux/pci.h>
20#include <linux/delay.h>
faead26d 21#include <linux/hardirq.h>
c6132da1 22#include <linux/scatterlist.h>
d285203c 23#include <linux/blk-mq.h>
f1569ff1 24#include <linux/ratelimit.h>
a8aa3978 25#include <asm/unaligned.h>
1da177e4
LT
26
27#include <scsi/scsi.h>
beb40487 28#include <scsi/scsi_cmnd.h>
1da177e4
LT
29#include <scsi/scsi_dbg.h>
30#include <scsi/scsi_device.h>
31#include <scsi/scsi_driver.h>
32#include <scsi/scsi_eh.h>
33#include <scsi/scsi_host.h>
7aa686d3 34#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
ee14c674 35#include <scsi/scsi_dh.h>
1da177e4 36
3b5382c4
CH
37#include <trace/events/scsi.h>
38
0eebd005 39#include "scsi_debugfs.h"
1da177e4
LT
40#include "scsi_priv.h"
41#include "scsi_logging.h"
42
92524fa1
ML
43/*
44 * Size of integrity metadata is usually small, 1 inline sg should
45 * cover normal cases.
46 */
3e99b3b1
ML
47#ifdef CONFIG_ARCH_NO_SG_CHAIN
48#define SCSI_INLINE_PROT_SG_CNT 0
49#define SCSI_INLINE_SG_CNT 0
50#else
92524fa1 51#define SCSI_INLINE_PROT_SG_CNT 1
3dccdf53 52#define SCSI_INLINE_SG_CNT 2
3e99b3b1 53#endif
3dccdf53 54
e9c787e6 55static struct kmem_cache *scsi_sdb_cache;
0a6ac4ee
CH
56static struct kmem_cache *scsi_sense_cache;
57static struct kmem_cache *scsi_sense_isadma_cache;
58static DEFINE_MUTEX(scsi_sense_cache_mutex);
1da177e4 59
a45a1f36
BVA
60static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
61
0a6ac4ee 62static inline struct kmem_cache *
8e688254 63scsi_select_sense_cache(bool unchecked_isa_dma)
0a6ac4ee 64{
8e688254 65 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
0a6ac4ee
CH
66}
67
8e688254
BVA
68static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69 unsigned char *sense_buffer)
0a6ac4ee 70{
8e688254
BVA
71 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
72 sense_buffer);
0a6ac4ee
CH
73}
74
8e688254 75static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
e9c787e6 76 gfp_t gfp_mask, int numa_node)
0a6ac4ee 77{
8e688254
BVA
78 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
79 gfp_mask, numa_node);
0a6ac4ee
CH
80}
81
82int scsi_init_sense_cache(struct Scsi_Host *shost)
83{
84 struct kmem_cache *cache;
85 int ret = 0;
86
f9b0530f 87 mutex_lock(&scsi_sense_cache_mutex);
8e688254 88 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
0a6ac4ee 89 if (cache)
f9b0530f 90 goto exit;
0a6ac4ee 91
0a6ac4ee
CH
92 if (shost->unchecked_isa_dma) {
93 scsi_sense_isadma_cache =
94 kmem_cache_create("scsi_sense_cache(DMA)",
0afe76e8
DW
95 SCSI_SENSE_BUFFERSIZE, 0,
96 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
0a6ac4ee
CH
97 if (!scsi_sense_isadma_cache)
98 ret = -ENOMEM;
99 } else {
100 scsi_sense_cache =
0afe76e8
DW
101 kmem_cache_create_usercopy("scsi_sense_cache",
102 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103 0, SCSI_SENSE_BUFFERSIZE, NULL);
0a6ac4ee
CH
104 if (!scsi_sense_cache)
105 ret = -ENOMEM;
106 }
f9b0530f 107 exit:
0a6ac4ee
CH
108 mutex_unlock(&scsi_sense_cache_mutex);
109 return ret;
110}
6f9a35e2 111
a488e749
JA
112/*
113 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114 * not change behaviour from the previous unplug mechanism, experimentation
115 * may prove this needs changing.
116 */
117#define SCSI_QUEUE_DELAY 3
118
de3e8bf3
CH
119static void
120scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
1da177e4
LT
121{
122 struct Scsi_Host *host = cmd->device->host;
123 struct scsi_device *device = cmd->device;
f0c0a376 124 struct scsi_target *starget = scsi_target(device);
1da177e4
LT
125
126 /*
d8c37e7b 127 * Set the appropriate busy bit for the device/host.
1da177e4
LT
128 *
129 * If the host/device isn't busy, assume that something actually
130 * completed, and that we should be able to queue a command now.
131 *
132 * Note that the prior mid-layer assumption that any host could
133 * always queue at least one command is now broken. The mid-layer
134 * will implement a user specifiable stall (see
135 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136 * if a command is requeued with no other commands outstanding
137 * either for the device or for the host.
138 */
f0c0a376
MC
139 switch (reason) {
140 case SCSI_MLQUEUE_HOST_BUSY:
cd9070c9 141 atomic_set(&host->host_blocked, host->max_host_blocked);
f0c0a376
MC
142 break;
143 case SCSI_MLQUEUE_DEVICE_BUSY:
573e5913 144 case SCSI_MLQUEUE_EH_RETRY:
cd9070c9
CH
145 atomic_set(&device->device_blocked,
146 device->max_device_blocked);
f0c0a376
MC
147 break;
148 case SCSI_MLQUEUE_TARGET_BUSY:
cd9070c9
CH
149 atomic_set(&starget->target_blocked,
150 starget->max_target_blocked);
f0c0a376
MC
151 break;
152 }
de3e8bf3
CH
153}
154
d285203c
CH
155static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
156{
a45a1f36
BVA
157 if (cmd->request->rq_flags & RQF_DONTPREP) {
158 cmd->request->rq_flags &= ~RQF_DONTPREP;
159 scsi_mq_uninit_cmd(cmd);
160 } else {
161 WARN_ON_ONCE(true);
162 }
2b053aca 163 blk_mq_requeue_request(cmd->request, true);
d285203c
CH
164}
165
de3e8bf3
CH
166/**
167 * __scsi_queue_insert - private queue insertion
168 * @cmd: The SCSI command being requeued
169 * @reason: The reason for the requeue
170 * @unbusy: Whether the queue should be unbusied
171 *
172 * This is a private queue insertion. The public interface
173 * scsi_queue_insert() always assumes the queue should be unbusied
174 * because it's always called before the completion. This function is
175 * for a requeue after completion, which should only occur in this
176 * file.
177 */
08640e81 178static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
de3e8bf3
CH
179{
180 struct scsi_device *device = cmd->device;
de3e8bf3
CH
181
182 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183 "Inserting command %p into mlqueue\n", cmd));
184
185 scsi_set_blocked(cmd, reason);
1da177e4 186
1da177e4
LT
187 /*
188 * Decrement the counters, since these commands are no longer
189 * active on the host/device.
190 */
4f5299ac 191 if (unbusy)
6eb045e0 192 scsi_device_unbusy(device, cmd);
1da177e4
LT
193
194 /*
a1bf9d1d 195 * Requeue this command. It will go before all other commands
b485462a
BVA
196 * that are already in the queue. Schedule requeue work under
197 * lock such that the kblockd_schedule_work() call happens
198 * before blk_cleanup_queue() finishes.
a488e749 199 */
644373a4 200 cmd->result = 0;
f664a3cc 201
f664a3cc 202 blk_mq_requeue_request(cmd->request, true);
1da177e4
LT
203}
204
4f5299ac
JB
205/*
206 * Function: scsi_queue_insert()
207 *
208 * Purpose: Insert a command in the midlevel queue.
209 *
210 * Arguments: cmd - command that we are adding to queue.
211 * reason - why we are inserting command to queue.
212 *
213 * Lock status: Assumed that lock is not held upon entry.
214 *
215 * Returns: Nothing.
216 *
217 * Notes: We do this for one of two cases. Either the host is busy
218 * and it cannot accept any more commands for the time being,
219 * or the device returned QUEUE_FULL and can accept no more
220 * commands.
221 * Notes: This could be called either from an interrupt context or a
222 * normal process context.
223 */
84feb166 224void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
4f5299ac 225{
08640e81 226 __scsi_queue_insert(cmd, reason, true);
4f5299ac 227}
e8064021 228
76aaf87b
CH
229
230/**
704f8392 231 * __scsi_execute - insert request and wait for the result
76aaf87b
CH
232 * @sdev: scsi device
233 * @cmd: scsi command
234 * @data_direction: data direction
235 * @buffer: data buffer
236 * @bufflen: len of buffer
237 * @sense: optional sense buffer
238 * @sshdr: optional decoded sense header
239 * @timeout: request timeout in seconds
240 * @retries: number of times to retry request
241 * @flags: flags for ->cmd_flags
242 * @rq_flags: flags for ->rq_flags
243 * @resid: optional residual length
244 *
17d5363b
CH
245 * Returns the scsi_cmnd result field if a command was executed, or a negative
246 * Linux error code if we didn't get that far.
76aaf87b 247 */
704f8392 248int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
33aa687d 249 int data_direction, void *buffer, unsigned bufflen,
3949e2f0
CH
250 unsigned char *sense, struct scsi_sense_hdr *sshdr,
251 int timeout, int retries, u64 flags, req_flags_t rq_flags,
252 int *resid)
39216033
JB
253{
254 struct request *req;
82ed4db4 255 struct scsi_request *rq;
39216033
JB
256 int ret = DRIVER_ERROR << 24;
257
ff005a06 258 req = blk_get_request(sdev->request_queue,
aebf526b 259 data_direction == DMA_TO_DEVICE ?
039c635f 260 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
a492f075 261 if (IS_ERR(req))
bfe159a5 262 return ret;
82ed4db4 263 rq = scsi_req(req);
39216033
JB
264
265 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
0eb0b63c 266 buffer, bufflen, GFP_NOIO))
39216033
JB
267 goto out;
268
82ed4db4
CH
269 rq->cmd_len = COMMAND_SIZE(cmd[0]);
270 memcpy(rq->cmd, cmd, rq->cmd_len);
64c7f1d1 271 rq->retries = retries;
39216033 272 req->timeout = timeout;
e8064021 273 req->cmd_flags |= flags;
039c635f 274 req->rq_flags |= rq_flags | RQF_QUIET;
39216033
JB
275
276 /*
277 * head injection *required* here otherwise quiesce won't work
278 */
279 blk_execute_rq(req->q, NULL, req, 1);
280
bdb2b8ca
AS
281 /*
282 * Some devices (USB mass-storage in particular) may transfer
283 * garbage data together with a residue indicating that the data
284 * is invalid. Prevent the garbage from being misinterpreted
285 * and prevent security leaks by zeroing out the excess data.
286 */
82ed4db4
CH
287 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
bdb2b8ca 289
f4f4e47e 290 if (resid)
82ed4db4
CH
291 *resid = rq->resid_len;
292 if (sense && rq->sense_len)
293 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
3949e2f0
CH
294 if (sshdr)
295 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
17d5363b 296 ret = rq->result;
39216033
JB
297 out:
298 blk_put_request(req);
299
300 return ret;
301}
704f8392 302EXPORT_SYMBOL(__scsi_execute);
39216033 303
1da177e4
LT
304/*
305 * Function: scsi_init_cmd_errh()
306 *
307 * Purpose: Initialize cmd fields related to error handling.
308 *
309 * Arguments: cmd - command that is ready to be queued.
310 *
1da177e4
LT
311 * Notes: This function has the job of initializing a number of
312 * fields related to error handling. Typically this will
313 * be called once for each command, as required.
314 */
631c228c 315static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
1da177e4 316{
30b0c37b 317 scsi_set_resid(cmd, 0);
b80ca4f7 318 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1da177e4 319 if (cmd->cmd_len == 0)
db4742dd 320 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1da177e4
LT
321}
322
3bd6f43f 323/*
6eb045e0
ML
324 * Wake up the error handler if necessary. Avoid as follows that the error
325 * handler is not woken up if host in-flight requests number ==
326 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327 * with an RCU read lock in this function to ensure that this function in
328 * its entirety either finishes before scsi_eh_scmd_add() increases the
3bd6f43f
BVA
329 * host_failed counter or that it notices the shost state change made by
330 * scsi_eh_scmd_add().
331 */
6eb045e0 332static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
1da177e4 333{
1da177e4
LT
334 unsigned long flags;
335
3bd6f43f 336 rcu_read_lock();
6eb045e0 337 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
3bd6f43f 338 if (unlikely(scsi_host_in_recovery(shost))) {
74665016 339 spin_lock_irqsave(shost->host_lock, flags);
3bd6f43f
BVA
340 if (shost->host_failed || shost->host_eh_scheduled)
341 scsi_eh_wakeup(shost);
74665016
CH
342 spin_unlock_irqrestore(shost->host_lock, flags);
343 }
3bd6f43f
BVA
344 rcu_read_unlock();
345}
346
6eb045e0 347void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
3bd6f43f
BVA
348{
349 struct Scsi_Host *shost = sdev->host;
350 struct scsi_target *starget = scsi_target(sdev);
351
6eb045e0 352 scsi_dec_host_busy(shost, cmd);
3bd6f43f
BVA
353
354 if (starget->can_queue > 0)
355 atomic_dec(&starget->target_busy);
74665016 356
71e75c97 357 atomic_dec(&sdev->device_busy);
1da177e4
LT
358}
359
d285203c
CH
360static void scsi_kick_queue(struct request_queue *q)
361{
f664a3cc 362 blk_mq_run_hw_queues(q, false);
d285203c
CH
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 (atomic_read(&shost->host_blocked) > 0)
434 return true;
435 if (shost->host_self_blocked)
436 return true;
437 return false;
9d112517
KU
438}
439
21a05df5 440static void scsi_starved_list_run(struct Scsi_Host *shost)
1da177e4 441{
2a3a59e5 442 LIST_HEAD(starved_list);
21a05df5 443 struct scsi_device *sdev;
1da177e4
LT
444 unsigned long flags;
445
1da177e4 446 spin_lock_irqsave(shost->host_lock, flags);
2a3a59e5
MC
447 list_splice_init(&shost->starved_list, &starved_list);
448
449 while (!list_empty(&starved_list)) {
e2eb7244
JB
450 struct request_queue *slq;
451
1da177e4
LT
452 /*
453 * As long as shost is accepting commands and we have
454 * starved queues, call blk_run_queue. scsi_request_fn
455 * drops the queue_lock and can add us back to the
456 * starved_list.
457 *
458 * host_lock protects the starved_list and starved_entry.
459 * scsi_request_fn must get the host_lock before checking
460 * or modifying starved_list or starved_entry.
461 */
2a3a59e5 462 if (scsi_host_is_busy(shost))
f0c0a376 463 break;
f0c0a376 464
2a3a59e5
MC
465 sdev = list_entry(starved_list.next,
466 struct scsi_device, starved_entry);
467 list_del_init(&sdev->starved_entry);
f0c0a376
MC
468 if (scsi_target_is_busy(scsi_target(sdev))) {
469 list_move_tail(&sdev->starved_entry,
470 &shost->starved_list);
471 continue;
472 }
473
e2eb7244
JB
474 /*
475 * Once we drop the host lock, a racing scsi_remove_device()
476 * call may remove the sdev from the starved list and destroy
477 * it and the queue. Mitigate by taking a reference to the
478 * queue and never touching the sdev again after we drop the
479 * host lock. Note: if __scsi_remove_device() invokes
480 * blk_cleanup_queue() before the queue is run from this
481 * function then blk_run_queue() will return immediately since
482 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
483 */
484 slq = sdev->request_queue;
485 if (!blk_get_queue(slq))
486 continue;
487 spin_unlock_irqrestore(shost->host_lock, flags);
488
d285203c 489 scsi_kick_queue(slq);
e2eb7244
JB
490 blk_put_queue(slq);
491
492 spin_lock_irqsave(shost->host_lock, flags);
1da177e4 493 }
2a3a59e5
MC
494 /* put any unprocessed entries back */
495 list_splice(&starved_list, &shost->starved_list);
1da177e4 496 spin_unlock_irqrestore(shost->host_lock, flags);
21a05df5
CH
497}
498
499/*
500 * Function: scsi_run_queue()
501 *
502 * Purpose: Select a proper request queue to serve next
503 *
504 * Arguments: q - last request's queue
505 *
506 * Returns: Nothing
507 *
508 * Notes: The previous command was completely finished, start
509 * a new one if possible.
510 */
511static void scsi_run_queue(struct request_queue *q)
512{
513 struct scsi_device *sdev = q->queuedata;
514
515 if (scsi_target(sdev)->single_lun)
516 scsi_single_lun_run(sdev);
517 if (!list_empty(&sdev->host->starved_list))
518 scsi_starved_list_run(sdev->host);
1da177e4 519
f664a3cc 520 blk_mq_run_hw_queues(q, false);
1da177e4
LT
521}
522
9937a5e2
JA
523void scsi_requeue_run_queue(struct work_struct *work)
524{
525 struct scsi_device *sdev;
526 struct request_queue *q;
527
528 sdev = container_of(work, struct scsi_device, requeue_work);
529 q = sdev->request_queue;
530 scsi_run_queue(q);
531}
532
1da177e4
LT
533void scsi_run_host_queues(struct Scsi_Host *shost)
534{
535 struct scsi_device *sdev;
536
537 shost_for_each_device(sdev, shost)
538 scsi_run_queue(sdev->request_queue);
539}
540
d285203c
CH
541static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
542{
57292b58 543 if (!blk_rq_is_passthrough(cmd->request)) {
d285203c
CH
544 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
545
546 if (drv->uninit_command)
547 drv->uninit_command(cmd);
548 }
549}
550
551static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
552{
553 if (cmd->sdb.table.nents)
3dccdf53
ML
554 sg_free_table_chained(&cmd->sdb.table,
555 SCSI_INLINE_SG_CNT);
d285203c 556 if (scsi_prot_sg_count(cmd))
92524fa1
ML
557 sg_free_table_chained(&cmd->prot_sdb->table,
558 SCSI_INLINE_PROT_SG_CNT);
d285203c
CH
559}
560
561static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
562{
d285203c
CH
563 scsi_mq_free_sgtables(cmd);
564 scsi_uninit_cmd(cmd);
d285203c
CH
565}
566
7e63b5a4 567/* Returns false when no more bytes to process, true if there are more */
2a842aca 568static bool scsi_end_request(struct request *req, blk_status_t error,
ae3d56d8 569 unsigned int bytes)
f6d47e74 570{
bed2213d 571 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
f6d47e74
CH
572 struct scsi_device *sdev = cmd->device;
573 struct request_queue *q = sdev->request_queue;
f6d47e74
CH
574
575 if (blk_update_request(req, error, bytes))
576 return true;
577
f6d47e74
CH
578 if (blk_queue_add_random(q))
579 add_disk_randomness(req->rq_disk);
580
64104f70
BVA
581 if (!blk_rq_is_scsi(req)) {
582 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
583 cmd->flags &= ~SCMD_INITIALIZED;
584 }
585
db983f6e
BVA
586 /*
587 * Calling rcu_barrier() is not necessary here because the
588 * SCSI error handler guarantees that the function called by
589 * call_rcu() has been called before scsi_end_request() is
590 * called.
591 */
592 destroy_rcu_head(&cmd->rcu);
593
f664a3cc
JA
594 /*
595 * In the MQ case the command gets freed by __blk_mq_end_request,
596 * so we have to do all cleanup that depends on it earlier.
597 *
598 * We also can't kick the queues from irq context, so we
599 * will have to defer it to a workqueue.
600 */
601 scsi_mq_uninit_cmd(cmd);
f81426a8 602
a78b03bc
JA
603 /*
604 * queue is still alive, so grab the ref for preventing it
605 * from being cleaned up during running queue.
606 */
607 percpu_ref_get(&q->q_usage_counter);
f81426a8 608
f664a3cc 609 __blk_mq_end_request(req, error);
d285203c 610
f664a3cc
JA
611 if (scsi_target(sdev)->single_lun ||
612 !list_empty(&sdev->host->starved_list))
613 kblockd_schedule_work(&sdev->requeue_work);
614 else
615 blk_mq_run_hw_queues(q, true);
f6d47e74 616
a78b03bc 617 percpu_ref_put(&q->q_usage_counter);
f6d47e74
CH
618 return false;
619}
620
0f7f6234 621/**
a77b32d8
BVA
622 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
623 * @cmd: SCSI command
0f7f6234
HR
624 * @result: scsi error code
625 *
a77b32d8
BVA
626 * Translate a SCSI result code into a blk_status_t value. May reset the host
627 * byte of @cmd->result.
0f7f6234 628 */
a77b32d8 629static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
63583cca 630{
2a842aca 631 switch (host_byte(result)) {
f4abab3f
BVA
632 case DID_OK:
633 /*
634 * Also check the other bytes than the status byte in result
635 * to handle the case when a SCSI LLD sets result to
636 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
637 */
638 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
639 return BLK_STS_OK;
640 return BLK_STS_IOERR;
63583cca 641 case DID_TRANSPORT_FAILFAST:
2a842aca 642 return BLK_STS_TRANSPORT;
63583cca 643 case DID_TARGET_FAILURE:
2082ebc4 644 set_host_byte(cmd, DID_OK);
2a842aca 645 return BLK_STS_TARGET;
63583cca 646 case DID_NEXUS_FAILURE:
4a067cf8 647 set_host_byte(cmd, DID_OK);
2a842aca 648 return BLK_STS_NEXUS;
a9d6ceb8
HR
649 case DID_ALLOC_FAILURE:
650 set_host_byte(cmd, DID_OK);
2a842aca 651 return BLK_STS_NOSPC;
7e782af5
HR
652 case DID_MEDIUM_ERROR:
653 set_host_byte(cmd, DID_OK);
2a842aca 654 return BLK_STS_MEDIUM;
63583cca 655 default:
2a842aca 656 return BLK_STS_IOERR;
63583cca 657 }
63583cca
HR
658}
659
4ae61c68
DG
660/* Helper for scsi_io_completion() when "reprep" action required. */
661static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
662 struct request_queue *q)
663{
664 /* A new command will be prepared and issued. */
f664a3cc 665 scsi_mq_requeue_cmd(cmd);
4ae61c68
DG
666}
667
da32baea
DG
668/* Helper for scsi_io_completion() when special action required. */
669static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
1da177e4 670{
165125e1 671 struct request_queue *q = cmd->device->request_queue;
1da177e4 672 struct request *req = cmd->request;
da32baea 673 int level = 0;
b60af5b0
AS
674 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
675 ACTION_DELAYED_RETRY} action;
ee60b2c5 676 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
da32baea
DG
677 struct scsi_sense_hdr sshdr;
678 bool sense_valid;
679 bool sense_current = true; /* false implies "deferred sense" */
680 blk_status_t blk_stat;
1da177e4 681
da32baea
DG
682 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
683 if (sense_valid)
684 sense_current = !scsi_sense_is_deferred(&sshdr);
03aba2f7 685
da32baea 686 blk_stat = scsi_result_to_blk_status(cmd, result);
3e695f89 687
b60af5b0
AS
688 if (host_byte(result) == DID_RESET) {
689 /* Third party bus reset or reset for error recovery
690 * reasons. Just retry the command and see what
691 * happens.
692 */
693 action = ACTION_RETRY;
da32baea 694 } else if (sense_valid && sense_current) {
1da177e4
LT
695 switch (sshdr.sense_key) {
696 case UNIT_ATTENTION:
697 if (cmd->device->removable) {
03aba2f7 698 /* Detected disc change. Set a bit
1da177e4
LT
699 * and quietly refuse further access.
700 */
701 cmd->device->changed = 1;
b60af5b0 702 action = ACTION_FAIL;
1da177e4 703 } else {
03aba2f7
LT
704 /* Must have been a power glitch, or a
705 * bus reset. Could not have been a
706 * media change, so we just retry the
b60af5b0 707 * command and see what happens.
03aba2f7 708 */
b60af5b0 709 action = ACTION_RETRY;
1da177e4
LT
710 }
711 break;
712 case ILLEGAL_REQUEST:
03aba2f7
LT
713 /* If we had an ILLEGAL REQUEST returned, then
714 * we may have performed an unsupported
715 * command. The only thing this should be
716 * would be a ten byte read where only a six
717 * byte read was supported. Also, on a system
718 * where READ CAPACITY failed, we may have
719 * read past the end of the disk.
720 */
26a68019
JA
721 if ((cmd->device->use_10_for_rw &&
722 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
1da177e4
LT
723 (cmd->cmnd[0] == READ_10 ||
724 cmd->cmnd[0] == WRITE_10)) {
b60af5b0 725 /* This will issue a new 6-byte command. */
1da177e4 726 cmd->device->use_10_for_rw = 0;
b60af5b0 727 action = ACTION_REPREP;
3e695f89 728 } else if (sshdr.asc == 0x10) /* DIX */ {
3e695f89 729 action = ACTION_FAIL;
da32baea 730 blk_stat = BLK_STS_PROTECTION;
c98a0eb0 731 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
5db44863 732 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
c98a0eb0 733 action = ACTION_FAIL;
da32baea 734 blk_stat = BLK_STS_TARGET;
b60af5b0
AS
735 } else
736 action = ACTION_FAIL;
737 break;
511e44f4 738 case ABORTED_COMMAND:
126c0982 739 action = ACTION_FAIL;
e6c11dbb 740 if (sshdr.asc == 0x10) /* DIF */
da32baea 741 blk_stat = BLK_STS_PROTECTION;
1da177e4
LT
742 break;
743 case NOT_READY:
03aba2f7 744 /* If the device is in the process of becoming
f3e93f73 745 * ready, or has a temporary blockage, retry.
1da177e4 746 */
f3e93f73
JB
747 if (sshdr.asc == 0x04) {
748 switch (sshdr.ascq) {
749 case 0x01: /* becoming ready */
750 case 0x04: /* format in progress */
751 case 0x05: /* rebuild in progress */
752 case 0x06: /* recalculation in progress */
753 case 0x07: /* operation in progress */
754 case 0x08: /* Long write in progress */
755 case 0x09: /* self test in progress */
d8705f11 756 case 0x14: /* space allocation in progress */
e37c7d9a
DG
757 case 0x1a: /* start stop unit in progress */
758 case 0x1b: /* sanitize in progress */
759 case 0x1d: /* configuration in progress */
760 case 0x24: /* depopulation in progress */
b60af5b0 761 action = ACTION_DELAYED_RETRY;
f3e93f73 762 break;
3dbf6a54 763 default:
3dbf6a54
AS
764 action = ACTION_FAIL;
765 break;
f3e93f73 766 }
e6c11dbb 767 } else
b60af5b0 768 action = ACTION_FAIL;
b60af5b0 769 break;
1da177e4 770 case VOLUME_OVERFLOW:
03aba2f7 771 /* See SSC3rXX or current. */
b60af5b0
AS
772 action = ACTION_FAIL;
773 break;
1da177e4 774 default:
b60af5b0 775 action = ACTION_FAIL;
1da177e4
LT
776 break;
777 }
e6c11dbb 778 } else
b60af5b0 779 action = ACTION_FAIL;
b60af5b0 780
ee60b2c5 781 if (action != ACTION_FAIL &&
e6c11dbb 782 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
ee60b2c5 783 action = ACTION_FAIL;
ee60b2c5 784
b60af5b0
AS
785 switch (action) {
786 case ACTION_FAIL:
787 /* Give up and fail the remainder of the request */
e8064021 788 if (!(req->rq_flags & RQF_QUIET)) {
f1569ff1
HR
789 static DEFINE_RATELIMIT_STATE(_rs,
790 DEFAULT_RATELIMIT_INTERVAL,
791 DEFAULT_RATELIMIT_BURST);
792
793 if (unlikely(scsi_logging_level))
da32baea
DG
794 level =
795 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
796 SCSI_LOG_MLCOMPLETE_BITS);
f1569ff1
HR
797
798 /*
799 * if logging is enabled the failure will be printed
800 * in scsi_log_completion(), so avoid duplicate messages
801 */
802 if (!level && __ratelimit(&_rs)) {
803 scsi_print_result(cmd, NULL, FAILED);
c65be1a6 804 if (driver_byte(result) == DRIVER_SENSE)
f1569ff1
HR
805 scsi_print_sense(cmd);
806 scsi_print_command(cmd);
807 }
3173d8c3 808 }
ae3d56d8 809 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
f6d47e74 810 return;
bc85dc50 811 /*FALLTHRU*/
b60af5b0 812 case ACTION_REPREP:
4ae61c68 813 scsi_io_completion_reprep(cmd, q);
b60af5b0
AS
814 break;
815 case ACTION_RETRY:
816 /* Retry the same command immediately */
08640e81 817 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
b60af5b0
AS
818 break;
819 case ACTION_DELAYED_RETRY:
820 /* Retry the same command after a delay */
08640e81 821 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
b60af5b0 822 break;
1da177e4
LT
823 }
824}
1da177e4 825
ab831084
DG
826/*
827 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
828 * new result that may suppress further error checking. Also modifies
829 * *blk_statp in some cases.
830 */
831static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
832 blk_status_t *blk_statp)
833{
834 bool sense_valid;
835 bool sense_current = true; /* false implies "deferred sense" */
836 struct request *req = cmd->request;
837 struct scsi_sense_hdr sshdr;
838
839 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
840 if (sense_valid)
841 sense_current = !scsi_sense_is_deferred(&sshdr);
842
843 if (blk_rq_is_passthrough(req)) {
844 if (sense_valid) {
845 /*
846 * SG_IO wants current and deferred errors
847 */
848 scsi_req(req)->sense_len =
849 min(8 + cmd->sense_buffer[7],
850 SCSI_SENSE_BUFFERSIZE);
851 }
852 if (sense_current)
853 *blk_statp = scsi_result_to_blk_status(cmd, result);
854 } else if (blk_rq_bytes(req) == 0 && sense_current) {
855 /*
856 * Flush commands do not transfers any data, and thus cannot use
857 * good_bytes != blk_rq_bytes(req) as the signal for an error.
858 * This sets *blk_statp explicitly for the problem case.
859 */
860 *blk_statp = scsi_result_to_blk_status(cmd, result);
861 }
862 /*
863 * Recovered errors need reporting, but they're always treated as
864 * success, so fiddle the result code here. For passthrough requests
865 * we already took a copy of the original into sreq->result which
866 * is what gets returned to the user
867 */
868 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
869 bool do_print = true;
870 /*
871 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
872 * skip print since caller wants ATA registers. Only occurs
873 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
874 */
875 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
876 do_print = false;
877 else if (req->rq_flags & RQF_QUIET)
878 do_print = false;
879 if (do_print)
880 scsi_print_sense(cmd);
881 result = 0;
882 /* for passthrough, *blk_statp may be set */
883 *blk_statp = BLK_STS_OK;
884 }
885 /*
886 * Another corner case: the SCSI status byte is non-zero but 'good'.
887 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
888 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
889 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
890 * intermediate statuses (both obsolete in SAM-4) as good.
891 */
892 if (status_byte(result) && scsi_status_is_good(result)) {
893 result = 0;
894 *blk_statp = BLK_STS_OK;
895 }
896 return result;
897}
898
1da177e4
LT
899/*
900 * Function: scsi_io_completion()
901 *
902 * Purpose: Completion processing for block device I/O requests.
903 *
904 * Arguments: cmd - command that is finished.
905 *
906 * Lock status: Assumed that no lock is held upon entry.
907 *
908 * Returns: Nothing
909 *
bc85dc50
CH
910 * Notes: We will finish off the specified number of sectors. If we
911 * are done, the command block will be released and the queue
912 * function will be goosed. If we are not done then we have to
b60af5b0 913 * figure out what to do next:
1da177e4 914 *
b60af5b0
AS
915 * a) We can call scsi_requeue_command(). The request
916 * will be unprepared and put back on the queue. Then
917 * a new command will be created for it. This should
918 * be used if we made forward progress, or if we want
919 * to switch from READ(10) to READ(6) for example.
1da177e4 920 *
bc85dc50 921 * b) We can call __scsi_queue_insert(). The request will
b60af5b0
AS
922 * be put back on the queue and retried using the same
923 * command as before, possibly after a delay.
924 *
1f7cbb8e
DG
925 * c) We can call scsi_end_request() with blk_stat other than
926 * BLK_STS_OK, to fail the remainder of the request.
1da177e4 927 */
03aba2f7 928void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1da177e4
LT
929{
930 int result = cmd->result;
165125e1 931 struct request_queue *q = cmd->device->request_queue;
1da177e4 932 struct request *req = cmd->request;
1f7cbb8e 933 blk_status_t blk_stat = BLK_STS_OK;
1da177e4 934
0d437906 935 if (unlikely(result)) /* a nz result may or may not be an error */
ab831084 936 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
631c228c 937
0d437906 938 if (unlikely(blk_rq_is_passthrough(req))) {
27c41973 939 /*
a77b32d8 940 * scsi_result_to_blk_status may have reset the host_byte
27c41973 941 */
17d5363b 942 scsi_req(req)->result = cmd->result;
8e1695a0 943 }
30b0c37b 944
1da177e4
LT
945 /*
946 * Next deal with any sectors which we were able to correctly
947 * handle.
948 */
91921e01
HR
949 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
950 "%u sectors total, %d bytes done.\n",
951 blk_rq_sectors(req), good_bytes));
d6b0c537 952
a9bddd74 953 /*
1f7cbb8e
DG
954 * Next deal with any sectors which we were able to correctly
955 * handle. Failed, zero length commands always need to drop down
956 * to retry code. Fast path should return in this block.
d6b0c537 957 */
0d437906 958 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
ae3d56d8 959 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1f7cbb8e
DG
960 return; /* no bytes remaining */
961 }
bc85dc50 962
0d437906
DG
963 /* Kill remainder if no retries. */
964 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
ae3d56d8 965 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
8e1695a0
DG
966 WARN_ONCE(true,
967 "Bytes remaining after failed, no-retry command");
f6d47e74 968 return;
bc85dc50
CH
969 }
970
971 /*
972 * If there had been no error, but we have leftover bytes in the
973 * requeues just queue the command up again.
d6b0c537 974 */
0d437906 975 if (likely(result == 0))
4ae61c68
DG
976 scsi_io_completion_reprep(cmd, q);
977 else
da32baea 978 scsi_io_completion_action(cmd, result);
1da177e4 979}
1da177e4 980
159b2cbf
CH
981static blk_status_t scsi_init_sgtable(struct request *req,
982 struct scsi_data_buffer *sdb)
1da177e4 983{
6f9a35e2 984 int count;
1da177e4
LT
985
986 /*
3b003157 987 * If sg table allocation fails, requeue request later.
1da177e4 988 */
f9d03f96 989 if (unlikely(sg_alloc_table_chained(&sdb->table,
4635873c 990 blk_rq_nr_phys_segments(req), sdb->table.sgl,
3dccdf53 991 SCSI_INLINE_SG_CNT)))
159b2cbf 992 return BLK_STS_RESOURCE;
1da177e4 993
1da177e4
LT
994 /*
995 * Next, walk the list, and fill in the addresses and sizes of
996 * each segment.
997 */
30b0c37b
BH
998 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
999 BUG_ON(count > sdb->table.nents);
1000 sdb->table.nents = count;
fd102b12 1001 sdb->length = blk_rq_payload_bytes(req);
159b2cbf 1002 return BLK_STS_OK;
1da177e4 1003}
6f9a35e2
BH
1004
1005/*
1006 * Function: scsi_init_io()
1007 *
1008 * Purpose: SCSI I/O initialize function.
1009 *
1010 * Arguments: cmd - Command descriptor we wish to initialize
1011 *
159b2cbf
CH
1012 * Returns: BLK_STS_OK on success
1013 * BLK_STS_RESOURCE if the failure is retryable
1014 * BLK_STS_IOERR if the failure is fatal
6f9a35e2 1015 */
159b2cbf 1016blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
6f9a35e2 1017{
13f05c8d 1018 struct request *rq = cmd->request;
159b2cbf 1019 blk_status_t ret;
13f05c8d 1020
fd3fc0b4 1021 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
159b2cbf 1022 return BLK_STS_IOERR;
635d98b1 1023
159b2cbf
CH
1024 ret = scsi_init_sgtable(rq, &cmd->sdb);
1025 if (ret)
1026 return ret;
6f9a35e2 1027
13f05c8d 1028 if (blk_integrity_rq(rq)) {
7027ad72
MP
1029 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1030 int ivecs, count;
1031
14784565 1032 if (WARN_ON_ONCE(!prot_sdb)) {
91724c20
EM
1033 /*
1034 * This can happen if someone (e.g. multipath)
1035 * queues a command to a device on an adapter
1036 * that does not support DIX.
1037 */
159b2cbf 1038 ret = BLK_STS_IOERR;
14784565 1039 goto out_free_sgtables;
91724c20
EM
1040 }
1041
13f05c8d 1042 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
7027ad72 1043
001d63be 1044 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
4635873c 1045 prot_sdb->table.sgl,
92524fa1 1046 SCSI_INLINE_PROT_SG_CNT)) {
159b2cbf 1047 ret = BLK_STS_RESOURCE;
14784565 1048 goto out_free_sgtables;
7027ad72
MP
1049 }
1050
13f05c8d 1051 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
7027ad72 1052 prot_sdb->table.sgl);
6f1d8a53
IS
1053 BUG_ON(count > ivecs);
1054 BUG_ON(count > queue_max_integrity_segments(rq->q));
7027ad72
MP
1055
1056 cmd->prot_sdb = prot_sdb;
1057 cmd->prot_sdb->table.nents = count;
1058 }
1059
159b2cbf 1060 return BLK_STS_OK;
14784565 1061out_free_sgtables:
f664a3cc 1062 scsi_mq_free_sgtables(cmd);
159b2cbf 1063 return ret;
6f9a35e2 1064}
bb52d82f 1065EXPORT_SYMBOL(scsi_init_io);
1da177e4 1066
ca18d6f7 1067/**
832889f5 1068 * scsi_initialize_rq - initialize struct scsi_cmnd partially
35c0506f 1069 * @rq: Request associated with the SCSI command to be initialized.
ca18d6f7 1070 *
832889f5
BVA
1071 * This function initializes the members of struct scsi_cmnd that must be
1072 * initialized before request processing starts and that won't be
1073 * reinitialized if a SCSI command is requeued.
1074 *
64104f70
BVA
1075 * Called from inside blk_get_request() for pass-through requests and from
1076 * inside scsi_init_command() for filesystem requests.
ca18d6f7 1077 */
e4c9470b 1078static void scsi_initialize_rq(struct request *rq)
ca18d6f7 1079{
c8d9cf22
BVA
1080 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1081
1082 scsi_req_init(&cmd->req);
3be8828f 1083 init_rcu_head(&cmd->rcu);
832889f5
BVA
1084 cmd->jiffies_at_alloc = jiffies;
1085 cmd->retries = 0;
ca18d6f7 1086}
ca18d6f7 1087
b7e9e1fb
ML
1088/*
1089 * Only called when the request isn't completed by SCSI, and not freed by
1090 * SCSI
1091 */
1092static void scsi_cleanup_rq(struct request *rq)
1093{
1094 if (rq->rq_flags & RQF_DONTPREP) {
1095 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1096 rq->rq_flags &= ~RQF_DONTPREP;
1097 }
1098}
1099
65ca846a 1100/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
e9c787e6 1101void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
3b003157 1102{
e9c787e6
CH
1103 void *buf = cmd->sense_buffer;
1104 void *prot = cmd->prot_sdb;
64104f70
BVA
1105 struct request *rq = blk_mq_rq_from_pdu(cmd);
1106 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
832889f5 1107 unsigned long jiffies_at_alloc;
65ca846a 1108 int retries, to_clear;
6eb045e0 1109 bool in_flight;
64104f70
BVA
1110
1111 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1112 flags |= SCMD_INITIALIZED;
1113 scsi_initialize_rq(rq);
1114 }
3b003157 1115
832889f5
BVA
1116 jiffies_at_alloc = cmd->jiffies_at_alloc;
1117 retries = cmd->retries;
6eb045e0 1118 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
65ca846a
BVA
1119 /*
1120 * Zero out the cmd, except for the embedded scsi_request. Only clear
1121 * the driver-private command data if the LLD does not supply a
1122 * function to initialize that data.
1123 */
1124 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1125 if (!dev->host->hostt->init_cmd_priv)
1126 to_clear += dev->host->hostt->cmd_size;
1127 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
3b003157 1128
e9c787e6
CH
1129 cmd->device = dev;
1130 cmd->sense_buffer = buf;
1131 cmd->prot_sdb = prot;
64104f70 1132 cmd->flags = flags;
e9c787e6 1133 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
832889f5
BVA
1134 cmd->jiffies_at_alloc = jiffies_at_alloc;
1135 cmd->retries = retries;
6eb045e0
ML
1136 if (in_flight)
1137 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
64a87b24 1138
3b003157
CH
1139}
1140
785ba83b
CH
1141static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1142 struct request *req)
7b16318d 1143{
bed2213d 1144 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
3b003157
CH
1145
1146 /*
aebf526b 1147 * Passthrough requests may transfer data, in which case they must
3b003157
CH
1148 * a bio attached to them. Or they might contain a SCSI command
1149 * that does not transfer data, in which case they may optionally
1150 * submit a request without an attached bio.
1151 */
1152 if (req->bio) {
159b2cbf
CH
1153 blk_status_t ret = scsi_init_io(cmd);
1154 if (unlikely(ret != BLK_STS_OK))
1155 return ret;
3b003157 1156 } else {
b0790410 1157 BUG_ON(blk_rq_bytes(req));
3b003157 1158
30b0c37b 1159 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
3b003157 1160 }
7b16318d 1161
82ed4db4
CH
1162 cmd->cmd_len = scsi_req(req)->cmd_len;
1163 cmd->cmnd = scsi_req(req)->cmd;
b0790410 1164 cmd->transfersize = blk_rq_bytes(req);
64c7f1d1 1165 cmd->allowed = scsi_req(req)->retries;
785ba83b 1166 return BLK_STS_OK;
7b16318d 1167}
7b16318d 1168
3b003157 1169/*
aebf526b 1170 * Setup a normal block command. These are simple request from filesystems
3868cf8e 1171 * that still need to be translated to SCSI CDBs from the ULD.
3b003157 1172 */
785ba83b
CH
1173static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1174 struct request *req)
1da177e4 1175{
bed2213d 1176 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
a6a8d9f8 1177
ee14c674 1178 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
4c1cb67c
CH
1179 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1180 if (ret != BLK_STS_OK)
1181 return ret;
a6a8d9f8
CS
1182 }
1183
82ed4db4 1184 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
64a87b24 1185 memset(cmd->cmnd, 0, BLK_MAX_CDB);
159b2cbf 1186 return scsi_cmd_to_driver(cmd)->init_command(cmd);
3b003157
CH
1187}
1188
785ba83b
CH
1189static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1190 struct request *req)
6af7a4ff 1191{
bed2213d 1192 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
6af7a4ff
CH
1193
1194 if (!blk_rq_bytes(req))
1195 cmd->sc_data_direction = DMA_NONE;
1196 else if (rq_data_dir(req) == WRITE)
1197 cmd->sc_data_direction = DMA_TO_DEVICE;
1198 else
1199 cmd->sc_data_direction = DMA_FROM_DEVICE;
1200
aebf526b
CH
1201 if (blk_rq_is_scsi(req))
1202 return scsi_setup_scsi_cmnd(sdev, req);
1203 else
6af7a4ff 1204 return scsi_setup_fs_cmnd(sdev, req);
6af7a4ff
CH
1205}
1206
c092d4ec 1207static blk_status_t
a1b73fc1 1208scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
3b003157 1209{
c092d4ec
CH
1210 switch (sdev->sdev_state) {
1211 case SDEV_OFFLINE:
1212 case SDEV_TRANSPORT_OFFLINE:
1213 /*
1214 * If the device is offline we refuse to process any
1215 * commands. The device must be brought online
1216 * before trying any recovery commands.
1217 */
b0962c53
EM
1218 if (!sdev->offline_already) {
1219 sdev->offline_already = true;
1220 sdev_printk(KERN_ERR, sdev,
1221 "rejecting I/O to offline device\n");
1222 }
c092d4ec
CH
1223 return BLK_STS_IOERR;
1224 case SDEV_DEL:
1225 /*
1226 * If the device is fully deleted, we refuse to
1227 * process any commands as well.
1228 */
1229 sdev_printk(KERN_ERR, sdev,
1230 "rejecting I/O to dead device\n");
1231 return BLK_STS_IOERR;
1232 case SDEV_BLOCK:
1233 case SDEV_CREATED_BLOCK:
1234 return BLK_STS_RESOURCE;
1235 case SDEV_QUIESCE:
1236 /*
1237 * If the devices is blocked we defer normal commands.
1238 */
1239 if (req && !(req->rq_flags & RQF_PREEMPT))
1240 return BLK_STS_RESOURCE;
1241 return BLK_STS_OK;
1242 default:
1243 /*
1244 * For any other not fully online state we only allow
1245 * special commands. In particular any user initiated
1246 * command is not allowed.
1247 */
1248 if (req && !(req->rq_flags & RQF_PREEMPT))
1249 return BLK_STS_IOERR;
1250 return BLK_STS_OK;
1da177e4 1251 }
7f9a6bc4 1252}
1da177e4 1253
1da177e4
LT
1254/*
1255 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1256 * return 0.
1257 *
1258 * Called with the queue_lock held.
1259 */
1260static inline int scsi_dev_queue_ready(struct request_queue *q,
1261 struct scsi_device *sdev)
1262{
71e75c97
CH
1263 unsigned int busy;
1264
1265 busy = atomic_inc_return(&sdev->device_busy) - 1;
cd9070c9 1266 if (atomic_read(&sdev->device_blocked)) {
71e75c97
CH
1267 if (busy)
1268 goto out_dec;
1269
1da177e4
LT
1270 /*
1271 * unblock after device_blocked iterates to zero
1272 */
f664a3cc 1273 if (atomic_dec_return(&sdev->device_blocked) > 0)
71e75c97 1274 goto out_dec;
71e75c97
CH
1275 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1276 "unblocking device at zero depth\n"));
1da177e4 1277 }
71e75c97
CH
1278
1279 if (busy >= sdev->queue_depth)
1280 goto out_dec;
1da177e4
LT
1281
1282 return 1;
71e75c97
CH
1283out_dec:
1284 atomic_dec(&sdev->device_busy);
1285 return 0;
1da177e4
LT
1286}
1287
f0c0a376
MC
1288/*
1289 * scsi_target_queue_ready: checks if there we can send commands to target
1290 * @sdev: scsi device on starget to check.
f0c0a376
MC
1291 */
1292static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1293 struct scsi_device *sdev)
1294{
1295 struct scsi_target *starget = scsi_target(sdev);
7ae65c0f 1296 unsigned int busy;
f0c0a376
MC
1297
1298 if (starget->single_lun) {
7ae65c0f 1299 spin_lock_irq(shost->host_lock);
f0c0a376 1300 if (starget->starget_sdev_user &&
7ae65c0f
CH
1301 starget->starget_sdev_user != sdev) {
1302 spin_unlock_irq(shost->host_lock);
1303 return 0;
1304 }
f0c0a376 1305 starget->starget_sdev_user = sdev;
7ae65c0f 1306 spin_unlock_irq(shost->host_lock);
f0c0a376
MC
1307 }
1308
2ccbb008
CH
1309 if (starget->can_queue <= 0)
1310 return 1;
1311
7ae65c0f 1312 busy = atomic_inc_return(&starget->target_busy) - 1;
cd9070c9 1313 if (atomic_read(&starget->target_blocked) > 0) {
7ae65c0f
CH
1314 if (busy)
1315 goto starved;
1316
f0c0a376
MC
1317 /*
1318 * unblock after target_blocked iterates to zero
1319 */
cd9070c9 1320 if (atomic_dec_return(&starget->target_blocked) > 0)
7ae65c0f 1321 goto out_dec;
cf68d334
CH
1322
1323 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1324 "unblocking target at zero depth\n"));
f0c0a376
MC
1325 }
1326
2ccbb008 1327 if (busy >= starget->can_queue)
7ae65c0f 1328 goto starved;
f0c0a376 1329
7ae65c0f
CH
1330 return 1;
1331
1332starved:
1333 spin_lock_irq(shost->host_lock);
1334 list_move_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1335 spin_unlock_irq(shost->host_lock);
7ae65c0f 1336out_dec:
2ccbb008
CH
1337 if (starget->can_queue > 0)
1338 atomic_dec(&starget->target_busy);
7ae65c0f 1339 return 0;
f0c0a376
MC
1340}
1341
1da177e4
LT
1342/*
1343 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1344 * return 0. We must end up running the queue again whenever 0 is
1345 * returned, else IO can hang.
1da177e4
LT
1346 */
1347static inline int scsi_host_queue_ready(struct request_queue *q,
1348 struct Scsi_Host *shost,
6eb045e0
ML
1349 struct scsi_device *sdev,
1350 struct scsi_cmnd *cmd)
1da177e4 1351{
939647ee 1352 if (scsi_host_in_recovery(shost))
74665016
CH
1353 return 0;
1354
cd9070c9 1355 if (atomic_read(&shost->host_blocked) > 0) {
6eb045e0 1356 if (scsi_host_busy(shost) > 0)
74665016
CH
1357 goto starved;
1358
1da177e4
LT
1359 /*
1360 * unblock after host_blocked iterates to zero
1361 */
cd9070c9 1362 if (atomic_dec_return(&shost->host_blocked) > 0)
74665016 1363 goto out_dec;
cf68d334
CH
1364
1365 SCSI_LOG_MLQUEUE(3,
1366 shost_printk(KERN_INFO, shost,
1367 "unblocking host at zero depth\n"));
1da177e4 1368 }
74665016 1369
74665016
CH
1370 if (shost->host_self_blocked)
1371 goto starved;
1da177e4
LT
1372
1373 /* We're OK to process the command, so we can't be starved */
74665016
CH
1374 if (!list_empty(&sdev->starved_entry)) {
1375 spin_lock_irq(shost->host_lock);
1376 if (!list_empty(&sdev->starved_entry))
1377 list_del_init(&sdev->starved_entry);
1378 spin_unlock_irq(shost->host_lock);
1379 }
1da177e4 1380
6eb045e0
ML
1381 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1382
74665016
CH
1383 return 1;
1384
1385starved:
1386 spin_lock_irq(shost->host_lock);
1387 if (list_empty(&sdev->starved_entry))
1388 list_add_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1389 spin_unlock_irq(shost->host_lock);
74665016 1390out_dec:
6eb045e0 1391 scsi_dec_host_busy(shost, cmd);
74665016 1392 return 0;
1da177e4
LT
1393}
1394
6c5121b7
KU
1395/*
1396 * Busy state exporting function for request stacking drivers.
1397 *
1398 * For efficiency, no lock is taken to check the busy state of
1399 * shost/starget/sdev, since the returned value is not guaranteed and
1400 * may be changed after request stacking drivers call the function,
1401 * regardless of taking lock or not.
1402 *
67bd9413
BVA
1403 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1404 * needs to return 'not busy'. Otherwise, request stacking drivers
1405 * may hold requests forever.
6c5121b7 1406 */
f664a3cc 1407static bool scsi_mq_lld_busy(struct request_queue *q)
6c5121b7
KU
1408{
1409 struct scsi_device *sdev = q->queuedata;
1410 struct Scsi_Host *shost;
6c5121b7 1411
3f3299d5 1412 if (blk_queue_dying(q))
f664a3cc 1413 return false;
6c5121b7
KU
1414
1415 shost = sdev->host;
6c5121b7 1416
b7e94a16
JN
1417 /*
1418 * Ignore host/starget busy state.
1419 * Since block layer does not have a concept of fairness across
1420 * multiple queues, congestion of host/starget needs to be handled
1421 * in SCSI layer.
1422 */
1423 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
f664a3cc 1424 return true;
e36e0c80 1425
f664a3cc 1426 return false;
1da177e4
LT
1427}
1428
1aea6434
JA
1429static void scsi_softirq_done(struct request *rq)
1430{
bed2213d 1431 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
242f9dcb 1432 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1aea6434
JA
1433 int disposition;
1434
1435 INIT_LIST_HEAD(&cmd->eh_entry);
1436
242f9dcb
JA
1437 atomic_inc(&cmd->device->iodone_cnt);
1438 if (cmd->result)
1439 atomic_inc(&cmd->device->ioerr_cnt);
1440
1aea6434
JA
1441 disposition = scsi_decide_disposition(cmd);
1442 if (disposition != SUCCESS &&
1443 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
463cdad8 1444 scmd_printk(KERN_ERR, cmd,
1aea6434
JA
1445 "timing out command, waited %lus\n",
1446 wait_for/HZ);
1447 disposition = SUCCESS;
1448 }
91921e01 1449
1aea6434
JA
1450 scsi_log_completion(cmd, disposition);
1451
1452 switch (disposition) {
1453 case SUCCESS:
1454 scsi_finish_command(cmd);
1455 break;
1456 case NEEDS_RETRY:
596f482a 1457 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1aea6434
JA
1458 break;
1459 case ADD_TO_MLQUEUE:
1460 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1461 break;
1462 default:
a0658632 1463 scsi_eh_scmd_add(cmd);
2171b6d0 1464 break;
1aea6434
JA
1465 }
1466}
1467
82042a2c
CH
1468/**
1469 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1470 * @cmd: command block we are dispatching.
1471 *
1472 * Return: nonzero return request was rejected and device's queue needs to be
1473 * plugged.
1474 */
1475static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1476{
1477 struct Scsi_Host *host = cmd->device->host;
1478 int rtn = 0;
1479
1480 atomic_inc(&cmd->device->iorequest_cnt);
1481
1482 /* check if the device is still usable */
1483 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1484 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1485 * returns an immediate error upwards, and signals
1486 * that the device is no longer present */
1487 cmd->result = DID_NO_CONNECT << 16;
1488 goto done;
1489 }
1490
1491 /* Check to see if the scsi lld made this device blocked. */
1492 if (unlikely(scsi_device_blocked(cmd->device))) {
1493 /*
1494 * in blocked state, the command is just put back on
1495 * the device queue. The suspend state has already
1496 * blocked the queue so future requests should not
1497 * occur until the device transitions out of the
1498 * suspend state.
1499 */
1500 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1501 "queuecommand : device blocked\n"));
1502 return SCSI_MLQUEUE_DEVICE_BUSY;
1503 }
1504
1505 /* Store the LUN value in cmnd, if needed. */
1506 if (cmd->device->lun_in_cdb)
1507 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1508 (cmd->device->lun << 5 & 0xe0);
1509
1510 scsi_log_send(cmd);
1511
1512 /*
1513 * Before we queue this command, check if the command
1514 * length exceeds what the host adapter can handle.
1515 */
1516 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1517 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1518 "queuecommand : command too long. "
1519 "cdb_size=%d host->max_cmd_len=%d\n",
1520 cmd->cmd_len, cmd->device->host->max_cmd_len));
1521 cmd->result = (DID_ABORT << 16);
1522 goto done;
1523 }
1524
1525 if (unlikely(host->shost_state == SHOST_DEL)) {
1526 cmd->result = (DID_NO_CONNECT << 16);
1527 goto done;
1528
1529 }
1530
1531 trace_scsi_dispatch_cmd_start(cmd);
1532 rtn = host->hostt->queuecommand(host, cmd);
1533 if (rtn) {
1534 trace_scsi_dispatch_cmd_error(cmd, rtn);
1535 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1536 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1537 rtn = SCSI_MLQUEUE_HOST_BUSY;
1538
1539 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1540 "queuecommand : request rejected\n"));
1541 }
1542
1543 return rtn;
1544 done:
1545 cmd->scsi_done(cmd);
1546 return 0;
1547}
1548
be4c186c 1549/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
3dccdf53 1550static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
be4c186c 1551{
3dccdf53 1552 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
be4c186c
BVA
1553 sizeof(struct scatterlist);
1554}
1555
785ba83b 1556static blk_status_t scsi_mq_prep_fn(struct request *req)
d285203c
CH
1557{
1558 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1559 struct scsi_device *sdev = req->q->queuedata;
1560 struct Scsi_Host *shost = sdev->host;
d285203c
CH
1561 struct scatterlist *sg;
1562
08f78436 1563 scsi_init_command(sdev, cmd);
d285203c 1564
d285203c 1565 cmd->request = req;
d285203c 1566 cmd->tag = req->tag;
d285203c
CH
1567 cmd->prot_op = SCSI_PROT_NORMAL;
1568
d285203c
CH
1569 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1570 cmd->sdb.table.sgl = sg;
1571
1572 if (scsi_host_get_prot(shost)) {
d285203c
CH
1573 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1574
1575 cmd->prot_sdb->table.sgl =
1576 (struct scatterlist *)(cmd->prot_sdb + 1);
1577 }
1578
fe052529
CH
1579 blk_mq_start_request(req);
1580
8fe8ffb1 1581 return scsi_setup_cmnd(sdev, req);
d285203c
CH
1582}
1583
1584static void scsi_mq_done(struct scsi_cmnd *cmd)
1585{
f1342709
KB
1586 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1587 return;
d285203c 1588 trace_scsi_dispatch_cmd_done(cmd);
f1342709
KB
1589
1590 /*
1591 * If the block layer didn't complete the request due to a timeout
1592 * injection, scsi must clear its internal completed state so that the
1593 * timeout handler will see it needs to escalate its own error
1594 * recovery.
1595 */
1596 if (unlikely(!blk_mq_complete_request(cmd->request)))
1597 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
d285203c
CH
1598}
1599
0df21c86 1600static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
d285203c 1601{
0df21c86
ML
1602 struct request_queue *q = hctx->queue;
1603 struct scsi_device *sdev = q->queuedata;
0df21c86 1604
0df21c86 1605 atomic_dec(&sdev->device_busy);
0df21c86
ML
1606}
1607
88022d72 1608static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
0df21c86
ML
1609{
1610 struct request_queue *q = hctx->queue;
d285203c 1611 struct scsi_device *sdev = q->queuedata;
d285203c 1612
b4fd63f4 1613 return scsi_dev_queue_ready(q, sdev);
0df21c86
ML
1614}
1615
fc17b653 1616static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
74c45052 1617 const struct blk_mq_queue_data *bd)
d285203c 1618{
74c45052 1619 struct request *req = bd->rq;
d285203c
CH
1620 struct request_queue *q = req->q;
1621 struct scsi_device *sdev = q->queuedata;
1622 struct Scsi_Host *shost = sdev->host;
1623 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
fc17b653 1624 blk_status_t ret;
d285203c
CH
1625 int reason;
1626
c092d4ec
CH
1627 /*
1628 * If the device is not in running state we will reject some or all
1629 * commands.
1630 */
1631 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1632 ret = scsi_prep_state_check(sdev, req);
1633 if (ret != BLK_STS_OK)
1634 goto out_put_budget;
1635 }
d285203c 1636
fc17b653 1637 ret = BLK_STS_RESOURCE;
d285203c 1638 if (!scsi_target_queue_ready(shost, sdev))
826a70a0 1639 goto out_put_budget;
6eb045e0 1640 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
d285203c
CH
1641 goto out_dec_target_busy;
1642
e8064021 1643 if (!(req->rq_flags & RQF_DONTPREP)) {
785ba83b 1644 ret = scsi_mq_prep_fn(req);
fc17b653 1645 if (ret != BLK_STS_OK)
d285203c 1646 goto out_dec_host_busy;
e8064021 1647 req->rq_flags |= RQF_DONTPREP;
fe052529 1648 } else {
cd464d83 1649 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
fe052529 1650 blk_mq_start_request(req);
d285203c
CH
1651 }
1652
8930a6c2 1653 cmd->flags &= SCMD_PRESERVED_FLAGS;
125c99bc
CH
1654 if (sdev->simple_tags)
1655 cmd->flags |= SCMD_TAGGED;
8930a6c2
PB
1656 if (bd->last)
1657 cmd->flags |= SCMD_LAST;
b1dd2aac 1658
d285203c
CH
1659 scsi_init_cmd_errh(cmd);
1660 cmd->scsi_done = scsi_mq_done;
1661
1662 reason = scsi_dispatch_cmd(cmd);
1663 if (reason) {
1664 scsi_set_blocked(cmd, reason);
fc17b653 1665 ret = BLK_STS_RESOURCE;
d285203c
CH
1666 goto out_dec_host_busy;
1667 }
1668
fc17b653 1669 return BLK_STS_OK;
d285203c
CH
1670
1671out_dec_host_busy:
6eb045e0 1672 scsi_dec_host_busy(shost, cmd);
d285203c
CH
1673out_dec_target_busy:
1674 if (scsi_target(sdev)->can_queue > 0)
1675 atomic_dec(&scsi_target(sdev)->target_busy);
0df21c86
ML
1676out_put_budget:
1677 scsi_mq_put_budget(hctx);
d285203c 1678 switch (ret) {
fc17b653
CH
1679 case BLK_STS_OK:
1680 break;
1681 case BLK_STS_RESOURCE:
86ff7c2a
ML
1682 if (atomic_read(&sdev->device_busy) ||
1683 scsi_device_blocked(sdev))
1684 ret = BLK_STS_DEV_RESOURCE;
d285203c 1685 break;
fc17b653 1686 default:
be549d49
JL
1687 if (unlikely(!scsi_device_online(sdev)))
1688 scsi_req(req)->result = DID_NO_CONNECT << 16;
1689 else
1690 scsi_req(req)->result = DID_ERROR << 16;
d285203c 1691 /*
be549d49 1692 * Make sure to release all allocated resources when
d285203c
CH
1693 * we hit an error, as we will never see this command
1694 * again.
1695 */
e8064021 1696 if (req->rq_flags & RQF_DONTPREP)
d285203c
CH
1697 scsi_mq_uninit_cmd(cmd);
1698 break;
d285203c
CH
1699 }
1700 return ret;
1701}
1702
0152fb6b
CH
1703static enum blk_eh_timer_return scsi_timeout(struct request *req,
1704 bool reserved)
1705{
1706 if (reserved)
1707 return BLK_EH_RESET_TIMER;
1708 return scsi_times_out(req);
1709}
1710
e7008ff5
BVA
1711static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1712 unsigned int hctx_idx, unsigned int numa_node)
d285203c 1713{
d6296d39 1714 struct Scsi_Host *shost = set->driver_data;
8e688254 1715 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
d285203c 1716 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
08f78436 1717 struct scatterlist *sg;
65ca846a 1718 int ret = 0;
d285203c 1719
8e688254
BVA
1720 if (unchecked_isa_dma)
1721 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1722 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1723 GFP_KERNEL, numa_node);
d285203c
CH
1724 if (!cmd->sense_buffer)
1725 return -ENOMEM;
82ed4db4 1726 cmd->req.sense = cmd->sense_buffer;
08f78436
BVA
1727
1728 if (scsi_host_get_prot(shost)) {
1729 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1730 shost->hostt->cmd_size;
3dccdf53 1731 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
08f78436
BVA
1732 }
1733
65ca846a
BVA
1734 if (shost->hostt->init_cmd_priv) {
1735 ret = shost->hostt->init_cmd_priv(shost, cmd);
1736 if (ret < 0)
1737 scsi_free_sense_buffer(unchecked_isa_dma,
1738 cmd->sense_buffer);
1739 }
1740
1741 return ret;
d285203c
CH
1742}
1743
e7008ff5
BVA
1744static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1745 unsigned int hctx_idx)
d285203c 1746{
65ca846a 1747 struct Scsi_Host *shost = set->driver_data;
d285203c
CH
1748 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1749
65ca846a
BVA
1750 if (shost->hostt->exit_cmd_priv)
1751 shost->hostt->exit_cmd_priv(shost, cmd);
8e688254
BVA
1752 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1753 cmd->sense_buffer);
d285203c
CH
1754}
1755
2d9c5c20
CH
1756static int scsi_map_queues(struct blk_mq_tag_set *set)
1757{
1758 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1759
1760 if (shost->hostt->map_queues)
1761 return shost->hostt->map_queues(shost);
99bbf484 1762 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
2d9c5c20
CH
1763}
1764
d48777a6 1765void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1da177e4 1766{
6f381fa3 1767 struct device *dev = shost->dma_dev;
1da177e4 1768
a8474ce2
JA
1769 /*
1770 * this limit is imposed by hardware restrictions
1771 */
8a78362c 1772 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
65e8617f 1773 SG_MAX_SEGMENTS));
a8474ce2 1774
13f05c8d
MP
1775 if (scsi_host_prot_dma(shost)) {
1776 shost->sg_prot_tablesize =
1777 min_not_zero(shost->sg_prot_tablesize,
1778 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1779 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1780 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1781 }
1782
1b5d9a6e
CH
1783 if (dev->dma_mask) {
1784 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1785 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1786 }
086fa5ff 1787 blk_queue_max_hw_sectors(q, shost->max_sectors);
21e07dba
CH
1788 if (shost->unchecked_isa_dma)
1789 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1da177e4 1790 blk_queue_segment_boundary(q, shost->dma_boundary);
99c84dbd 1791 dma_set_seg_boundary(dev, shost->dma_boundary);
1da177e4 1792
a8cf59a6 1793 blk_queue_max_segment_size(q, shost->max_segment_size);
7ad388d8
CH
1794 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1795 dma_set_max_seg_size(dev, queue_max_segment_size(q));
465ff318
JB
1796
1797 /*
90addc6b
HC
1798 * Set a reasonable default alignment: The larger of 32-byte (dword),
1799 * which is a common minimum for HBAs, and the minimum DMA alignment,
1800 * which is set by the platform.
1801 *
1802 * Devices that require a bigger alignment can increase it later.
465ff318 1803 */
90addc6b 1804 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
d285203c 1805}
d48777a6 1806EXPORT_SYMBOL_GPL(__scsi_init_queue);
465ff318 1807
8930a6c2
PB
1808static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1809 .get_budget = scsi_mq_get_budget,
1810 .put_budget = scsi_mq_put_budget,
1811 .queue_rq = scsi_queue_rq,
1812 .complete = scsi_softirq_done,
1813 .timeout = scsi_timeout,
1814#ifdef CONFIG_BLK_DEBUG_FS
1815 .show_rq = scsi_show_rq,
1816#endif
1817 .init_request = scsi_mq_init_request,
1818 .exit_request = scsi_mq_exit_request,
1819 .initialize_rq_fn = scsi_initialize_rq,
82a9ac71 1820 .cleanup_rq = scsi_cleanup_rq,
8930a6c2
PB
1821 .busy = scsi_mq_lld_busy,
1822 .map_queues = scsi_map_queues,
1823};
1824
1825
1826static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1827{
1828 struct request_queue *q = hctx->queue;
1829 struct scsi_device *sdev = q->queuedata;
1830 struct Scsi_Host *shost = sdev->host;
1831
1832 shost->hostt->commit_rqs(shost, hctx->queue_num);
1833}
1834
f363b089 1835static const struct blk_mq_ops scsi_mq_ops = {
0df21c86
ML
1836 .get_budget = scsi_mq_get_budget,
1837 .put_budget = scsi_mq_put_budget,
d285203c 1838 .queue_rq = scsi_queue_rq,
8930a6c2 1839 .commit_rqs = scsi_commit_rqs,
d285203c 1840 .complete = scsi_softirq_done,
0152fb6b 1841 .timeout = scsi_timeout,
0eebd005
BVA
1842#ifdef CONFIG_BLK_DEBUG_FS
1843 .show_rq = scsi_show_rq,
1844#endif
e7008ff5
BVA
1845 .init_request = scsi_mq_init_request,
1846 .exit_request = scsi_mq_exit_request,
ca18d6f7 1847 .initialize_rq_fn = scsi_initialize_rq,
b7e9e1fb 1848 .cleanup_rq = scsi_cleanup_rq,
3a7ea2c4 1849 .busy = scsi_mq_lld_busy,
2d9c5c20 1850 .map_queues = scsi_map_queues,
d285203c
CH
1851};
1852
1853struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1854{
1855 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1856 if (IS_ERR(sdev->request_queue))
1857 return NULL;
1858
1859 sdev->request_queue->queuedata = sdev;
1860 __scsi_init_queue(sdev->host, sdev->request_queue);
17cb960f 1861 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
d285203c
CH
1862 return sdev->request_queue;
1863}
1864
1865int scsi_mq_setup_tags(struct Scsi_Host *shost)
1866{
be4c186c 1867 unsigned int cmd_size, sgl_size;
d285203c 1868
9393c8de
MS
1869 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1870 scsi_mq_inline_sgl_size(shost));
d285203c
CH
1871 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1872 if (scsi_host_get_prot(shost))
92524fa1
ML
1873 cmd_size += sizeof(struct scsi_data_buffer) +
1874 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
d285203c
CH
1875
1876 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
8930a6c2
PB
1877 if (shost->hostt->commit_rqs)
1878 shost->tag_set.ops = &scsi_mq_ops;
1879 else
1880 shost->tag_set.ops = &scsi_mq_ops_no_commit;
efec4b90 1881 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
d285203c
CH
1882 shost->tag_set.queue_depth = shost->can_queue;
1883 shost->tag_set.cmd_size = cmd_size;
1884 shost->tag_set.numa_node = NUMA_NO_NODE;
56d18f62 1885 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
24391c0d
SL
1886 shost->tag_set.flags |=
1887 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
d285203c
CH
1888 shost->tag_set.driver_data = shost;
1889
1890 return blk_mq_alloc_tag_set(&shost->tag_set);
1891}
1892
1893void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1894{
1895 blk_mq_free_tag_set(&shost->tag_set);
1896}
1897
857de6e0
HR
1898/**
1899 * scsi_device_from_queue - return sdev associated with a request_queue
1900 * @q: The request queue to return the sdev from
1901 *
1902 * Return the sdev associated with a request queue or NULL if the
1903 * request_queue does not reference a SCSI device.
1904 */
1905struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1906{
1907 struct scsi_device *sdev = NULL;
1908
6b6fa7a5
SM
1909 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1910 q->mq_ops == &scsi_mq_ops)
857de6e0
HR
1911 sdev = q->queuedata;
1912 if (!sdev || !get_device(&sdev->sdev_gendev))
1913 sdev = NULL;
1914
1915 return sdev;
1916}
1917EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1918
1da177e4
LT
1919/*
1920 * Function: scsi_block_requests()
1921 *
1922 * Purpose: Utility function used by low-level drivers to prevent further
1923 * commands from being queued to the device.
1924 *
1925 * Arguments: shost - Host in question
1926 *
1927 * Returns: Nothing
1928 *
1929 * Lock status: No locks are assumed held.
1930 *
1931 * Notes: There is no timer nor any other means by which the requests
1932 * get unblocked other than the low-level driver calling
1933 * scsi_unblock_requests().
1934 */
1935void scsi_block_requests(struct Scsi_Host *shost)
1936{
1937 shost->host_self_blocked = 1;
1938}
1939EXPORT_SYMBOL(scsi_block_requests);
1940
1941/*
1942 * Function: scsi_unblock_requests()
1943 *
1944 * Purpose: Utility function used by low-level drivers to allow further
1945 * commands from being queued to the device.
1946 *
1947 * Arguments: shost - Host in question
1948 *
1949 * Returns: Nothing
1950 *
1951 * Lock status: No locks are assumed held.
1952 *
1953 * Notes: There is no timer nor any other means by which the requests
1954 * get unblocked other than the low-level driver calling
1955 * scsi_unblock_requests().
1956 *
1957 * This is done as an API function so that changes to the
1958 * internals of the scsi mid-layer won't require wholesale
1959 * changes to drivers that use this feature.
1960 */
1961void scsi_unblock_requests(struct Scsi_Host *shost)
1962{
1963 shost->host_self_blocked = 0;
1964 scsi_run_host_queues(shost);
1965}
1966EXPORT_SYMBOL(scsi_unblock_requests);
1967
1968int __init scsi_init_queue(void)
1969{
6362abd3
MP
1970 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1971 sizeof(struct scsi_data_buffer),
1972 0, 0, NULL);
1973 if (!scsi_sdb_cache) {
1974 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
f078727b 1975 return -ENOMEM;
6f9a35e2
BH
1976 }
1977
1da177e4
LT
1978 return 0;
1979}
1980
1981void scsi_exit_queue(void)
1982{
0a6ac4ee
CH
1983 kmem_cache_destroy(scsi_sense_cache);
1984 kmem_cache_destroy(scsi_sense_isadma_cache);
6362abd3 1985 kmem_cache_destroy(scsi_sdb_cache);
1da177e4 1986}
5baba830
JB
1987
1988/**
1989 * scsi_mode_select - issue a mode select
1990 * @sdev: SCSI device to be queried
1991 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1992 * @sp: Save page bit (0 == don't save, 1 == save)
1993 * @modepage: mode page being requested
1994 * @buffer: request buffer (may not be smaller than eight bytes)
1995 * @len: length of request buffer.
1996 * @timeout: command timeout
1997 * @retries: number of retries before failing
1998 * @data: returns a structure abstracting the mode header data
eb44820c 1999 * @sshdr: place to put sense data (or NULL if no sense to be collected).
5baba830
JB
2000 * must be SCSI_SENSE_BUFFERSIZE big.
2001 *
2002 * Returns zero if successful; negative error number or scsi
2003 * status on error
2004 *
2005 */
2006int
2007scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2008 unsigned char *buffer, int len, int timeout, int retries,
2009 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2010{
2011 unsigned char cmd[10];
2012 unsigned char *real_buffer;
2013 int ret;
2014
2015 memset(cmd, 0, sizeof(cmd));
2016 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2017
2018 if (sdev->use_10_for_ms) {
2019 if (len > 65535)
2020 return -EINVAL;
2021 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2022 if (!real_buffer)
2023 return -ENOMEM;
2024 memcpy(real_buffer + 8, buffer, len);
2025 len += 8;
2026 real_buffer[0] = 0;
2027 real_buffer[1] = 0;
2028 real_buffer[2] = data->medium_type;
2029 real_buffer[3] = data->device_specific;
2030 real_buffer[4] = data->longlba ? 0x01 : 0;
2031 real_buffer[5] = 0;
2032 real_buffer[6] = data->block_descriptor_length >> 8;
2033 real_buffer[7] = data->block_descriptor_length;
2034
2035 cmd[0] = MODE_SELECT_10;
2036 cmd[7] = len >> 8;
2037 cmd[8] = len;
2038 } else {
2039 if (len > 255 || data->block_descriptor_length > 255 ||
2040 data->longlba)
2041 return -EINVAL;
2042
2043 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2044 if (!real_buffer)
2045 return -ENOMEM;
2046 memcpy(real_buffer + 4, buffer, len);
2047 len += 4;
2048 real_buffer[0] = 0;
2049 real_buffer[1] = data->medium_type;
2050 real_buffer[2] = data->device_specific;
2051 real_buffer[3] = data->block_descriptor_length;
2052
2053
2054 cmd[0] = MODE_SELECT;
2055 cmd[4] = len;
2056 }
2057
2058 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
f4f4e47e 2059 sshdr, timeout, retries, NULL);
5baba830
JB
2060 kfree(real_buffer);
2061 return ret;
2062}
2063EXPORT_SYMBOL_GPL(scsi_mode_select);
2064
1da177e4 2065/**
eb44820c 2066 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1cf72699 2067 * @sdev: SCSI device to be queried
1da177e4
LT
2068 * @dbd: set if mode sense will allow block descriptors to be returned
2069 * @modepage: mode page being requested
2070 * @buffer: request buffer (may not be smaller than eight bytes)
2071 * @len: length of request buffer.
2072 * @timeout: command timeout
2073 * @retries: number of retries before failing
2074 * @data: returns a structure abstracting the mode header data
eb44820c 2075 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1cf72699 2076 * must be SCSI_SENSE_BUFFERSIZE big.
1da177e4
LT
2077 *
2078 * Returns zero if unsuccessful, or the header offset (either 4
2079 * or 8 depending on whether a six or ten byte command was
2080 * issued) if successful.
eb44820c 2081 */
1da177e4 2082int
1cf72699 2083scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1da177e4 2084 unsigned char *buffer, int len, int timeout, int retries,
5baba830
JB
2085 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2086{
1da177e4
LT
2087 unsigned char cmd[12];
2088 int use_10_for_ms;
2089 int header_length;
0ae80ba9 2090 int result, retry_count = retries;
ea73a9f2 2091 struct scsi_sense_hdr my_sshdr;
1da177e4
LT
2092
2093 memset(data, 0, sizeof(*data));
2094 memset(&cmd[0], 0, 12);
0ec96913
CG
2095
2096 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
1da177e4
LT
2097 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2098 cmd[2] = modepage;
2099
ea73a9f2
JB
2100 /* caller might not be interested in sense, but we need it */
2101 if (!sshdr)
2102 sshdr = &my_sshdr;
2103
1da177e4 2104 retry:
1cf72699 2105 use_10_for_ms = sdev->use_10_for_ms;
1da177e4
LT
2106
2107 if (use_10_for_ms) {
2108 if (len < 8)
2109 len = 8;
2110
2111 cmd[0] = MODE_SENSE_10;
2112 cmd[8] = len;
2113 header_length = 8;
2114 } else {
2115 if (len < 4)
2116 len = 4;
2117
2118 cmd[0] = MODE_SENSE;
2119 cmd[4] = len;
2120 header_length = 4;
2121 }
2122
1da177e4
LT
2123 memset(buffer, 0, len);
2124
1cf72699 2125 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
f4f4e47e 2126 sshdr, timeout, retries, NULL);
1da177e4
LT
2127
2128 /* This code looks awful: what it's doing is making sure an
2129 * ILLEGAL REQUEST sense return identifies the actual command
2130 * byte as the problem. MODE_SENSE commands can return
2131 * ILLEGAL REQUEST if the code page isn't supported */
2132
1cf72699 2133 if (use_10_for_ms && !scsi_status_is_good(result) &&
c65be1a6 2134 driver_byte(result) == DRIVER_SENSE) {
ea73a9f2
JB
2135 if (scsi_sense_valid(sshdr)) {
2136 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2137 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1da177e4
LT
2138 /*
2139 * Invalid command operation code
2140 */
1cf72699 2141 sdev->use_10_for_ms = 0;
1da177e4
LT
2142 goto retry;
2143 }
2144 }
2145 }
2146
1cf72699 2147 if(scsi_status_is_good(result)) {
6d73c851
AV
2148 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2149 (modepage == 6 || modepage == 8))) {
2150 /* Initio breakage? */
2151 header_length = 0;
2152 data->length = 13;
2153 data->medium_type = 0;
2154 data->device_specific = 0;
2155 data->longlba = 0;
2156 data->block_descriptor_length = 0;
2157 } else if(use_10_for_ms) {
1da177e4
LT
2158 data->length = buffer[0]*256 + buffer[1] + 2;
2159 data->medium_type = buffer[2];
2160 data->device_specific = buffer[3];
2161 data->longlba = buffer[4] & 0x01;
2162 data->block_descriptor_length = buffer[6]*256
2163 + buffer[7];
2164 } else {
2165 data->length = buffer[0] + 1;
2166 data->medium_type = buffer[1];
2167 data->device_specific = buffer[2];
2168 data->block_descriptor_length = buffer[3];
2169 }
6d73c851 2170 data->header_length = header_length;
0ae80ba9
HR
2171 } else if ((status_byte(result) == CHECK_CONDITION) &&
2172 scsi_sense_valid(sshdr) &&
2173 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2174 retry_count--;
2175 goto retry;
1da177e4
LT
2176 }
2177
1cf72699 2178 return result;
1da177e4
LT
2179}
2180EXPORT_SYMBOL(scsi_mode_sense);
2181
001aac25
JB
2182/**
2183 * scsi_test_unit_ready - test if unit is ready
2184 * @sdev: scsi device to change the state of.
2185 * @timeout: command timeout
2186 * @retries: number of retries before failing
74a78ebd 2187 * @sshdr: outpout pointer for decoded sense information.
001aac25
JB
2188 *
2189 * Returns zero if unsuccessful or an error if TUR failed. For
9f8a2c23 2190 * removable media, UNIT_ATTENTION sets ->changed flag.
001aac25 2191 **/
1da177e4 2192int
001aac25 2193scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
74a78ebd 2194 struct scsi_sense_hdr *sshdr)
1da177e4 2195{
1da177e4
LT
2196 char cmd[] = {
2197 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2198 };
2199 int result;
001aac25 2200
001aac25
JB
2201 /* try to eat the UNIT_ATTENTION if there are enough retries */
2202 do {
2203 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
9b91fd34 2204 timeout, 1, NULL);
32c356d7
JB
2205 if (sdev->removable && scsi_sense_valid(sshdr) &&
2206 sshdr->sense_key == UNIT_ATTENTION)
2207 sdev->changed = 1;
2208 } while (scsi_sense_valid(sshdr) &&
2209 sshdr->sense_key == UNIT_ATTENTION && --retries);
001aac25 2210
1da177e4
LT
2211 return result;
2212}
2213EXPORT_SYMBOL(scsi_test_unit_ready);
2214
2215/**
eb44820c 2216 * scsi_device_set_state - Take the given device through the device state model.
1da177e4
LT
2217 * @sdev: scsi device to change the state of.
2218 * @state: state to change to.
2219 *
23cb27fd 2220 * Returns zero if successful or an error if the requested
1da177e4 2221 * transition is illegal.
eb44820c 2222 */
1da177e4
LT
2223int
2224scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2225{
2226 enum scsi_device_state oldstate = sdev->sdev_state;
2227
2228 if (state == oldstate)
2229 return 0;
2230
2231 switch (state) {
2232 case SDEV_CREATED:
6f4267e3
JB
2233 switch (oldstate) {
2234 case SDEV_CREATED_BLOCK:
2235 break;
2236 default:
2237 goto illegal;
2238 }
2239 break;
1da177e4
LT
2240
2241 case SDEV_RUNNING:
2242 switch (oldstate) {
2243 case SDEV_CREATED:
2244 case SDEV_OFFLINE:
1b8d2620 2245 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2246 case SDEV_QUIESCE:
2247 case SDEV_BLOCK:
2248 break;
2249 default:
2250 goto illegal;
2251 }
2252 break;
2253
2254 case SDEV_QUIESCE:
2255 switch (oldstate) {
2256 case SDEV_RUNNING:
2257 case SDEV_OFFLINE:
1b8d2620 2258 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2259 break;
2260 default:
2261 goto illegal;
2262 }
2263 break;
2264
2265 case SDEV_OFFLINE:
1b8d2620 2266 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2267 switch (oldstate) {
2268 case SDEV_CREATED:
2269 case SDEV_RUNNING:
2270 case SDEV_QUIESCE:
2271 case SDEV_BLOCK:
2272 break;
2273 default:
2274 goto illegal;
2275 }
2276 break;
2277
2278 case SDEV_BLOCK:
2279 switch (oldstate) {
1da177e4 2280 case SDEV_RUNNING:
6f4267e3 2281 case SDEV_CREATED_BLOCK:
a33e5bfb 2282 case SDEV_OFFLINE:
6f4267e3
JB
2283 break;
2284 default:
2285 goto illegal;
2286 }
2287 break;
2288
2289 case SDEV_CREATED_BLOCK:
2290 switch (oldstate) {
2291 case SDEV_CREATED:
1da177e4
LT
2292 break;
2293 default:
2294 goto illegal;
2295 }
2296 break;
2297
2298 case SDEV_CANCEL:
2299 switch (oldstate) {
2300 case SDEV_CREATED:
2301 case SDEV_RUNNING:
9ea72909 2302 case SDEV_QUIESCE:
1da177e4 2303 case SDEV_OFFLINE:
1b8d2620 2304 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2305 break;
2306 default:
2307 goto illegal;
2308 }
2309 break;
2310
2311 case SDEV_DEL:
2312 switch (oldstate) {
309bd271
BK
2313 case SDEV_CREATED:
2314 case SDEV_RUNNING:
2315 case SDEV_OFFLINE:
1b8d2620 2316 case SDEV_TRANSPORT_OFFLINE:
1da177e4 2317 case SDEV_CANCEL:
255ee932 2318 case SDEV_BLOCK:
0516c08d 2319 case SDEV_CREATED_BLOCK:
1da177e4
LT
2320 break;
2321 default:
2322 goto illegal;
2323 }
2324 break;
2325
2326 }
b0962c53 2327 sdev->offline_already = false;
1da177e4
LT
2328 sdev->sdev_state = state;
2329 return 0;
2330
2331 illegal:
91921e01 2332 SCSI_LOG_ERROR_RECOVERY(1,
9ccfc756 2333 sdev_printk(KERN_ERR, sdev,
91921e01 2334 "Illegal state transition %s->%s",
9ccfc756
JB
2335 scsi_device_state_name(oldstate),
2336 scsi_device_state_name(state))
1da177e4
LT
2337 );
2338 return -EINVAL;
2339}
2340EXPORT_SYMBOL(scsi_device_set_state);
2341
a341cd0f
JG
2342/**
2343 * sdev_evt_emit - emit a single SCSI device uevent
2344 * @sdev: associated SCSI device
2345 * @evt: event to emit
2346 *
2347 * Send a single uevent (scsi_event) to the associated scsi_device.
2348 */
2349static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2350{
2351 int idx = 0;
2352 char *envp[3];
2353
2354 switch (evt->evt_type) {
2355 case SDEV_EVT_MEDIA_CHANGE:
2356 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2357 break;
279afdfe 2358 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
d3d32891 2359 scsi_rescan_device(&sdev->sdev_gendev);
279afdfe
EM
2360 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2361 break;
2362 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2363 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2364 break;
2365 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2366 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2367 break;
2368 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2369 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2370 break;
2371 case SDEV_EVT_LUN_CHANGE_REPORTED:
2372 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2373 break;
14c3e677
HR
2374 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2375 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2376 break;
cf3431bb
HR
2377 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2378 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2379 break;
a341cd0f
JG
2380 default:
2381 /* do nothing */
2382 break;
2383 }
2384
2385 envp[idx++] = NULL;
2386
2387 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2388}
2389
2390/**
2391 * sdev_evt_thread - send a uevent for each scsi event
2392 * @work: work struct for scsi_device
2393 *
2394 * Dispatch queued events to their associated scsi_device kobjects
2395 * as uevents.
2396 */
2397void scsi_evt_thread(struct work_struct *work)
2398{
2399 struct scsi_device *sdev;
279afdfe 2400 enum scsi_device_event evt_type;
a341cd0f
JG
2401 LIST_HEAD(event_list);
2402
2403 sdev = container_of(work, struct scsi_device, event_work);
2404
279afdfe
EM
2405 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2406 if (test_and_clear_bit(evt_type, sdev->pending_events))
2407 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2408
a341cd0f
JG
2409 while (1) {
2410 struct scsi_event *evt;
2411 struct list_head *this, *tmp;
2412 unsigned long flags;
2413
2414 spin_lock_irqsave(&sdev->list_lock, flags);
2415 list_splice_init(&sdev->event_list, &event_list);
2416 spin_unlock_irqrestore(&sdev->list_lock, flags);
2417
2418 if (list_empty(&event_list))
2419 break;
2420
2421 list_for_each_safe(this, tmp, &event_list) {
2422 evt = list_entry(this, struct scsi_event, node);
2423 list_del(&evt->node);
2424 scsi_evt_emit(sdev, evt);
2425 kfree(evt);
2426 }
2427 }
2428}
2429
2430/**
2431 * sdev_evt_send - send asserted event to uevent thread
2432 * @sdev: scsi_device event occurred on
2433 * @evt: event to send
2434 *
2435 * Assert scsi device event asynchronously.
2436 */
2437void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2438{
2439 unsigned long flags;
2440
4d1566ed
KS
2441#if 0
2442 /* FIXME: currently this check eliminates all media change events
2443 * for polled devices. Need to update to discriminate between AN
2444 * and polled events */
a341cd0f
JG
2445 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2446 kfree(evt);
2447 return;
2448 }
4d1566ed 2449#endif
a341cd0f
JG
2450
2451 spin_lock_irqsave(&sdev->list_lock, flags);
2452 list_add_tail(&evt->node, &sdev->event_list);
2453 schedule_work(&sdev->event_work);
2454 spin_unlock_irqrestore(&sdev->list_lock, flags);
2455}
2456EXPORT_SYMBOL_GPL(sdev_evt_send);
2457
2458/**
2459 * sdev_evt_alloc - allocate a new scsi event
2460 * @evt_type: type of event to allocate
2461 * @gfpflags: GFP flags for allocation
2462 *
2463 * Allocates and returns a new scsi_event.
2464 */
2465struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2466 gfp_t gfpflags)
2467{
2468 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2469 if (!evt)
2470 return NULL;
2471
2472 evt->evt_type = evt_type;
2473 INIT_LIST_HEAD(&evt->node);
2474
2475 /* evt_type-specific initialization, if any */
2476 switch (evt_type) {
2477 case SDEV_EVT_MEDIA_CHANGE:
279afdfe
EM
2478 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2479 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2480 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2481 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2482 case SDEV_EVT_LUN_CHANGE_REPORTED:
14c3e677 2483 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
cf3431bb 2484 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
a341cd0f
JG
2485 default:
2486 /* do nothing */
2487 break;
2488 }
2489
2490 return evt;
2491}
2492EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2493
2494/**
2495 * sdev_evt_send_simple - send asserted event to uevent thread
2496 * @sdev: scsi_device event occurred on
2497 * @evt_type: type of event to send
2498 * @gfpflags: GFP flags for allocation
2499 *
2500 * Assert scsi device event asynchronously, given an event type.
2501 */
2502void sdev_evt_send_simple(struct scsi_device *sdev,
2503 enum scsi_device_event evt_type, gfp_t gfpflags)
2504{
2505 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2506 if (!evt) {
2507 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2508 evt_type);
2509 return;
2510 }
2511
2512 sdev_evt_send(sdev, evt);
2513}
2514EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2515
1da177e4
LT
2516/**
2517 * scsi_device_quiesce - Block user issued commands.
2518 * @sdev: scsi device to quiesce.
2519 *
2520 * This works by trying to transition to the SDEV_QUIESCE state
2521 * (which must be a legal transition). When the device is in this
2522 * state, only special requests will be accepted, all others will
2523 * be deferred. Since special requests may also be requeued requests,
2524 * a successful return doesn't guarantee the device will be
2525 * totally quiescent.
2526 *
2527 * Must be called with user context, may sleep.
2528 *
2529 * Returns zero if unsuccessful or an error if not.
eb44820c 2530 */
1da177e4
LT
2531int
2532scsi_device_quiesce(struct scsi_device *sdev)
2533{
3a0a5299 2534 struct request_queue *q = sdev->request_queue;
0db6ca8a
BVA
2535 int err;
2536
3a0a5299
BVA
2537 /*
2538 * It is allowed to call scsi_device_quiesce() multiple times from
2539 * the same context but concurrent scsi_device_quiesce() calls are
2540 * not allowed.
2541 */
2542 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2543
cd84a62e
BVA
2544 if (sdev->quiesced_by == current)
2545 return 0;
2546
2547 blk_set_pm_only(q);
3a0a5299
BVA
2548
2549 blk_mq_freeze_queue(q);
2550 /*
cd84a62e 2551 * Ensure that the effect of blk_set_pm_only() will be visible
3a0a5299
BVA
2552 * for percpu_ref_tryget() callers that occur after the queue
2553 * unfreeze even if the queue was already frozen before this function
2554 * was called. See also https://lwn.net/Articles/573497/.
2555 */
2556 synchronize_rcu();
2557 blk_mq_unfreeze_queue(q);
2558
0db6ca8a
BVA
2559 mutex_lock(&sdev->state_mutex);
2560 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
3a0a5299
BVA
2561 if (err == 0)
2562 sdev->quiesced_by = current;
2563 else
cd84a62e 2564 blk_clear_pm_only(q);
0db6ca8a
BVA
2565 mutex_unlock(&sdev->state_mutex);
2566
3a0a5299 2567 return err;
1da177e4
LT
2568}
2569EXPORT_SYMBOL(scsi_device_quiesce);
2570
2571/**
2572 * scsi_device_resume - Restart user issued commands to a quiesced device.
2573 * @sdev: scsi device to resume.
2574 *
2575 * Moves the device from quiesced back to running and restarts the
2576 * queues.
2577 *
2578 * Must be called with user context, may sleep.
eb44820c 2579 */
a7a20d10 2580void scsi_device_resume(struct scsi_device *sdev)
1da177e4 2581{
a7a20d10
DW
2582 /* check if the device state was mutated prior to resume, and if
2583 * so assume the state is being managed elsewhere (for example
2584 * device deleted during suspend)
2585 */
0db6ca8a 2586 mutex_lock(&sdev->state_mutex);
17605afa
BVA
2587 if (sdev->quiesced_by) {
2588 sdev->quiesced_by = NULL;
2589 blk_clear_pm_only(sdev->request_queue);
2590 }
3a0a5299
BVA
2591 if (sdev->sdev_state == SDEV_QUIESCE)
2592 scsi_device_set_state(sdev, SDEV_RUNNING);
0db6ca8a 2593 mutex_unlock(&sdev->state_mutex);
1da177e4
LT
2594}
2595EXPORT_SYMBOL(scsi_device_resume);
2596
2597static void
2598device_quiesce_fn(struct scsi_device *sdev, void *data)
2599{
2600 scsi_device_quiesce(sdev);
2601}
2602
2603void
2604scsi_target_quiesce(struct scsi_target *starget)
2605{
2606 starget_for_each_device(starget, NULL, device_quiesce_fn);
2607}
2608EXPORT_SYMBOL(scsi_target_quiesce);
2609
2610static void
2611device_resume_fn(struct scsi_device *sdev, void *data)
2612{
2613 scsi_device_resume(sdev);
2614}
2615
2616void
2617scsi_target_resume(struct scsi_target *starget)
2618{
2619 starget_for_each_device(starget, NULL, device_resume_fn);
2620}
2621EXPORT_SYMBOL(scsi_target_resume);
2622
2623/**
551eb598
BVA
2624 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2625 * @sdev: device to block
1da177e4 2626 *
551eb598 2627 * Pause SCSI command processing on the specified device. Does not sleep.
1da177e4 2628 *
551eb598 2629 * Returns zero if successful or a negative error code upon failure.
669f0441 2630 *
551eb598
BVA
2631 * Notes:
2632 * This routine transitions the device to the SDEV_BLOCK state (which must be
2633 * a legal transition). When the device is in this state, command processing
2634 * is paused until the device leaves the SDEV_BLOCK state. See also
2635 * scsi_internal_device_unblock_nowait().
eb44820c 2636 */
551eb598 2637int scsi_internal_device_block_nowait(struct scsi_device *sdev)
1da177e4 2638{
165125e1 2639 struct request_queue *q = sdev->request_queue;
1da177e4
LT
2640 int err = 0;
2641
2642 err = scsi_device_set_state(sdev, SDEV_BLOCK);
6f4267e3
JB
2643 if (err) {
2644 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2645
2646 if (err)
2647 return err;
2648 }
1da177e4
LT
2649
2650 /*
2651 * The device has transitioned to SDEV_BLOCK. Stop the
2652 * block layer from calling the midlayer with this device's
2653 * request queue.
2654 */
f664a3cc 2655 blk_mq_quiesce_queue_nowait(q);
1da177e4
LT
2656 return 0;
2657}
551eb598
BVA
2658EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2659
1da177e4 2660/**
551eb598
BVA
2661 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2662 * @sdev: device to block
2663 *
2664 * Pause SCSI command processing on the specified device and wait until all
2665 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
1da177e4 2666 *
551eb598 2667 * Returns zero if successful or a negative error code upon failure.
1da177e4 2668 *
551eb598
BVA
2669 * Note:
2670 * This routine transitions the device to the SDEV_BLOCK state (which must be
2671 * a legal transition). When the device is in this state, command processing
2672 * is paused until the device leaves the SDEV_BLOCK state. See also
2673 * scsi_internal_device_unblock().
eb44820c 2674 */
551eb598 2675static int scsi_internal_device_block(struct scsi_device *sdev)
1da177e4 2676{
551eb598
BVA
2677 struct request_queue *q = sdev->request_queue;
2678 int err;
2679
0db6ca8a 2680 mutex_lock(&sdev->state_mutex);
551eb598 2681 err = scsi_internal_device_block_nowait(sdev);
f664a3cc
JA
2682 if (err == 0)
2683 blk_mq_quiesce_queue(q);
0db6ca8a
BVA
2684 mutex_unlock(&sdev->state_mutex);
2685
551eb598
BVA
2686 return err;
2687}
1da177e4 2688
66483a4a
BVA
2689void scsi_start_queue(struct scsi_device *sdev)
2690{
2691 struct request_queue *q = sdev->request_queue;
5d9fb5cc 2692
f664a3cc 2693 blk_mq_unquiesce_queue(q);
66483a4a
BVA
2694}
2695
1da177e4 2696/**
43f7571b 2697 * scsi_internal_device_unblock_nowait - resume a device after a block request
1da177e4 2698 * @sdev: device to resume
43f7571b 2699 * @new_state: state to set the device to after unblocking
1da177e4 2700 *
43f7571b
BVA
2701 * Restart the device queue for a previously suspended SCSI device. Does not
2702 * sleep.
1da177e4 2703 *
43f7571b 2704 * Returns zero if successful or a negative error code upon failure.
1da177e4 2705 *
43f7571b
BVA
2706 * Notes:
2707 * This routine transitions the device to the SDEV_RUNNING state or to one of
2708 * the offline states (which must be a legal transition) allowing the midlayer
2709 * to goose the queue for this device.
eb44820c 2710 */
43f7571b
BVA
2711int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2712 enum scsi_device_state new_state)
1da177e4 2713{
09addb1d
BVA
2714 switch (new_state) {
2715 case SDEV_RUNNING:
2716 case SDEV_TRANSPORT_OFFLINE:
2717 break;
2718 default:
2719 return -EINVAL;
2720 }
2721
5d9fb5cc
MC
2722 /*
2723 * Try to transition the scsi device to SDEV_RUNNING or one of the
2724 * offlined states and goose the device queue if successful.
1da177e4 2725 */
8cd1ec78
HR
2726 switch (sdev->sdev_state) {
2727 case SDEV_BLOCK:
2728 case SDEV_TRANSPORT_OFFLINE:
5d9fb5cc 2729 sdev->sdev_state = new_state;
8cd1ec78
HR
2730 break;
2731 case SDEV_CREATED_BLOCK:
5d9fb5cc
MC
2732 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2733 new_state == SDEV_OFFLINE)
2734 sdev->sdev_state = new_state;
2735 else
2736 sdev->sdev_state = SDEV_CREATED;
8cd1ec78
HR
2737 break;
2738 case SDEV_CANCEL:
2739 case SDEV_OFFLINE:
2740 break;
2741 default:
5c10e63c 2742 return -EINVAL;
8cd1ec78 2743 }
66483a4a 2744 scsi_start_queue(sdev);
1da177e4
LT
2745
2746 return 0;
2747}
43f7571b
BVA
2748EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2749
2750/**
2751 * scsi_internal_device_unblock - resume a device after a block request
2752 * @sdev: device to resume
2753 * @new_state: state to set the device to after unblocking
2754 *
2755 * Restart the device queue for a previously suspended SCSI device. May sleep.
2756 *
2757 * Returns zero if successful or a negative error code upon failure.
2758 *
2759 * Notes:
2760 * This routine transitions the device to the SDEV_RUNNING state or to one of
2761 * the offline states (which must be a legal transition) allowing the midlayer
2762 * to goose the queue for this device.
2763 */
2764static int scsi_internal_device_unblock(struct scsi_device *sdev,
2765 enum scsi_device_state new_state)
2766{
0db6ca8a
BVA
2767 int ret;
2768
2769 mutex_lock(&sdev->state_mutex);
2770 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2771 mutex_unlock(&sdev->state_mutex);
2772
2773 return ret;
43f7571b 2774}
1da177e4
LT
2775
2776static void
2777device_block(struct scsi_device *sdev, void *data)
2778{
94ef80a5
BVA
2779 int ret;
2780
2781 ret = scsi_internal_device_block(sdev);
2782
2783 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2784 dev_name(&sdev->sdev_gendev), ret);
1da177e4
LT
2785}
2786
2787static int
2788target_block(struct device *dev, void *data)
2789{
2790 if (scsi_is_target_device(dev))
2791 starget_for_each_device(to_scsi_target(dev), NULL,
2792 device_block);
2793 return 0;
2794}
2795
2796void
2797scsi_target_block(struct device *dev)
2798{
2799 if (scsi_is_target_device(dev))
2800 starget_for_each_device(to_scsi_target(dev), NULL,
2801 device_block);
2802 else
2803 device_for_each_child(dev, NULL, target_block);
2804}
2805EXPORT_SYMBOL_GPL(scsi_target_block);
2806
2807static void
2808device_unblock(struct scsi_device *sdev, void *data)
2809{
5d9fb5cc 2810 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
1da177e4
LT
2811}
2812
2813static int
2814target_unblock(struct device *dev, void *data)
2815{
2816 if (scsi_is_target_device(dev))
5d9fb5cc 2817 starget_for_each_device(to_scsi_target(dev), data,
1da177e4
LT
2818 device_unblock);
2819 return 0;
2820}
2821
2822void
5d9fb5cc 2823scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
1da177e4
LT
2824{
2825 if (scsi_is_target_device(dev))
5d9fb5cc 2826 starget_for_each_device(to_scsi_target(dev), &new_state,
1da177e4
LT
2827 device_unblock);
2828 else
5d9fb5cc 2829 device_for_each_child(dev, &new_state, target_unblock);
1da177e4
LT
2830}
2831EXPORT_SYMBOL_GPL(scsi_target_unblock);
cdb8c2a6 2832
2bb95584
HR
2833int
2834scsi_host_block(struct Scsi_Host *shost)
2835{
2836 struct scsi_device *sdev;
2837 int ret = 0;
2838
2839 shost_for_each_device(sdev, shost) {
2840 ret = scsi_internal_device_block(sdev);
2841 if (ret)
2842 break;
2843 }
2844 return ret;
2845}
2846EXPORT_SYMBOL_GPL(scsi_host_block);
2847
2848int
2849scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2850{
2851 struct scsi_device *sdev;
2852 int ret = 0;
2853
2854 shost_for_each_device(sdev, shost) {
2855 ret = scsi_internal_device_unblock(sdev, new_state);
2856 if (ret)
2857 break;
2858 }
2859 return ret;
2860}
2861EXPORT_SYMBOL_GPL(scsi_host_unblock);
2862
cdb8c2a6
GL
2863/**
2864 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
eb44820c 2865 * @sgl: scatter-gather list
cdb8c2a6
GL
2866 * @sg_count: number of segments in sg
2867 * @offset: offset in bytes into sg, on return offset into the mapped area
2868 * @len: bytes to map, on return number of bytes mapped
2869 *
2870 * Returns virtual address of the start of the mapped page
2871 */
c6132da1 2872void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
cdb8c2a6
GL
2873 size_t *offset, size_t *len)
2874{
2875 int i;
2876 size_t sg_len = 0, len_complete = 0;
c6132da1 2877 struct scatterlist *sg;
cdb8c2a6
GL
2878 struct page *page;
2879
22cfefb5
AM
2880 WARN_ON(!irqs_disabled());
2881
c6132da1 2882 for_each_sg(sgl, sg, sg_count, i) {
cdb8c2a6 2883 len_complete = sg_len; /* Complete sg-entries */
c6132da1 2884 sg_len += sg->length;
cdb8c2a6
GL
2885 if (sg_len > *offset)
2886 break;
2887 }
2888
2889 if (unlikely(i == sg_count)) {
169e1a2a
AM
2890 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2891 "elements %d\n",
cadbd4a5 2892 __func__, sg_len, *offset, sg_count);
cdb8c2a6
GL
2893 WARN_ON(1);
2894 return NULL;
2895 }
2896
2897 /* Offset starting from the beginning of first page in this sg-entry */
c6132da1 2898 *offset = *offset - len_complete + sg->offset;
cdb8c2a6
GL
2899
2900 /* Assumption: contiguous pages can be accessed as "page + i" */
45711f1a 2901 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
cdb8c2a6
GL
2902 *offset &= ~PAGE_MASK;
2903
2904 /* Bytes in this sg-entry from *offset to the end of the page */
2905 sg_len = PAGE_SIZE - *offset;
2906 if (*len > sg_len)
2907 *len = sg_len;
2908
77dfce07 2909 return kmap_atomic(page);
cdb8c2a6
GL
2910}
2911EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2912
2913/**
eb44820c 2914 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
cdb8c2a6
GL
2915 * @virt: virtual address to be unmapped
2916 */
2917void scsi_kunmap_atomic_sg(void *virt)
2918{
77dfce07 2919 kunmap_atomic(virt);
cdb8c2a6
GL
2920}
2921EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
6f4c827e
AL
2922
2923void sdev_disable_disk_events(struct scsi_device *sdev)
2924{
2925 atomic_inc(&sdev->disk_events_disable_depth);
2926}
2927EXPORT_SYMBOL(sdev_disable_disk_events);
2928
2929void sdev_enable_disk_events(struct scsi_device *sdev)
2930{
2931 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2932 return;
2933 atomic_dec(&sdev->disk_events_disable_depth);
2934}
2935EXPORT_SYMBOL(sdev_enable_disk_events);
9983bed3
HR
2936
2937/**
2938 * scsi_vpd_lun_id - return a unique device identification
2939 * @sdev: SCSI device
2940 * @id: buffer for the identification
2941 * @id_len: length of the buffer
2942 *
2943 * Copies a unique device identification into @id based
2944 * on the information in the VPD page 0x83 of the device.
2945 * The string will be formatted as a SCSI name string.
2946 *
2947 * Returns the length of the identification or error on failure.
2948 * If the identifier is longer than the supplied buffer the actual
2949 * identifier length is returned and the buffer is not zero-padded.
2950 */
2951int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2952{
2953 u8 cur_id_type = 0xff;
2954 u8 cur_id_size = 0;
ccf1e004
BVA
2955 const unsigned char *d, *cur_id_str;
2956 const struct scsi_vpd *vpd_pg83;
9983bed3
HR
2957 int id_size = -EINVAL;
2958
2959 rcu_read_lock();
2960 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2961 if (!vpd_pg83) {
2962 rcu_read_unlock();
2963 return -ENXIO;
2964 }
2965
2966 /*
2967 * Look for the correct descriptor.
2968 * Order of preference for lun descriptor:
2969 * - SCSI name string
2970 * - NAA IEEE Registered Extended
2971 * - EUI-64 based 16-byte
2972 * - EUI-64 based 12-byte
2973 * - NAA IEEE Registered
2974 * - NAA IEEE Extended
d230823a 2975 * - T10 Vendor ID
9983bed3
HR
2976 * as longer descriptors reduce the likelyhood
2977 * of identification clashes.
2978 */
2979
2980 /* The id string must be at least 20 bytes + terminating NULL byte */
2981 if (id_len < 21) {
2982 rcu_read_unlock();
2983 return -EINVAL;
2984 }
2985
2986 memset(id, 0, id_len);
ccf1e004
BVA
2987 d = vpd_pg83->data + 4;
2988 while (d < vpd_pg83->data + vpd_pg83->len) {
9983bed3
HR
2989 /* Skip designators not referring to the LUN */
2990 if ((d[1] & 0x30) != 0x00)
2991 goto next_desig;
2992
2993 switch (d[1] & 0xf) {
d230823a
HR
2994 case 0x1:
2995 /* T10 Vendor ID */
2996 if (cur_id_size > d[3])
2997 break;
2998 /* Prefer anything */
2999 if (cur_id_type > 0x01 && cur_id_type != 0xff)
3000 break;
3001 cur_id_size = d[3];
3002 if (cur_id_size + 4 > id_len)
3003 cur_id_size = id_len - 4;
3004 cur_id_str = d + 4;
3005 cur_id_type = d[1] & 0xf;
3006 id_size = snprintf(id, id_len, "t10.%*pE",
3007 cur_id_size, cur_id_str);
3008 break;
9983bed3
HR
3009 case 0x2:
3010 /* EUI-64 */
3011 if (cur_id_size > d[3])
3012 break;
3013 /* Prefer NAA IEEE Registered Extended */
3014 if (cur_id_type == 0x3 &&
3015 cur_id_size == d[3])
3016 break;
3017 cur_id_size = d[3];
3018 cur_id_str = d + 4;
3019 cur_id_type = d[1] & 0xf;
3020 switch (cur_id_size) {
3021 case 8:
3022 id_size = snprintf(id, id_len,
3023 "eui.%8phN",
3024 cur_id_str);
3025 break;
3026 case 12:
3027 id_size = snprintf(id, id_len,
3028 "eui.%12phN",
3029 cur_id_str);
3030 break;
3031 case 16:
3032 id_size = snprintf(id, id_len,
3033 "eui.%16phN",
3034 cur_id_str);
3035 break;
3036 default:
3037 cur_id_size = 0;
3038 break;
3039 }
3040 break;
3041 case 0x3:
3042 /* NAA */
3043 if (cur_id_size > d[3])
3044 break;
3045 cur_id_size = d[3];
3046 cur_id_str = d + 4;
3047 cur_id_type = d[1] & 0xf;
3048 switch (cur_id_size) {
3049 case 8:
3050 id_size = snprintf(id, id_len,
3051 "naa.%8phN",
3052 cur_id_str);
3053 break;
3054 case 16:
3055 id_size = snprintf(id, id_len,
3056 "naa.%16phN",
3057 cur_id_str);
3058 break;
3059 default:
3060 cur_id_size = 0;
3061 break;
3062 }
3063 break;
3064 case 0x8:
3065 /* SCSI name string */
3066 if (cur_id_size + 4 > d[3])
3067 break;
3068 /* Prefer others for truncated descriptor */
3069 if (cur_id_size && d[3] > id_len)
3070 break;
3071 cur_id_size = id_size = d[3];
3072 cur_id_str = d + 4;
3073 cur_id_type = d[1] & 0xf;
3074 if (cur_id_size >= id_len)
3075 cur_id_size = id_len - 1;
3076 memcpy(id, cur_id_str, cur_id_size);
3077 /* Decrease priority for truncated descriptor */
3078 if (cur_id_size != id_size)
3079 cur_id_size = 6;
3080 break;
3081 default:
3082 break;
3083 }
3084next_desig:
3085 d += d[3] + 4;
3086 }
3087 rcu_read_unlock();
3088
3089 return id_size;
3090}
3091EXPORT_SYMBOL(scsi_vpd_lun_id);
a8aa3978
HR
3092
3093/*
3094 * scsi_vpd_tpg_id - return a target port group identifier
3095 * @sdev: SCSI device
3096 *
3097 * Returns the Target Port Group identifier from the information
3098 * froom VPD page 0x83 of the device.
3099 *
3100 * Returns the identifier or error on failure.
3101 */
3102int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3103{
ccf1e004
BVA
3104 const unsigned char *d;
3105 const struct scsi_vpd *vpd_pg83;
a8aa3978
HR
3106 int group_id = -EAGAIN, rel_port = -1;
3107
3108 rcu_read_lock();
3109 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3110 if (!vpd_pg83) {
3111 rcu_read_unlock();
3112 return -ENXIO;
3113 }
3114
ccf1e004
BVA
3115 d = vpd_pg83->data + 4;
3116 while (d < vpd_pg83->data + vpd_pg83->len) {
a8aa3978
HR
3117 switch (d[1] & 0xf) {
3118 case 0x4:
3119 /* Relative target port */
3120 rel_port = get_unaligned_be16(&d[6]);
3121 break;
3122 case 0x5:
3123 /* Target port group */
3124 group_id = get_unaligned_be16(&d[6]);
3125 break;
3126 default:
3127 break;
3128 }
3129 d += d[3] + 4;
3130 }
3131 rcu_read_unlock();
3132
3133 if (group_id >= 0 && rel_id && rel_port != -1)
3134 *rel_id = rel_port;
3135
3136 return group_id;
3137}
3138EXPORT_SYMBOL(scsi_vpd_tpg_id);