1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12#include <linux/bio.h>
13#include <linux/bitops.h>
14#include <linux/blkdev.h>
15#include <linux/completion.h>
16#include <linux/kernel.h>
17#include <linux/export.h>
18#include <linux/init.h>
19#include <linux/pci.h>
20#include <linux/delay.h>
21#include <linux/hardirq.h>
22#include <linux/scatterlist.h>
23#include <linux/blk-mq.h>
24#include <linux/blk-integrity.h>
25#include <linux/ratelimit.h>
26#include <linux/unaligned.h>
27
28#include <scsi/scsi.h>
29#include <scsi/scsi_cmnd.h>
30#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_driver.h>
33#include <scsi/scsi_eh.h>
34#include <scsi/scsi_host.h>
35#include <scsi/scsi_transport.h> /* scsi_init_limits() */
36#include <scsi/scsi_dh.h>
37
38#include <trace/events/scsi.h>
39
40#include "scsi_debugfs.h"
41#include "scsi_priv.h"
42#include "scsi_logging.h"
43
44/*
45 * Size of integrity metadata is usually small, 1 inline sg should
46 * cover normal cases.
47 */
48#ifdef CONFIG_ARCH_NO_SG_CHAIN
49#define SCSI_INLINE_PROT_SG_CNT 0
50#define SCSI_INLINE_SG_CNT 0
51#else
52#define SCSI_INLINE_PROT_SG_CNT 1
53#define SCSI_INLINE_SG_CNT 2
54#endif
55
56static struct kmem_cache *scsi_sense_cache;
57static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
61int scsi_init_sense_cache(struct Scsi_Host *shost)
62{
63 int ret = 0;
64
65 mutex_lock(lock: &scsi_sense_cache_mutex);
66 if (!scsi_sense_cache) {
67 scsi_sense_cache =
68 kmem_cache_create_usercopy(name: "scsi_sense_cache",
69 SCSI_SENSE_BUFFERSIZE, align: 0, SLAB_HWCACHE_ALIGN,
70 useroffset: 0, SCSI_SENSE_BUFFERSIZE, NULL);
71 if (!scsi_sense_cache)
72 ret = -ENOMEM;
73 }
74 mutex_unlock(lock: &scsi_sense_cache_mutex);
75 return ret;
76}
77
78static void
79scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80{
81 struct Scsi_Host *host = cmd->device->host;
82 struct scsi_device *device = cmd->device;
83 struct scsi_target *starget = scsi_target(sdev: device);
84
85 /*
86 * Set the appropriate busy bit for the device/host.
87 *
88 * If the host/device isn't busy, assume that something actually
89 * completed, and that we should be able to queue a command now.
90 *
91 * Note that the prior mid-layer assumption that any host could
92 * always queue at least one command is now broken. The mid-layer
93 * will implement a user specifiable stall (see
94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 * if a command is requeued with no other commands outstanding
96 * either for the device or for the host.
97 */
98 switch (reason) {
99 case SCSI_MLQUEUE_HOST_BUSY:
100 atomic_set(v: &host->host_blocked, i: host->max_host_blocked);
101 break;
102 case SCSI_MLQUEUE_DEVICE_BUSY:
103 case SCSI_MLQUEUE_EH_RETRY:
104 atomic_set(v: &device->device_blocked,
105 i: device->max_device_blocked);
106 break;
107 case SCSI_MLQUEUE_TARGET_BUSY:
108 atomic_set(v: &starget->target_blocked,
109 i: starget->max_target_blocked);
110 break;
111 }
112}
113
114static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115{
116 struct request *rq = scsi_cmd_to_rq(scmd: cmd);
117
118 if (rq->rq_flags & RQF_DONTPREP) {
119 rq->rq_flags &= ~RQF_DONTPREP;
120 scsi_mq_uninit_cmd(cmd);
121 } else {
122 WARN_ON_ONCE(true);
123 }
124
125 blk_mq_requeue_request(rq, kick_requeue_list: false);
126 if (!scsi_host_in_recovery(shost: cmd->device->host))
127 blk_mq_delay_kick_requeue_list(q: rq->q, msecs);
128}
129
130/**
131 * __scsi_queue_insert - private queue insertion
132 * @cmd: The SCSI command being requeued
133 * @reason: The reason for the requeue
134 * @unbusy: Whether the queue should be unbusied
135 *
136 * This is a private queue insertion. The public interface
137 * scsi_queue_insert() always assumes the queue should be unbusied
138 * because it's always called before the completion. This function is
139 * for a requeue after completion, which should only occur in this
140 * file.
141 */
142static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143{
144 struct scsi_device *device = cmd->device;
145
146 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 "Inserting command %p into mlqueue\n", cmd));
148
149 scsi_set_blocked(cmd, reason);
150
151 /*
152 * Decrement the counters, since these commands are no longer
153 * active on the host/device.
154 */
155 if (unbusy)
156 scsi_device_unbusy(sdev: device, cmd);
157
158 /*
159 * Requeue this command. It will go before all other commands
160 * that are already in the queue. Schedule requeue work under
161 * lock such that the kblockd_schedule_work() call happens
162 * before blk_mq_destroy_queue() finishes.
163 */
164 cmd->result = 0;
165
166 blk_mq_requeue_request(rq: scsi_cmd_to_rq(scmd: cmd),
167 kick_requeue_list: !scsi_host_in_recovery(shost: cmd->device->host));
168}
169
170/**
171 * scsi_queue_insert - Reinsert a command in the queue.
172 * @cmd: command that we are adding to queue.
173 * @reason: why we are inserting command to queue.
174 *
175 * We do this for one of two cases. Either the host is busy and it cannot accept
176 * any more commands for the time being, or the device returned QUEUE_FULL and
177 * can accept no more commands.
178 *
179 * Context: This could be called either from an interrupt context or a normal
180 * process context.
181 */
182void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183{
184 __scsi_queue_insert(cmd, reason, unbusy: true);
185}
186
187/**
188 * scsi_failures_reset_retries - reset all failures to zero
189 * @failures: &struct scsi_failures with specific failure modes set
190 */
191void scsi_failures_reset_retries(struct scsi_failures *failures)
192{
193 struct scsi_failure *failure;
194
195 failures->total_retries = 0;
196
197 for (failure = failures->failure_definitions; failure->result;
198 failure++)
199 failure->retries = 0;
200}
201EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);
202
203/**
204 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.
205 * @scmd: scsi_cmnd to check.
206 * @failures: scsi_failures struct that lists failures to check for.
207 *
208 * Returns -EAGAIN if the caller should retry else 0.
209 */
210static int scsi_check_passthrough(struct scsi_cmnd *scmd,
211 struct scsi_failures *failures)
212{
213 struct scsi_failure *failure;
214 struct scsi_sense_hdr sshdr;
215 enum sam_status status;
216
217 if (!scmd->result)
218 return 0;
219
220 if (!failures)
221 return 0;
222
223 for (failure = failures->failure_definitions; failure->result;
224 failure++) {
225 if (failure->result == SCMD_FAILURE_RESULT_ANY)
226 goto maybe_retry;
227
228 if (host_byte(scmd->result) &&
229 host_byte(scmd->result) == host_byte(failure->result))
230 goto maybe_retry;
231
232 status = status_byte(scmd->result);
233 if (!status)
234 continue;
235
236 if (failure->result == SCMD_FAILURE_STAT_ANY &&
237 !scsi_status_is_good(status: scmd->result))
238 goto maybe_retry;
239
240 if (status != status_byte(failure->result))
241 continue;
242
243 if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||
244 failure->sense == SCMD_FAILURE_SENSE_ANY)
245 goto maybe_retry;
246
247 if (!scsi_command_normalize_sense(cmd: scmd, sshdr: &sshdr))
248 return 0;
249
250 if (failure->sense != sshdr.sense_key)
251 continue;
252
253 if (failure->asc == SCMD_FAILURE_ASC_ANY)
254 goto maybe_retry;
255
256 if (failure->asc != sshdr.asc)
257 continue;
258
259 if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||
260 failure->ascq == sshdr.ascq)
261 goto maybe_retry;
262 }
263
264 return 0;
265
266maybe_retry:
267 if (failure->allowed) {
268 if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||
269 ++failure->retries <= failure->allowed)
270 return -EAGAIN;
271 } else {
272 if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||
273 ++failures->total_retries <= failures->total_allowed)
274 return -EAGAIN;
275 }
276
277 return 0;
278}
279
280/**
281 * scsi_execute_cmd - insert request and wait for the result
282 * @sdev: scsi_device
283 * @cmd: scsi command
284 * @opf: block layer request cmd_flags
285 * @buffer: data buffer
286 * @bufflen: len of buffer
287 * @timeout: request timeout in HZ
288 * @ml_retries: number of times SCSI midlayer will retry request
289 * @args: Optional args. See struct definition for field descriptions
290 *
291 * Returns the scsi_cmnd result field if a command was executed, or a negative
292 * Linux error code if we didn't get that far.
293 */
294int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
295 blk_opf_t opf, void *buffer, unsigned int bufflen,
296 int timeout, int ml_retries,
297 const struct scsi_exec_args *args)
298{
299 static const struct scsi_exec_args default_args;
300 struct request *req;
301 struct scsi_cmnd *scmd;
302 int ret;
303
304 if (!args)
305 args = &default_args;
306 else if (WARN_ON_ONCE(args->sense &&
307 args->sense_len != SCSI_SENSE_BUFFERSIZE))
308 return -EINVAL;
309
310retry:
311 req = scsi_alloc_request(q: sdev->request_queue, opf, flags: args->req_flags);
312 if (IS_ERR(ptr: req))
313 return PTR_ERR(ptr: req);
314
315 if (bufflen) {
316 ret = blk_rq_map_kern(rq: req, kbuf: buffer, len: bufflen, GFP_NOIO);
317 if (ret)
318 goto out;
319 }
320 scmd = blk_mq_rq_to_pdu(rq: req);
321 scmd->cmd_len = COMMAND_SIZE(cmd[0]);
322 memcpy(to: scmd->cmnd, from: cmd, len: scmd->cmd_len);
323 scmd->allowed = ml_retries;
324 scmd->flags |= args->scmd_flags;
325 req->timeout = timeout;
326 req->rq_flags |= RQF_QUIET;
327
328 /*
329 * head injection *required* here otherwise quiesce won't work
330 */
331 blk_execute_rq(rq: req, at_head: true);
332
333 if (scsi_check_passthrough(scmd, failures: args->failures) == -EAGAIN) {
334 blk_mq_free_request(rq: req);
335 goto retry;
336 }
337
338 /*
339 * Some devices (USB mass-storage in particular) may transfer
340 * garbage data together with a residue indicating that the data
341 * is invalid. Prevent the garbage from being misinterpreted
342 * and prevent security leaks by zeroing out the excess data.
343 */
344 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
345 memset(s: buffer + bufflen - scmd->resid_len, c: 0, n: scmd->resid_len);
346
347 if (args->resid)
348 *args->resid = scmd->resid_len;
349 if (args->sense)
350 memcpy(to: args->sense, from: scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
351 if (args->sshdr)
352 scsi_normalize_sense(sense_buffer: scmd->sense_buffer, sb_len: scmd->sense_len,
353 sshdr: args->sshdr);
354
355 ret = scmd->result;
356 out:
357 blk_mq_free_request(rq: req);
358
359 return ret;
360}
361EXPORT_SYMBOL(scsi_execute_cmd);
362
363/*
364 * Wake up the error handler if necessary. Avoid as follows that the error
365 * handler is not woken up if host in-flight requests number ==
366 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
367 * with an RCU read lock in this function to ensure that this function in
368 * its entirety either finishes before scsi_eh_scmd_add() increases the
369 * host_failed counter or that it notices the shost state change made by
370 * scsi_eh_scmd_add().
371 */
372static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
373{
374 unsigned long flags;
375
376 rcu_read_lock();
377 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
378 if (unlikely(scsi_host_in_recovery(shost))) {
379 unsigned int busy = scsi_host_busy(shost);
380
381 spin_lock_irqsave(shost->host_lock, flags);
382 if (shost->host_failed || shost->host_eh_scheduled)
383 scsi_eh_wakeup(shost, busy);
384 spin_unlock_irqrestore(lock: shost->host_lock, flags);
385 }
386 rcu_read_unlock();
387}
388
389void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
390{
391 struct Scsi_Host *shost = sdev->host;
392 struct scsi_target *starget = scsi_target(sdev);
393
394 scsi_dec_host_busy(shost, cmd);
395
396 if (starget->can_queue > 0)
397 atomic_dec(v: &starget->target_busy);
398
399 sbitmap_put(sb: &sdev->budget_map, bitnr: cmd->budget_token);
400 cmd->budget_token = -1;
401}
402
403/*
404 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
405 * interrupts disabled.
406 */
407static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
408{
409 struct scsi_device *current_sdev = data;
410
411 if (sdev != current_sdev)
412 blk_mq_run_hw_queues(q: sdev->request_queue, async: true);
413}
414
415/*
416 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
417 * and call blk_run_queue for all the scsi_devices on the target -
418 * including current_sdev first.
419 *
420 * Called with *no* scsi locks held.
421 */
422static void scsi_single_lun_run(struct scsi_device *current_sdev)
423{
424 struct Scsi_Host *shost = current_sdev->host;
425 struct scsi_target *starget = scsi_target(sdev: current_sdev);
426 unsigned long flags;
427
428 spin_lock_irqsave(shost->host_lock, flags);
429 starget->starget_sdev_user = NULL;
430 spin_unlock_irqrestore(lock: shost->host_lock, flags);
431
432 /*
433 * Call blk_run_queue for all LUNs on the target, starting with
434 * current_sdev. We race with others (to set starget_sdev_user),
435 * but in most cases, we will be first. Ideally, each LU on the
436 * target would get some limited time or requests on the target.
437 */
438 blk_mq_run_hw_queues(q: current_sdev->request_queue,
439 async: shost->queuecommand_may_block);
440
441 spin_lock_irqsave(shost->host_lock, flags);
442 if (!starget->starget_sdev_user)
443 __starget_for_each_device(starget, current_sdev,
444 fn: scsi_kick_sdev_queue);
445 spin_unlock_irqrestore(lock: shost->host_lock, flags);
446}
447
448static inline bool scsi_device_is_busy(struct scsi_device *sdev)
449{
450 if (scsi_device_busy(sdev) >= sdev->queue_depth)
451 return true;
452 if (atomic_read(v: &sdev->device_blocked) > 0)
453 return true;
454 return false;
455}
456
457static inline bool scsi_target_is_busy(struct scsi_target *starget)
458{
459 if (starget->can_queue > 0) {
460 if (atomic_read(v: &starget->target_busy) >= starget->can_queue)
461 return true;
462 if (atomic_read(v: &starget->target_blocked) > 0)
463 return true;
464 }
465 return false;
466}
467
468static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
469{
470 if (atomic_read(v: &shost->host_blocked) > 0)
471 return true;
472 if (shost->host_self_blocked)
473 return true;
474 return false;
475}
476
477static void scsi_starved_list_run(struct Scsi_Host *shost)
478{
479 LIST_HEAD(starved_list);
480 struct scsi_device *sdev;
481 unsigned long flags;
482
483 spin_lock_irqsave(shost->host_lock, flags);
484 list_splice_init(list: &shost->starved_list, head: &starved_list);
485
486 while (!list_empty(head: &starved_list)) {
487 struct request_queue *slq;
488
489 /*
490 * As long as shost is accepting commands and we have
491 * starved queues, call blk_run_queue. scsi_request_fn
492 * drops the queue_lock and can add us back to the
493 * starved_list.
494 *
495 * host_lock protects the starved_list and starved_entry.
496 * scsi_request_fn must get the host_lock before checking
497 * or modifying starved_list or starved_entry.
498 */
499 if (scsi_host_is_busy(shost))
500 break;
501
502 sdev = list_entry(starved_list.next,
503 struct scsi_device, starved_entry);
504 list_del_init(entry: &sdev->starved_entry);
505 if (scsi_target_is_busy(starget: scsi_target(sdev))) {
506 list_move_tail(list: &sdev->starved_entry,
507 head: &shost->starved_list);
508 continue;
509 }
510
511 /*
512 * Once we drop the host lock, a racing scsi_remove_device()
513 * call may remove the sdev from the starved list and destroy
514 * it and the queue. Mitigate by taking a reference to the
515 * queue and never touching the sdev again after we drop the
516 * host lock. Note: if __scsi_remove_device() invokes
517 * blk_mq_destroy_queue() before the queue is run from this
518 * function then blk_run_queue() will return immediately since
519 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
520 */
521 slq = sdev->request_queue;
522 if (!blk_get_queue(slq))
523 continue;
524 spin_unlock_irqrestore(lock: shost->host_lock, flags);
525
526 blk_mq_run_hw_queues(q: slq, async: false);
527 blk_put_queue(slq);
528
529 spin_lock_irqsave(shost->host_lock, flags);
530 }
531 /* put any unprocessed entries back */
532 list_splice(list: &starved_list, head: &shost->starved_list);
533 spin_unlock_irqrestore(lock: shost->host_lock, flags);
534}
535
536/**
537 * scsi_run_queue - Select a proper request queue to serve next.
538 * @q: last request's queue
539 *
540 * The previous command was completely finished, start a new one if possible.
541 */
542static void scsi_run_queue(struct request_queue *q)
543{
544 struct scsi_device *sdev = q->queuedata;
545
546 if (scsi_target(sdev)->single_lun)
547 scsi_single_lun_run(current_sdev: sdev);
548 if (!list_empty(head: &sdev->host->starved_list))
549 scsi_starved_list_run(shost: sdev->host);
550
551 /* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
552 blk_mq_kick_requeue_list(q);
553}
554
555void scsi_requeue_run_queue(struct work_struct *work)
556{
557 struct scsi_device *sdev;
558 struct request_queue *q;
559
560 sdev = container_of(work, struct scsi_device, requeue_work);
561 q = sdev->request_queue;
562 scsi_run_queue(q);
563}
564
565void scsi_run_host_queues(struct Scsi_Host *shost)
566{
567 struct scsi_device *sdev;
568
569 shost_for_each_device(sdev, shost)
570 scsi_run_queue(q: sdev->request_queue);
571}
572
573static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
574{
575 if (!blk_rq_is_passthrough(rq: scsi_cmd_to_rq(scmd: cmd))) {
576 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
577
578 if (drv->uninit_command)
579 drv->uninit_command(cmd);
580 }
581}
582
583void scsi_free_sgtables(struct scsi_cmnd *cmd)
584{
585 if (cmd->sdb.table.nents)
586 sg_free_table_chained(table: &cmd->sdb.table,
587 SCSI_INLINE_SG_CNT);
588 if (scsi_prot_sg_count(cmd))
589 sg_free_table_chained(table: &cmd->prot_sdb->table,
590 SCSI_INLINE_PROT_SG_CNT);
591}
592EXPORT_SYMBOL_GPL(scsi_free_sgtables);
593
594static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
595{
596 scsi_free_sgtables(cmd);
597 scsi_uninit_cmd(cmd);
598}
599
600static void scsi_run_queue_async(struct scsi_device *sdev)
601{
602 if (scsi_host_in_recovery(shost: sdev->host))
603 return;
604
605 if (scsi_target(sdev)->single_lun ||
606 !list_empty(head: &sdev->host->starved_list)) {
607 kblockd_schedule_work(work: &sdev->requeue_work);
608 } else {
609 /*
610 * smp_mb() present in sbitmap_queue_clear() or implied in
611 * .end_io is for ordering writing .device_busy in
612 * scsi_device_unbusy() and reading sdev->restarts.
613 */
614 int old = atomic_read(v: &sdev->restarts);
615
616 /*
617 * ->restarts has to be kept as non-zero if new budget
618 * contention occurs.
619 *
620 * No need to run queue when either another re-run
621 * queue wins in updating ->restarts or a new budget
622 * contention occurs.
623 */
624 if (old && atomic_cmpxchg(v: &sdev->restarts, old, new: 0) == old)
625 blk_mq_run_hw_queues(q: sdev->request_queue, async: true);
626 }
627}
628
629/* Returns false when no more bytes to process, true if there are more */
630static bool scsi_end_request(struct request *req, blk_status_t error,
631 unsigned int bytes)
632{
633 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
634 struct scsi_device *sdev = cmd->device;
635 struct request_queue *q = sdev->request_queue;
636
637 if (blk_update_request(rq: req, error, nr_bytes: bytes))
638 return true;
639
640 if (q->limits.features & BLK_FEAT_ADD_RANDOM)
641 add_disk_randomness(disk: req->q->disk);
642
643 WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
644 !(cmd->flags & SCMD_INITIALIZED));
645 cmd->flags = 0;
646
647 /*
648 * Calling rcu_barrier() is not necessary here because the
649 * SCSI error handler guarantees that the function called by
650 * call_rcu() has been called before scsi_end_request() is
651 * called.
652 */
653 destroy_rcu_head(head: &cmd->rcu);
654
655 /*
656 * In the MQ case the command gets freed by __blk_mq_end_request,
657 * so we have to do all cleanup that depends on it earlier.
658 *
659 * We also can't kick the queues from irq context, so we
660 * will have to defer it to a workqueue.
661 */
662 scsi_mq_uninit_cmd(cmd);
663
664 /*
665 * queue is still alive, so grab the ref for preventing it
666 * from being cleaned up during running queue.
667 */
668 percpu_ref_get(ref: &q->q_usage_counter);
669
670 __blk_mq_end_request(rq: req, error);
671
672 scsi_run_queue_async(sdev);
673
674 percpu_ref_put(ref: &q->q_usage_counter);
675 return false;
676}
677
678/**
679 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
680 * @result: scsi error code
681 *
682 * Translate a SCSI result code into a blk_status_t value.
683 */
684static blk_status_t scsi_result_to_blk_status(int result)
685{
686 /*
687 * Check the scsi-ml byte first in case we converted a host or status
688 * byte.
689 */
690 switch (scsi_ml_byte(result)) {
691 case SCSIML_STAT_OK:
692 break;
693 case SCSIML_STAT_RESV_CONFLICT:
694 return BLK_STS_RESV_CONFLICT;
695 case SCSIML_STAT_NOSPC:
696 return BLK_STS_NOSPC;
697 case SCSIML_STAT_MED_ERROR:
698 return BLK_STS_MEDIUM;
699 case SCSIML_STAT_TGT_FAILURE:
700 return BLK_STS_TARGET;
701 case SCSIML_STAT_DL_TIMEOUT:
702 return BLK_STS_DURATION_LIMIT;
703 }
704
705 switch (host_byte(result)) {
706 case DID_OK:
707 if (scsi_status_is_good(status: result))
708 return BLK_STS_OK;
709 return BLK_STS_IOERR;
710 case DID_TRANSPORT_FAILFAST:
711 case DID_TRANSPORT_MARGINAL:
712 return BLK_STS_TRANSPORT;
713 default:
714 return BLK_STS_IOERR;
715 }
716}
717
718/**
719 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
720 * @rq: request to examine
721 *
722 * Description:
723 * A request could be merge of IOs which require different failure
724 * handling. This function determines the number of bytes which
725 * can be failed from the beginning of the request without
726 * crossing into area which need to be retried further.
727 *
728 * Return:
729 * The number of bytes to fail.
730 */
731static unsigned int scsi_rq_err_bytes(const struct request *rq)
732{
733 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
734 unsigned int bytes = 0;
735 struct bio *bio;
736
737 if (!(rq->rq_flags & RQF_MIXED_MERGE))
738 return blk_rq_bytes(rq);
739
740 /*
741 * Currently the only 'mixing' which can happen is between
742 * different fastfail types. We can safely fail portions
743 * which have all the failfast bits that the first one has -
744 * the ones which are at least as eager to fail as the first
745 * one.
746 */
747 for (bio = rq->bio; bio; bio = bio->bi_next) {
748 if ((bio->bi_opf & ff) != ff)
749 break;
750 bytes += bio->bi_iter.bi_size;
751 }
752
753 /* this could lead to infinite loop */
754 BUG_ON(blk_rq_bytes(rq) && !bytes);
755 return bytes;
756}
757
758static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
759{
760 struct request *req = scsi_cmd_to_rq(scmd: cmd);
761 unsigned long wait_for;
762
763 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
764 return false;
765
766 wait_for = (cmd->allowed + 1) * req->timeout;
767 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
768 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
769 wait_for/HZ);
770 return true;
771 }
772 return false;
773}
774
775/*
776 * When ALUA transition state is returned, reprep the cmd to
777 * use the ALUA handler's transition timeout. Delay the reprep
778 * 1 sec to avoid aggressive retries of the target in that
779 * state.
780 */
781#define ALUA_TRANSITION_REPREP_DELAY 1000
782
783/* Helper for scsi_io_completion() when special action required. */
784static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
785{
786 struct request *req = scsi_cmd_to_rq(scmd: cmd);
787 int level = 0;
788 enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
789 ACTION_RETRY, ACTION_DELAYED_RETRY} action;
790 struct scsi_sense_hdr sshdr;
791 bool sense_valid;
792 bool sense_current = true; /* false implies "deferred sense" */
793 blk_status_t blk_stat;
794
795 sense_valid = scsi_command_normalize_sense(cmd, sshdr: &sshdr);
796 if (sense_valid)
797 sense_current = !scsi_sense_is_deferred(sshdr: &sshdr);
798
799 blk_stat = scsi_result_to_blk_status(result);
800
801 if (host_byte(result) == DID_RESET) {
802 /* Third party bus reset or reset for error recovery
803 * reasons. Just retry the command and see what
804 * happens.
805 */
806 action = ACTION_RETRY;
807 } else if (sense_valid && sense_current) {
808 switch (sshdr.sense_key) {
809 case UNIT_ATTENTION:
810 if (cmd->device->removable) {
811 /* Detected disc change. Set a bit
812 * and quietly refuse further access.
813 */
814 cmd->device->changed = 1;
815 action = ACTION_FAIL;
816 } else {
817 /* Must have been a power glitch, or a
818 * bus reset. Could not have been a
819 * media change, so we just retry the
820 * command and see what happens.
821 */
822 action = ACTION_RETRY;
823 }
824 break;
825 case ILLEGAL_REQUEST:
826 /* If we had an ILLEGAL REQUEST returned, then
827 * we may have performed an unsupported
828 * command. The only thing this should be
829 * would be a ten byte read where only a six
830 * byte read was supported. Also, on a system
831 * where READ CAPACITY failed, we may have
832 * read past the end of the disk.
833 */
834 if ((cmd->device->use_10_for_rw &&
835 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
836 (cmd->cmnd[0] == READ_10 ||
837 cmd->cmnd[0] == WRITE_10)) {
838 /* This will issue a new 6-byte command. */
839 cmd->device->use_10_for_rw = 0;
840 action = ACTION_REPREP;
841 } else if (sshdr.asc == 0x10) /* DIX */ {
842 action = ACTION_FAIL;
843 blk_stat = BLK_STS_PROTECTION;
844 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
845 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
846 action = ACTION_FAIL;
847 blk_stat = BLK_STS_TARGET;
848 } else
849 action = ACTION_FAIL;
850 break;
851 case ABORTED_COMMAND:
852 action = ACTION_FAIL;
853 if (sshdr.asc == 0x10) /* DIF */
854 blk_stat = BLK_STS_PROTECTION;
855 break;
856 case NOT_READY:
857 /* If the device is in the process of becoming
858 * ready, or has a temporary blockage, retry.
859 */
860 if (sshdr.asc == 0x04) {
861 switch (sshdr.ascq) {
862 case 0x01: /* becoming ready */
863 case 0x04: /* format in progress */
864 case 0x05: /* rebuild in progress */
865 case 0x06: /* recalculation in progress */
866 case 0x07: /* operation in progress */
867 case 0x08: /* Long write in progress */
868 case 0x09: /* self test in progress */
869 case 0x11: /* notify (enable spinup) required */
870 case 0x14: /* space allocation in progress */
871 case 0x1a: /* start stop unit in progress */
872 case 0x1b: /* sanitize in progress */
873 case 0x1d: /* configuration in progress */
874 action = ACTION_DELAYED_RETRY;
875 break;
876 case 0x0a: /* ALUA state transition */
877 action = ACTION_DELAYED_REPREP;
878 break;
879 /*
880 * Depopulation might take many hours,
881 * thus it is not worthwhile to retry.
882 */
883 case 0x24: /* depopulation in progress */
884 case 0x25: /* depopulation restore in progress */
885 fallthrough;
886 default:
887 action = ACTION_FAIL;
888 break;
889 }
890 } else
891 action = ACTION_FAIL;
892 break;
893 case VOLUME_OVERFLOW:
894 /* See SSC3rXX or current. */
895 action = ACTION_FAIL;
896 break;
897 case DATA_PROTECT:
898 action = ACTION_FAIL;
899 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
900 (sshdr.asc == 0x55 &&
901 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
902 /* Insufficient zone resources */
903 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
904 }
905 break;
906 case COMPLETED:
907 fallthrough;
908 default:
909 action = ACTION_FAIL;
910 break;
911 }
912 } else
913 action = ACTION_FAIL;
914
915 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
916 action = ACTION_FAIL;
917
918 switch (action) {
919 case ACTION_FAIL:
920 /* Give up and fail the remainder of the request */
921 if (!(req->rq_flags & RQF_QUIET)) {
922 static DEFINE_RATELIMIT_STATE(_rs,
923 DEFAULT_RATELIMIT_INTERVAL,
924 DEFAULT_RATELIMIT_BURST);
925
926 if (unlikely(scsi_logging_level))
927 level =
928 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
929 SCSI_LOG_MLCOMPLETE_BITS);
930
931 /*
932 * if logging is enabled the failure will be printed
933 * in scsi_log_completion(), so avoid duplicate messages
934 */
935 if (!level && __ratelimit(&_rs)) {
936 scsi_print_result(cmd, NULL, FAILED);
937 if (sense_valid)
938 scsi_print_sense(cmd);
939 scsi_print_command(cmd);
940 }
941 }
942 if (!scsi_end_request(req, error: blk_stat, bytes: scsi_rq_err_bytes(rq: req)))
943 return;
944 fallthrough;
945 case ACTION_REPREP:
946 scsi_mq_requeue_cmd(cmd, msecs: 0);
947 break;
948 case ACTION_DELAYED_REPREP:
949 scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
950 break;
951 case ACTION_RETRY:
952 /* Retry the same command immediately */
953 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, unbusy: false);
954 break;
955 case ACTION_DELAYED_RETRY:
956 /* Retry the same command after a delay */
957 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, unbusy: false);
958 break;
959 }
960}
961
962/*
963 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
964 * new result that may suppress further error checking. Also modifies
965 * *blk_statp in some cases.
966 */
967static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
968 blk_status_t *blk_statp)
969{
970 bool sense_valid;
971 bool sense_current = true; /* false implies "deferred sense" */
972 struct request *req = scsi_cmd_to_rq(scmd: cmd);
973 struct scsi_sense_hdr sshdr;
974
975 sense_valid = scsi_command_normalize_sense(cmd, sshdr: &sshdr);
976 if (sense_valid)
977 sense_current = !scsi_sense_is_deferred(sshdr: &sshdr);
978
979 if (blk_rq_is_passthrough(rq: req)) {
980 if (sense_valid) {
981 /*
982 * SG_IO wants current and deferred errors
983 */
984 cmd->sense_len = min(8 + cmd->sense_buffer[7],
985 SCSI_SENSE_BUFFERSIZE);
986 }
987 if (sense_current)
988 *blk_statp = scsi_result_to_blk_status(result);
989 } else if (blk_rq_bytes(rq: req) == 0 && sense_current) {
990 /*
991 * Flush commands do not transfers any data, and thus cannot use
992 * good_bytes != blk_rq_bytes(req) as the signal for an error.
993 * This sets *blk_statp explicitly for the problem case.
994 */
995 *blk_statp = scsi_result_to_blk_status(result);
996 }
997 /*
998 * Recovered errors need reporting, but they're always treated as
999 * success, so fiddle the result code here. For passthrough requests
1000 * we already took a copy of the original into sreq->result which
1001 * is what gets returned to the user
1002 */
1003 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
1004 bool do_print = true;
1005 /*
1006 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
1007 * skip print since caller wants ATA registers. Only occurs
1008 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
1009 */
1010 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1011 do_print = false;
1012 else if (req->rq_flags & RQF_QUIET)
1013 do_print = false;
1014 if (do_print)
1015 scsi_print_sense(cmd);
1016 result = 0;
1017 /* for passthrough, *blk_statp may be set */
1018 *blk_statp = BLK_STS_OK;
1019 }
1020 /*
1021 * Another corner case: the SCSI status byte is non-zero but 'good'.
1022 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1023 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1024 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1025 * intermediate statuses (both obsolete in SAM-4) as good.
1026 */
1027 if ((result & 0xff) && scsi_status_is_good(status: result)) {
1028 result = 0;
1029 *blk_statp = BLK_STS_OK;
1030 }
1031 return result;
1032}
1033
1034/**
1035 * scsi_io_completion - Completion processing for SCSI commands.
1036 * @cmd: command that is finished.
1037 * @good_bytes: number of processed bytes.
1038 *
1039 * We will finish off the specified number of sectors. If we are done, the
1040 * command block will be released and the queue function will be goosed. If we
1041 * are not done then we have to figure out what to do next:
1042 *
1043 * a) We can call scsi_mq_requeue_cmd(). The request will be
1044 * unprepared and put back on the queue. Then a new command will
1045 * be created for it. This should be used if we made forward
1046 * progress, or if we want to switch from READ(10) to READ(6) for
1047 * example.
1048 *
1049 * b) We can call scsi_io_completion_action(). The request will be
1050 * put back on the queue and retried using the same command as
1051 * before, possibly after a delay.
1052 *
1053 * c) We can call scsi_end_request() with blk_stat other than
1054 * BLK_STS_OK, to fail the remainder of the request.
1055 */
1056void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1057{
1058 int result = cmd->result;
1059 struct request *req = scsi_cmd_to_rq(scmd: cmd);
1060 blk_status_t blk_stat = BLK_STS_OK;
1061
1062 if (unlikely(result)) /* a nz result may or may not be an error */
1063 result = scsi_io_completion_nz_result(cmd, result, blk_statp: &blk_stat);
1064
1065 /*
1066 * Next deal with any sectors which we were able to correctly
1067 * handle.
1068 */
1069 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1070 "%u sectors total, %d bytes done.\n",
1071 blk_rq_sectors(req), good_bytes));
1072
1073 /*
1074 * Failed, zero length commands always need to drop down
1075 * to retry code. Fast path should return in this block.
1076 */
1077 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1078 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1079 return; /* no bytes remaining */
1080 }
1081
1082 /* Kill remainder if no retries. */
1083 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1084 if (scsi_end_request(req, error: blk_stat, bytes: blk_rq_bytes(rq: req)))
1085 WARN_ONCE(true,
1086 "Bytes remaining after failed, no-retry command");
1087 return;
1088 }
1089
1090 /*
1091 * If there had been no error, but we have leftover bytes in the
1092 * request just queue the command up again.
1093 */
1094 if (likely(result == 0))
1095 scsi_mq_requeue_cmd(cmd, msecs: 0);
1096 else
1097 scsi_io_completion_action(cmd, result);
1098}
1099
1100static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1101 struct request *rq)
1102{
1103 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1104 !op_is_write(op: req_op(req: rq)) &&
1105 sdev->host->hostt->dma_need_drain(rq);
1106}
1107
1108/**
1109 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1110 * @cmd: SCSI command data structure to initialize.
1111 *
1112 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1113 * for @cmd.
1114 *
1115 * Returns:
1116 * * BLK_STS_OK - on success
1117 * * BLK_STS_RESOURCE - if the failure is retryable
1118 * * BLK_STS_IOERR - if the failure is fatal
1119 */
1120blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1121{
1122 struct scsi_device *sdev = cmd->device;
1123 struct request *rq = scsi_cmd_to_rq(scmd: cmd);
1124 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1125 struct scatterlist *last_sg = NULL;
1126 blk_status_t ret;
1127 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1128 int count;
1129
1130 if (WARN_ON_ONCE(!nr_segs))
1131 return BLK_STS_IOERR;
1132
1133 /*
1134 * Make sure there is space for the drain. The driver must adjust
1135 * max_hw_segments to be prepared for this.
1136 */
1137 if (need_drain)
1138 nr_segs++;
1139
1140 /*
1141 * If sg table allocation fails, requeue request later.
1142 */
1143 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1144 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1145 return BLK_STS_RESOURCE;
1146
1147 /*
1148 * Next, walk the list, and fill in the addresses and sizes of
1149 * each segment.
1150 */
1151 count = __blk_rq_map_sg(rq, sglist: cmd->sdb.table.sgl, last_sg: &last_sg);
1152
1153 if (blk_rq_bytes(rq) & rq->q->limits.dma_pad_mask) {
1154 unsigned int pad_len =
1155 (rq->q->limits.dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1156
1157 last_sg->length += pad_len;
1158 cmd->extra_len += pad_len;
1159 }
1160
1161 if (need_drain) {
1162 sg_unmark_end(sg: last_sg);
1163 last_sg = sg_next(sg: last_sg);
1164 sg_set_buf(sg: last_sg, buf: sdev->dma_drain_buf, buflen: sdev->dma_drain_len);
1165 sg_mark_end(sg: last_sg);
1166
1167 cmd->extra_len += sdev->dma_drain_len;
1168 count++;
1169 }
1170
1171 BUG_ON(count > cmd->sdb.table.nents);
1172 cmd->sdb.table.nents = count;
1173 cmd->sdb.length = blk_rq_payload_bytes(rq);
1174
1175 if (blk_integrity_rq(rq)) {
1176 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1177
1178 if (WARN_ON_ONCE(!prot_sdb)) {
1179 /*
1180 * This can happen if someone (e.g. multipath)
1181 * queues a command to a device on an adapter
1182 * that does not support DIX.
1183 */
1184 ret = BLK_STS_IOERR;
1185 goto out_free_sgtables;
1186 }
1187
1188 if (sg_alloc_table_chained(table: &prot_sdb->table,
1189 nents: rq->nr_integrity_segments,
1190 first_chunk: prot_sdb->table.sgl,
1191 SCSI_INLINE_PROT_SG_CNT)) {
1192 ret = BLK_STS_RESOURCE;
1193 goto out_free_sgtables;
1194 }
1195
1196 count = blk_rq_map_integrity_sg(q: rq, s: prot_sdb->table.sgl);
1197 cmd->prot_sdb = prot_sdb;
1198 cmd->prot_sdb->table.nents = count;
1199 }
1200
1201 return BLK_STS_OK;
1202out_free_sgtables:
1203 scsi_free_sgtables(cmd);
1204 return ret;
1205}
1206EXPORT_SYMBOL(scsi_alloc_sgtables);
1207
1208/**
1209 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1210 * @rq: Request associated with the SCSI command to be initialized.
1211 *
1212 * This function initializes the members of struct scsi_cmnd that must be
1213 * initialized before request processing starts and that won't be
1214 * reinitialized if a SCSI command is requeued.
1215 */
1216static void scsi_initialize_rq(struct request *rq)
1217{
1218 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1219
1220 memset(s: cmd->cmnd, c: 0, n: sizeof(cmd->cmnd));
1221 cmd->cmd_len = MAX_COMMAND_SIZE;
1222 cmd->sense_len = 0;
1223 init_rcu_head(head: &cmd->rcu);
1224 cmd->jiffies_at_alloc = jiffies;
1225 cmd->retries = 0;
1226}
1227
1228/**
1229 * scsi_alloc_request - allocate a block request and partially
1230 * initialize its &scsi_cmnd
1231 * @q: the device's request queue
1232 * @opf: the request operation code
1233 * @flags: block layer allocation flags
1234 *
1235 * Return: &struct request pointer on success or %NULL on failure
1236 */
1237struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1238 blk_mq_req_flags_t flags)
1239{
1240 struct request *rq;
1241
1242 rq = blk_mq_alloc_request(q, opf, flags);
1243 if (!IS_ERR(ptr: rq))
1244 scsi_initialize_rq(rq);
1245 return rq;
1246}
1247EXPORT_SYMBOL_GPL(scsi_alloc_request);
1248
1249/*
1250 * Only called when the request isn't completed by SCSI, and not freed by
1251 * SCSI
1252 */
1253static void scsi_cleanup_rq(struct request *rq)
1254{
1255 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1256
1257 cmd->flags = 0;
1258
1259 if (rq->rq_flags & RQF_DONTPREP) {
1260 scsi_mq_uninit_cmd(cmd);
1261 rq->rq_flags &= ~RQF_DONTPREP;
1262 }
1263}
1264
1265/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1266void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1267{
1268 struct request *rq = scsi_cmd_to_rq(scmd: cmd);
1269
1270 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1271 cmd->flags |= SCMD_INITIALIZED;
1272 scsi_initialize_rq(rq);
1273 }
1274
1275 cmd->device = dev;
1276 INIT_LIST_HEAD(list: &cmd->eh_entry);
1277 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1278}
1279
1280static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1281 struct request *req)
1282{
1283 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
1284
1285 /*
1286 * Passthrough requests may transfer data, in which case they must
1287 * a bio attached to them. Or they might contain a SCSI command
1288 * that does not transfer data, in which case they may optionally
1289 * submit a request without an attached bio.
1290 */
1291 if (req->bio) {
1292 blk_status_t ret = scsi_alloc_sgtables(cmd);
1293 if (unlikely(ret != BLK_STS_OK))
1294 return ret;
1295 } else {
1296 BUG_ON(blk_rq_bytes(req));
1297
1298 memset(s: &cmd->sdb, c: 0, n: sizeof(cmd->sdb));
1299 }
1300
1301 cmd->transfersize = blk_rq_bytes(rq: req);
1302 return BLK_STS_OK;
1303}
1304
1305static blk_status_t
1306scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1307{
1308 switch (sdev->sdev_state) {
1309 case SDEV_CREATED:
1310 return BLK_STS_OK;
1311 case SDEV_OFFLINE:
1312 case SDEV_TRANSPORT_OFFLINE:
1313 /*
1314 * If the device is offline we refuse to process any
1315 * commands. The device must be brought online
1316 * before trying any recovery commands.
1317 */
1318 if (!sdev->offline_already) {
1319 sdev->offline_already = true;
1320 sdev_printk(KERN_ERR, sdev,
1321 "rejecting I/O to offline device\n");
1322 }
1323 return BLK_STS_IOERR;
1324 case SDEV_DEL:
1325 /*
1326 * If the device is fully deleted, we refuse to
1327 * process any commands as well.
1328 */
1329 sdev_printk(KERN_ERR, sdev,
1330 "rejecting I/O to dead device\n");
1331 return BLK_STS_IOERR;
1332 case SDEV_BLOCK:
1333 case SDEV_CREATED_BLOCK:
1334 return BLK_STS_RESOURCE;
1335 case SDEV_QUIESCE:
1336 /*
1337 * If the device is blocked we only accept power management
1338 * commands.
1339 */
1340 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1341 return BLK_STS_RESOURCE;
1342 return BLK_STS_OK;
1343 default:
1344 /*
1345 * For any other not fully online state we only allow
1346 * power management commands.
1347 */
1348 if (req && !(req->rq_flags & RQF_PM))
1349 return BLK_STS_OFFLINE;
1350 return BLK_STS_OK;
1351 }
1352}
1353
1354/*
1355 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1356 * and return the token else return -1.
1357 */
1358static inline int scsi_dev_queue_ready(struct request_queue *q,
1359 struct scsi_device *sdev)
1360{
1361 int token;
1362
1363 token = sbitmap_get(sb: &sdev->budget_map);
1364 if (token < 0)
1365 return -1;
1366
1367 if (!atomic_read(v: &sdev->device_blocked))
1368 return token;
1369
1370 /*
1371 * Only unblock if no other commands are pending and
1372 * if device_blocked has decreased to zero
1373 */
1374 if (scsi_device_busy(sdev) > 1 ||
1375 atomic_dec_return(v: &sdev->device_blocked) > 0) {
1376 sbitmap_put(sb: &sdev->budget_map, bitnr: token);
1377 return -1;
1378 }
1379
1380 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1381 "unblocking device at zero depth\n"));
1382
1383 return token;
1384}
1385
1386/*
1387 * scsi_target_queue_ready: checks if there we can send commands to target
1388 * @sdev: scsi device on starget to check.
1389 */
1390static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1391 struct scsi_device *sdev)
1392{
1393 struct scsi_target *starget = scsi_target(sdev);
1394 unsigned int busy;
1395
1396 if (starget->single_lun) {
1397 spin_lock_irq(lock: shost->host_lock);
1398 if (starget->starget_sdev_user &&
1399 starget->starget_sdev_user != sdev) {
1400 spin_unlock_irq(lock: shost->host_lock);
1401 return 0;
1402 }
1403 starget->starget_sdev_user = sdev;
1404 spin_unlock_irq(lock: shost->host_lock);
1405 }
1406
1407 if (starget->can_queue <= 0)
1408 return 1;
1409
1410 busy = atomic_inc_return(v: &starget->target_busy) - 1;
1411 if (atomic_read(v: &starget->target_blocked) > 0) {
1412 if (busy)
1413 goto starved;
1414
1415 /*
1416 * unblock after target_blocked iterates to zero
1417 */
1418 if (atomic_dec_return(v: &starget->target_blocked) > 0)
1419 goto out_dec;
1420
1421 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1422 "unblocking target at zero depth\n"));
1423 }
1424
1425 if (busy >= starget->can_queue)
1426 goto starved;
1427
1428 return 1;
1429
1430starved:
1431 spin_lock_irq(lock: shost->host_lock);
1432 list_move_tail(list: &sdev->starved_entry, head: &shost->starved_list);
1433 spin_unlock_irq(lock: shost->host_lock);
1434out_dec:
1435 if (starget->can_queue > 0)
1436 atomic_dec(v: &starget->target_busy);
1437 return 0;
1438}
1439
1440/*
1441 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1442 * return 0. We must end up running the queue again whenever 0 is
1443 * returned, else IO can hang.
1444 */
1445static inline int scsi_host_queue_ready(struct request_queue *q,
1446 struct Scsi_Host *shost,
1447 struct scsi_device *sdev,
1448 struct scsi_cmnd *cmd)
1449{
1450 if (atomic_read(v: &shost->host_blocked) > 0) {
1451 if (scsi_host_busy(shost) > 0)
1452 goto starved;
1453
1454 /*
1455 * unblock after host_blocked iterates to zero
1456 */
1457 if (atomic_dec_return(v: &shost->host_blocked) > 0)
1458 goto out_dec;
1459
1460 SCSI_LOG_MLQUEUE(3,
1461 shost_printk(KERN_INFO, shost,
1462 "unblocking host at zero depth\n"));
1463 }
1464
1465 if (shost->host_self_blocked)
1466 goto starved;
1467
1468 /* We're OK to process the command, so we can't be starved */
1469 if (!list_empty(head: &sdev->starved_entry)) {
1470 spin_lock_irq(lock: shost->host_lock);
1471 if (!list_empty(head: &sdev->starved_entry))
1472 list_del_init(entry: &sdev->starved_entry);
1473 spin_unlock_irq(lock: shost->host_lock);
1474 }
1475
1476 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1477
1478 return 1;
1479
1480starved:
1481 spin_lock_irq(lock: shost->host_lock);
1482 if (list_empty(head: &sdev->starved_entry))
1483 list_add_tail(new: &sdev->starved_entry, head: &shost->starved_list);
1484 spin_unlock_irq(lock: shost->host_lock);
1485out_dec:
1486 scsi_dec_host_busy(shost, cmd);
1487 return 0;
1488}
1489
1490/*
1491 * Busy state exporting function for request stacking drivers.
1492 *
1493 * For efficiency, no lock is taken to check the busy state of
1494 * shost/starget/sdev, since the returned value is not guaranteed and
1495 * may be changed after request stacking drivers call the function,
1496 * regardless of taking lock or not.
1497 *
1498 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1499 * needs to return 'not busy'. Otherwise, request stacking drivers
1500 * may hold requests forever.
1501 */
1502static bool scsi_mq_lld_busy(struct request_queue *q)
1503{
1504 struct scsi_device *sdev = q->queuedata;
1505 struct Scsi_Host *shost;
1506
1507 if (blk_queue_dying(q))
1508 return false;
1509
1510 shost = sdev->host;
1511
1512 /*
1513 * Ignore host/starget busy state.
1514 * Since block layer does not have a concept of fairness across
1515 * multiple queues, congestion of host/starget needs to be handled
1516 * in SCSI layer.
1517 */
1518 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1519 return true;
1520
1521 return false;
1522}
1523
1524/*
1525 * Block layer request completion callback. May be called from interrupt
1526 * context.
1527 */
1528static void scsi_complete(struct request *rq)
1529{
1530 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1531 enum scsi_disposition disposition;
1532
1533 INIT_LIST_HEAD(list: &cmd->eh_entry);
1534
1535 atomic_inc(v: &cmd->device->iodone_cnt);
1536 if (cmd->result)
1537 atomic_inc(v: &cmd->device->ioerr_cnt);
1538
1539 disposition = scsi_decide_disposition(cmd);
1540 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1541 disposition = SUCCESS;
1542
1543 scsi_log_completion(cmd, disposition);
1544
1545 switch (disposition) {
1546 case SUCCESS:
1547 scsi_finish_command(cmd);
1548 break;
1549 case NEEDS_RETRY:
1550 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1551 break;
1552 case ADD_TO_MLQUEUE:
1553 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1554 break;
1555 default:
1556 scsi_eh_scmd_add(cmd);
1557 break;
1558 }
1559}
1560
1561/**
1562 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1563 * @cmd: command block we are dispatching.
1564 *
1565 * Return: nonzero return request was rejected and device's queue needs to be
1566 * plugged.
1567 */
1568static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1569{
1570 struct Scsi_Host *host = cmd->device->host;
1571 int rtn = 0;
1572
1573 atomic_inc(v: &cmd->device->iorequest_cnt);
1574
1575 /* check if the device is still usable */
1576 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1577 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1578 * returns an immediate error upwards, and signals
1579 * that the device is no longer present */
1580 cmd->result = DID_NO_CONNECT << 16;
1581 goto done;
1582 }
1583
1584 /* Check to see if the scsi lld made this device blocked. */
1585 if (unlikely(scsi_device_blocked(cmd->device))) {
1586 /*
1587 * in blocked state, the command is just put back on
1588 * the device queue. The suspend state has already
1589 * blocked the queue so future requests should not
1590 * occur until the device transitions out of the
1591 * suspend state.
1592 */
1593 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1594 "queuecommand : device blocked\n"));
1595 atomic_dec(v: &cmd->device->iorequest_cnt);
1596 return SCSI_MLQUEUE_DEVICE_BUSY;
1597 }
1598
1599 /* Store the LUN value in cmnd, if needed. */
1600 if (cmd->device->lun_in_cdb)
1601 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1602 (cmd->device->lun << 5 & 0xe0);
1603
1604 scsi_log_send(cmd);
1605
1606 /*
1607 * Before we queue this command, check if the command
1608 * length exceeds what the host adapter can handle.
1609 */
1610 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1611 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1612 "queuecommand : command too long. "
1613 "cdb_size=%d host->max_cmd_len=%d\n",
1614 cmd->cmd_len, cmd->device->host->max_cmd_len));
1615 cmd->result = (DID_ABORT << 16);
1616 goto done;
1617 }
1618
1619 if (unlikely(host->shost_state == SHOST_DEL)) {
1620 cmd->result = (DID_NO_CONNECT << 16);
1621 goto done;
1622
1623 }
1624
1625 trace_scsi_dispatch_cmd_start(cmd);
1626 rtn = host->hostt->queuecommand(host, cmd);
1627 if (rtn) {
1628 atomic_dec(v: &cmd->device->iorequest_cnt);
1629 trace_scsi_dispatch_cmd_error(cmd, rtn);
1630 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1631 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1632 rtn = SCSI_MLQUEUE_HOST_BUSY;
1633
1634 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1635 "queuecommand : request rejected\n"));
1636 }
1637
1638 return rtn;
1639 done:
1640 scsi_done(cmd);
1641 return 0;
1642}
1643
1644/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1645static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1646{
1647 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1648 sizeof(struct scatterlist);
1649}
1650
1651static blk_status_t scsi_prepare_cmd(struct request *req)
1652{
1653 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
1654 struct scsi_device *sdev = req->q->queuedata;
1655 struct Scsi_Host *shost = sdev->host;
1656 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1657 struct scatterlist *sg;
1658
1659 scsi_init_command(dev: sdev, cmd);
1660
1661 cmd->eh_eflags = 0;
1662 cmd->prot_type = 0;
1663 cmd->prot_flags = 0;
1664 cmd->submitter = 0;
1665 memset(s: &cmd->sdb, c: 0, n: sizeof(cmd->sdb));
1666 cmd->underflow = 0;
1667 cmd->transfersize = 0;
1668 cmd->host_scribble = NULL;
1669 cmd->result = 0;
1670 cmd->extra_len = 0;
1671 cmd->state = 0;
1672 if (in_flight)
1673 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1674
1675 cmd->prot_op = SCSI_PROT_NORMAL;
1676 if (blk_rq_bytes(rq: req))
1677 cmd->sc_data_direction = rq_dma_dir(req);
1678 else
1679 cmd->sc_data_direction = DMA_NONE;
1680
1681 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1682 cmd->sdb.table.sgl = sg;
1683
1684 if (scsi_host_get_prot(shost)) {
1685 memset(s: cmd->prot_sdb, c: 0, n: sizeof(struct scsi_data_buffer));
1686
1687 cmd->prot_sdb->table.sgl =
1688 (struct scatterlist *)(cmd->prot_sdb + 1);
1689 }
1690
1691 /*
1692 * Special handling for passthrough commands, which don't go to the ULP
1693 * at all:
1694 */
1695 if (blk_rq_is_passthrough(rq: req))
1696 return scsi_setup_scsi_cmnd(sdev, req);
1697
1698 if (sdev->handler && sdev->handler->prep_fn) {
1699 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1700
1701 if (ret != BLK_STS_OK)
1702 return ret;
1703 }
1704
1705 /* Usually overridden by the ULP */
1706 cmd->allowed = 0;
1707 memset(s: cmd->cmnd, c: 0, n: sizeof(cmd->cmnd));
1708 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1709}
1710
1711static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1712{
1713 struct request *req = scsi_cmd_to_rq(scmd: cmd);
1714
1715 switch (cmd->submitter) {
1716 case SUBMITTED_BY_BLOCK_LAYER:
1717 break;
1718 case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1719 return scsi_eh_done(scmd: cmd);
1720 case SUBMITTED_BY_SCSI_RESET_IOCTL:
1721 return;
1722 }
1723
1724 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1725 return;
1726 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1727 return;
1728 trace_scsi_dispatch_cmd_done(cmd);
1729
1730 if (complete_directly)
1731 blk_mq_complete_request_direct(rq: req, complete: scsi_complete);
1732 else
1733 blk_mq_complete_request(rq: req);
1734}
1735
1736void scsi_done(struct scsi_cmnd *cmd)
1737{
1738 scsi_done_internal(cmd, complete_directly: false);
1739}
1740EXPORT_SYMBOL(scsi_done);
1741
1742void scsi_done_direct(struct scsi_cmnd *cmd)
1743{
1744 scsi_done_internal(cmd, complete_directly: true);
1745}
1746EXPORT_SYMBOL(scsi_done_direct);
1747
1748static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1749{
1750 struct scsi_device *sdev = q->queuedata;
1751
1752 sbitmap_put(sb: &sdev->budget_map, bitnr: budget_token);
1753}
1754
1755/*
1756 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1757 * not change behaviour from the previous unplug mechanism, experimentation
1758 * may prove this needs changing.
1759 */
1760#define SCSI_QUEUE_DELAY 3
1761
1762static int scsi_mq_get_budget(struct request_queue *q)
1763{
1764 struct scsi_device *sdev = q->queuedata;
1765 int token = scsi_dev_queue_ready(q, sdev);
1766
1767 if (token >= 0)
1768 return token;
1769
1770 atomic_inc(v: &sdev->restarts);
1771
1772 /*
1773 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1774 * .restarts must be incremented before .device_busy is read because the
1775 * code in scsi_run_queue_async() depends on the order of these operations.
1776 */
1777 smp_mb__after_atomic();
1778
1779 /*
1780 * If all in-flight requests originated from this LUN are completed
1781 * before reading .device_busy, sdev->device_busy will be observed as
1782 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1783 * soon. Otherwise, completion of one of these requests will observe
1784 * the .restarts flag, and the request queue will be run for handling
1785 * this request, see scsi_end_request().
1786 */
1787 if (unlikely(scsi_device_busy(sdev) == 0 &&
1788 !scsi_device_blocked(sdev)))
1789 blk_mq_delay_run_hw_queues(q: sdev->request_queue, SCSI_QUEUE_DELAY);
1790 return -1;
1791}
1792
1793static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1794{
1795 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
1796
1797 cmd->budget_token = token;
1798}
1799
1800static int scsi_mq_get_rq_budget_token(struct request *req)
1801{
1802 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
1803
1804 return cmd->budget_token;
1805}
1806
1807static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1808 const struct blk_mq_queue_data *bd)
1809{
1810 struct request *req = bd->rq;
1811 struct request_queue *q = req->q;
1812 struct scsi_device *sdev = q->queuedata;
1813 struct Scsi_Host *shost = sdev->host;
1814 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq: req);
1815 blk_status_t ret;
1816 int reason;
1817
1818 WARN_ON_ONCE(cmd->budget_token < 0);
1819
1820 /*
1821 * If the device is not in running state we will reject some or all
1822 * commands.
1823 */
1824 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1825 ret = scsi_device_state_check(sdev, req);
1826 if (ret != BLK_STS_OK)
1827 goto out_put_budget;
1828 }
1829
1830 ret = BLK_STS_RESOURCE;
1831 if (!scsi_target_queue_ready(shost, sdev))
1832 goto out_put_budget;
1833 if (unlikely(scsi_host_in_recovery(shost))) {
1834 if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1835 ret = BLK_STS_OFFLINE;
1836 goto out_dec_target_busy;
1837 }
1838 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1839 goto out_dec_target_busy;
1840
1841 /*
1842 * Only clear the driver-private command data if the LLD does not supply
1843 * a function to initialize that data.
1844 */
1845 if (shost->hostt->cmd_size && !shost->hostt->init_cmd_priv)
1846 memset(s: scsi_cmd_priv(cmd), c: 0, n: shost->hostt->cmd_size);
1847
1848 if (!(req->rq_flags & RQF_DONTPREP)) {
1849 ret = scsi_prepare_cmd(req);
1850 if (ret != BLK_STS_OK)
1851 goto out_dec_host_busy;
1852 req->rq_flags |= RQF_DONTPREP;
1853 } else {
1854 clear_bit(SCMD_STATE_COMPLETE, addr: &cmd->state);
1855 }
1856
1857 cmd->flags &= SCMD_PRESERVED_FLAGS;
1858 if (sdev->simple_tags)
1859 cmd->flags |= SCMD_TAGGED;
1860 if (bd->last)
1861 cmd->flags |= SCMD_LAST;
1862
1863 scsi_set_resid(cmd, resid: 0);
1864 memset(s: cmd->sense_buffer, c: 0, SCSI_SENSE_BUFFERSIZE);
1865 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1866
1867 blk_mq_start_request(rq: req);
1868 reason = scsi_dispatch_cmd(cmd);
1869 if (reason) {
1870 scsi_set_blocked(cmd, reason);
1871 ret = BLK_STS_RESOURCE;
1872 goto out_dec_host_busy;
1873 }
1874
1875 return BLK_STS_OK;
1876
1877out_dec_host_busy:
1878 scsi_dec_host_busy(shost, cmd);
1879out_dec_target_busy:
1880 if (scsi_target(sdev)->can_queue > 0)
1881 atomic_dec(v: &scsi_target(sdev)->target_busy);
1882out_put_budget:
1883 scsi_mq_put_budget(q, budget_token: cmd->budget_token);
1884 cmd->budget_token = -1;
1885 switch (ret) {
1886 case BLK_STS_OK:
1887 break;
1888 case BLK_STS_RESOURCE:
1889 if (scsi_device_blocked(sdev))
1890 ret = BLK_STS_DEV_RESOURCE;
1891 break;
1892 case BLK_STS_AGAIN:
1893 cmd->result = DID_BUS_BUSY << 16;
1894 if (req->rq_flags & RQF_DONTPREP)
1895 scsi_mq_uninit_cmd(cmd);
1896 break;
1897 default:
1898 if (unlikely(!scsi_device_online(sdev)))
1899 cmd->result = DID_NO_CONNECT << 16;
1900 else
1901 cmd->result = DID_ERROR << 16;
1902 /*
1903 * Make sure to release all allocated resources when
1904 * we hit an error, as we will never see this command
1905 * again.
1906 */
1907 if (req->rq_flags & RQF_DONTPREP)
1908 scsi_mq_uninit_cmd(cmd);
1909 scsi_run_queue_async(sdev);
1910 break;
1911 }
1912 return ret;
1913}
1914
1915static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1916 unsigned int hctx_idx, unsigned int numa_node)
1917{
1918 struct Scsi_Host *shost = set->driver_data;
1919 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1920 struct scatterlist *sg;
1921 int ret = 0;
1922
1923 cmd->sense_buffer =
1924 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1925 if (!cmd->sense_buffer)
1926 return -ENOMEM;
1927
1928 if (scsi_host_get_prot(shost)) {
1929 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1930 shost->hostt->cmd_size;
1931 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1932 }
1933
1934 if (shost->hostt->init_cmd_priv) {
1935 ret = shost->hostt->init_cmd_priv(shost, cmd);
1936 if (ret < 0)
1937 kmem_cache_free(s: scsi_sense_cache, objp: cmd->sense_buffer);
1938 }
1939
1940 return ret;
1941}
1942
1943static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1944 unsigned int hctx_idx)
1945{
1946 struct Scsi_Host *shost = set->driver_data;
1947 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1948
1949 if (shost->hostt->exit_cmd_priv)
1950 shost->hostt->exit_cmd_priv(shost, cmd);
1951 kmem_cache_free(s: scsi_sense_cache, objp: cmd->sense_buffer);
1952}
1953
1954
1955static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1956{
1957 struct Scsi_Host *shost = hctx->driver_data;
1958
1959 if (shost->hostt->mq_poll)
1960 return shost->hostt->mq_poll(shost, hctx->queue_num);
1961
1962 return 0;
1963}
1964
1965static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1966 unsigned int hctx_idx)
1967{
1968 struct Scsi_Host *shost = data;
1969
1970 hctx->driver_data = shost;
1971 return 0;
1972}
1973
1974static void scsi_map_queues(struct blk_mq_tag_set *set)
1975{
1976 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1977
1978 if (shost->hostt->map_queues)
1979 return shost->hostt->map_queues(shost);
1980 blk_mq_map_queues(qmap: &set->map[HCTX_TYPE_DEFAULT]);
1981}
1982
1983void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
1984{
1985 struct device *dev = shost->dma_dev;
1986
1987 memset(s: lim, c: 0, n: sizeof(*lim));
1988 lim->max_segments =
1989 min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);
1990
1991 if (scsi_host_prot_dma(shost)) {
1992 shost->sg_prot_tablesize =
1993 min_not_zero(shost->sg_prot_tablesize,
1994 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1995 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1996 lim->max_integrity_segments = shost->sg_prot_tablesize;
1997 }
1998
1999 lim->max_hw_sectors = shost->max_sectors;
2000 lim->seg_boundary_mask = shost->dma_boundary;
2001 lim->max_segment_size = shost->max_segment_size;
2002 lim->virt_boundary_mask = shost->virt_boundary_mask;
2003 lim->dma_alignment = max_t(unsigned int,
2004 shost->dma_alignment, dma_get_cache_alignment() - 1);
2005
2006 /*
2007 * Propagate the DMA formation properties to the dma-mapping layer as
2008 * a courtesy service to the LLDDs. This needs to check that the buses
2009 * actually support the DMA API first, though.
2010 */
2011 if (dev->dma_parms) {
2012 dma_set_seg_boundary(dev, mask: shost->dma_boundary);
2013 dma_set_max_seg_size(dev, size: shost->max_segment_size);
2014 }
2015}
2016EXPORT_SYMBOL_GPL(scsi_init_limits);
2017
2018static const struct blk_mq_ops scsi_mq_ops_no_commit = {
2019 .get_budget = scsi_mq_get_budget,
2020 .put_budget = scsi_mq_put_budget,
2021 .queue_rq = scsi_queue_rq,
2022 .complete = scsi_complete,
2023 .timeout = scsi_timeout,
2024#ifdef CONFIG_BLK_DEBUG_FS
2025 .show_rq = scsi_show_rq,
2026#endif
2027 .init_request = scsi_mq_init_request,
2028 .exit_request = scsi_mq_exit_request,
2029 .cleanup_rq = scsi_cleanup_rq,
2030 .busy = scsi_mq_lld_busy,
2031 .map_queues = scsi_map_queues,
2032 .init_hctx = scsi_init_hctx,
2033 .poll = scsi_mq_poll,
2034 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2035 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2036};
2037
2038
2039static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2040{
2041 struct Scsi_Host *shost = hctx->driver_data;
2042
2043 shost->hostt->commit_rqs(shost, hctx->queue_num);
2044}
2045
2046static const struct blk_mq_ops scsi_mq_ops = {
2047 .get_budget = scsi_mq_get_budget,
2048 .put_budget = scsi_mq_put_budget,
2049 .queue_rq = scsi_queue_rq,
2050 .commit_rqs = scsi_commit_rqs,
2051 .complete = scsi_complete,
2052 .timeout = scsi_timeout,
2053#ifdef CONFIG_BLK_DEBUG_FS
2054 .show_rq = scsi_show_rq,
2055#endif
2056 .init_request = scsi_mq_init_request,
2057 .exit_request = scsi_mq_exit_request,
2058 .cleanup_rq = scsi_cleanup_rq,
2059 .busy = scsi_mq_lld_busy,
2060 .map_queues = scsi_map_queues,
2061 .init_hctx = scsi_init_hctx,
2062 .poll = scsi_mq_poll,
2063 .set_rq_budget_token = scsi_mq_set_rq_budget_token,
2064 .get_rq_budget_token = scsi_mq_get_rq_budget_token,
2065};
2066
2067int scsi_mq_setup_tags(struct Scsi_Host *shost)
2068{
2069 unsigned int cmd_size, sgl_size;
2070 struct blk_mq_tag_set *tag_set = &shost->tag_set;
2071
2072 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2073 scsi_mq_inline_sgl_size(shost));
2074 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2075 if (scsi_host_get_prot(shost))
2076 cmd_size += sizeof(struct scsi_data_buffer) +
2077 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2078
2079 memset(s: tag_set, c: 0, n: sizeof(*tag_set));
2080 if (shost->hostt->commit_rqs)
2081 tag_set->ops = &scsi_mq_ops;
2082 else
2083 tag_set->ops = &scsi_mq_ops_no_commit;
2084 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2085 tag_set->nr_maps = shost->nr_maps ? : 1;
2086 tag_set->queue_depth = shost->can_queue;
2087 tag_set->cmd_size = cmd_size;
2088 tag_set->numa_node = dev_to_node(dev: shost->dma_dev);
2089 if (shost->hostt->tag_alloc_policy_rr)
2090 tag_set->flags |= BLK_MQ_F_TAG_RR;
2091 if (shost->queuecommand_may_block)
2092 tag_set->flags |= BLK_MQ_F_BLOCKING;
2093 tag_set->driver_data = shost;
2094 if (shost->host_tagset)
2095 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2096
2097 return blk_mq_alloc_tag_set(set: tag_set);
2098}
2099
2100void scsi_mq_free_tags(struct kref *kref)
2101{
2102 struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2103 tagset_refcnt);
2104
2105 blk_mq_free_tag_set(set: &shost->tag_set);
2106 complete(&shost->tagset_freed);
2107}
2108
2109/**
2110 * scsi_device_from_queue - return sdev associated with a request_queue
2111 * @q: The request queue to return the sdev from
2112 *
2113 * Return the sdev associated with a request queue or NULL if the
2114 * request_queue does not reference a SCSI device.
2115 */
2116struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2117{
2118 struct scsi_device *sdev = NULL;
2119
2120 if (q->mq_ops == &scsi_mq_ops_no_commit ||
2121 q->mq_ops == &scsi_mq_ops)
2122 sdev = q->queuedata;
2123 if (!sdev || !get_device(dev: &sdev->sdev_gendev))
2124 sdev = NULL;
2125
2126 return sdev;
2127}
2128/*
2129 * pktcdvd should have been integrated into the SCSI layers, but for historical
2130 * reasons like the old IDE driver it isn't. This export allows it to safely
2131 * probe if a given device is a SCSI one and only attach to that.
2132 */
2133#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2134EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2135#endif
2136
2137/**
2138 * scsi_block_requests - Utility function used by low-level drivers to prevent
2139 * further commands from being queued to the device.
2140 * @shost: host in question
2141 *
2142 * There is no timer nor any other means by which the requests get unblocked
2143 * other than the low-level driver calling scsi_unblock_requests().
2144 */
2145void scsi_block_requests(struct Scsi_Host *shost)
2146{
2147 shost->host_self_blocked = 1;
2148}
2149EXPORT_SYMBOL(scsi_block_requests);
2150
2151/**
2152 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2153 * further commands to be queued to the device.
2154 * @shost: host in question
2155 *
2156 * There is no timer nor any other means by which the requests get unblocked
2157 * other than the low-level driver calling scsi_unblock_requests(). This is done
2158 * as an API function so that changes to the internals of the scsi mid-layer
2159 * won't require wholesale changes to drivers that use this feature.
2160 */
2161void scsi_unblock_requests(struct Scsi_Host *shost)
2162{
2163 shost->host_self_blocked = 0;
2164 scsi_run_host_queues(shost);
2165}
2166EXPORT_SYMBOL(scsi_unblock_requests);
2167
2168void scsi_exit_queue(void)
2169{
2170 kmem_cache_destroy(s: scsi_sense_cache);
2171}
2172
2173/**
2174 * scsi_mode_select - issue a mode select
2175 * @sdev: SCSI device to be queried
2176 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2177 * @sp: Save page bit (0 == don't save, 1 == save)
2178 * @buffer: request buffer (may not be smaller than eight bytes)
2179 * @len: length of request buffer.
2180 * @timeout: command timeout
2181 * @retries: number of retries before failing
2182 * @data: returns a structure abstracting the mode header data
2183 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2184 * must be SCSI_SENSE_BUFFERSIZE big.
2185 *
2186 * Returns zero if successful; negative error number or scsi
2187 * status on error
2188 *
2189 */
2190int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2191 unsigned char *buffer, int len, int timeout, int retries,
2192 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2193{
2194 unsigned char cmd[10];
2195 unsigned char *real_buffer;
2196 const struct scsi_exec_args exec_args = {
2197 .sshdr = sshdr,
2198 };
2199 int ret;
2200
2201 memset(s: cmd, c: 0, n: sizeof(cmd));
2202 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2203
2204 /*
2205 * Use MODE SELECT(10) if the device asked for it or if the mode page
2206 * and the mode select header cannot fit within the maximumm 255 bytes
2207 * of the MODE SELECT(6) command.
2208 */
2209 if (sdev->use_10_for_ms ||
2210 len + 4 > 255 ||
2211 data->block_descriptor_length > 255) {
2212 if (len > 65535 - 8)
2213 return -EINVAL;
2214 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2215 if (!real_buffer)
2216 return -ENOMEM;
2217 memcpy(to: real_buffer + 8, from: buffer, len);
2218 len += 8;
2219 real_buffer[0] = 0;
2220 real_buffer[1] = 0;
2221 real_buffer[2] = data->medium_type;
2222 real_buffer[3] = data->device_specific;
2223 real_buffer[4] = data->longlba ? 0x01 : 0;
2224 real_buffer[5] = 0;
2225 put_unaligned_be16(val: data->block_descriptor_length,
2226 p: &real_buffer[6]);
2227
2228 cmd[0] = MODE_SELECT_10;
2229 put_unaligned_be16(val: len, p: &cmd[7]);
2230 } else {
2231 if (data->longlba)
2232 return -EINVAL;
2233
2234 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2235 if (!real_buffer)
2236 return -ENOMEM;
2237 memcpy(to: real_buffer + 4, from: buffer, len);
2238 len += 4;
2239 real_buffer[0] = 0;
2240 real_buffer[1] = data->medium_type;
2241 real_buffer[2] = data->device_specific;
2242 real_buffer[3] = data->block_descriptor_length;
2243
2244 cmd[0] = MODE_SELECT;
2245 cmd[4] = len;
2246 }
2247
2248 ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2249 timeout, retries, &exec_args);
2250 kfree(objp: real_buffer);
2251 return ret;
2252}
2253EXPORT_SYMBOL_GPL(scsi_mode_select);
2254
2255/**
2256 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2257 * @sdev: SCSI device to be queried
2258 * @dbd: set to prevent mode sense from returning block descriptors
2259 * @modepage: mode page being requested
2260 * @subpage: sub-page of the mode page being requested
2261 * @buffer: request buffer (may not be smaller than eight bytes)
2262 * @len: length of request buffer.
2263 * @timeout: command timeout
2264 * @retries: number of retries before failing
2265 * @data: returns a structure abstracting the mode header data
2266 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2267 * must be SCSI_SENSE_BUFFERSIZE big.
2268 *
2269 * Returns zero if successful, or a negative error number on failure
2270 */
2271int
2272scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2273 unsigned char *buffer, int len, int timeout, int retries,
2274 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2275{
2276 unsigned char cmd[12];
2277 int use_10_for_ms;
2278 int header_length;
2279 int result;
2280 struct scsi_sense_hdr my_sshdr;
2281 struct scsi_failure failure_defs[] = {
2282 {
2283 .sense = UNIT_ATTENTION,
2284 .asc = SCMD_FAILURE_ASC_ANY,
2285 .ascq = SCMD_FAILURE_ASCQ_ANY,
2286 .allowed = retries,
2287 .result = SAM_STAT_CHECK_CONDITION,
2288 },
2289 {}
2290 };
2291 struct scsi_failures failures = {
2292 .failure_definitions = failure_defs,
2293 };
2294 const struct scsi_exec_args exec_args = {
2295 /* caller might not be interested in sense, but we need it */
2296 .sshdr = sshdr ? : &my_sshdr,
2297 .failures = &failures,
2298 };
2299
2300 memset(s: data, c: 0, n: sizeof(*data));
2301 memset(s: &cmd[0], c: 0, n: 12);
2302
2303 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2304 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2305 cmd[2] = modepage;
2306 cmd[3] = subpage;
2307
2308 sshdr = exec_args.sshdr;
2309
2310 retry:
2311 use_10_for_ms = sdev->use_10_for_ms || len > 255;
2312
2313 if (use_10_for_ms) {
2314 if (len < 8 || len > 65535)
2315 return -EINVAL;
2316
2317 cmd[0] = MODE_SENSE_10;
2318 put_unaligned_be16(val: len, p: &cmd[7]);
2319 header_length = 8;
2320 } else {
2321 if (len < 4)
2322 return -EINVAL;
2323
2324 cmd[0] = MODE_SENSE;
2325 cmd[4] = len;
2326 header_length = 4;
2327 }
2328
2329 memset(s: buffer, c: 0, n: len);
2330
2331 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2332 timeout, retries, &exec_args);
2333 if (result < 0)
2334 return result;
2335
2336 /* This code looks awful: what it's doing is making sure an
2337 * ILLEGAL REQUEST sense return identifies the actual command
2338 * byte as the problem. MODE_SENSE commands can return
2339 * ILLEGAL REQUEST if the code page isn't supported */
2340
2341 if (!scsi_status_is_good(status: result)) {
2342 if (scsi_sense_valid(sshdr)) {
2343 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2344 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2345 /*
2346 * Invalid command operation code: retry using
2347 * MODE SENSE(6) if this was a MODE SENSE(10)
2348 * request, except if the request mode page is
2349 * too large for MODE SENSE single byte
2350 * allocation length field.
2351 */
2352 if (use_10_for_ms) {
2353 if (len > 255)
2354 return -EIO;
2355 sdev->use_10_for_ms = 0;
2356 goto retry;
2357 }
2358 }
2359 }
2360 return -EIO;
2361 }
2362 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2363 (modepage == 6 || modepage == 8))) {
2364 /* Initio breakage? */
2365 header_length = 0;
2366 data->length = 13;
2367 data->medium_type = 0;
2368 data->device_specific = 0;
2369 data->longlba = 0;
2370 data->block_descriptor_length = 0;
2371 } else if (use_10_for_ms) {
2372 data->length = get_unaligned_be16(p: &buffer[0]) + 2;
2373 data->medium_type = buffer[2];
2374 data->device_specific = buffer[3];
2375 data->longlba = buffer[4] & 0x01;
2376 data->block_descriptor_length = get_unaligned_be16(p: &buffer[6]);
2377 } else {
2378 data->length = buffer[0] + 1;
2379 data->medium_type = buffer[1];
2380 data->device_specific = buffer[2];
2381 data->block_descriptor_length = buffer[3];
2382 }
2383 data->header_length = header_length;
2384
2385 return 0;
2386}
2387EXPORT_SYMBOL(scsi_mode_sense);
2388
2389/**
2390 * scsi_test_unit_ready - test if unit is ready
2391 * @sdev: scsi device to change the state of.
2392 * @timeout: command timeout
2393 * @retries: number of retries before failing
2394 * @sshdr: outpout pointer for decoded sense information.
2395 *
2396 * Returns zero if unsuccessful or an error if TUR failed. For
2397 * removable media, UNIT_ATTENTION sets ->changed flag.
2398 **/
2399int
2400scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2401 struct scsi_sense_hdr *sshdr)
2402{
2403 char cmd[] = {
2404 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2405 };
2406 const struct scsi_exec_args exec_args = {
2407 .sshdr = sshdr,
2408 };
2409 int result;
2410
2411 /* try to eat the UNIT_ATTENTION if there are enough retries */
2412 do {
2413 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2414 timeout, 1, &exec_args);
2415 if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2416 sshdr->sense_key == UNIT_ATTENTION)
2417 sdev->changed = 1;
2418 } while (result > 0 && scsi_sense_valid(sshdr) &&
2419 sshdr->sense_key == UNIT_ATTENTION && --retries);
2420
2421 return result;
2422}
2423EXPORT_SYMBOL(scsi_test_unit_ready);
2424
2425/**
2426 * scsi_device_set_state - Take the given device through the device state model.
2427 * @sdev: scsi device to change the state of.
2428 * @state: state to change to.
2429 *
2430 * Returns zero if successful or an error if the requested
2431 * transition is illegal.
2432 */
2433int
2434scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2435{
2436 enum scsi_device_state oldstate = sdev->sdev_state;
2437
2438 if (state == oldstate)
2439 return 0;
2440
2441 switch (state) {
2442 case SDEV_CREATED:
2443 switch (oldstate) {
2444 case SDEV_CREATED_BLOCK:
2445 break;
2446 default:
2447 goto illegal;
2448 }
2449 break;
2450
2451 case SDEV_RUNNING:
2452 switch (oldstate) {
2453 case SDEV_CREATED:
2454 case SDEV_OFFLINE:
2455 case SDEV_TRANSPORT_OFFLINE:
2456 case SDEV_QUIESCE:
2457 case SDEV_BLOCK:
2458 break;
2459 default:
2460 goto illegal;
2461 }
2462 break;
2463
2464 case SDEV_QUIESCE:
2465 switch (oldstate) {
2466 case SDEV_RUNNING:
2467 case SDEV_OFFLINE:
2468 case SDEV_TRANSPORT_OFFLINE:
2469 break;
2470 default:
2471 goto illegal;
2472 }
2473 break;
2474
2475 case SDEV_OFFLINE:
2476 case SDEV_TRANSPORT_OFFLINE:
2477 switch (oldstate) {
2478 case SDEV_CREATED:
2479 case SDEV_RUNNING:
2480 case SDEV_QUIESCE:
2481 case SDEV_BLOCK:
2482 break;
2483 default:
2484 goto illegal;
2485 }
2486 break;
2487
2488 case SDEV_BLOCK:
2489 switch (oldstate) {
2490 case SDEV_RUNNING:
2491 case SDEV_CREATED_BLOCK:
2492 case SDEV_QUIESCE:
2493 case SDEV_OFFLINE:
2494 break;
2495 default:
2496 goto illegal;
2497 }
2498 break;
2499
2500 case SDEV_CREATED_BLOCK:
2501 switch (oldstate) {
2502 case SDEV_CREATED:
2503 break;
2504 default:
2505 goto illegal;
2506 }
2507 break;
2508
2509 case SDEV_CANCEL:
2510 switch (oldstate) {
2511 case SDEV_CREATED:
2512 case SDEV_RUNNING:
2513 case SDEV_QUIESCE:
2514 case SDEV_OFFLINE:
2515 case SDEV_TRANSPORT_OFFLINE:
2516 break;
2517 default:
2518 goto illegal;
2519 }
2520 break;
2521
2522 case SDEV_DEL:
2523 switch (oldstate) {
2524 case SDEV_CREATED:
2525 case SDEV_RUNNING:
2526 case SDEV_OFFLINE:
2527 case SDEV_TRANSPORT_OFFLINE:
2528 case SDEV_CANCEL:
2529 case SDEV_BLOCK:
2530 case SDEV_CREATED_BLOCK:
2531 break;
2532 default:
2533 goto illegal;
2534 }
2535 break;
2536
2537 }
2538 sdev->offline_already = false;
2539 sdev->sdev_state = state;
2540 return 0;
2541
2542 illegal:
2543 SCSI_LOG_ERROR_RECOVERY(1,
2544 sdev_printk(KERN_ERR, sdev,
2545 "Illegal state transition %s->%s",
2546 scsi_device_state_name(oldstate),
2547 scsi_device_state_name(state))
2548 );
2549 return -EINVAL;
2550}
2551EXPORT_SYMBOL(scsi_device_set_state);
2552
2553/**
2554 * scsi_evt_emit - emit a single SCSI device uevent
2555 * @sdev: associated SCSI device
2556 * @evt: event to emit
2557 *
2558 * Send a single uevent (scsi_event) to the associated scsi_device.
2559 */
2560static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2561{
2562 int idx = 0;
2563 char *envp[3];
2564
2565 switch (evt->evt_type) {
2566 case SDEV_EVT_MEDIA_CHANGE:
2567 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2568 break;
2569 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2570 scsi_rescan_device(sdev);
2571 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2572 break;
2573 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2574 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2575 break;
2576 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2577 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2578 break;
2579 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2580 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2581 break;
2582 case SDEV_EVT_LUN_CHANGE_REPORTED:
2583 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2584 break;
2585 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2586 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2587 break;
2588 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2589 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2590 break;
2591 default:
2592 /* do nothing */
2593 break;
2594 }
2595
2596 envp[idx++] = NULL;
2597
2598 kobject_uevent_env(kobj: &sdev->sdev_gendev.kobj, action: KOBJ_CHANGE, envp);
2599}
2600
2601/**
2602 * scsi_evt_thread - send a uevent for each scsi event
2603 * @work: work struct for scsi_device
2604 *
2605 * Dispatch queued events to their associated scsi_device kobjects
2606 * as uevents.
2607 */
2608void scsi_evt_thread(struct work_struct *work)
2609{
2610 struct scsi_device *sdev;
2611 enum scsi_device_event evt_type;
2612 LIST_HEAD(event_list);
2613
2614 sdev = container_of(work, struct scsi_device, event_work);
2615
2616 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2617 if (test_and_clear_bit(nr: evt_type, addr: sdev->pending_events))
2618 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2619
2620 while (1) {
2621 struct scsi_event *evt;
2622 struct list_head *this, *tmp;
2623 unsigned long flags;
2624
2625 spin_lock_irqsave(&sdev->list_lock, flags);
2626 list_splice_init(list: &sdev->event_list, head: &event_list);
2627 spin_unlock_irqrestore(lock: &sdev->list_lock, flags);
2628
2629 if (list_empty(head: &event_list))
2630 break;
2631
2632 list_for_each_safe(this, tmp, &event_list) {
2633 evt = list_entry(this, struct scsi_event, node);
2634 list_del(entry: &evt->node);
2635 scsi_evt_emit(sdev, evt);
2636 kfree(objp: evt);
2637 }
2638 }
2639}
2640
2641/**
2642 * sdev_evt_send - send asserted event to uevent thread
2643 * @sdev: scsi_device event occurred on
2644 * @evt: event to send
2645 *
2646 * Assert scsi device event asynchronously.
2647 */
2648void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2649{
2650 unsigned long flags;
2651
2652#if 0
2653 /* FIXME: currently this check eliminates all media change events
2654 * for polled devices. Need to update to discriminate between AN
2655 * and polled events */
2656 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2657 kfree(evt);
2658 return;
2659 }
2660#endif
2661
2662 spin_lock_irqsave(&sdev->list_lock, flags);
2663 list_add_tail(new: &evt->node, head: &sdev->event_list);
2664 schedule_work(work: &sdev->event_work);
2665 spin_unlock_irqrestore(lock: &sdev->list_lock, flags);
2666}
2667EXPORT_SYMBOL_GPL(sdev_evt_send);
2668
2669/**
2670 * sdev_evt_alloc - allocate a new scsi event
2671 * @evt_type: type of event to allocate
2672 * @gfpflags: GFP flags for allocation
2673 *
2674 * Allocates and returns a new scsi_event.
2675 */
2676struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2677 gfp_t gfpflags)
2678{
2679 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2680 if (!evt)
2681 return NULL;
2682
2683 evt->evt_type = evt_type;
2684 INIT_LIST_HEAD(list: &evt->node);
2685
2686 /* evt_type-specific initialization, if any */
2687 switch (evt_type) {
2688 case SDEV_EVT_MEDIA_CHANGE:
2689 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2690 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2691 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2692 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2693 case SDEV_EVT_LUN_CHANGE_REPORTED:
2694 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2695 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2696 default:
2697 /* do nothing */
2698 break;
2699 }
2700
2701 return evt;
2702}
2703EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2704
2705/**
2706 * sdev_evt_send_simple - send asserted event to uevent thread
2707 * @sdev: scsi_device event occurred on
2708 * @evt_type: type of event to send
2709 * @gfpflags: GFP flags for allocation
2710 *
2711 * Assert scsi device event asynchronously, given an event type.
2712 */
2713void sdev_evt_send_simple(struct scsi_device *sdev,
2714 enum scsi_device_event evt_type, gfp_t gfpflags)
2715{
2716 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2717 if (!evt) {
2718 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2719 evt_type);
2720 return;
2721 }
2722
2723 sdev_evt_send(sdev, evt);
2724}
2725EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2726
2727/**
2728 * scsi_device_quiesce - Block all commands except power management.
2729 * @sdev: scsi device to quiesce.
2730 *
2731 * This works by trying to transition to the SDEV_QUIESCE state
2732 * (which must be a legal transition). When the device is in this
2733 * state, only power management requests will be accepted, all others will
2734 * be deferred.
2735 *
2736 * Must be called with user context, may sleep.
2737 *
2738 * Returns zero if unsuccessful or an error if not.
2739 */
2740int
2741scsi_device_quiesce(struct scsi_device *sdev)
2742{
2743 struct request_queue *q = sdev->request_queue;
2744 unsigned int memflags;
2745 int err;
2746
2747 /*
2748 * It is allowed to call scsi_device_quiesce() multiple times from
2749 * the same context but concurrent scsi_device_quiesce() calls are
2750 * not allowed.
2751 */
2752 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2753
2754 if (sdev->quiesced_by == current)
2755 return 0;
2756
2757 blk_set_pm_only(q);
2758
2759 memflags = blk_mq_freeze_queue(q);
2760 /*
2761 * Ensure that the effect of blk_set_pm_only() will be visible
2762 * for percpu_ref_tryget() callers that occur after the queue
2763 * unfreeze even if the queue was already frozen before this function
2764 * was called. See also https://lwn.net/Articles/573497/.
2765 */
2766 synchronize_rcu();
2767 blk_mq_unfreeze_queue(q, memflags);
2768
2769 mutex_lock(lock: &sdev->state_mutex);
2770 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2771 if (err == 0)
2772 sdev->quiesced_by = current;
2773 else
2774 blk_clear_pm_only(q);
2775 mutex_unlock(lock: &sdev->state_mutex);
2776
2777 return err;
2778}
2779EXPORT_SYMBOL(scsi_device_quiesce);
2780
2781/**
2782 * scsi_device_resume - Restart user issued commands to a quiesced device.
2783 * @sdev: scsi device to resume.
2784 *
2785 * Moves the device from quiesced back to running and restarts the
2786 * queues.
2787 *
2788 * Must be called with user context, may sleep.
2789 */
2790void scsi_device_resume(struct scsi_device *sdev)
2791{
2792 /* check if the device state was mutated prior to resume, and if
2793 * so assume the state is being managed elsewhere (for example
2794 * device deleted during suspend)
2795 */
2796 mutex_lock(lock: &sdev->state_mutex);
2797 if (sdev->sdev_state == SDEV_QUIESCE)
2798 scsi_device_set_state(sdev, SDEV_RUNNING);
2799 if (sdev->quiesced_by) {
2800 sdev->quiesced_by = NULL;
2801 blk_clear_pm_only(q: sdev->request_queue);
2802 }
2803 mutex_unlock(lock: &sdev->state_mutex);
2804}
2805EXPORT_SYMBOL(scsi_device_resume);
2806
2807static void
2808device_quiesce_fn(struct scsi_device *sdev, void *data)
2809{
2810 scsi_device_quiesce(sdev);
2811}
2812
2813void
2814scsi_target_quiesce(struct scsi_target *starget)
2815{
2816 starget_for_each_device(starget, NULL, fn: device_quiesce_fn);
2817}
2818EXPORT_SYMBOL(scsi_target_quiesce);
2819
2820static void
2821device_resume_fn(struct scsi_device *sdev, void *data)
2822{
2823 scsi_device_resume(sdev);
2824}
2825
2826void
2827scsi_target_resume(struct scsi_target *starget)
2828{
2829 starget_for_each_device(starget, NULL, fn: device_resume_fn);
2830}
2831EXPORT_SYMBOL(scsi_target_resume);
2832
2833static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2834{
2835 if (scsi_device_set_state(sdev, SDEV_BLOCK))
2836 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2837
2838 return 0;
2839}
2840
2841void scsi_start_queue(struct scsi_device *sdev)
2842{
2843 if (cmpxchg(&sdev->queue_stopped, 1, 0))
2844 blk_mq_unquiesce_queue(q: sdev->request_queue);
2845}
2846
2847static void scsi_stop_queue(struct scsi_device *sdev)
2848{
2849 /*
2850 * The atomic variable of ->queue_stopped covers that
2851 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2852 *
2853 * The caller needs to wait until quiesce is done.
2854 */
2855 if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2856 blk_mq_quiesce_queue_nowait(q: sdev->request_queue);
2857}
2858
2859/**
2860 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2861 * @sdev: device to block
2862 *
2863 * Pause SCSI command processing on the specified device. Does not sleep.
2864 *
2865 * Returns zero if successful or a negative error code upon failure.
2866 *
2867 * Notes:
2868 * This routine transitions the device to the SDEV_BLOCK state (which must be
2869 * a legal transition). When the device is in this state, command processing
2870 * is paused until the device leaves the SDEV_BLOCK state. See also
2871 * scsi_internal_device_unblock_nowait().
2872 */
2873int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2874{
2875 int ret = __scsi_internal_device_block_nowait(sdev);
2876
2877 /*
2878 * The device has transitioned to SDEV_BLOCK. Stop the
2879 * block layer from calling the midlayer with this device's
2880 * request queue.
2881 */
2882 if (!ret)
2883 scsi_stop_queue(sdev);
2884 return ret;
2885}
2886EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2887
2888/**
2889 * scsi_device_block - try to transition to the SDEV_BLOCK state
2890 * @sdev: device to block
2891 * @data: dummy argument, ignored
2892 *
2893 * Pause SCSI command processing on the specified device. Callers must wait
2894 * until all ongoing scsi_queue_rq() calls have finished after this function
2895 * returns.
2896 *
2897 * Note:
2898 * This routine transitions the device to the SDEV_BLOCK state (which must be
2899 * a legal transition). When the device is in this state, command processing
2900 * is paused until the device leaves the SDEV_BLOCK state. See also
2901 * scsi_internal_device_unblock().
2902 */
2903static void scsi_device_block(struct scsi_device *sdev, void *data)
2904{
2905 int err;
2906 enum scsi_device_state state;
2907
2908 mutex_lock(lock: &sdev->state_mutex);
2909 err = __scsi_internal_device_block_nowait(sdev);
2910 state = sdev->sdev_state;
2911 if (err == 0)
2912 /*
2913 * scsi_stop_queue() must be called with the state_mutex
2914 * held. Otherwise a simultaneous scsi_start_queue() call
2915 * might unquiesce the queue before we quiesce it.
2916 */
2917 scsi_stop_queue(sdev);
2918
2919 mutex_unlock(lock: &sdev->state_mutex);
2920
2921 WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2922 __func__, dev_name(&sdev->sdev_gendev), state);
2923}
2924
2925/**
2926 * scsi_internal_device_unblock_nowait - resume a device after a block request
2927 * @sdev: device to resume
2928 * @new_state: state to set the device to after unblocking
2929 *
2930 * Restart the device queue for a previously suspended SCSI device. Does not
2931 * sleep.
2932 *
2933 * Returns zero if successful or a negative error code upon failure.
2934 *
2935 * Notes:
2936 * This routine transitions the device to the SDEV_RUNNING state or to one of
2937 * the offline states (which must be a legal transition) allowing the midlayer
2938 * to goose the queue for this device.
2939 */
2940int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2941 enum scsi_device_state new_state)
2942{
2943 switch (new_state) {
2944 case SDEV_RUNNING:
2945 case SDEV_TRANSPORT_OFFLINE:
2946 break;
2947 default:
2948 return -EINVAL;
2949 }
2950
2951 /*
2952 * Try to transition the scsi device to SDEV_RUNNING or one of the
2953 * offlined states and goose the device queue if successful.
2954 */
2955 switch (sdev->sdev_state) {
2956 case SDEV_BLOCK:
2957 case SDEV_TRANSPORT_OFFLINE:
2958 sdev->sdev_state = new_state;
2959 break;
2960 case SDEV_CREATED_BLOCK:
2961 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2962 new_state == SDEV_OFFLINE)
2963 sdev->sdev_state = new_state;
2964 else
2965 sdev->sdev_state = SDEV_CREATED;
2966 break;
2967 case SDEV_CANCEL:
2968 case SDEV_OFFLINE:
2969 break;
2970 default:
2971 return -EINVAL;
2972 }
2973 scsi_start_queue(sdev);
2974
2975 return 0;
2976}
2977EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2978
2979/**
2980 * scsi_internal_device_unblock - resume a device after a block request
2981 * @sdev: device to resume
2982 * @new_state: state to set the device to after unblocking
2983 *
2984 * Restart the device queue for a previously suspended SCSI device. May sleep.
2985 *
2986 * Returns zero if successful or a negative error code upon failure.
2987 *
2988 * Notes:
2989 * This routine transitions the device to the SDEV_RUNNING state or to one of
2990 * the offline states (which must be a legal transition) allowing the midlayer
2991 * to goose the queue for this device.
2992 */
2993static int scsi_internal_device_unblock(struct scsi_device *sdev,
2994 enum scsi_device_state new_state)
2995{
2996 int ret;
2997
2998 mutex_lock(lock: &sdev->state_mutex);
2999 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
3000 mutex_unlock(lock: &sdev->state_mutex);
3001
3002 return ret;
3003}
3004
3005static int
3006target_block(struct device *dev, void *data)
3007{
3008 if (scsi_is_target_device(dev))
3009 starget_for_each_device(to_scsi_target(dev), NULL,
3010 fn: scsi_device_block);
3011 return 0;
3012}
3013
3014/**
3015 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3016 * @dev: a parent device of one or more scsi_target devices
3017 * @shost: the Scsi_Host to which this device belongs
3018 *
3019 * Iterate over all children of @dev, which should be scsi_target devices,
3020 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3021 * ongoing scsi_queue_rq() calls to finish. May sleep.
3022 *
3023 * Note:
3024 * @dev must not itself be a scsi_target device.
3025 */
3026void
3027scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3028{
3029 WARN_ON_ONCE(scsi_is_target_device(dev));
3030 device_for_each_child(parent: dev, NULL, fn: target_block);
3031 blk_mq_wait_quiesce_done(set: &shost->tag_set);
3032}
3033EXPORT_SYMBOL_GPL(scsi_block_targets);
3034
3035static void
3036device_unblock(struct scsi_device *sdev, void *data)
3037{
3038 scsi_internal_device_unblock(sdev, new_state: *(enum scsi_device_state *)data);
3039}
3040
3041static int
3042target_unblock(struct device *dev, void *data)
3043{
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), data,
3046 fn: device_unblock);
3047 return 0;
3048}
3049
3050void
3051scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3052{
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), &new_state,
3055 fn: device_unblock);
3056 else
3057 device_for_each_child(parent: dev, data: &new_state, fn: target_unblock);
3058}
3059EXPORT_SYMBOL_GPL(scsi_target_unblock);
3060
3061/**
3062 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3063 * @shost: device to block
3064 *
3065 * Pause SCSI command processing for all logical units associated with the SCSI
3066 * host and wait until pending scsi_queue_rq() calls have finished.
3067 *
3068 * Returns zero if successful or a negative error code upon failure.
3069 */
3070int
3071scsi_host_block(struct Scsi_Host *shost)
3072{
3073 struct scsi_device *sdev;
3074 int ret;
3075
3076 /*
3077 * Call scsi_internal_device_block_nowait so we can avoid
3078 * calling synchronize_rcu() for each LUN.
3079 */
3080 shost_for_each_device(sdev, shost) {
3081 mutex_lock(lock: &sdev->state_mutex);
3082 ret = scsi_internal_device_block_nowait(sdev);
3083 mutex_unlock(lock: &sdev->state_mutex);
3084 if (ret) {
3085 scsi_device_put(sdev);
3086 return ret;
3087 }
3088 }
3089
3090 /* Wait for ongoing scsi_queue_rq() calls to finish. */
3091 blk_mq_wait_quiesce_done(set: &shost->tag_set);
3092
3093 return 0;
3094}
3095EXPORT_SYMBOL_GPL(scsi_host_block);
3096
3097int
3098scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3099{
3100 struct scsi_device *sdev;
3101 int ret = 0;
3102
3103 shost_for_each_device(sdev, shost) {
3104 ret = scsi_internal_device_unblock(sdev, new_state);
3105 if (ret) {
3106 scsi_device_put(sdev);
3107 break;
3108 }
3109 }
3110 return ret;
3111}
3112EXPORT_SYMBOL_GPL(scsi_host_unblock);
3113
3114/**
3115 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3116 * @sgl: scatter-gather list
3117 * @sg_count: number of segments in sg
3118 * @offset: offset in bytes into sg, on return offset into the mapped area
3119 * @len: bytes to map, on return number of bytes mapped
3120 *
3121 * Returns virtual address of the start of the mapped page
3122 */
3123void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3124 size_t *offset, size_t *len)
3125{
3126 int i;
3127 size_t sg_len = 0, len_complete = 0;
3128 struct scatterlist *sg;
3129 struct page *page;
3130
3131 WARN_ON(!irqs_disabled());
3132
3133 for_each_sg(sgl, sg, sg_count, i) {
3134 len_complete = sg_len; /* Complete sg-entries */
3135 sg_len += sg->length;
3136 if (sg_len > *offset)
3137 break;
3138 }
3139
3140 if (unlikely(i == sg_count)) {
3141 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3142 "elements %d\n",
3143 __func__, sg_len, *offset, sg_count);
3144 WARN_ON(1);
3145 return NULL;
3146 }
3147
3148 /* Offset starting from the beginning of first page in this sg-entry */
3149 *offset = *offset - len_complete + sg->offset;
3150
3151 page = sg_page(sg) + (*offset >> PAGE_SHIFT);
3152 *offset &= ~PAGE_MASK;
3153
3154 /* Bytes in this sg-entry from *offset to the end of the page */
3155 sg_len = PAGE_SIZE - *offset;
3156 if (*len > sg_len)
3157 *len = sg_len;
3158
3159 return kmap_atomic(page);
3160}
3161EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3162
3163/**
3164 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3165 * @virt: virtual address to be unmapped
3166 */
3167void scsi_kunmap_atomic_sg(void *virt)
3168{
3169 kunmap_atomic(virt);
3170}
3171EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3172
3173void sdev_disable_disk_events(struct scsi_device *sdev)
3174{
3175 atomic_inc(v: &sdev->disk_events_disable_depth);
3176}
3177EXPORT_SYMBOL(sdev_disable_disk_events);
3178
3179void sdev_enable_disk_events(struct scsi_device *sdev)
3180{
3181 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3182 return;
3183 atomic_dec(v: &sdev->disk_events_disable_depth);
3184}
3185EXPORT_SYMBOL(sdev_enable_disk_events);
3186
3187static unsigned char designator_prio(const unsigned char *d)
3188{
3189 if (d[1] & 0x30)
3190 /* not associated with LUN */
3191 return 0;
3192
3193 if (d[3] == 0)
3194 /* invalid length */
3195 return 0;
3196
3197 /*
3198 * Order of preference for lun descriptor:
3199 * - SCSI name string
3200 * - NAA IEEE Registered Extended
3201 * - EUI-64 based 16-byte
3202 * - EUI-64 based 12-byte
3203 * - NAA IEEE Registered
3204 * - NAA IEEE Extended
3205 * - EUI-64 based 8-byte
3206 * - SCSI name string (truncated)
3207 * - T10 Vendor ID
3208 * as longer descriptors reduce the likelyhood
3209 * of identification clashes.
3210 */
3211
3212 switch (d[1] & 0xf) {
3213 case 8:
3214 /* SCSI name string, variable-length UTF-8 */
3215 return 9;
3216 case 3:
3217 switch (d[4] >> 4) {
3218 case 6:
3219 /* NAA registered extended */
3220 return 8;
3221 case 5:
3222 /* NAA registered */
3223 return 5;
3224 case 4:
3225 /* NAA extended */
3226 return 4;
3227 case 3:
3228 /* NAA locally assigned */
3229 return 1;
3230 default:
3231 break;
3232 }
3233 break;
3234 case 2:
3235 switch (d[3]) {
3236 case 16:
3237 /* EUI64-based, 16 byte */
3238 return 7;
3239 case 12:
3240 /* EUI64-based, 12 byte */
3241 return 6;
3242 case 8:
3243 /* EUI64-based, 8 byte */
3244 return 3;
3245 default:
3246 break;
3247 }
3248 break;
3249 case 1:
3250 /* T10 vendor ID */
3251 return 1;
3252 default:
3253 break;
3254 }
3255
3256 return 0;
3257}
3258
3259/**
3260 * scsi_vpd_lun_id - return a unique device identification
3261 * @sdev: SCSI device
3262 * @id: buffer for the identification
3263 * @id_len: length of the buffer
3264 *
3265 * Copies a unique device identification into @id based
3266 * on the information in the VPD page 0x83 of the device.
3267 * The string will be formatted as a SCSI name string.
3268 *
3269 * Returns the length of the identification or error on failure.
3270 * If the identifier is longer than the supplied buffer the actual
3271 * identifier length is returned and the buffer is not zero-padded.
3272 */
3273int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3274{
3275 u8 cur_id_prio = 0;
3276 u8 cur_id_size = 0;
3277 const unsigned char *d, *cur_id_str;
3278 const struct scsi_vpd *vpd_pg83;
3279 int id_size = -EINVAL;
3280
3281 rcu_read_lock();
3282 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3283 if (!vpd_pg83) {
3284 rcu_read_unlock();
3285 return -ENXIO;
3286 }
3287
3288 /* The id string must be at least 20 bytes + terminating NULL byte */
3289 if (id_len < 21) {
3290 rcu_read_unlock();
3291 return -EINVAL;
3292 }
3293
3294 memset(s: id, c: 0, n: id_len);
3295 for (d = vpd_pg83->data + 4;
3296 d < vpd_pg83->data + vpd_pg83->len;
3297 d += d[3] + 4) {
3298 u8 prio = designator_prio(d);
3299
3300 if (prio == 0 || cur_id_prio > prio)
3301 continue;
3302
3303 switch (d[1] & 0xf) {
3304 case 0x1:
3305 /* T10 Vendor ID */
3306 if (cur_id_size > d[3])
3307 break;
3308 cur_id_prio = prio;
3309 cur_id_size = d[3];
3310 if (cur_id_size + 4 > id_len)
3311 cur_id_size = id_len - 4;
3312 cur_id_str = d + 4;
3313 id_size = snprintf(buf: id, size: id_len, fmt: "t10.%*pE",
3314 cur_id_size, cur_id_str);
3315 break;
3316 case 0x2:
3317 /* EUI-64 */
3318 cur_id_prio = prio;
3319 cur_id_size = d[3];
3320 cur_id_str = d + 4;
3321 switch (cur_id_size) {
3322 case 8:
3323 id_size = snprintf(buf: id, size: id_len,
3324 fmt: "eui.%8phN",
3325 cur_id_str);
3326 break;
3327 case 12:
3328 id_size = snprintf(buf: id, size: id_len,
3329 fmt: "eui.%12phN",
3330 cur_id_str);
3331 break;
3332 case 16:
3333 id_size = snprintf(buf: id, size: id_len,
3334 fmt: "eui.%16phN",
3335 cur_id_str);
3336 break;
3337 default:
3338 break;
3339 }
3340 break;
3341 case 0x3:
3342 /* NAA */
3343 cur_id_prio = prio;
3344 cur_id_size = d[3];
3345 cur_id_str = d + 4;
3346 switch (cur_id_size) {
3347 case 8:
3348 id_size = snprintf(buf: id, size: id_len,
3349 fmt: "naa.%8phN",
3350 cur_id_str);
3351 break;
3352 case 16:
3353 id_size = snprintf(buf: id, size: id_len,
3354 fmt: "naa.%16phN",
3355 cur_id_str);
3356 break;
3357 default:
3358 break;
3359 }
3360 break;
3361 case 0x8:
3362 /* SCSI name string */
3363 if (cur_id_size > d[3])
3364 break;
3365 /* Prefer others for truncated descriptor */
3366 if (d[3] > id_len) {
3367 prio = 2;
3368 if (cur_id_prio > prio)
3369 break;
3370 }
3371 cur_id_prio = prio;
3372 cur_id_size = id_size = d[3];
3373 cur_id_str = d + 4;
3374 if (cur_id_size >= id_len)
3375 cur_id_size = id_len - 1;
3376 memcpy(to: id, from: cur_id_str, len: cur_id_size);
3377 break;
3378 default:
3379 break;
3380 }
3381 }
3382 rcu_read_unlock();
3383
3384 return id_size;
3385}
3386EXPORT_SYMBOL(scsi_vpd_lun_id);
3387
3388/**
3389 * scsi_vpd_tpg_id - return a target port group identifier
3390 * @sdev: SCSI device
3391 * @rel_id: pointer to return relative target port in if not %NULL
3392 *
3393 * Returns the Target Port Group identifier from the information
3394 * from VPD page 0x83 of the device.
3395 * Optionally sets @rel_id to the relative target port on success.
3396 *
3397 * Return: the identifier or error on failure.
3398 */
3399int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3400{
3401 const unsigned char *d;
3402 const struct scsi_vpd *vpd_pg83;
3403 int group_id = -EAGAIN, rel_port = -1;
3404
3405 rcu_read_lock();
3406 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3407 if (!vpd_pg83) {
3408 rcu_read_unlock();
3409 return -ENXIO;
3410 }
3411
3412 d = vpd_pg83->data + 4;
3413 while (d < vpd_pg83->data + vpd_pg83->len) {
3414 switch (d[1] & 0xf) {
3415 case 0x4:
3416 /* Relative target port */
3417 rel_port = get_unaligned_be16(p: &d[6]);
3418 break;
3419 case 0x5:
3420 /* Target port group */
3421 group_id = get_unaligned_be16(p: &d[6]);
3422 break;
3423 default:
3424 break;
3425 }
3426 d += d[3] + 4;
3427 }
3428 rcu_read_unlock();
3429
3430 if (group_id >= 0 && rel_id && rel_port != -1)
3431 *rel_id = rel_port;
3432
3433 return group_id;
3434}
3435EXPORT_SYMBOL(scsi_vpd_tpg_id);
3436
3437/**
3438 * scsi_build_sense - build sense data for a command
3439 * @scmd: scsi command for which the sense should be formatted
3440 * @desc: Sense format (non-zero == descriptor format,
3441 * 0 == fixed format)
3442 * @key: Sense key
3443 * @asc: Additional sense code
3444 * @ascq: Additional sense code qualifier
3445 *
3446 **/
3447void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3448{
3449 scsi_build_sense_buffer(desc, buf: scmd->sense_buffer, key, asc, ascq);
3450 scmd->result = SAM_STAT_CHECK_CONDITION;
3451}
3452EXPORT_SYMBOL_GPL(scsi_build_sense);
3453
3454#ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3455#include "scsi_lib_test.c"
3456#endif
3457