]> git.ipfire.org Git - thirdparty/linux.git/blame - drivers/scsi/scsi_lib.c
scsi: lib/sg_pool.c: improve APIs for allocating sg pool
[thirdparty/linux.git] / drivers / scsi / scsi_lib.c
CommitLineData
1da177e4 1/*
d285203c
CH
2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
1da177e4
LT
4 *
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
9 */
10
11#include <linux/bio.h>
d3f46f39 12#include <linux/bitops.h>
1da177e4
LT
13#include <linux/blkdev.h>
14#include <linux/completion.h>
15#include <linux/kernel.h>
09703660 16#include <linux/export.h>
1da177e4
LT
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
faead26d 20#include <linux/hardirq.h>
c6132da1 21#include <linux/scatterlist.h>
d285203c 22#include <linux/blk-mq.h>
f1569ff1 23#include <linux/ratelimit.h>
a8aa3978 24#include <asm/unaligned.h>
1da177e4
LT
25
26#include <scsi/scsi.h>
beb40487 27#include <scsi/scsi_cmnd.h>
1da177e4
LT
28#include <scsi/scsi_dbg.h>
29#include <scsi/scsi_device.h>
30#include <scsi/scsi_driver.h>
31#include <scsi/scsi_eh.h>
32#include <scsi/scsi_host.h>
7aa686d3 33#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
ee14c674 34#include <scsi/scsi_dh.h>
1da177e4 35
3b5382c4
CH
36#include <trace/events/scsi.h>
37
0eebd005 38#include "scsi_debugfs.h"
1da177e4
LT
39#include "scsi_priv.h"
40#include "scsi_logging.h"
41
e9c787e6 42static struct kmem_cache *scsi_sdb_cache;
0a6ac4ee
CH
43static struct kmem_cache *scsi_sense_cache;
44static struct kmem_cache *scsi_sense_isadma_cache;
45static DEFINE_MUTEX(scsi_sense_cache_mutex);
1da177e4 46
a45a1f36
BVA
47static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
48
0a6ac4ee 49static inline struct kmem_cache *
8e688254 50scsi_select_sense_cache(bool unchecked_isa_dma)
0a6ac4ee 51{
8e688254 52 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
0a6ac4ee
CH
53}
54
8e688254
BVA
55static void scsi_free_sense_buffer(bool unchecked_isa_dma,
56 unsigned char *sense_buffer)
0a6ac4ee 57{
8e688254
BVA
58 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
59 sense_buffer);
0a6ac4ee
CH
60}
61
8e688254 62static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
e9c787e6 63 gfp_t gfp_mask, int numa_node)
0a6ac4ee 64{
8e688254
BVA
65 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
66 gfp_mask, numa_node);
0a6ac4ee
CH
67}
68
69int scsi_init_sense_cache(struct Scsi_Host *shost)
70{
71 struct kmem_cache *cache;
72 int ret = 0;
73
8e688254 74 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
0a6ac4ee
CH
75 if (cache)
76 return 0;
77
78 mutex_lock(&scsi_sense_cache_mutex);
79 if (shost->unchecked_isa_dma) {
80 scsi_sense_isadma_cache =
81 kmem_cache_create("scsi_sense_cache(DMA)",
0afe76e8
DW
82 SCSI_SENSE_BUFFERSIZE, 0,
83 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
0a6ac4ee
CH
84 if (!scsi_sense_isadma_cache)
85 ret = -ENOMEM;
86 } else {
87 scsi_sense_cache =
0afe76e8
DW
88 kmem_cache_create_usercopy("scsi_sense_cache",
89 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
90 0, SCSI_SENSE_BUFFERSIZE, NULL);
0a6ac4ee
CH
91 if (!scsi_sense_cache)
92 ret = -ENOMEM;
93 }
94
95 mutex_unlock(&scsi_sense_cache_mutex);
96 return ret;
97}
6f9a35e2 98
a488e749
JA
99/*
100 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
101 * not change behaviour from the previous unplug mechanism, experimentation
102 * may prove this needs changing.
103 */
104#define SCSI_QUEUE_DELAY 3
105
de3e8bf3
CH
106static void
107scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
1da177e4
LT
108{
109 struct Scsi_Host *host = cmd->device->host;
110 struct scsi_device *device = cmd->device;
f0c0a376 111 struct scsi_target *starget = scsi_target(device);
1da177e4
LT
112
113 /*
d8c37e7b 114 * Set the appropriate busy bit for the device/host.
1da177e4
LT
115 *
116 * If the host/device isn't busy, assume that something actually
117 * completed, and that we should be able to queue a command now.
118 *
119 * Note that the prior mid-layer assumption that any host could
120 * always queue at least one command is now broken. The mid-layer
121 * will implement a user specifiable stall (see
122 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
123 * if a command is requeued with no other commands outstanding
124 * either for the device or for the host.
125 */
f0c0a376
MC
126 switch (reason) {
127 case SCSI_MLQUEUE_HOST_BUSY:
cd9070c9 128 atomic_set(&host->host_blocked, host->max_host_blocked);
f0c0a376
MC
129 break;
130 case SCSI_MLQUEUE_DEVICE_BUSY:
573e5913 131 case SCSI_MLQUEUE_EH_RETRY:
cd9070c9
CH
132 atomic_set(&device->device_blocked,
133 device->max_device_blocked);
f0c0a376
MC
134 break;
135 case SCSI_MLQUEUE_TARGET_BUSY:
cd9070c9
CH
136 atomic_set(&starget->target_blocked,
137 starget->max_target_blocked);
f0c0a376
MC
138 break;
139 }
de3e8bf3
CH
140}
141
d285203c
CH
142static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
143{
a45a1f36
BVA
144 if (cmd->request->rq_flags & RQF_DONTPREP) {
145 cmd->request->rq_flags &= ~RQF_DONTPREP;
146 scsi_mq_uninit_cmd(cmd);
147 } else {
148 WARN_ON_ONCE(true);
149 }
2b053aca 150 blk_mq_requeue_request(cmd->request, true);
d285203c
CH
151}
152
de3e8bf3
CH
153/**
154 * __scsi_queue_insert - private queue insertion
155 * @cmd: The SCSI command being requeued
156 * @reason: The reason for the requeue
157 * @unbusy: Whether the queue should be unbusied
158 *
159 * This is a private queue insertion. The public interface
160 * scsi_queue_insert() always assumes the queue should be unbusied
161 * because it's always called before the completion. This function is
162 * for a requeue after completion, which should only occur in this
163 * file.
164 */
08640e81 165static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
de3e8bf3
CH
166{
167 struct scsi_device *device = cmd->device;
de3e8bf3
CH
168
169 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
170 "Inserting command %p into mlqueue\n", cmd));
171
172 scsi_set_blocked(cmd, reason);
1da177e4 173
1da177e4
LT
174 /*
175 * Decrement the counters, since these commands are no longer
176 * active on the host/device.
177 */
4f5299ac
JB
178 if (unbusy)
179 scsi_device_unbusy(device);
1da177e4
LT
180
181 /*
a1bf9d1d 182 * Requeue this command. It will go before all other commands
b485462a
BVA
183 * that are already in the queue. Schedule requeue work under
184 * lock such that the kblockd_schedule_work() call happens
185 * before blk_cleanup_queue() finishes.
a488e749 186 */
644373a4 187 cmd->result = 0;
f664a3cc 188
f664a3cc 189 blk_mq_requeue_request(cmd->request, true);
1da177e4
LT
190}
191
4f5299ac
JB
192/*
193 * Function: scsi_queue_insert()
194 *
195 * Purpose: Insert a command in the midlevel queue.
196 *
197 * Arguments: cmd - command that we are adding to queue.
198 * reason - why we are inserting command to queue.
199 *
200 * Lock status: Assumed that lock is not held upon entry.
201 *
202 * Returns: Nothing.
203 *
204 * Notes: We do this for one of two cases. Either the host is busy
205 * and it cannot accept any more commands for the time being,
206 * or the device returned QUEUE_FULL and can accept no more
207 * commands.
208 * Notes: This could be called either from an interrupt context or a
209 * normal process context.
210 */
84feb166 211void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
4f5299ac 212{
08640e81 213 __scsi_queue_insert(cmd, reason, true);
4f5299ac 214}
e8064021 215
76aaf87b
CH
216
217/**
704f8392 218 * __scsi_execute - insert request and wait for the result
76aaf87b
CH
219 * @sdev: scsi device
220 * @cmd: scsi command
221 * @data_direction: data direction
222 * @buffer: data buffer
223 * @bufflen: len of buffer
224 * @sense: optional sense buffer
225 * @sshdr: optional decoded sense header
226 * @timeout: request timeout in seconds
227 * @retries: number of times to retry request
228 * @flags: flags for ->cmd_flags
229 * @rq_flags: flags for ->rq_flags
230 * @resid: optional residual length
231 *
17d5363b
CH
232 * Returns the scsi_cmnd result field if a command was executed, or a negative
233 * Linux error code if we didn't get that far.
76aaf87b 234 */
704f8392 235int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
33aa687d 236 int data_direction, void *buffer, unsigned bufflen,
3949e2f0
CH
237 unsigned char *sense, struct scsi_sense_hdr *sshdr,
238 int timeout, int retries, u64 flags, req_flags_t rq_flags,
239 int *resid)
39216033
JB
240{
241 struct request *req;
82ed4db4 242 struct scsi_request *rq;
39216033
JB
243 int ret = DRIVER_ERROR << 24;
244
ff005a06 245 req = blk_get_request(sdev->request_queue,
aebf526b 246 data_direction == DMA_TO_DEVICE ?
039c635f 247 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
a492f075 248 if (IS_ERR(req))
bfe159a5 249 return ret;
82ed4db4 250 rq = scsi_req(req);
39216033
JB
251
252 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
0eb0b63c 253 buffer, bufflen, GFP_NOIO))
39216033
JB
254 goto out;
255
82ed4db4
CH
256 rq->cmd_len = COMMAND_SIZE(cmd[0]);
257 memcpy(rq->cmd, cmd, rq->cmd_len);
64c7f1d1 258 rq->retries = retries;
39216033 259 req->timeout = timeout;
e8064021 260 req->cmd_flags |= flags;
039c635f 261 req->rq_flags |= rq_flags | RQF_QUIET;
39216033
JB
262
263 /*
264 * head injection *required* here otherwise quiesce won't work
265 */
266 blk_execute_rq(req->q, NULL, req, 1);
267
bdb2b8ca
AS
268 /*
269 * Some devices (USB mass-storage in particular) may transfer
270 * garbage data together with a residue indicating that the data
271 * is invalid. Prevent the garbage from being misinterpreted
272 * and prevent security leaks by zeroing out the excess data.
273 */
82ed4db4
CH
274 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
275 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
bdb2b8ca 276
f4f4e47e 277 if (resid)
82ed4db4
CH
278 *resid = rq->resid_len;
279 if (sense && rq->sense_len)
280 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
3949e2f0
CH
281 if (sshdr)
282 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
17d5363b 283 ret = rq->result;
39216033
JB
284 out:
285 blk_put_request(req);
286
287 return ret;
288}
704f8392 289EXPORT_SYMBOL(__scsi_execute);
39216033 290
1da177e4
LT
291/*
292 * Function: scsi_init_cmd_errh()
293 *
294 * Purpose: Initialize cmd fields related to error handling.
295 *
296 * Arguments: cmd - command that is ready to be queued.
297 *
1da177e4
LT
298 * Notes: This function has the job of initializing a number of
299 * fields related to error handling. Typically this will
300 * be called once for each command, as required.
301 */
631c228c 302static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
1da177e4 303{
30b0c37b 304 scsi_set_resid(cmd, 0);
b80ca4f7 305 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1da177e4 306 if (cmd->cmd_len == 0)
db4742dd 307 cmd->cmd_len = scsi_command_size(cmd->cmnd);
1da177e4
LT
308}
309
3bd6f43f
BVA
310/*
311 * Decrement the host_busy counter and wake up the error handler if necessary.
312 * Avoid as follows that the error handler is not woken up if shost->host_busy
313 * == shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
314 * with an RCU read lock in this function to ensure that this function in its
315 * entirety either finishes before scsi_eh_scmd_add() increases the
316 * host_failed counter or that it notices the shost state change made by
317 * scsi_eh_scmd_add().
318 */
319static void scsi_dec_host_busy(struct Scsi_Host *shost)
1da177e4 320{
1da177e4
LT
321 unsigned long flags;
322
3bd6f43f 323 rcu_read_lock();
d772a65d 324 atomic_dec(&shost->host_busy);
3bd6f43f 325 if (unlikely(scsi_host_in_recovery(shost))) {
74665016 326 spin_lock_irqsave(shost->host_lock, flags);
3bd6f43f
BVA
327 if (shost->host_failed || shost->host_eh_scheduled)
328 scsi_eh_wakeup(shost);
74665016
CH
329 spin_unlock_irqrestore(shost->host_lock, flags);
330 }
3bd6f43f
BVA
331 rcu_read_unlock();
332}
333
334void scsi_device_unbusy(struct scsi_device *sdev)
335{
336 struct Scsi_Host *shost = sdev->host;
337 struct scsi_target *starget = scsi_target(sdev);
338
339 scsi_dec_host_busy(shost);
340
341 if (starget->can_queue > 0)
342 atomic_dec(&starget->target_busy);
74665016 343
71e75c97 344 atomic_dec(&sdev->device_busy);
1da177e4
LT
345}
346
d285203c
CH
347static void scsi_kick_queue(struct request_queue *q)
348{
f664a3cc 349 blk_mq_run_hw_queues(q, false);
d285203c
CH
350}
351
1da177e4
LT
352/*
353 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
354 * and call blk_run_queue for all the scsi_devices on the target -
355 * including current_sdev first.
356 *
357 * Called with *no* scsi locks held.
358 */
359static void scsi_single_lun_run(struct scsi_device *current_sdev)
360{
361 struct Scsi_Host *shost = current_sdev->host;
362 struct scsi_device *sdev, *tmp;
363 struct scsi_target *starget = scsi_target(current_sdev);
364 unsigned long flags;
365
366 spin_lock_irqsave(shost->host_lock, flags);
367 starget->starget_sdev_user = NULL;
368 spin_unlock_irqrestore(shost->host_lock, flags);
369
370 /*
371 * Call blk_run_queue for all LUNs on the target, starting with
372 * current_sdev. We race with others (to set starget_sdev_user),
373 * but in most cases, we will be first. Ideally, each LU on the
374 * target would get some limited time or requests on the target.
375 */
d285203c 376 scsi_kick_queue(current_sdev->request_queue);
1da177e4
LT
377
378 spin_lock_irqsave(shost->host_lock, flags);
379 if (starget->starget_sdev_user)
380 goto out;
381 list_for_each_entry_safe(sdev, tmp, &starget->devices,
382 same_target_siblings) {
383 if (sdev == current_sdev)
384 continue;
385 if (scsi_device_get(sdev))
386 continue;
387
388 spin_unlock_irqrestore(shost->host_lock, flags);
d285203c 389 scsi_kick_queue(sdev->request_queue);
1da177e4
LT
390 spin_lock_irqsave(shost->host_lock, flags);
391
392 scsi_device_put(sdev);
393 }
394 out:
395 spin_unlock_irqrestore(shost->host_lock, flags);
396}
397
cd9070c9 398static inline bool scsi_device_is_busy(struct scsi_device *sdev)
9d112517 399{
cd9070c9
CH
400 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
401 return true;
402 if (atomic_read(&sdev->device_blocked) > 0)
403 return true;
404 return false;
9d112517
KU
405}
406
cd9070c9 407static inline bool scsi_target_is_busy(struct scsi_target *starget)
f0c0a376 408{
2ccbb008
CH
409 if (starget->can_queue > 0) {
410 if (atomic_read(&starget->target_busy) >= starget->can_queue)
411 return true;
412 if (atomic_read(&starget->target_blocked) > 0)
413 return true;
414 }
cd9070c9 415 return false;
f0c0a376
MC
416}
417
cd9070c9 418static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
9d112517 419{
d772a65d 420 if (shost->can_queue > 0 &&
cd9070c9
CH
421 atomic_read(&shost->host_busy) >= shost->can_queue)
422 return true;
423 if (atomic_read(&shost->host_blocked) > 0)
424 return true;
425 if (shost->host_self_blocked)
426 return true;
427 return false;
9d112517
KU
428}
429
21a05df5 430static void scsi_starved_list_run(struct Scsi_Host *shost)
1da177e4 431{
2a3a59e5 432 LIST_HEAD(starved_list);
21a05df5 433 struct scsi_device *sdev;
1da177e4
LT
434 unsigned long flags;
435
1da177e4 436 spin_lock_irqsave(shost->host_lock, flags);
2a3a59e5
MC
437 list_splice_init(&shost->starved_list, &starved_list);
438
439 while (!list_empty(&starved_list)) {
e2eb7244
JB
440 struct request_queue *slq;
441
1da177e4
LT
442 /*
443 * As long as shost is accepting commands and we have
444 * starved queues, call blk_run_queue. scsi_request_fn
445 * drops the queue_lock and can add us back to the
446 * starved_list.
447 *
448 * host_lock protects the starved_list and starved_entry.
449 * scsi_request_fn must get the host_lock before checking
450 * or modifying starved_list or starved_entry.
451 */
2a3a59e5 452 if (scsi_host_is_busy(shost))
f0c0a376 453 break;
f0c0a376 454
2a3a59e5
MC
455 sdev = list_entry(starved_list.next,
456 struct scsi_device, starved_entry);
457 list_del_init(&sdev->starved_entry);
f0c0a376
MC
458 if (scsi_target_is_busy(scsi_target(sdev))) {
459 list_move_tail(&sdev->starved_entry,
460 &shost->starved_list);
461 continue;
462 }
463
e2eb7244
JB
464 /*
465 * Once we drop the host lock, a racing scsi_remove_device()
466 * call may remove the sdev from the starved list and destroy
467 * it and the queue. Mitigate by taking a reference to the
468 * queue and never touching the sdev again after we drop the
469 * host lock. Note: if __scsi_remove_device() invokes
470 * blk_cleanup_queue() before the queue is run from this
471 * function then blk_run_queue() will return immediately since
472 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
473 */
474 slq = sdev->request_queue;
475 if (!blk_get_queue(slq))
476 continue;
477 spin_unlock_irqrestore(shost->host_lock, flags);
478
d285203c 479 scsi_kick_queue(slq);
e2eb7244
JB
480 blk_put_queue(slq);
481
482 spin_lock_irqsave(shost->host_lock, flags);
1da177e4 483 }
2a3a59e5
MC
484 /* put any unprocessed entries back */
485 list_splice(&starved_list, &shost->starved_list);
1da177e4 486 spin_unlock_irqrestore(shost->host_lock, flags);
21a05df5
CH
487}
488
489/*
490 * Function: scsi_run_queue()
491 *
492 * Purpose: Select a proper request queue to serve next
493 *
494 * Arguments: q - last request's queue
495 *
496 * Returns: Nothing
497 *
498 * Notes: The previous command was completely finished, start
499 * a new one if possible.
500 */
501static void scsi_run_queue(struct request_queue *q)
502{
503 struct scsi_device *sdev = q->queuedata;
504
505 if (scsi_target(sdev)->single_lun)
506 scsi_single_lun_run(sdev);
507 if (!list_empty(&sdev->host->starved_list))
508 scsi_starved_list_run(sdev->host);
1da177e4 509
f664a3cc 510 blk_mq_run_hw_queues(q, false);
1da177e4
LT
511}
512
9937a5e2
JA
513void scsi_requeue_run_queue(struct work_struct *work)
514{
515 struct scsi_device *sdev;
516 struct request_queue *q;
517
518 sdev = container_of(work, struct scsi_device, requeue_work);
519 q = sdev->request_queue;
520 scsi_run_queue(q);
521}
522
1da177e4
LT
523void scsi_run_host_queues(struct Scsi_Host *shost)
524{
525 struct scsi_device *sdev;
526
527 shost_for_each_device(sdev, shost)
528 scsi_run_queue(sdev->request_queue);
529}
530
d285203c
CH
531static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
532{
57292b58 533 if (!blk_rq_is_passthrough(cmd->request)) {
d285203c
CH
534 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
535
536 if (drv->uninit_command)
537 drv->uninit_command(cmd);
538 }
539}
540
541static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
542{
543 if (cmd->sdb.table.nents)
4635873c 544 sg_free_table_chained(&cmd->sdb.table, SG_CHUNK_SIZE);
d285203c 545 if (scsi_prot_sg_count(cmd))
4635873c 546 sg_free_table_chained(&cmd->prot_sdb->table, SG_CHUNK_SIZE);
d285203c
CH
547}
548
549static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
550{
d285203c
CH
551 scsi_mq_free_sgtables(cmd);
552 scsi_uninit_cmd(cmd);
2dd6fb59 553 scsi_del_cmd_from_list(cmd);
d285203c
CH
554}
555
7e63b5a4 556/* Returns false when no more bytes to process, true if there are more */
2a842aca 557static bool scsi_end_request(struct request *req, blk_status_t error,
ae3d56d8 558 unsigned int bytes)
f6d47e74 559{
bed2213d 560 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
f6d47e74
CH
561 struct scsi_device *sdev = cmd->device;
562 struct request_queue *q = sdev->request_queue;
f6d47e74
CH
563
564 if (blk_update_request(req, error, bytes))
565 return true;
566
f6d47e74
CH
567 if (blk_queue_add_random(q))
568 add_disk_randomness(req->rq_disk);
569
64104f70
BVA
570 if (!blk_rq_is_scsi(req)) {
571 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
572 cmd->flags &= ~SCMD_INITIALIZED;
573 }
574
db983f6e
BVA
575 /*
576 * Calling rcu_barrier() is not necessary here because the
577 * SCSI error handler guarantees that the function called by
578 * call_rcu() has been called before scsi_end_request() is
579 * called.
580 */
581 destroy_rcu_head(&cmd->rcu);
582
f664a3cc
JA
583 /*
584 * In the MQ case the command gets freed by __blk_mq_end_request,
585 * so we have to do all cleanup that depends on it earlier.
586 *
587 * We also can't kick the queues from irq context, so we
588 * will have to defer it to a workqueue.
589 */
590 scsi_mq_uninit_cmd(cmd);
f81426a8 591
a78b03bc
JA
592 /*
593 * queue is still alive, so grab the ref for preventing it
594 * from being cleaned up during running queue.
595 */
596 percpu_ref_get(&q->q_usage_counter);
f81426a8 597
f664a3cc 598 __blk_mq_end_request(req, error);
d285203c 599
f664a3cc
JA
600 if (scsi_target(sdev)->single_lun ||
601 !list_empty(&sdev->host->starved_list))
602 kblockd_schedule_work(&sdev->requeue_work);
603 else
604 blk_mq_run_hw_queues(q, true);
f6d47e74 605
a78b03bc 606 percpu_ref_put(&q->q_usage_counter);
f6d47e74
CH
607 return false;
608}
609
0f7f6234 610/**
a77b32d8
BVA
611 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
612 * @cmd: SCSI command
0f7f6234
HR
613 * @result: scsi error code
614 *
a77b32d8
BVA
615 * Translate a SCSI result code into a blk_status_t value. May reset the host
616 * byte of @cmd->result.
0f7f6234 617 */
a77b32d8 618static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
63583cca 619{
2a842aca 620 switch (host_byte(result)) {
f4abab3f
BVA
621 case DID_OK:
622 /*
623 * Also check the other bytes than the status byte in result
624 * to handle the case when a SCSI LLD sets result to
625 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
626 */
627 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
628 return BLK_STS_OK;
629 return BLK_STS_IOERR;
63583cca 630 case DID_TRANSPORT_FAILFAST:
2a842aca 631 return BLK_STS_TRANSPORT;
63583cca 632 case DID_TARGET_FAILURE:
2082ebc4 633 set_host_byte(cmd, DID_OK);
2a842aca 634 return BLK_STS_TARGET;
63583cca 635 case DID_NEXUS_FAILURE:
4a067cf8 636 set_host_byte(cmd, DID_OK);
2a842aca 637 return BLK_STS_NEXUS;
a9d6ceb8
HR
638 case DID_ALLOC_FAILURE:
639 set_host_byte(cmd, DID_OK);
2a842aca 640 return BLK_STS_NOSPC;
7e782af5
HR
641 case DID_MEDIUM_ERROR:
642 set_host_byte(cmd, DID_OK);
2a842aca 643 return BLK_STS_MEDIUM;
63583cca 644 default:
2a842aca 645 return BLK_STS_IOERR;
63583cca 646 }
63583cca
HR
647}
648
4ae61c68
DG
649/* Helper for scsi_io_completion() when "reprep" action required. */
650static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
651 struct request_queue *q)
652{
653 /* A new command will be prepared and issued. */
f664a3cc 654 scsi_mq_requeue_cmd(cmd);
4ae61c68
DG
655}
656
da32baea
DG
657/* Helper for scsi_io_completion() when special action required. */
658static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
1da177e4 659{
165125e1 660 struct request_queue *q = cmd->device->request_queue;
1da177e4 661 struct request *req = cmd->request;
da32baea 662 int level = 0;
b60af5b0
AS
663 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
664 ACTION_DELAYED_RETRY} action;
ee60b2c5 665 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
da32baea
DG
666 struct scsi_sense_hdr sshdr;
667 bool sense_valid;
668 bool sense_current = true; /* false implies "deferred sense" */
669 blk_status_t blk_stat;
1da177e4 670
da32baea
DG
671 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
672 if (sense_valid)
673 sense_current = !scsi_sense_is_deferred(&sshdr);
03aba2f7 674
da32baea 675 blk_stat = scsi_result_to_blk_status(cmd, result);
3e695f89 676
b60af5b0
AS
677 if (host_byte(result) == DID_RESET) {
678 /* Third party bus reset or reset for error recovery
679 * reasons. Just retry the command and see what
680 * happens.
681 */
682 action = ACTION_RETRY;
da32baea 683 } else if (sense_valid && sense_current) {
1da177e4
LT
684 switch (sshdr.sense_key) {
685 case UNIT_ATTENTION:
686 if (cmd->device->removable) {
03aba2f7 687 /* Detected disc change. Set a bit
1da177e4
LT
688 * and quietly refuse further access.
689 */
690 cmd->device->changed = 1;
b60af5b0 691 action = ACTION_FAIL;
1da177e4 692 } else {
03aba2f7
LT
693 /* Must have been a power glitch, or a
694 * bus reset. Could not have been a
695 * media change, so we just retry the
b60af5b0 696 * command and see what happens.
03aba2f7 697 */
b60af5b0 698 action = ACTION_RETRY;
1da177e4
LT
699 }
700 break;
701 case ILLEGAL_REQUEST:
03aba2f7
LT
702 /* If we had an ILLEGAL REQUEST returned, then
703 * we may have performed an unsupported
704 * command. The only thing this should be
705 * would be a ten byte read where only a six
706 * byte read was supported. Also, on a system
707 * where READ CAPACITY failed, we may have
708 * read past the end of the disk.
709 */
26a68019
JA
710 if ((cmd->device->use_10_for_rw &&
711 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
1da177e4
LT
712 (cmd->cmnd[0] == READ_10 ||
713 cmd->cmnd[0] == WRITE_10)) {
b60af5b0 714 /* This will issue a new 6-byte command. */
1da177e4 715 cmd->device->use_10_for_rw = 0;
b60af5b0 716 action = ACTION_REPREP;
3e695f89 717 } else if (sshdr.asc == 0x10) /* DIX */ {
3e695f89 718 action = ACTION_FAIL;
da32baea 719 blk_stat = BLK_STS_PROTECTION;
c98a0eb0 720 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
5db44863 721 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
c98a0eb0 722 action = ACTION_FAIL;
da32baea 723 blk_stat = BLK_STS_TARGET;
b60af5b0
AS
724 } else
725 action = ACTION_FAIL;
726 break;
511e44f4 727 case ABORTED_COMMAND:
126c0982 728 action = ACTION_FAIL;
e6c11dbb 729 if (sshdr.asc == 0x10) /* DIF */
da32baea 730 blk_stat = BLK_STS_PROTECTION;
1da177e4
LT
731 break;
732 case NOT_READY:
03aba2f7 733 /* If the device is in the process of becoming
f3e93f73 734 * ready, or has a temporary blockage, retry.
1da177e4 735 */
f3e93f73
JB
736 if (sshdr.asc == 0x04) {
737 switch (sshdr.ascq) {
738 case 0x01: /* becoming ready */
739 case 0x04: /* format in progress */
740 case 0x05: /* rebuild in progress */
741 case 0x06: /* recalculation in progress */
742 case 0x07: /* operation in progress */
743 case 0x08: /* Long write in progress */
744 case 0x09: /* self test in progress */
d8705f11 745 case 0x14: /* space allocation in progress */
e37c7d9a
DG
746 case 0x1a: /* start stop unit in progress */
747 case 0x1b: /* sanitize in progress */
748 case 0x1d: /* configuration in progress */
749 case 0x24: /* depopulation in progress */
b60af5b0 750 action = ACTION_DELAYED_RETRY;
f3e93f73 751 break;
3dbf6a54 752 default:
3dbf6a54
AS
753 action = ACTION_FAIL;
754 break;
f3e93f73 755 }
e6c11dbb 756 } else
b60af5b0 757 action = ACTION_FAIL;
b60af5b0 758 break;
1da177e4 759 case VOLUME_OVERFLOW:
03aba2f7 760 /* See SSC3rXX or current. */
b60af5b0
AS
761 action = ACTION_FAIL;
762 break;
1da177e4 763 default:
b60af5b0 764 action = ACTION_FAIL;
1da177e4
LT
765 break;
766 }
e6c11dbb 767 } else
b60af5b0 768 action = ACTION_FAIL;
b60af5b0 769
ee60b2c5 770 if (action != ACTION_FAIL &&
e6c11dbb 771 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
ee60b2c5 772 action = ACTION_FAIL;
ee60b2c5 773
b60af5b0
AS
774 switch (action) {
775 case ACTION_FAIL:
776 /* Give up and fail the remainder of the request */
e8064021 777 if (!(req->rq_flags & RQF_QUIET)) {
f1569ff1
HR
778 static DEFINE_RATELIMIT_STATE(_rs,
779 DEFAULT_RATELIMIT_INTERVAL,
780 DEFAULT_RATELIMIT_BURST);
781
782 if (unlikely(scsi_logging_level))
da32baea
DG
783 level =
784 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
785 SCSI_LOG_MLCOMPLETE_BITS);
f1569ff1
HR
786
787 /*
788 * if logging is enabled the failure will be printed
789 * in scsi_log_completion(), so avoid duplicate messages
790 */
791 if (!level && __ratelimit(&_rs)) {
792 scsi_print_result(cmd, NULL, FAILED);
c65be1a6 793 if (driver_byte(result) == DRIVER_SENSE)
f1569ff1
HR
794 scsi_print_sense(cmd);
795 scsi_print_command(cmd);
796 }
3173d8c3 797 }
ae3d56d8 798 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
f6d47e74 799 return;
bc85dc50 800 /*FALLTHRU*/
b60af5b0 801 case ACTION_REPREP:
4ae61c68 802 scsi_io_completion_reprep(cmd, q);
b60af5b0
AS
803 break;
804 case ACTION_RETRY:
805 /* Retry the same command immediately */
08640e81 806 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
b60af5b0
AS
807 break;
808 case ACTION_DELAYED_RETRY:
809 /* Retry the same command after a delay */
08640e81 810 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
b60af5b0 811 break;
1da177e4
LT
812 }
813}
1da177e4 814
ab831084
DG
815/*
816 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
817 * new result that may suppress further error checking. Also modifies
818 * *blk_statp in some cases.
819 */
820static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
821 blk_status_t *blk_statp)
822{
823 bool sense_valid;
824 bool sense_current = true; /* false implies "deferred sense" */
825 struct request *req = cmd->request;
826 struct scsi_sense_hdr sshdr;
827
828 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
829 if (sense_valid)
830 sense_current = !scsi_sense_is_deferred(&sshdr);
831
832 if (blk_rq_is_passthrough(req)) {
833 if (sense_valid) {
834 /*
835 * SG_IO wants current and deferred errors
836 */
837 scsi_req(req)->sense_len =
838 min(8 + cmd->sense_buffer[7],
839 SCSI_SENSE_BUFFERSIZE);
840 }
841 if (sense_current)
842 *blk_statp = scsi_result_to_blk_status(cmd, result);
843 } else if (blk_rq_bytes(req) == 0 && sense_current) {
844 /*
845 * Flush commands do not transfers any data, and thus cannot use
846 * good_bytes != blk_rq_bytes(req) as the signal for an error.
847 * This sets *blk_statp explicitly for the problem case.
848 */
849 *blk_statp = scsi_result_to_blk_status(cmd, result);
850 }
851 /*
852 * Recovered errors need reporting, but they're always treated as
853 * success, so fiddle the result code here. For passthrough requests
854 * we already took a copy of the original into sreq->result which
855 * is what gets returned to the user
856 */
857 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
858 bool do_print = true;
859 /*
860 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
861 * skip print since caller wants ATA registers. Only occurs
862 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
863 */
864 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
865 do_print = false;
866 else if (req->rq_flags & RQF_QUIET)
867 do_print = false;
868 if (do_print)
869 scsi_print_sense(cmd);
870 result = 0;
871 /* for passthrough, *blk_statp may be set */
872 *blk_statp = BLK_STS_OK;
873 }
874 /*
875 * Another corner case: the SCSI status byte is non-zero but 'good'.
876 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
877 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
878 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
879 * intermediate statuses (both obsolete in SAM-4) as good.
880 */
881 if (status_byte(result) && scsi_status_is_good(result)) {
882 result = 0;
883 *blk_statp = BLK_STS_OK;
884 }
885 return result;
886}
887
1da177e4
LT
888/*
889 * Function: scsi_io_completion()
890 *
891 * Purpose: Completion processing for block device I/O requests.
892 *
893 * Arguments: cmd - command that is finished.
894 *
895 * Lock status: Assumed that no lock is held upon entry.
896 *
897 * Returns: Nothing
898 *
bc85dc50
CH
899 * Notes: We will finish off the specified number of sectors. If we
900 * are done, the command block will be released and the queue
901 * function will be goosed. If we are not done then we have to
b60af5b0 902 * figure out what to do next:
1da177e4 903 *
b60af5b0
AS
904 * a) We can call scsi_requeue_command(). The request
905 * will be unprepared and put back on the queue. Then
906 * a new command will be created for it. This should
907 * be used if we made forward progress, or if we want
908 * to switch from READ(10) to READ(6) for example.
1da177e4 909 *
bc85dc50 910 * b) We can call __scsi_queue_insert(). The request will
b60af5b0
AS
911 * be put back on the queue and retried using the same
912 * command as before, possibly after a delay.
913 *
1f7cbb8e
DG
914 * c) We can call scsi_end_request() with blk_stat other than
915 * BLK_STS_OK, to fail the remainder of the request.
1da177e4 916 */
03aba2f7 917void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1da177e4
LT
918{
919 int result = cmd->result;
165125e1 920 struct request_queue *q = cmd->device->request_queue;
1da177e4 921 struct request *req = cmd->request;
1f7cbb8e 922 blk_status_t blk_stat = BLK_STS_OK;
1da177e4 923
0d437906 924 if (unlikely(result)) /* a nz result may or may not be an error */
ab831084 925 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
631c228c 926
0d437906 927 if (unlikely(blk_rq_is_passthrough(req))) {
27c41973 928 /*
a77b32d8 929 * scsi_result_to_blk_status may have reset the host_byte
27c41973 930 */
17d5363b 931 scsi_req(req)->result = cmd->result;
8e1695a0 932 }
30b0c37b 933
1da177e4
LT
934 /*
935 * Next deal with any sectors which we were able to correctly
936 * handle.
937 */
91921e01
HR
938 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
939 "%u sectors total, %d bytes done.\n",
940 blk_rq_sectors(req), good_bytes));
d6b0c537 941
a9bddd74 942 /*
1f7cbb8e
DG
943 * Next deal with any sectors which we were able to correctly
944 * handle. Failed, zero length commands always need to drop down
945 * to retry code. Fast path should return in this block.
d6b0c537 946 */
0d437906 947 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
ae3d56d8 948 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1f7cbb8e
DG
949 return; /* no bytes remaining */
950 }
bc85dc50 951
0d437906
DG
952 /* Kill remainder if no retries. */
953 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
ae3d56d8 954 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
8e1695a0
DG
955 WARN_ONCE(true,
956 "Bytes remaining after failed, no-retry command");
f6d47e74 957 return;
bc85dc50
CH
958 }
959
960 /*
961 * If there had been no error, but we have leftover bytes in the
962 * requeues just queue the command up again.
d6b0c537 963 */
0d437906 964 if (likely(result == 0))
4ae61c68
DG
965 scsi_io_completion_reprep(cmd, q);
966 else
da32baea 967 scsi_io_completion_action(cmd, result);
1da177e4 968}
1da177e4 969
159b2cbf
CH
970static blk_status_t scsi_init_sgtable(struct request *req,
971 struct scsi_data_buffer *sdb)
1da177e4 972{
6f9a35e2 973 int count;
1da177e4
LT
974
975 /*
3b003157 976 * If sg table allocation fails, requeue request later.
1da177e4 977 */
f9d03f96 978 if (unlikely(sg_alloc_table_chained(&sdb->table,
4635873c
ML
979 blk_rq_nr_phys_segments(req), sdb->table.sgl,
980 SG_CHUNK_SIZE)))
159b2cbf 981 return BLK_STS_RESOURCE;
1da177e4 982
1da177e4
LT
983 /*
984 * Next, walk the list, and fill in the addresses and sizes of
985 * each segment.
986 */
30b0c37b
BH
987 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
988 BUG_ON(count > sdb->table.nents);
989 sdb->table.nents = count;
fd102b12 990 sdb->length = blk_rq_payload_bytes(req);
159b2cbf 991 return BLK_STS_OK;
1da177e4 992}
6f9a35e2
BH
993
994/*
995 * Function: scsi_init_io()
996 *
997 * Purpose: SCSI I/O initialize function.
998 *
999 * Arguments: cmd - Command descriptor we wish to initialize
1000 *
159b2cbf
CH
1001 * Returns: BLK_STS_OK on success
1002 * BLK_STS_RESOURCE if the failure is retryable
1003 * BLK_STS_IOERR if the failure is fatal
6f9a35e2 1004 */
159b2cbf 1005blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
6f9a35e2 1006{
13f05c8d 1007 struct request *rq = cmd->request;
159b2cbf 1008 blk_status_t ret;
13f05c8d 1009
fd3fc0b4 1010 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
159b2cbf 1011 return BLK_STS_IOERR;
635d98b1 1012
159b2cbf
CH
1013 ret = scsi_init_sgtable(rq, &cmd->sdb);
1014 if (ret)
1015 return ret;
6f9a35e2 1016
13f05c8d 1017 if (blk_integrity_rq(rq)) {
7027ad72
MP
1018 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1019 int ivecs, count;
1020
14784565 1021 if (WARN_ON_ONCE(!prot_sdb)) {
91724c20
EM
1022 /*
1023 * This can happen if someone (e.g. multipath)
1024 * queues a command to a device on an adapter
1025 * that does not support DIX.
1026 */
159b2cbf 1027 ret = BLK_STS_IOERR;
14784565 1028 goto out_free_sgtables;
91724c20
EM
1029 }
1030
13f05c8d 1031 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
7027ad72 1032
001d63be 1033 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
4635873c
ML
1034 prot_sdb->table.sgl,
1035 SG_CHUNK_SIZE)) {
159b2cbf 1036 ret = BLK_STS_RESOURCE;
14784565 1037 goto out_free_sgtables;
7027ad72
MP
1038 }
1039
13f05c8d 1040 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
7027ad72 1041 prot_sdb->table.sgl);
6f1d8a53
IS
1042 BUG_ON(count > ivecs);
1043 BUG_ON(count > queue_max_integrity_segments(rq->q));
7027ad72
MP
1044
1045 cmd->prot_sdb = prot_sdb;
1046 cmd->prot_sdb->table.nents = count;
1047 }
1048
159b2cbf 1049 return BLK_STS_OK;
14784565 1050out_free_sgtables:
f664a3cc 1051 scsi_mq_free_sgtables(cmd);
159b2cbf 1052 return ret;
6f9a35e2 1053}
bb52d82f 1054EXPORT_SYMBOL(scsi_init_io);
1da177e4 1055
ca18d6f7 1056/**
832889f5 1057 * scsi_initialize_rq - initialize struct scsi_cmnd partially
35c0506f 1058 * @rq: Request associated with the SCSI command to be initialized.
ca18d6f7 1059 *
832889f5
BVA
1060 * This function initializes the members of struct scsi_cmnd that must be
1061 * initialized before request processing starts and that won't be
1062 * reinitialized if a SCSI command is requeued.
1063 *
64104f70
BVA
1064 * Called from inside blk_get_request() for pass-through requests and from
1065 * inside scsi_init_command() for filesystem requests.
ca18d6f7 1066 */
e4c9470b 1067static void scsi_initialize_rq(struct request *rq)
ca18d6f7 1068{
c8d9cf22
BVA
1069 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1070
1071 scsi_req_init(&cmd->req);
3be8828f 1072 init_rcu_head(&cmd->rcu);
832889f5
BVA
1073 cmd->jiffies_at_alloc = jiffies;
1074 cmd->retries = 0;
ca18d6f7 1075}
ca18d6f7 1076
2dd6fb59
BVA
1077/* Add a command to the list used by the aacraid and dpt_i2o drivers */
1078void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1079{
1080 struct scsi_device *sdev = cmd->device;
1081 struct Scsi_Host *shost = sdev->host;
1082 unsigned long flags;
1083
1084 if (shost->use_cmd_list) {
1085 spin_lock_irqsave(&sdev->list_lock, flags);
1086 list_add_tail(&cmd->list, &sdev->cmd_list);
1087 spin_unlock_irqrestore(&sdev->list_lock, flags);
1088 }
1089}
1090
1091/* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1092void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1093{
1094 struct scsi_device *sdev = cmd->device;
1095 struct Scsi_Host *shost = sdev->host;
1096 unsigned long flags;
1097
1098 if (shost->use_cmd_list) {
1099 spin_lock_irqsave(&sdev->list_lock, flags);
1100 BUG_ON(list_empty(&cmd->list));
1101 list_del_init(&cmd->list);
1102 spin_unlock_irqrestore(&sdev->list_lock, flags);
1103 }
1104}
1105
ca18d6f7 1106/* Called after a request has been started. */
e9c787e6 1107void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
3b003157 1108{
e9c787e6
CH
1109 void *buf = cmd->sense_buffer;
1110 void *prot = cmd->prot_sdb;
64104f70
BVA
1111 struct request *rq = blk_mq_rq_from_pdu(cmd);
1112 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
832889f5
BVA
1113 unsigned long jiffies_at_alloc;
1114 int retries;
64104f70
BVA
1115
1116 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1117 flags |= SCMD_INITIALIZED;
1118 scsi_initialize_rq(rq);
1119 }
3b003157 1120
832889f5
BVA
1121 jiffies_at_alloc = cmd->jiffies_at_alloc;
1122 retries = cmd->retries;
82ed4db4
CH
1123 /* zero out the cmd, except for the embedded scsi_request */
1124 memset((char *)cmd + sizeof(cmd->req), 0,
ee524236 1125 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
3b003157 1126
e9c787e6
CH
1127 cmd->device = dev;
1128 cmd->sense_buffer = buf;
1129 cmd->prot_sdb = prot;
64104f70 1130 cmd->flags = flags;
e9c787e6 1131 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
832889f5
BVA
1132 cmd->jiffies_at_alloc = jiffies_at_alloc;
1133 cmd->retries = retries;
64a87b24 1134
2dd6fb59 1135 scsi_add_cmd_to_list(cmd);
3b003157
CH
1136}
1137
785ba83b
CH
1138static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1139 struct request *req)
7b16318d 1140{
bed2213d 1141 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
3b003157
CH
1142
1143 /*
aebf526b 1144 * Passthrough requests may transfer data, in which case they must
3b003157
CH
1145 * a bio attached to them. Or they might contain a SCSI command
1146 * that does not transfer data, in which case they may optionally
1147 * submit a request without an attached bio.
1148 */
1149 if (req->bio) {
159b2cbf
CH
1150 blk_status_t ret = scsi_init_io(cmd);
1151 if (unlikely(ret != BLK_STS_OK))
1152 return ret;
3b003157 1153 } else {
b0790410 1154 BUG_ON(blk_rq_bytes(req));
3b003157 1155
30b0c37b 1156 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
3b003157 1157 }
7b16318d 1158
82ed4db4
CH
1159 cmd->cmd_len = scsi_req(req)->cmd_len;
1160 cmd->cmnd = scsi_req(req)->cmd;
b0790410 1161 cmd->transfersize = blk_rq_bytes(req);
64c7f1d1 1162 cmd->allowed = scsi_req(req)->retries;
785ba83b 1163 return BLK_STS_OK;
7b16318d 1164}
7b16318d 1165
3b003157 1166/*
aebf526b 1167 * Setup a normal block command. These are simple request from filesystems
3868cf8e 1168 * that still need to be translated to SCSI CDBs from the ULD.
3b003157 1169 */
785ba83b
CH
1170static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1171 struct request *req)
1da177e4 1172{
bed2213d 1173 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
a6a8d9f8 1174
ee14c674 1175 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
4c1cb67c
CH
1176 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1177 if (ret != BLK_STS_OK)
1178 return ret;
a6a8d9f8
CS
1179 }
1180
82ed4db4 1181 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
64a87b24 1182 memset(cmd->cmnd, 0, BLK_MAX_CDB);
159b2cbf 1183 return scsi_cmd_to_driver(cmd)->init_command(cmd);
3b003157
CH
1184}
1185
785ba83b
CH
1186static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1187 struct request *req)
6af7a4ff 1188{
bed2213d 1189 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
6af7a4ff
CH
1190
1191 if (!blk_rq_bytes(req))
1192 cmd->sc_data_direction = DMA_NONE;
1193 else if (rq_data_dir(req) == WRITE)
1194 cmd->sc_data_direction = DMA_TO_DEVICE;
1195 else
1196 cmd->sc_data_direction = DMA_FROM_DEVICE;
1197
aebf526b
CH
1198 if (blk_rq_is_scsi(req))
1199 return scsi_setup_scsi_cmnd(sdev, req);
1200 else
6af7a4ff 1201 return scsi_setup_fs_cmnd(sdev, req);
6af7a4ff
CH
1202}
1203
c092d4ec 1204static blk_status_t
a1b73fc1 1205scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
3b003157 1206{
c092d4ec
CH
1207 switch (sdev->sdev_state) {
1208 case SDEV_OFFLINE:
1209 case SDEV_TRANSPORT_OFFLINE:
1210 /*
1211 * If the device is offline we refuse to process any
1212 * commands. The device must be brought online
1213 * before trying any recovery commands.
1214 */
1215 sdev_printk(KERN_ERR, sdev,
1216 "rejecting I/O to offline device\n");
1217 return BLK_STS_IOERR;
1218 case SDEV_DEL:
1219 /*
1220 * If the device is fully deleted, we refuse to
1221 * process any commands as well.
1222 */
1223 sdev_printk(KERN_ERR, sdev,
1224 "rejecting I/O to dead device\n");
1225 return BLK_STS_IOERR;
1226 case SDEV_BLOCK:
1227 case SDEV_CREATED_BLOCK:
1228 return BLK_STS_RESOURCE;
1229 case SDEV_QUIESCE:
1230 /*
1231 * If the devices is blocked we defer normal commands.
1232 */
1233 if (req && !(req->rq_flags & RQF_PREEMPT))
1234 return BLK_STS_RESOURCE;
1235 return BLK_STS_OK;
1236 default:
1237 /*
1238 * For any other not fully online state we only allow
1239 * special commands. In particular any user initiated
1240 * command is not allowed.
1241 */
1242 if (req && !(req->rq_flags & RQF_PREEMPT))
1243 return BLK_STS_IOERR;
1244 return BLK_STS_OK;
1da177e4 1245 }
7f9a6bc4 1246}
1da177e4 1247
1da177e4
LT
1248/*
1249 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1250 * return 0.
1251 *
1252 * Called with the queue_lock held.
1253 */
1254static inline int scsi_dev_queue_ready(struct request_queue *q,
1255 struct scsi_device *sdev)
1256{
71e75c97
CH
1257 unsigned int busy;
1258
1259 busy = atomic_inc_return(&sdev->device_busy) - 1;
cd9070c9 1260 if (atomic_read(&sdev->device_blocked)) {
71e75c97
CH
1261 if (busy)
1262 goto out_dec;
1263
1da177e4
LT
1264 /*
1265 * unblock after device_blocked iterates to zero
1266 */
f664a3cc 1267 if (atomic_dec_return(&sdev->device_blocked) > 0)
71e75c97 1268 goto out_dec;
71e75c97
CH
1269 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1270 "unblocking device at zero depth\n"));
1da177e4 1271 }
71e75c97
CH
1272
1273 if (busy >= sdev->queue_depth)
1274 goto out_dec;
1da177e4
LT
1275
1276 return 1;
71e75c97
CH
1277out_dec:
1278 atomic_dec(&sdev->device_busy);
1279 return 0;
1da177e4
LT
1280}
1281
f0c0a376
MC
1282/*
1283 * scsi_target_queue_ready: checks if there we can send commands to target
1284 * @sdev: scsi device on starget to check.
f0c0a376
MC
1285 */
1286static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1287 struct scsi_device *sdev)
1288{
1289 struct scsi_target *starget = scsi_target(sdev);
7ae65c0f 1290 unsigned int busy;
f0c0a376
MC
1291
1292 if (starget->single_lun) {
7ae65c0f 1293 spin_lock_irq(shost->host_lock);
f0c0a376 1294 if (starget->starget_sdev_user &&
7ae65c0f
CH
1295 starget->starget_sdev_user != sdev) {
1296 spin_unlock_irq(shost->host_lock);
1297 return 0;
1298 }
f0c0a376 1299 starget->starget_sdev_user = sdev;
7ae65c0f 1300 spin_unlock_irq(shost->host_lock);
f0c0a376
MC
1301 }
1302
2ccbb008
CH
1303 if (starget->can_queue <= 0)
1304 return 1;
1305
7ae65c0f 1306 busy = atomic_inc_return(&starget->target_busy) - 1;
cd9070c9 1307 if (atomic_read(&starget->target_blocked) > 0) {
7ae65c0f
CH
1308 if (busy)
1309 goto starved;
1310
f0c0a376
MC
1311 /*
1312 * unblock after target_blocked iterates to zero
1313 */
cd9070c9 1314 if (atomic_dec_return(&starget->target_blocked) > 0)
7ae65c0f 1315 goto out_dec;
cf68d334
CH
1316
1317 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1318 "unblocking target at zero depth\n"));
f0c0a376
MC
1319 }
1320
2ccbb008 1321 if (busy >= starget->can_queue)
7ae65c0f 1322 goto starved;
f0c0a376 1323
7ae65c0f
CH
1324 return 1;
1325
1326starved:
1327 spin_lock_irq(shost->host_lock);
1328 list_move_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1329 spin_unlock_irq(shost->host_lock);
7ae65c0f 1330out_dec:
2ccbb008
CH
1331 if (starget->can_queue > 0)
1332 atomic_dec(&starget->target_busy);
7ae65c0f 1333 return 0;
f0c0a376
MC
1334}
1335
1da177e4
LT
1336/*
1337 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1338 * return 0. We must end up running the queue again whenever 0 is
1339 * returned, else IO can hang.
1da177e4
LT
1340 */
1341static inline int scsi_host_queue_ready(struct request_queue *q,
1342 struct Scsi_Host *shost,
1343 struct scsi_device *sdev)
1344{
74665016 1345 unsigned int busy;
cf68d334 1346
939647ee 1347 if (scsi_host_in_recovery(shost))
74665016
CH
1348 return 0;
1349
d772a65d 1350 busy = atomic_inc_return(&shost->host_busy) - 1;
cd9070c9 1351 if (atomic_read(&shost->host_blocked) > 0) {
d772a65d 1352 if (busy)
74665016
CH
1353 goto starved;
1354
1da177e4
LT
1355 /*
1356 * unblock after host_blocked iterates to zero
1357 */
cd9070c9 1358 if (atomic_dec_return(&shost->host_blocked) > 0)
74665016 1359 goto out_dec;
cf68d334
CH
1360
1361 SCSI_LOG_MLQUEUE(3,
1362 shost_printk(KERN_INFO, shost,
1363 "unblocking host at zero depth\n"));
1da177e4 1364 }
74665016 1365
d772a65d 1366 if (shost->can_queue > 0 && busy >= shost->can_queue)
74665016
CH
1367 goto starved;
1368 if (shost->host_self_blocked)
1369 goto starved;
1da177e4
LT
1370
1371 /* We're OK to process the command, so we can't be starved */
74665016
CH
1372 if (!list_empty(&sdev->starved_entry)) {
1373 spin_lock_irq(shost->host_lock);
1374 if (!list_empty(&sdev->starved_entry))
1375 list_del_init(&sdev->starved_entry);
1376 spin_unlock_irq(shost->host_lock);
1377 }
1da177e4 1378
74665016
CH
1379 return 1;
1380
1381starved:
1382 spin_lock_irq(shost->host_lock);
1383 if (list_empty(&sdev->starved_entry))
1384 list_add_tail(&sdev->starved_entry, &shost->starved_list);
cf68d334 1385 spin_unlock_irq(shost->host_lock);
74665016 1386out_dec:
3bd6f43f 1387 scsi_dec_host_busy(shost);
74665016 1388 return 0;
1da177e4
LT
1389}
1390
6c5121b7
KU
1391/*
1392 * Busy state exporting function for request stacking drivers.
1393 *
1394 * For efficiency, no lock is taken to check the busy state of
1395 * shost/starget/sdev, since the returned value is not guaranteed and
1396 * may be changed after request stacking drivers call the function,
1397 * regardless of taking lock or not.
1398 *
67bd9413
BVA
1399 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1400 * needs to return 'not busy'. Otherwise, request stacking drivers
1401 * may hold requests forever.
6c5121b7 1402 */
f664a3cc 1403static bool scsi_mq_lld_busy(struct request_queue *q)
6c5121b7
KU
1404{
1405 struct scsi_device *sdev = q->queuedata;
1406 struct Scsi_Host *shost;
6c5121b7 1407
3f3299d5 1408 if (blk_queue_dying(q))
f664a3cc 1409 return false;
6c5121b7
KU
1410
1411 shost = sdev->host;
6c5121b7 1412
b7e94a16
JN
1413 /*
1414 * Ignore host/starget busy state.
1415 * Since block layer does not have a concept of fairness across
1416 * multiple queues, congestion of host/starget needs to be handled
1417 * in SCSI layer.
1418 */
1419 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
f664a3cc 1420 return true;
e36e0c80 1421
f664a3cc 1422 return false;
1da177e4
LT
1423}
1424
1aea6434
JA
1425static void scsi_softirq_done(struct request *rq)
1426{
bed2213d 1427 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
242f9dcb 1428 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1aea6434
JA
1429 int disposition;
1430
1431 INIT_LIST_HEAD(&cmd->eh_entry);
1432
242f9dcb
JA
1433 atomic_inc(&cmd->device->iodone_cnt);
1434 if (cmd->result)
1435 atomic_inc(&cmd->device->ioerr_cnt);
1436
1aea6434
JA
1437 disposition = scsi_decide_disposition(cmd);
1438 if (disposition != SUCCESS &&
1439 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1440 sdev_printk(KERN_ERR, cmd->device,
1441 "timing out command, waited %lus\n",
1442 wait_for/HZ);
1443 disposition = SUCCESS;
1444 }
91921e01 1445
1aea6434
JA
1446 scsi_log_completion(cmd, disposition);
1447
1448 switch (disposition) {
1449 case SUCCESS:
1450 scsi_finish_command(cmd);
1451 break;
1452 case NEEDS_RETRY:
596f482a 1453 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1aea6434
JA
1454 break;
1455 case ADD_TO_MLQUEUE:
1456 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1457 break;
1458 default:
a0658632 1459 scsi_eh_scmd_add(cmd);
2171b6d0 1460 break;
1aea6434
JA
1461 }
1462}
1463
82042a2c
CH
1464/**
1465 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1466 * @cmd: command block we are dispatching.
1467 *
1468 * Return: nonzero return request was rejected and device's queue needs to be
1469 * plugged.
1470 */
1471static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1472{
1473 struct Scsi_Host *host = cmd->device->host;
1474 int rtn = 0;
1475
1476 atomic_inc(&cmd->device->iorequest_cnt);
1477
1478 /* check if the device is still usable */
1479 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1480 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1481 * returns an immediate error upwards, and signals
1482 * that the device is no longer present */
1483 cmd->result = DID_NO_CONNECT << 16;
1484 goto done;
1485 }
1486
1487 /* Check to see if the scsi lld made this device blocked. */
1488 if (unlikely(scsi_device_blocked(cmd->device))) {
1489 /*
1490 * in blocked state, the command is just put back on
1491 * the device queue. The suspend state has already
1492 * blocked the queue so future requests should not
1493 * occur until the device transitions out of the
1494 * suspend state.
1495 */
1496 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1497 "queuecommand : device blocked\n"));
1498 return SCSI_MLQUEUE_DEVICE_BUSY;
1499 }
1500
1501 /* Store the LUN value in cmnd, if needed. */
1502 if (cmd->device->lun_in_cdb)
1503 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1504 (cmd->device->lun << 5 & 0xe0);
1505
1506 scsi_log_send(cmd);
1507
1508 /*
1509 * Before we queue this command, check if the command
1510 * length exceeds what the host adapter can handle.
1511 */
1512 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1513 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1514 "queuecommand : command too long. "
1515 "cdb_size=%d host->max_cmd_len=%d\n",
1516 cmd->cmd_len, cmd->device->host->max_cmd_len));
1517 cmd->result = (DID_ABORT << 16);
1518 goto done;
1519 }
1520
1521 if (unlikely(host->shost_state == SHOST_DEL)) {
1522 cmd->result = (DID_NO_CONNECT << 16);
1523 goto done;
1524
1525 }
1526
1527 trace_scsi_dispatch_cmd_start(cmd);
1528 rtn = host->hostt->queuecommand(host, cmd);
1529 if (rtn) {
1530 trace_scsi_dispatch_cmd_error(cmd, rtn);
1531 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1532 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1533 rtn = SCSI_MLQUEUE_HOST_BUSY;
1534
1535 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1536 "queuecommand : request rejected\n"));
1537 }
1538
1539 return rtn;
1540 done:
1541 cmd->scsi_done(cmd);
1542 return 0;
1543}
1544
be4c186c
BVA
1545/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1546static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
1547{
1548 return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
1549 sizeof(struct scatterlist);
1550}
1551
785ba83b 1552static blk_status_t scsi_mq_prep_fn(struct request *req)
d285203c
CH
1553{
1554 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1555 struct scsi_device *sdev = req->q->queuedata;
1556 struct Scsi_Host *shost = sdev->host;
d285203c
CH
1557 struct scatterlist *sg;
1558
08f78436 1559 scsi_init_command(sdev, cmd);
d285203c 1560
d285203c 1561 cmd->request = req;
d285203c 1562 cmd->tag = req->tag;
d285203c
CH
1563 cmd->prot_op = SCSI_PROT_NORMAL;
1564
d285203c
CH
1565 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1566 cmd->sdb.table.sgl = sg;
1567
1568 if (scsi_host_get_prot(shost)) {
d285203c
CH
1569 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1570
1571 cmd->prot_sdb->table.sgl =
1572 (struct scatterlist *)(cmd->prot_sdb + 1);
1573 }
1574
fe052529
CH
1575 blk_mq_start_request(req);
1576
8fe8ffb1 1577 return scsi_setup_cmnd(sdev, req);
d285203c
CH
1578}
1579
1580static void scsi_mq_done(struct scsi_cmnd *cmd)
1581{
f1342709
KB
1582 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1583 return;
d285203c 1584 trace_scsi_dispatch_cmd_done(cmd);
f1342709
KB
1585
1586 /*
1587 * If the block layer didn't complete the request due to a timeout
1588 * injection, scsi must clear its internal completed state so that the
1589 * timeout handler will see it needs to escalate its own error
1590 * recovery.
1591 */
1592 if (unlikely(!blk_mq_complete_request(cmd->request)))
1593 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
d285203c
CH
1594}
1595
0df21c86 1596static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
d285203c 1597{
0df21c86
ML
1598 struct request_queue *q = hctx->queue;
1599 struct scsi_device *sdev = q->queuedata;
0df21c86 1600
0df21c86 1601 atomic_dec(&sdev->device_busy);
0df21c86
ML
1602}
1603
88022d72 1604static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
0df21c86
ML
1605{
1606 struct request_queue *q = hctx->queue;
d285203c 1607 struct scsi_device *sdev = q->queuedata;
d285203c 1608
18c4f0a4
ML
1609 if (scsi_dev_queue_ready(q, sdev))
1610 return true;
0df21c86 1611
7e70aa78
ML
1612 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1613 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
88022d72 1614 return false;
0df21c86
ML
1615}
1616
fc17b653 1617static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
74c45052 1618 const struct blk_mq_queue_data *bd)
d285203c 1619{
74c45052 1620 struct request *req = bd->rq;
d285203c
CH
1621 struct request_queue *q = req->q;
1622 struct scsi_device *sdev = q->queuedata;
1623 struct Scsi_Host *shost = sdev->host;
1624 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
fc17b653 1625 blk_status_t ret;
d285203c
CH
1626 int reason;
1627
c092d4ec
CH
1628 /*
1629 * If the device is not in running state we will reject some or all
1630 * commands.
1631 */
1632 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1633 ret = scsi_prep_state_check(sdev, req);
1634 if (ret != BLK_STS_OK)
1635 goto out_put_budget;
1636 }
d285203c 1637
fc17b653 1638 ret = BLK_STS_RESOURCE;
d285203c 1639 if (!scsi_target_queue_ready(shost, sdev))
826a70a0 1640 goto out_put_budget;
d285203c
CH
1641 if (!scsi_host_queue_ready(q, shost, sdev))
1642 goto out_dec_target_busy;
1643
e8064021 1644 if (!(req->rq_flags & RQF_DONTPREP)) {
785ba83b 1645 ret = scsi_mq_prep_fn(req);
fc17b653 1646 if (ret != BLK_STS_OK)
d285203c 1647 goto out_dec_host_busy;
e8064021 1648 req->rq_flags |= RQF_DONTPREP;
fe052529 1649 } else {
cd464d83 1650 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
fe052529 1651 blk_mq_start_request(req);
d285203c
CH
1652 }
1653
125c99bc
CH
1654 if (sdev->simple_tags)
1655 cmd->flags |= SCMD_TAGGED;
b1dd2aac 1656 else
125c99bc 1657 cmd->flags &= ~SCMD_TAGGED;
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:
3bd6f43f 1672 scsi_dec_host_busy(shost);
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;
d285203c 1718
8e688254
BVA
1719 if (unchecked_isa_dma)
1720 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1721 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1722 GFP_KERNEL, numa_node);
d285203c
CH
1723 if (!cmd->sense_buffer)
1724 return -ENOMEM;
82ed4db4 1725 cmd->req.sense = cmd->sense_buffer;
08f78436
BVA
1726
1727 if (scsi_host_get_prot(shost)) {
1728 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1729 shost->hostt->cmd_size;
1730 cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
1731 }
1732
d285203c
CH
1733 return 0;
1734}
1735
e7008ff5
BVA
1736static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1737 unsigned int hctx_idx)
d285203c
CH
1738{
1739 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1740
8e688254
BVA
1741 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1742 cmd->sense_buffer);
d285203c
CH
1743}
1744
2d9c5c20
CH
1745static int scsi_map_queues(struct blk_mq_tag_set *set)
1746{
1747 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1748
1749 if (shost->hostt->map_queues)
1750 return shost->hostt->map_queues(shost);
99bbf484 1751 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
2d9c5c20
CH
1752}
1753
d48777a6 1754void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1da177e4 1755{
6f381fa3 1756 struct device *dev = shost->dma_dev;
1da177e4 1757
a8474ce2
JA
1758 /*
1759 * this limit is imposed by hardware restrictions
1760 */
8a78362c 1761 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
65e8617f 1762 SG_MAX_SEGMENTS));
a8474ce2 1763
13f05c8d
MP
1764 if (scsi_host_prot_dma(shost)) {
1765 shost->sg_prot_tablesize =
1766 min_not_zero(shost->sg_prot_tablesize,
1767 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1768 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1769 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1770 }
1771
086fa5ff 1772 blk_queue_max_hw_sectors(q, shost->max_sectors);
21e07dba
CH
1773 if (shost->unchecked_isa_dma)
1774 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1da177e4 1775 blk_queue_segment_boundary(q, shost->dma_boundary);
99c84dbd 1776 dma_set_seg_boundary(dev, shost->dma_boundary);
1da177e4 1777
a8cf59a6
CH
1778 blk_queue_max_segment_size(q, shost->max_segment_size);
1779 dma_set_max_seg_size(dev, shost->max_segment_size);
465ff318
JB
1780
1781 /*
90addc6b
HC
1782 * Set a reasonable default alignment: The larger of 32-byte (dword),
1783 * which is a common minimum for HBAs, and the minimum DMA alignment,
1784 * which is set by the platform.
1785 *
1786 * Devices that require a bigger alignment can increase it later.
465ff318 1787 */
90addc6b 1788 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
d285203c 1789}
d48777a6 1790EXPORT_SYMBOL_GPL(__scsi_init_queue);
465ff318 1791
f363b089 1792static const struct blk_mq_ops scsi_mq_ops = {
0df21c86
ML
1793 .get_budget = scsi_mq_get_budget,
1794 .put_budget = scsi_mq_put_budget,
d285203c
CH
1795 .queue_rq = scsi_queue_rq,
1796 .complete = scsi_softirq_done,
0152fb6b 1797 .timeout = scsi_timeout,
0eebd005
BVA
1798#ifdef CONFIG_BLK_DEBUG_FS
1799 .show_rq = scsi_show_rq,
1800#endif
e7008ff5
BVA
1801 .init_request = scsi_mq_init_request,
1802 .exit_request = scsi_mq_exit_request,
ca18d6f7 1803 .initialize_rq_fn = scsi_initialize_rq,
3a7ea2c4 1804 .busy = scsi_mq_lld_busy,
2d9c5c20 1805 .map_queues = scsi_map_queues,
d285203c
CH
1806};
1807
1808struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1809{
1810 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1811 if (IS_ERR(sdev->request_queue))
1812 return NULL;
1813
1814 sdev->request_queue->queuedata = sdev;
1815 __scsi_init_queue(sdev->host, sdev->request_queue);
17cb960f 1816 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
d285203c
CH
1817 return sdev->request_queue;
1818}
1819
1820int scsi_mq_setup_tags(struct Scsi_Host *shost)
1821{
be4c186c 1822 unsigned int cmd_size, sgl_size;
d285203c 1823
be4c186c 1824 sgl_size = scsi_mq_sgl_size(shost);
d285203c
CH
1825 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1826 if (scsi_host_get_prot(shost))
1827 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
1828
1829 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1830 shost->tag_set.ops = &scsi_mq_ops;
efec4b90 1831 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
d285203c
CH
1832 shost->tag_set.queue_depth = shost->can_queue;
1833 shost->tag_set.cmd_size = cmd_size;
1834 shost->tag_set.numa_node = NUMA_NO_NODE;
56d18f62 1835 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
24391c0d
SL
1836 shost->tag_set.flags |=
1837 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
d285203c
CH
1838 shost->tag_set.driver_data = shost;
1839
1840 return blk_mq_alloc_tag_set(&shost->tag_set);
1841}
1842
1843void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1844{
1845 blk_mq_free_tag_set(&shost->tag_set);
1846}
1847
857de6e0
HR
1848/**
1849 * scsi_device_from_queue - return sdev associated with a request_queue
1850 * @q: The request queue to return the sdev from
1851 *
1852 * Return the sdev associated with a request queue or NULL if the
1853 * request_queue does not reference a SCSI device.
1854 */
1855struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1856{
1857 struct scsi_device *sdev = NULL;
1858
f664a3cc 1859 if (q->mq_ops == &scsi_mq_ops)
857de6e0
HR
1860 sdev = q->queuedata;
1861 if (!sdev || !get_device(&sdev->sdev_gendev))
1862 sdev = NULL;
1863
1864 return sdev;
1865}
1866EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1867
1da177e4
LT
1868/*
1869 * Function: scsi_block_requests()
1870 *
1871 * Purpose: Utility function used by low-level drivers to prevent further
1872 * commands from being queued to the device.
1873 *
1874 * Arguments: shost - Host in question
1875 *
1876 * Returns: Nothing
1877 *
1878 * Lock status: No locks are assumed held.
1879 *
1880 * Notes: There is no timer nor any other means by which the requests
1881 * get unblocked other than the low-level driver calling
1882 * scsi_unblock_requests().
1883 */
1884void scsi_block_requests(struct Scsi_Host *shost)
1885{
1886 shost->host_self_blocked = 1;
1887}
1888EXPORT_SYMBOL(scsi_block_requests);
1889
1890/*
1891 * Function: scsi_unblock_requests()
1892 *
1893 * Purpose: Utility function used by low-level drivers to allow further
1894 * commands from being queued to the device.
1895 *
1896 * Arguments: shost - Host in question
1897 *
1898 * Returns: Nothing
1899 *
1900 * Lock status: No locks are assumed held.
1901 *
1902 * Notes: There is no timer nor any other means by which the requests
1903 * get unblocked other than the low-level driver calling
1904 * scsi_unblock_requests().
1905 *
1906 * This is done as an API function so that changes to the
1907 * internals of the scsi mid-layer won't require wholesale
1908 * changes to drivers that use this feature.
1909 */
1910void scsi_unblock_requests(struct Scsi_Host *shost)
1911{
1912 shost->host_self_blocked = 0;
1913 scsi_run_host_queues(shost);
1914}
1915EXPORT_SYMBOL(scsi_unblock_requests);
1916
1917int __init scsi_init_queue(void)
1918{
6362abd3
MP
1919 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1920 sizeof(struct scsi_data_buffer),
1921 0, 0, NULL);
1922 if (!scsi_sdb_cache) {
1923 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
f078727b 1924 return -ENOMEM;
6f9a35e2
BH
1925 }
1926
1da177e4
LT
1927 return 0;
1928}
1929
1930void scsi_exit_queue(void)
1931{
0a6ac4ee
CH
1932 kmem_cache_destroy(scsi_sense_cache);
1933 kmem_cache_destroy(scsi_sense_isadma_cache);
6362abd3 1934 kmem_cache_destroy(scsi_sdb_cache);
1da177e4 1935}
5baba830
JB
1936
1937/**
1938 * scsi_mode_select - issue a mode select
1939 * @sdev: SCSI device to be queried
1940 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1941 * @sp: Save page bit (0 == don't save, 1 == save)
1942 * @modepage: mode page being requested
1943 * @buffer: request buffer (may not be smaller than eight bytes)
1944 * @len: length of request buffer.
1945 * @timeout: command timeout
1946 * @retries: number of retries before failing
1947 * @data: returns a structure abstracting the mode header data
eb44820c 1948 * @sshdr: place to put sense data (or NULL if no sense to be collected).
5baba830
JB
1949 * must be SCSI_SENSE_BUFFERSIZE big.
1950 *
1951 * Returns zero if successful; negative error number or scsi
1952 * status on error
1953 *
1954 */
1955int
1956scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1957 unsigned char *buffer, int len, int timeout, int retries,
1958 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1959{
1960 unsigned char cmd[10];
1961 unsigned char *real_buffer;
1962 int ret;
1963
1964 memset(cmd, 0, sizeof(cmd));
1965 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1966
1967 if (sdev->use_10_for_ms) {
1968 if (len > 65535)
1969 return -EINVAL;
1970 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1971 if (!real_buffer)
1972 return -ENOMEM;
1973 memcpy(real_buffer + 8, buffer, len);
1974 len += 8;
1975 real_buffer[0] = 0;
1976 real_buffer[1] = 0;
1977 real_buffer[2] = data->medium_type;
1978 real_buffer[3] = data->device_specific;
1979 real_buffer[4] = data->longlba ? 0x01 : 0;
1980 real_buffer[5] = 0;
1981 real_buffer[6] = data->block_descriptor_length >> 8;
1982 real_buffer[7] = data->block_descriptor_length;
1983
1984 cmd[0] = MODE_SELECT_10;
1985 cmd[7] = len >> 8;
1986 cmd[8] = len;
1987 } else {
1988 if (len > 255 || data->block_descriptor_length > 255 ||
1989 data->longlba)
1990 return -EINVAL;
1991
1992 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1993 if (!real_buffer)
1994 return -ENOMEM;
1995 memcpy(real_buffer + 4, buffer, len);
1996 len += 4;
1997 real_buffer[0] = 0;
1998 real_buffer[1] = data->medium_type;
1999 real_buffer[2] = data->device_specific;
2000 real_buffer[3] = data->block_descriptor_length;
2001
2002
2003 cmd[0] = MODE_SELECT;
2004 cmd[4] = len;
2005 }
2006
2007 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
f4f4e47e 2008 sshdr, timeout, retries, NULL);
5baba830
JB
2009 kfree(real_buffer);
2010 return ret;
2011}
2012EXPORT_SYMBOL_GPL(scsi_mode_select);
2013
1da177e4 2014/**
eb44820c 2015 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1cf72699 2016 * @sdev: SCSI device to be queried
1da177e4
LT
2017 * @dbd: set if mode sense will allow block descriptors to be returned
2018 * @modepage: mode page being requested
2019 * @buffer: request buffer (may not be smaller than eight bytes)
2020 * @len: length of request buffer.
2021 * @timeout: command timeout
2022 * @retries: number of retries before failing
2023 * @data: returns a structure abstracting the mode header data
eb44820c 2024 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1cf72699 2025 * must be SCSI_SENSE_BUFFERSIZE big.
1da177e4
LT
2026 *
2027 * Returns zero if unsuccessful, or the header offset (either 4
2028 * or 8 depending on whether a six or ten byte command was
2029 * issued) if successful.
eb44820c 2030 */
1da177e4 2031int
1cf72699 2032scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1da177e4 2033 unsigned char *buffer, int len, int timeout, int retries,
5baba830
JB
2034 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2035{
1da177e4
LT
2036 unsigned char cmd[12];
2037 int use_10_for_ms;
2038 int header_length;
0ae80ba9 2039 int result, retry_count = retries;
ea73a9f2 2040 struct scsi_sense_hdr my_sshdr;
1da177e4
LT
2041
2042 memset(data, 0, sizeof(*data));
2043 memset(&cmd[0], 0, 12);
2044 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2045 cmd[2] = modepage;
2046
ea73a9f2
JB
2047 /* caller might not be interested in sense, but we need it */
2048 if (!sshdr)
2049 sshdr = &my_sshdr;
2050
1da177e4 2051 retry:
1cf72699 2052 use_10_for_ms = sdev->use_10_for_ms;
1da177e4
LT
2053
2054 if (use_10_for_ms) {
2055 if (len < 8)
2056 len = 8;
2057
2058 cmd[0] = MODE_SENSE_10;
2059 cmd[8] = len;
2060 header_length = 8;
2061 } else {
2062 if (len < 4)
2063 len = 4;
2064
2065 cmd[0] = MODE_SENSE;
2066 cmd[4] = len;
2067 header_length = 4;
2068 }
2069
1da177e4
LT
2070 memset(buffer, 0, len);
2071
1cf72699 2072 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
f4f4e47e 2073 sshdr, timeout, retries, NULL);
1da177e4
LT
2074
2075 /* This code looks awful: what it's doing is making sure an
2076 * ILLEGAL REQUEST sense return identifies the actual command
2077 * byte as the problem. MODE_SENSE commands can return
2078 * ILLEGAL REQUEST if the code page isn't supported */
2079
1cf72699 2080 if (use_10_for_ms && !scsi_status_is_good(result) &&
c65be1a6 2081 driver_byte(result) == DRIVER_SENSE) {
ea73a9f2
JB
2082 if (scsi_sense_valid(sshdr)) {
2083 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2084 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1da177e4
LT
2085 /*
2086 * Invalid command operation code
2087 */
1cf72699 2088 sdev->use_10_for_ms = 0;
1da177e4
LT
2089 goto retry;
2090 }
2091 }
2092 }
2093
1cf72699 2094 if(scsi_status_is_good(result)) {
6d73c851
AV
2095 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2096 (modepage == 6 || modepage == 8))) {
2097 /* Initio breakage? */
2098 header_length = 0;
2099 data->length = 13;
2100 data->medium_type = 0;
2101 data->device_specific = 0;
2102 data->longlba = 0;
2103 data->block_descriptor_length = 0;
2104 } else if(use_10_for_ms) {
1da177e4
LT
2105 data->length = buffer[0]*256 + buffer[1] + 2;
2106 data->medium_type = buffer[2];
2107 data->device_specific = buffer[3];
2108 data->longlba = buffer[4] & 0x01;
2109 data->block_descriptor_length = buffer[6]*256
2110 + buffer[7];
2111 } else {
2112 data->length = buffer[0] + 1;
2113 data->medium_type = buffer[1];
2114 data->device_specific = buffer[2];
2115 data->block_descriptor_length = buffer[3];
2116 }
6d73c851 2117 data->header_length = header_length;
0ae80ba9
HR
2118 } else if ((status_byte(result) == CHECK_CONDITION) &&
2119 scsi_sense_valid(sshdr) &&
2120 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2121 retry_count--;
2122 goto retry;
1da177e4
LT
2123 }
2124
1cf72699 2125 return result;
1da177e4
LT
2126}
2127EXPORT_SYMBOL(scsi_mode_sense);
2128
001aac25
JB
2129/**
2130 * scsi_test_unit_ready - test if unit is ready
2131 * @sdev: scsi device to change the state of.
2132 * @timeout: command timeout
2133 * @retries: number of retries before failing
74a78ebd 2134 * @sshdr: outpout pointer for decoded sense information.
001aac25
JB
2135 *
2136 * Returns zero if unsuccessful or an error if TUR failed. For
9f8a2c23 2137 * removable media, UNIT_ATTENTION sets ->changed flag.
001aac25 2138 **/
1da177e4 2139int
001aac25 2140scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
74a78ebd 2141 struct scsi_sense_hdr *sshdr)
1da177e4 2142{
1da177e4
LT
2143 char cmd[] = {
2144 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2145 };
2146 int result;
001aac25 2147
001aac25
JB
2148 /* try to eat the UNIT_ATTENTION if there are enough retries */
2149 do {
2150 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
9b91fd34 2151 timeout, 1, NULL);
32c356d7
JB
2152 if (sdev->removable && scsi_sense_valid(sshdr) &&
2153 sshdr->sense_key == UNIT_ATTENTION)
2154 sdev->changed = 1;
2155 } while (scsi_sense_valid(sshdr) &&
2156 sshdr->sense_key == UNIT_ATTENTION && --retries);
001aac25 2157
1da177e4
LT
2158 return result;
2159}
2160EXPORT_SYMBOL(scsi_test_unit_ready);
2161
2162/**
eb44820c 2163 * scsi_device_set_state - Take the given device through the device state model.
1da177e4
LT
2164 * @sdev: scsi device to change the state of.
2165 * @state: state to change to.
2166 *
23cb27fd 2167 * Returns zero if successful or an error if the requested
1da177e4 2168 * transition is illegal.
eb44820c 2169 */
1da177e4
LT
2170int
2171scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2172{
2173 enum scsi_device_state oldstate = sdev->sdev_state;
2174
2175 if (state == oldstate)
2176 return 0;
2177
2178 switch (state) {
2179 case SDEV_CREATED:
6f4267e3
JB
2180 switch (oldstate) {
2181 case SDEV_CREATED_BLOCK:
2182 break;
2183 default:
2184 goto illegal;
2185 }
2186 break;
1da177e4
LT
2187
2188 case SDEV_RUNNING:
2189 switch (oldstate) {
2190 case SDEV_CREATED:
2191 case SDEV_OFFLINE:
1b8d2620 2192 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2193 case SDEV_QUIESCE:
2194 case SDEV_BLOCK:
2195 break;
2196 default:
2197 goto illegal;
2198 }
2199 break;
2200
2201 case SDEV_QUIESCE:
2202 switch (oldstate) {
2203 case SDEV_RUNNING:
2204 case SDEV_OFFLINE:
1b8d2620 2205 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2206 break;
2207 default:
2208 goto illegal;
2209 }
2210 break;
2211
2212 case SDEV_OFFLINE:
1b8d2620 2213 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2214 switch (oldstate) {
2215 case SDEV_CREATED:
2216 case SDEV_RUNNING:
2217 case SDEV_QUIESCE:
2218 case SDEV_BLOCK:
2219 break;
2220 default:
2221 goto illegal;
2222 }
2223 break;
2224
2225 case SDEV_BLOCK:
2226 switch (oldstate) {
1da177e4 2227 case SDEV_RUNNING:
6f4267e3 2228 case SDEV_CREATED_BLOCK:
a33e5bfb 2229 case SDEV_OFFLINE:
6f4267e3
JB
2230 break;
2231 default:
2232 goto illegal;
2233 }
2234 break;
2235
2236 case SDEV_CREATED_BLOCK:
2237 switch (oldstate) {
2238 case SDEV_CREATED:
1da177e4
LT
2239 break;
2240 default:
2241 goto illegal;
2242 }
2243 break;
2244
2245 case SDEV_CANCEL:
2246 switch (oldstate) {
2247 case SDEV_CREATED:
2248 case SDEV_RUNNING:
9ea72909 2249 case SDEV_QUIESCE:
1da177e4 2250 case SDEV_OFFLINE:
1b8d2620 2251 case SDEV_TRANSPORT_OFFLINE:
1da177e4
LT
2252 break;
2253 default:
2254 goto illegal;
2255 }
2256 break;
2257
2258 case SDEV_DEL:
2259 switch (oldstate) {
309bd271
BK
2260 case SDEV_CREATED:
2261 case SDEV_RUNNING:
2262 case SDEV_OFFLINE:
1b8d2620 2263 case SDEV_TRANSPORT_OFFLINE:
1da177e4 2264 case SDEV_CANCEL:
255ee932 2265 case SDEV_BLOCK:
0516c08d 2266 case SDEV_CREATED_BLOCK:
1da177e4
LT
2267 break;
2268 default:
2269 goto illegal;
2270 }
2271 break;
2272
2273 }
2274 sdev->sdev_state = state;
2275 return 0;
2276
2277 illegal:
91921e01 2278 SCSI_LOG_ERROR_RECOVERY(1,
9ccfc756 2279 sdev_printk(KERN_ERR, sdev,
91921e01 2280 "Illegal state transition %s->%s",
9ccfc756
JB
2281 scsi_device_state_name(oldstate),
2282 scsi_device_state_name(state))
1da177e4
LT
2283 );
2284 return -EINVAL;
2285}
2286EXPORT_SYMBOL(scsi_device_set_state);
2287
a341cd0f
JG
2288/**
2289 * sdev_evt_emit - emit a single SCSI device uevent
2290 * @sdev: associated SCSI device
2291 * @evt: event to emit
2292 *
2293 * Send a single uevent (scsi_event) to the associated scsi_device.
2294 */
2295static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2296{
2297 int idx = 0;
2298 char *envp[3];
2299
2300 switch (evt->evt_type) {
2301 case SDEV_EVT_MEDIA_CHANGE:
2302 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2303 break;
279afdfe 2304 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
d3d32891 2305 scsi_rescan_device(&sdev->sdev_gendev);
279afdfe
EM
2306 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2307 break;
2308 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2309 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2310 break;
2311 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2312 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2313 break;
2314 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2315 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2316 break;
2317 case SDEV_EVT_LUN_CHANGE_REPORTED:
2318 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2319 break;
14c3e677
HR
2320 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2321 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2322 break;
cf3431bb
HR
2323 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2324 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2325 break;
a341cd0f
JG
2326 default:
2327 /* do nothing */
2328 break;
2329 }
2330
2331 envp[idx++] = NULL;
2332
2333 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2334}
2335
2336/**
2337 * sdev_evt_thread - send a uevent for each scsi event
2338 * @work: work struct for scsi_device
2339 *
2340 * Dispatch queued events to their associated scsi_device kobjects
2341 * as uevents.
2342 */
2343void scsi_evt_thread(struct work_struct *work)
2344{
2345 struct scsi_device *sdev;
279afdfe 2346 enum scsi_device_event evt_type;
a341cd0f
JG
2347 LIST_HEAD(event_list);
2348
2349 sdev = container_of(work, struct scsi_device, event_work);
2350
279afdfe
EM
2351 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2352 if (test_and_clear_bit(evt_type, sdev->pending_events))
2353 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2354
a341cd0f
JG
2355 while (1) {
2356 struct scsi_event *evt;
2357 struct list_head *this, *tmp;
2358 unsigned long flags;
2359
2360 spin_lock_irqsave(&sdev->list_lock, flags);
2361 list_splice_init(&sdev->event_list, &event_list);
2362 spin_unlock_irqrestore(&sdev->list_lock, flags);
2363
2364 if (list_empty(&event_list))
2365 break;
2366
2367 list_for_each_safe(this, tmp, &event_list) {
2368 evt = list_entry(this, struct scsi_event, node);
2369 list_del(&evt->node);
2370 scsi_evt_emit(sdev, evt);
2371 kfree(evt);
2372 }
2373 }
2374}
2375
2376/**
2377 * sdev_evt_send - send asserted event to uevent thread
2378 * @sdev: scsi_device event occurred on
2379 * @evt: event to send
2380 *
2381 * Assert scsi device event asynchronously.
2382 */
2383void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2384{
2385 unsigned long flags;
2386
4d1566ed
KS
2387#if 0
2388 /* FIXME: currently this check eliminates all media change events
2389 * for polled devices. Need to update to discriminate between AN
2390 * and polled events */
a341cd0f
JG
2391 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2392 kfree(evt);
2393 return;
2394 }
4d1566ed 2395#endif
a341cd0f
JG
2396
2397 spin_lock_irqsave(&sdev->list_lock, flags);
2398 list_add_tail(&evt->node, &sdev->event_list);
2399 schedule_work(&sdev->event_work);
2400 spin_unlock_irqrestore(&sdev->list_lock, flags);
2401}
2402EXPORT_SYMBOL_GPL(sdev_evt_send);
2403
2404/**
2405 * sdev_evt_alloc - allocate a new scsi event
2406 * @evt_type: type of event to allocate
2407 * @gfpflags: GFP flags for allocation
2408 *
2409 * Allocates and returns a new scsi_event.
2410 */
2411struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2412 gfp_t gfpflags)
2413{
2414 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2415 if (!evt)
2416 return NULL;
2417
2418 evt->evt_type = evt_type;
2419 INIT_LIST_HEAD(&evt->node);
2420
2421 /* evt_type-specific initialization, if any */
2422 switch (evt_type) {
2423 case SDEV_EVT_MEDIA_CHANGE:
279afdfe
EM
2424 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2425 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2426 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2427 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2428 case SDEV_EVT_LUN_CHANGE_REPORTED:
14c3e677 2429 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
cf3431bb 2430 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
a341cd0f
JG
2431 default:
2432 /* do nothing */
2433 break;
2434 }
2435
2436 return evt;
2437}
2438EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2439
2440/**
2441 * sdev_evt_send_simple - send asserted event to uevent thread
2442 * @sdev: scsi_device event occurred on
2443 * @evt_type: type of event to send
2444 * @gfpflags: GFP flags for allocation
2445 *
2446 * Assert scsi device event asynchronously, given an event type.
2447 */
2448void sdev_evt_send_simple(struct scsi_device *sdev,
2449 enum scsi_device_event evt_type, gfp_t gfpflags)
2450{
2451 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2452 if (!evt) {
2453 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2454 evt_type);
2455 return;
2456 }
2457
2458 sdev_evt_send(sdev, evt);
2459}
2460EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2461
1da177e4
LT
2462/**
2463 * scsi_device_quiesce - Block user issued commands.
2464 * @sdev: scsi device to quiesce.
2465 *
2466 * This works by trying to transition to the SDEV_QUIESCE state
2467 * (which must be a legal transition). When the device is in this
2468 * state, only special requests will be accepted, all others will
2469 * be deferred. Since special requests may also be requeued requests,
2470 * a successful return doesn't guarantee the device will be
2471 * totally quiescent.
2472 *
2473 * Must be called with user context, may sleep.
2474 *
2475 * Returns zero if unsuccessful or an error if not.
eb44820c 2476 */
1da177e4
LT
2477int
2478scsi_device_quiesce(struct scsi_device *sdev)
2479{
3a0a5299 2480 struct request_queue *q = sdev->request_queue;
0db6ca8a
BVA
2481 int err;
2482
3a0a5299
BVA
2483 /*
2484 * It is allowed to call scsi_device_quiesce() multiple times from
2485 * the same context but concurrent scsi_device_quiesce() calls are
2486 * not allowed.
2487 */
2488 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2489
cd84a62e
BVA
2490 if (sdev->quiesced_by == current)
2491 return 0;
2492
2493 blk_set_pm_only(q);
3a0a5299
BVA
2494
2495 blk_mq_freeze_queue(q);
2496 /*
cd84a62e 2497 * Ensure that the effect of blk_set_pm_only() will be visible
3a0a5299
BVA
2498 * for percpu_ref_tryget() callers that occur after the queue
2499 * unfreeze even if the queue was already frozen before this function
2500 * was called. See also https://lwn.net/Articles/573497/.
2501 */
2502 synchronize_rcu();
2503 blk_mq_unfreeze_queue(q);
2504
0db6ca8a
BVA
2505 mutex_lock(&sdev->state_mutex);
2506 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
3a0a5299
BVA
2507 if (err == 0)
2508 sdev->quiesced_by = current;
2509 else
cd84a62e 2510 blk_clear_pm_only(q);
0db6ca8a
BVA
2511 mutex_unlock(&sdev->state_mutex);
2512
3a0a5299 2513 return err;
1da177e4
LT
2514}
2515EXPORT_SYMBOL(scsi_device_quiesce);
2516
2517/**
2518 * scsi_device_resume - Restart user issued commands to a quiesced device.
2519 * @sdev: scsi device to resume.
2520 *
2521 * Moves the device from quiesced back to running and restarts the
2522 * queues.
2523 *
2524 * Must be called with user context, may sleep.
eb44820c 2525 */
a7a20d10 2526void scsi_device_resume(struct scsi_device *sdev)
1da177e4 2527{
a7a20d10
DW
2528 /* check if the device state was mutated prior to resume, and if
2529 * so assume the state is being managed elsewhere (for example
2530 * device deleted during suspend)
2531 */
0db6ca8a 2532 mutex_lock(&sdev->state_mutex);
17605afa
BVA
2533 if (sdev->quiesced_by) {
2534 sdev->quiesced_by = NULL;
2535 blk_clear_pm_only(sdev->request_queue);
2536 }
3a0a5299
BVA
2537 if (sdev->sdev_state == SDEV_QUIESCE)
2538 scsi_device_set_state(sdev, SDEV_RUNNING);
0db6ca8a 2539 mutex_unlock(&sdev->state_mutex);
1da177e4
LT
2540}
2541EXPORT_SYMBOL(scsi_device_resume);
2542
2543static void
2544device_quiesce_fn(struct scsi_device *sdev, void *data)
2545{
2546 scsi_device_quiesce(sdev);
2547}
2548
2549void
2550scsi_target_quiesce(struct scsi_target *starget)
2551{
2552 starget_for_each_device(starget, NULL, device_quiesce_fn);
2553}
2554EXPORT_SYMBOL(scsi_target_quiesce);
2555
2556static void
2557device_resume_fn(struct scsi_device *sdev, void *data)
2558{
2559 scsi_device_resume(sdev);
2560}
2561
2562void
2563scsi_target_resume(struct scsi_target *starget)
2564{
2565 starget_for_each_device(starget, NULL, device_resume_fn);
2566}
2567EXPORT_SYMBOL(scsi_target_resume);
2568
2569/**
551eb598
BVA
2570 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2571 * @sdev: device to block
1da177e4 2572 *
551eb598 2573 * Pause SCSI command processing on the specified device. Does not sleep.
1da177e4 2574 *
551eb598 2575 * Returns zero if successful or a negative error code upon failure.
669f0441 2576 *
551eb598
BVA
2577 * Notes:
2578 * This routine transitions the device to the SDEV_BLOCK state (which must be
2579 * a legal transition). When the device is in this state, command processing
2580 * is paused until the device leaves the SDEV_BLOCK state. See also
2581 * scsi_internal_device_unblock_nowait().
eb44820c 2582 */
551eb598 2583int scsi_internal_device_block_nowait(struct scsi_device *sdev)
1da177e4 2584{
165125e1 2585 struct request_queue *q = sdev->request_queue;
1da177e4
LT
2586 int err = 0;
2587
2588 err = scsi_device_set_state(sdev, SDEV_BLOCK);
6f4267e3
JB
2589 if (err) {
2590 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2591
2592 if (err)
2593 return err;
2594 }
1da177e4
LT
2595
2596 /*
2597 * The device has transitioned to SDEV_BLOCK. Stop the
2598 * block layer from calling the midlayer with this device's
2599 * request queue.
2600 */
f664a3cc 2601 blk_mq_quiesce_queue_nowait(q);
1da177e4
LT
2602 return 0;
2603}
551eb598
BVA
2604EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2605
1da177e4 2606/**
551eb598
BVA
2607 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2608 * @sdev: device to block
2609 *
2610 * Pause SCSI command processing on the specified device and wait until all
2611 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
1da177e4 2612 *
551eb598 2613 * Returns zero if successful or a negative error code upon failure.
1da177e4 2614 *
551eb598
BVA
2615 * Note:
2616 * This routine transitions the device to the SDEV_BLOCK state (which must be
2617 * a legal transition). When the device is in this state, command processing
2618 * is paused until the device leaves the SDEV_BLOCK state. See also
2619 * scsi_internal_device_unblock().
1da177e4 2620 *
551eb598
BVA
2621 * To do: avoid that scsi_send_eh_cmnd() calls queuecommand() after
2622 * scsi_internal_device_block() has blocked a SCSI device and also
2623 * remove the rport mutex lock and unlock calls from srp_queuecommand().
eb44820c 2624 */
551eb598 2625static int scsi_internal_device_block(struct scsi_device *sdev)
1da177e4 2626{
551eb598
BVA
2627 struct request_queue *q = sdev->request_queue;
2628 int err;
2629
0db6ca8a 2630 mutex_lock(&sdev->state_mutex);
551eb598 2631 err = scsi_internal_device_block_nowait(sdev);
f664a3cc
JA
2632 if (err == 0)
2633 blk_mq_quiesce_queue(q);
0db6ca8a
BVA
2634 mutex_unlock(&sdev->state_mutex);
2635
551eb598
BVA
2636 return err;
2637}
1da177e4 2638
66483a4a
BVA
2639void scsi_start_queue(struct scsi_device *sdev)
2640{
2641 struct request_queue *q = sdev->request_queue;
5d9fb5cc 2642
f664a3cc 2643 blk_mq_unquiesce_queue(q);
66483a4a
BVA
2644}
2645
1da177e4 2646/**
43f7571b 2647 * scsi_internal_device_unblock_nowait - resume a device after a block request
1da177e4 2648 * @sdev: device to resume
43f7571b 2649 * @new_state: state to set the device to after unblocking
1da177e4 2650 *
43f7571b
BVA
2651 * Restart the device queue for a previously suspended SCSI device. Does not
2652 * sleep.
1da177e4 2653 *
43f7571b 2654 * Returns zero if successful or a negative error code upon failure.
1da177e4 2655 *
43f7571b
BVA
2656 * Notes:
2657 * This routine transitions the device to the SDEV_RUNNING state or to one of
2658 * the offline states (which must be a legal transition) allowing the midlayer
2659 * to goose the queue for this device.
eb44820c 2660 */
43f7571b
BVA
2661int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2662 enum scsi_device_state new_state)
1da177e4 2663{
5d9fb5cc
MC
2664 /*
2665 * Try to transition the scsi device to SDEV_RUNNING or one of the
2666 * offlined states and goose the device queue if successful.
1da177e4 2667 */
8cd1ec78
HR
2668 switch (sdev->sdev_state) {
2669 case SDEV_BLOCK:
2670 case SDEV_TRANSPORT_OFFLINE:
5d9fb5cc 2671 sdev->sdev_state = new_state;
8cd1ec78
HR
2672 break;
2673 case SDEV_CREATED_BLOCK:
5d9fb5cc
MC
2674 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2675 new_state == SDEV_OFFLINE)
2676 sdev->sdev_state = new_state;
2677 else
2678 sdev->sdev_state = SDEV_CREATED;
8cd1ec78
HR
2679 break;
2680 case SDEV_CANCEL:
2681 case SDEV_OFFLINE:
2682 break;
2683 default:
5c10e63c 2684 return -EINVAL;
8cd1ec78 2685 }
66483a4a 2686 scsi_start_queue(sdev);
1da177e4
LT
2687
2688 return 0;
2689}
43f7571b
BVA
2690EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2691
2692/**
2693 * scsi_internal_device_unblock - resume a device after a block request
2694 * @sdev: device to resume
2695 * @new_state: state to set the device to after unblocking
2696 *
2697 * Restart the device queue for a previously suspended SCSI device. May sleep.
2698 *
2699 * Returns zero if successful or a negative error code upon failure.
2700 *
2701 * Notes:
2702 * This routine transitions the device to the SDEV_RUNNING state or to one of
2703 * the offline states (which must be a legal transition) allowing the midlayer
2704 * to goose the queue for this device.
2705 */
2706static int scsi_internal_device_unblock(struct scsi_device *sdev,
2707 enum scsi_device_state new_state)
2708{
0db6ca8a
BVA
2709 int ret;
2710
2711 mutex_lock(&sdev->state_mutex);
2712 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2713 mutex_unlock(&sdev->state_mutex);
2714
2715 return ret;
43f7571b 2716}
1da177e4
LT
2717
2718static void
2719device_block(struct scsi_device *sdev, void *data)
2720{
551eb598 2721 scsi_internal_device_block(sdev);
1da177e4
LT
2722}
2723
2724static int
2725target_block(struct device *dev, void *data)
2726{
2727 if (scsi_is_target_device(dev))
2728 starget_for_each_device(to_scsi_target(dev), NULL,
2729 device_block);
2730 return 0;
2731}
2732
2733void
2734scsi_target_block(struct device *dev)
2735{
2736 if (scsi_is_target_device(dev))
2737 starget_for_each_device(to_scsi_target(dev), NULL,
2738 device_block);
2739 else
2740 device_for_each_child(dev, NULL, target_block);
2741}
2742EXPORT_SYMBOL_GPL(scsi_target_block);
2743
2744static void
2745device_unblock(struct scsi_device *sdev, void *data)
2746{
5d9fb5cc 2747 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
1da177e4
LT
2748}
2749
2750static int
2751target_unblock(struct device *dev, void *data)
2752{
2753 if (scsi_is_target_device(dev))
5d9fb5cc 2754 starget_for_each_device(to_scsi_target(dev), data,
1da177e4
LT
2755 device_unblock);
2756 return 0;
2757}
2758
2759void
5d9fb5cc 2760scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
1da177e4
LT
2761{
2762 if (scsi_is_target_device(dev))
5d9fb5cc 2763 starget_for_each_device(to_scsi_target(dev), &new_state,
1da177e4
LT
2764 device_unblock);
2765 else
5d9fb5cc 2766 device_for_each_child(dev, &new_state, target_unblock);
1da177e4
LT
2767}
2768EXPORT_SYMBOL_GPL(scsi_target_unblock);
cdb8c2a6
GL
2769
2770/**
2771 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
eb44820c 2772 * @sgl: scatter-gather list
cdb8c2a6
GL
2773 * @sg_count: number of segments in sg
2774 * @offset: offset in bytes into sg, on return offset into the mapped area
2775 * @len: bytes to map, on return number of bytes mapped
2776 *
2777 * Returns virtual address of the start of the mapped page
2778 */
c6132da1 2779void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
cdb8c2a6
GL
2780 size_t *offset, size_t *len)
2781{
2782 int i;
2783 size_t sg_len = 0, len_complete = 0;
c6132da1 2784 struct scatterlist *sg;
cdb8c2a6
GL
2785 struct page *page;
2786
22cfefb5
AM
2787 WARN_ON(!irqs_disabled());
2788
c6132da1 2789 for_each_sg(sgl, sg, sg_count, i) {
cdb8c2a6 2790 len_complete = sg_len; /* Complete sg-entries */
c6132da1 2791 sg_len += sg->length;
cdb8c2a6
GL
2792 if (sg_len > *offset)
2793 break;
2794 }
2795
2796 if (unlikely(i == sg_count)) {
169e1a2a
AM
2797 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2798 "elements %d\n",
cadbd4a5 2799 __func__, sg_len, *offset, sg_count);
cdb8c2a6
GL
2800 WARN_ON(1);
2801 return NULL;
2802 }
2803
2804 /* Offset starting from the beginning of first page in this sg-entry */
c6132da1 2805 *offset = *offset - len_complete + sg->offset;
cdb8c2a6
GL
2806
2807 /* Assumption: contiguous pages can be accessed as "page + i" */
45711f1a 2808 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
cdb8c2a6
GL
2809 *offset &= ~PAGE_MASK;
2810
2811 /* Bytes in this sg-entry from *offset to the end of the page */
2812 sg_len = PAGE_SIZE - *offset;
2813 if (*len > sg_len)
2814 *len = sg_len;
2815
77dfce07 2816 return kmap_atomic(page);
cdb8c2a6
GL
2817}
2818EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2819
2820/**
eb44820c 2821 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
cdb8c2a6
GL
2822 * @virt: virtual address to be unmapped
2823 */
2824void scsi_kunmap_atomic_sg(void *virt)
2825{
77dfce07 2826 kunmap_atomic(virt);
cdb8c2a6
GL
2827}
2828EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
6f4c827e
AL
2829
2830void sdev_disable_disk_events(struct scsi_device *sdev)
2831{
2832 atomic_inc(&sdev->disk_events_disable_depth);
2833}
2834EXPORT_SYMBOL(sdev_disable_disk_events);
2835
2836void sdev_enable_disk_events(struct scsi_device *sdev)
2837{
2838 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2839 return;
2840 atomic_dec(&sdev->disk_events_disable_depth);
2841}
2842EXPORT_SYMBOL(sdev_enable_disk_events);
9983bed3
HR
2843
2844/**
2845 * scsi_vpd_lun_id - return a unique device identification
2846 * @sdev: SCSI device
2847 * @id: buffer for the identification
2848 * @id_len: length of the buffer
2849 *
2850 * Copies a unique device identification into @id based
2851 * on the information in the VPD page 0x83 of the device.
2852 * The string will be formatted as a SCSI name string.
2853 *
2854 * Returns the length of the identification or error on failure.
2855 * If the identifier is longer than the supplied buffer the actual
2856 * identifier length is returned and the buffer is not zero-padded.
2857 */
2858int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2859{
2860 u8 cur_id_type = 0xff;
2861 u8 cur_id_size = 0;
ccf1e004
BVA
2862 const unsigned char *d, *cur_id_str;
2863 const struct scsi_vpd *vpd_pg83;
9983bed3
HR
2864 int id_size = -EINVAL;
2865
2866 rcu_read_lock();
2867 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2868 if (!vpd_pg83) {
2869 rcu_read_unlock();
2870 return -ENXIO;
2871 }
2872
2873 /*
2874 * Look for the correct descriptor.
2875 * Order of preference for lun descriptor:
2876 * - SCSI name string
2877 * - NAA IEEE Registered Extended
2878 * - EUI-64 based 16-byte
2879 * - EUI-64 based 12-byte
2880 * - NAA IEEE Registered
2881 * - NAA IEEE Extended
d230823a 2882 * - T10 Vendor ID
9983bed3
HR
2883 * as longer descriptors reduce the likelyhood
2884 * of identification clashes.
2885 */
2886
2887 /* The id string must be at least 20 bytes + terminating NULL byte */
2888 if (id_len < 21) {
2889 rcu_read_unlock();
2890 return -EINVAL;
2891 }
2892
2893 memset(id, 0, id_len);
ccf1e004
BVA
2894 d = vpd_pg83->data + 4;
2895 while (d < vpd_pg83->data + vpd_pg83->len) {
9983bed3
HR
2896 /* Skip designators not referring to the LUN */
2897 if ((d[1] & 0x30) != 0x00)
2898 goto next_desig;
2899
2900 switch (d[1] & 0xf) {
d230823a
HR
2901 case 0x1:
2902 /* T10 Vendor ID */
2903 if (cur_id_size > d[3])
2904 break;
2905 /* Prefer anything */
2906 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2907 break;
2908 cur_id_size = d[3];
2909 if (cur_id_size + 4 > id_len)
2910 cur_id_size = id_len - 4;
2911 cur_id_str = d + 4;
2912 cur_id_type = d[1] & 0xf;
2913 id_size = snprintf(id, id_len, "t10.%*pE",
2914 cur_id_size, cur_id_str);
2915 break;
9983bed3
HR
2916 case 0x2:
2917 /* EUI-64 */
2918 if (cur_id_size > d[3])
2919 break;
2920 /* Prefer NAA IEEE Registered Extended */
2921 if (cur_id_type == 0x3 &&
2922 cur_id_size == d[3])
2923 break;
2924 cur_id_size = d[3];
2925 cur_id_str = d + 4;
2926 cur_id_type = d[1] & 0xf;
2927 switch (cur_id_size) {
2928 case 8:
2929 id_size = snprintf(id, id_len,
2930 "eui.%8phN",
2931 cur_id_str);
2932 break;
2933 case 12:
2934 id_size = snprintf(id, id_len,
2935 "eui.%12phN",
2936 cur_id_str);
2937 break;
2938 case 16:
2939 id_size = snprintf(id, id_len,
2940 "eui.%16phN",
2941 cur_id_str);
2942 break;
2943 default:
2944 cur_id_size = 0;
2945 break;
2946 }
2947 break;
2948 case 0x3:
2949 /* NAA */
2950 if (cur_id_size > d[3])
2951 break;
2952 cur_id_size = d[3];
2953 cur_id_str = d + 4;
2954 cur_id_type = d[1] & 0xf;
2955 switch (cur_id_size) {
2956 case 8:
2957 id_size = snprintf(id, id_len,
2958 "naa.%8phN",
2959 cur_id_str);
2960 break;
2961 case 16:
2962 id_size = snprintf(id, id_len,
2963 "naa.%16phN",
2964 cur_id_str);
2965 break;
2966 default:
2967 cur_id_size = 0;
2968 break;
2969 }
2970 break;
2971 case 0x8:
2972 /* SCSI name string */
2973 if (cur_id_size + 4 > d[3])
2974 break;
2975 /* Prefer others for truncated descriptor */
2976 if (cur_id_size && d[3] > id_len)
2977 break;
2978 cur_id_size = id_size = d[3];
2979 cur_id_str = d + 4;
2980 cur_id_type = d[1] & 0xf;
2981 if (cur_id_size >= id_len)
2982 cur_id_size = id_len - 1;
2983 memcpy(id, cur_id_str, cur_id_size);
2984 /* Decrease priority for truncated descriptor */
2985 if (cur_id_size != id_size)
2986 cur_id_size = 6;
2987 break;
2988 default:
2989 break;
2990 }
2991next_desig:
2992 d += d[3] + 4;
2993 }
2994 rcu_read_unlock();
2995
2996 return id_size;
2997}
2998EXPORT_SYMBOL(scsi_vpd_lun_id);
a8aa3978
HR
2999
3000/*
3001 * scsi_vpd_tpg_id - return a target port group identifier
3002 * @sdev: SCSI device
3003 *
3004 * Returns the Target Port Group identifier from the information
3005 * froom VPD page 0x83 of the device.
3006 *
3007 * Returns the identifier or error on failure.
3008 */
3009int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3010{
ccf1e004
BVA
3011 const unsigned char *d;
3012 const struct scsi_vpd *vpd_pg83;
a8aa3978
HR
3013 int group_id = -EAGAIN, rel_port = -1;
3014
3015 rcu_read_lock();
3016 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3017 if (!vpd_pg83) {
3018 rcu_read_unlock();
3019 return -ENXIO;
3020 }
3021
ccf1e004
BVA
3022 d = vpd_pg83->data + 4;
3023 while (d < vpd_pg83->data + vpd_pg83->len) {
a8aa3978
HR
3024 switch (d[1] & 0xf) {
3025 case 0x4:
3026 /* Relative target port */
3027 rel_port = get_unaligned_be16(&d[6]);
3028 break;
3029 case 0x5:
3030 /* Target port group */
3031 group_id = get_unaligned_be16(&d[6]);
3032 break;
3033 default:
3034 break;
3035 }
3036 d += d[3] + 4;
3037 }
3038 rcu_read_unlock();
3039
3040 if (group_id >= 0 && rel_id && rel_port != -1)
3041 *rel_id = rel_port;
3042
3043 return group_id;
3044}
3045EXPORT_SYMBOL(scsi_vpd_tpg_id);